xref: /linux/arch/arm64/kvm/arm.c (revision 2d38f43930266b9a3835f8bceb5a32b7fba4a52a)
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 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
50 
51 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
52 
53 DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
54 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
55 
56 DECLARE_KVM_NVHE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
57 
58 static bool vgic_present, kvm_arm_initialised;
59 
60 static DEFINE_PER_CPU(unsigned char, kvm_hyp_initialized);
61 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
62 
63 bool is_kvm_arm_initialised(void)
64 {
65 	return kvm_arm_initialised;
66 }
67 
68 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
69 {
70 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
71 }
72 
73 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
74 			    struct kvm_enable_cap *cap)
75 {
76 	int r;
77 	u64 new_cap;
78 
79 	if (cap->flags)
80 		return -EINVAL;
81 
82 	switch (cap->cap) {
83 	case KVM_CAP_ARM_NISV_TO_USER:
84 		r = 0;
85 		set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
86 			&kvm->arch.flags);
87 		break;
88 	case KVM_CAP_ARM_MTE:
89 		mutex_lock(&kvm->lock);
90 		if (!system_supports_mte() || kvm->created_vcpus) {
91 			r = -EINVAL;
92 		} else {
93 			r = 0;
94 			set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
95 		}
96 		mutex_unlock(&kvm->lock);
97 		break;
98 	case KVM_CAP_ARM_SYSTEM_SUSPEND:
99 		r = 0;
100 		set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags);
101 		break;
102 	case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
103 		new_cap = cap->args[0];
104 
105 		mutex_lock(&kvm->slots_lock);
106 		/*
107 		 * To keep things simple, allow changing the chunk
108 		 * size only when no memory slots have been created.
109 		 */
110 		if (!kvm_are_all_memslots_empty(kvm)) {
111 			r = -EINVAL;
112 		} else if (new_cap && !kvm_is_block_size_supported(new_cap)) {
113 			r = -EINVAL;
114 		} else {
115 			r = 0;
116 			kvm->arch.mmu.split_page_chunk_size = new_cap;
117 		}
118 		mutex_unlock(&kvm->slots_lock);
119 		break;
120 	default:
121 		r = -EINVAL;
122 		break;
123 	}
124 
125 	return r;
126 }
127 
128 static int kvm_arm_default_max_vcpus(void)
129 {
130 	return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
131 }
132 
133 /**
134  * kvm_arch_init_vm - initializes a VM data structure
135  * @kvm:	pointer to the KVM struct
136  */
137 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
138 {
139 	int ret;
140 
141 	mutex_init(&kvm->arch.config_lock);
142 
143 #ifdef CONFIG_LOCKDEP
144 	/* Clue in lockdep that the config_lock must be taken inside kvm->lock */
145 	mutex_lock(&kvm->lock);
146 	mutex_lock(&kvm->arch.config_lock);
147 	mutex_unlock(&kvm->arch.config_lock);
148 	mutex_unlock(&kvm->lock);
149 #endif
150 
151 	ret = kvm_share_hyp(kvm, kvm + 1);
152 	if (ret)
153 		return ret;
154 
155 	ret = pkvm_init_host_vm(kvm);
156 	if (ret)
157 		goto err_unshare_kvm;
158 
159 	if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL_ACCOUNT)) {
160 		ret = -ENOMEM;
161 		goto err_unshare_kvm;
162 	}
163 	cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
164 
165 	ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type);
166 	if (ret)
167 		goto err_free_cpumask;
168 
169 	kvm_vgic_early_init(kvm);
170 
171 	kvm_timer_init_vm(kvm);
172 
173 	/* The maximum number of VCPUs is limited by the host's GIC model */
174 	kvm->max_vcpus = kvm_arm_default_max_vcpus();
175 
176 	kvm_arm_init_hypercalls(kvm);
177 
178 	bitmap_zero(kvm->arch.vcpu_features, KVM_VCPU_MAX_FEATURES);
179 
180 	return 0;
181 
182 err_free_cpumask:
183 	free_cpumask_var(kvm->arch.supported_cpus);
184 err_unshare_kvm:
185 	kvm_unshare_hyp(kvm, kvm + 1);
186 	return ret;
187 }
188 
189 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
190 {
191 	return VM_FAULT_SIGBUS;
192 }
193 
194 void kvm_arch_create_vm_debugfs(struct kvm *kvm)
195 {
196 	kvm_sys_regs_create_debugfs(kvm);
197 }
198 
199 /**
200  * kvm_arch_destroy_vm - destroy the VM data structure
201  * @kvm:	pointer to the KVM struct
202  */
203 void kvm_arch_destroy_vm(struct kvm *kvm)
204 {
205 	bitmap_free(kvm->arch.pmu_filter);
206 	free_cpumask_var(kvm->arch.supported_cpus);
207 
208 	kvm_vgic_destroy(kvm);
209 
210 	if (is_protected_kvm_enabled())
211 		pkvm_destroy_hyp_vm(kvm);
212 
213 	kfree(kvm->arch.mpidr_data);
214 	kfree(kvm->arch.sysreg_masks);
215 	kvm_destroy_vcpus(kvm);
216 
217 	kvm_unshare_hyp(kvm, kvm + 1);
218 
219 	kvm_arm_teardown_hypercalls(kvm);
220 }
221 
222 static bool kvm_has_full_ptr_auth(void)
223 {
224 	bool apa, gpa, api, gpi, apa3, gpa3;
225 	u64 isar1, isar2, val;
226 
227 	/*
228 	 * Check that:
229 	 *
230 	 * - both Address and Generic auth are implemented for a given
231          *   algorithm (Q5, IMPDEF or Q3)
232 	 * - only a single algorithm is implemented.
233 	 */
234 	if (!system_has_full_ptr_auth())
235 		return false;
236 
237 	isar1 = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
238 	isar2 = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
239 
240 	apa = !!FIELD_GET(ID_AA64ISAR1_EL1_APA_MASK, isar1);
241 	val = FIELD_GET(ID_AA64ISAR1_EL1_GPA_MASK, isar1);
242 	gpa = (val == ID_AA64ISAR1_EL1_GPA_IMP);
243 
244 	api = !!FIELD_GET(ID_AA64ISAR1_EL1_API_MASK, isar1);
245 	val = FIELD_GET(ID_AA64ISAR1_EL1_GPI_MASK, isar1);
246 	gpi = (val == ID_AA64ISAR1_EL1_GPI_IMP);
247 
248 	apa3 = !!FIELD_GET(ID_AA64ISAR2_EL1_APA3_MASK, isar2);
249 	val  = FIELD_GET(ID_AA64ISAR2_EL1_GPA3_MASK, isar2);
250 	gpa3 = (val == ID_AA64ISAR2_EL1_GPA3_IMP);
251 
252 	return (apa == gpa && api == gpi && apa3 == gpa3 &&
253 		(apa + api + apa3) == 1);
254 }
255 
256 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
257 {
258 	int r;
259 	switch (ext) {
260 	case KVM_CAP_IRQCHIP:
261 		r = vgic_present;
262 		break;
263 	case KVM_CAP_IOEVENTFD:
264 	case KVM_CAP_USER_MEMORY:
265 	case KVM_CAP_SYNC_MMU:
266 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
267 	case KVM_CAP_ONE_REG:
268 	case KVM_CAP_ARM_PSCI:
269 	case KVM_CAP_ARM_PSCI_0_2:
270 	case KVM_CAP_READONLY_MEM:
271 	case KVM_CAP_MP_STATE:
272 	case KVM_CAP_IMMEDIATE_EXIT:
273 	case KVM_CAP_VCPU_EVENTS:
274 	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
275 	case KVM_CAP_ARM_NISV_TO_USER:
276 	case KVM_CAP_ARM_INJECT_EXT_DABT:
277 	case KVM_CAP_SET_GUEST_DEBUG:
278 	case KVM_CAP_VCPU_ATTRIBUTES:
279 	case KVM_CAP_PTP_KVM:
280 	case KVM_CAP_ARM_SYSTEM_SUSPEND:
281 	case KVM_CAP_IRQFD_RESAMPLE:
282 	case KVM_CAP_COUNTER_OFFSET:
283 		r = 1;
284 		break;
285 	case KVM_CAP_SET_GUEST_DEBUG2:
286 		return KVM_GUESTDBG_VALID_MASK;
287 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
288 		r = 1;
289 		break;
290 	case KVM_CAP_NR_VCPUS:
291 		/*
292 		 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
293 		 * architectures, as it does not always bound it to
294 		 * KVM_CAP_MAX_VCPUS. It should not matter much because
295 		 * this is just an advisory value.
296 		 */
297 		r = min_t(unsigned int, num_online_cpus(),
298 			  kvm_arm_default_max_vcpus());
299 		break;
300 	case KVM_CAP_MAX_VCPUS:
301 	case KVM_CAP_MAX_VCPU_ID:
302 		if (kvm)
303 			r = kvm->max_vcpus;
304 		else
305 			r = kvm_arm_default_max_vcpus();
306 		break;
307 	case KVM_CAP_MSI_DEVID:
308 		if (!kvm)
309 			r = -EINVAL;
310 		else
311 			r = kvm->arch.vgic.msis_require_devid;
312 		break;
313 	case KVM_CAP_ARM_USER_IRQ:
314 		/*
315 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
316 		 * (bump this number if adding more devices)
317 		 */
318 		r = 1;
319 		break;
320 	case KVM_CAP_ARM_MTE:
321 		r = system_supports_mte();
322 		break;
323 	case KVM_CAP_STEAL_TIME:
324 		r = kvm_arm_pvtime_supported();
325 		break;
326 	case KVM_CAP_ARM_EL1_32BIT:
327 		r = cpus_have_final_cap(ARM64_HAS_32BIT_EL1);
328 		break;
329 	case KVM_CAP_GUEST_DEBUG_HW_BPS:
330 		r = get_num_brps();
331 		break;
332 	case KVM_CAP_GUEST_DEBUG_HW_WPS:
333 		r = get_num_wrps();
334 		break;
335 	case KVM_CAP_ARM_PMU_V3:
336 		r = kvm_arm_support_pmu_v3();
337 		break;
338 	case KVM_CAP_ARM_INJECT_SERROR_ESR:
339 		r = cpus_have_final_cap(ARM64_HAS_RAS_EXTN);
340 		break;
341 	case KVM_CAP_ARM_VM_IPA_SIZE:
342 		r = get_kvm_ipa_limit();
343 		break;
344 	case KVM_CAP_ARM_SVE:
345 		r = system_supports_sve();
346 		break;
347 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
348 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
349 		r = kvm_has_full_ptr_auth();
350 		break;
351 	case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
352 		if (kvm)
353 			r = kvm->arch.mmu.split_page_chunk_size;
354 		else
355 			r = KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT;
356 		break;
357 	case KVM_CAP_ARM_SUPPORTED_BLOCK_SIZES:
358 		r = kvm_supported_block_sizes();
359 		break;
360 	case KVM_CAP_ARM_SUPPORTED_REG_MASK_RANGES:
361 		r = BIT(0);
362 		break;
363 	default:
364 		r = 0;
365 	}
366 
367 	return r;
368 }
369 
370 long kvm_arch_dev_ioctl(struct file *filp,
371 			unsigned int ioctl, unsigned long arg)
372 {
373 	return -EINVAL;
374 }
375 
376 struct kvm *kvm_arch_alloc_vm(void)
377 {
378 	size_t sz = sizeof(struct kvm);
379 
380 	if (!has_vhe())
381 		return kzalloc(sz, GFP_KERNEL_ACCOUNT);
382 
383 	return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
384 }
385 
386 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
387 {
388 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
389 		return -EBUSY;
390 
391 	if (id >= kvm->max_vcpus)
392 		return -EINVAL;
393 
394 	return 0;
395 }
396 
397 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
398 {
399 	int err;
400 
401 	spin_lock_init(&vcpu->arch.mp_state_lock);
402 
403 #ifdef CONFIG_LOCKDEP
404 	/* Inform lockdep that the config_lock is acquired after vcpu->mutex */
405 	mutex_lock(&vcpu->mutex);
406 	mutex_lock(&vcpu->kvm->arch.config_lock);
407 	mutex_unlock(&vcpu->kvm->arch.config_lock);
408 	mutex_unlock(&vcpu->mutex);
409 #endif
410 
411 	/* Force users to call KVM_ARM_VCPU_INIT */
412 	vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
413 
414 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
415 
416 	/* Set up the timer */
417 	kvm_timer_vcpu_init(vcpu);
418 
419 	kvm_pmu_vcpu_init(vcpu);
420 
421 	kvm_arm_reset_debug_ptr(vcpu);
422 
423 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
424 
425 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
426 
427 	err = kvm_vgic_vcpu_init(vcpu);
428 	if (err)
429 		return err;
430 
431 	return kvm_share_hyp(vcpu, vcpu + 1);
432 }
433 
434 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
435 {
436 }
437 
438 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
439 {
440 	if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
441 		static_branch_dec(&userspace_irqchip_in_use);
442 
443 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
444 	kvm_timer_vcpu_terminate(vcpu);
445 	kvm_pmu_vcpu_destroy(vcpu);
446 	kvm_vgic_vcpu_destroy(vcpu);
447 	kvm_arm_vcpu_destroy(vcpu);
448 }
449 
450 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
451 {
452 
453 }
454 
455 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
456 {
457 
458 }
459 
460 static void vcpu_set_pauth_traps(struct kvm_vcpu *vcpu)
461 {
462 	if (vcpu_has_ptrauth(vcpu)) {
463 		/*
464 		 * Either we're running running an L2 guest, and the API/APK
465 		 * bits come from L1's HCR_EL2, or API/APK are both set.
466 		 */
467 		if (unlikely(vcpu_has_nv(vcpu) && !is_hyp_ctxt(vcpu))) {
468 			u64 val;
469 
470 			val = __vcpu_sys_reg(vcpu, HCR_EL2);
471 			val &= (HCR_API | HCR_APK);
472 			vcpu->arch.hcr_el2 &= ~(HCR_API | HCR_APK);
473 			vcpu->arch.hcr_el2 |= val;
474 		} else {
475 			vcpu->arch.hcr_el2 |= (HCR_API | HCR_APK);
476 		}
477 
478 		/*
479 		 * Save the host keys if there is any chance for the guest
480 		 * to use pauth, as the entry code will reload the guest
481 		 * keys in that case.
482 		 * Protected mode is the exception to that rule, as the
483 		 * entry into the EL2 code eagerly switch back and forth
484 		 * between host and hyp keys (and kvm_hyp_ctxt is out of
485 		 * reach anyway).
486 		 */
487 		if (is_protected_kvm_enabled())
488 			return;
489 
490 		if (vcpu->arch.hcr_el2 & (HCR_API | HCR_APK)) {
491 			struct kvm_cpu_context *ctxt;
492 			ctxt = this_cpu_ptr_hyp_sym(kvm_hyp_ctxt);
493 			ptrauth_save_keys(ctxt);
494 		}
495 	}
496 }
497 
498 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
499 {
500 	struct kvm_s2_mmu *mmu;
501 	int *last_ran;
502 
503 	mmu = vcpu->arch.hw_mmu;
504 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
505 
506 	/*
507 	 * We guarantee that both TLBs and I-cache are private to each
508 	 * vcpu. If detecting that a vcpu from the same VM has
509 	 * previously run on the same physical CPU, call into the
510 	 * hypervisor code to nuke the relevant contexts.
511 	 *
512 	 * We might get preempted before the vCPU actually runs, but
513 	 * over-invalidation doesn't affect correctness.
514 	 */
515 	if (*last_ran != vcpu->vcpu_idx) {
516 		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
517 		*last_ran = vcpu->vcpu_idx;
518 	}
519 
520 	vcpu->cpu = cpu;
521 
522 	kvm_vgic_load(vcpu);
523 	kvm_timer_vcpu_load(vcpu);
524 	if (has_vhe())
525 		kvm_vcpu_load_vhe(vcpu);
526 	kvm_arch_vcpu_load_fp(vcpu);
527 	kvm_vcpu_pmu_restore_guest(vcpu);
528 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
529 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
530 
531 	if (single_task_running())
532 		vcpu_clear_wfx_traps(vcpu);
533 	else
534 		vcpu_set_wfx_traps(vcpu);
535 
536 	vcpu_set_pauth_traps(vcpu);
537 
538 	kvm_arch_vcpu_load_debug_state_flags(vcpu);
539 
540 	if (!cpumask_test_cpu(cpu, vcpu->kvm->arch.supported_cpus))
541 		vcpu_set_on_unsupported_cpu(vcpu);
542 }
543 
544 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
545 {
546 	kvm_arch_vcpu_put_debug_state_flags(vcpu);
547 	kvm_arch_vcpu_put_fp(vcpu);
548 	if (has_vhe())
549 		kvm_vcpu_put_vhe(vcpu);
550 	kvm_timer_vcpu_put(vcpu);
551 	kvm_vgic_put(vcpu);
552 	kvm_vcpu_pmu_restore_host(vcpu);
553 	kvm_arm_vmid_clear_active();
554 
555 	vcpu_clear_on_unsupported_cpu(vcpu);
556 	vcpu->cpu = -1;
557 }
558 
559 static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
560 {
561 	WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
562 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
563 	kvm_vcpu_kick(vcpu);
564 }
565 
566 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
567 {
568 	spin_lock(&vcpu->arch.mp_state_lock);
569 	__kvm_arm_vcpu_power_off(vcpu);
570 	spin_unlock(&vcpu->arch.mp_state_lock);
571 }
572 
573 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
574 {
575 	return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED;
576 }
577 
578 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
579 {
580 	WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_SUSPENDED);
581 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
582 	kvm_vcpu_kick(vcpu);
583 }
584 
585 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
586 {
587 	return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED;
588 }
589 
590 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
591 				    struct kvm_mp_state *mp_state)
592 {
593 	*mp_state = READ_ONCE(vcpu->arch.mp_state);
594 
595 	return 0;
596 }
597 
598 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
599 				    struct kvm_mp_state *mp_state)
600 {
601 	int ret = 0;
602 
603 	spin_lock(&vcpu->arch.mp_state_lock);
604 
605 	switch (mp_state->mp_state) {
606 	case KVM_MP_STATE_RUNNABLE:
607 		WRITE_ONCE(vcpu->arch.mp_state, *mp_state);
608 		break;
609 	case KVM_MP_STATE_STOPPED:
610 		__kvm_arm_vcpu_power_off(vcpu);
611 		break;
612 	case KVM_MP_STATE_SUSPENDED:
613 		kvm_arm_vcpu_suspend(vcpu);
614 		break;
615 	default:
616 		ret = -EINVAL;
617 	}
618 
619 	spin_unlock(&vcpu->arch.mp_state_lock);
620 
621 	return ret;
622 }
623 
624 /**
625  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
626  * @v:		The VCPU pointer
627  *
628  * If the guest CPU is not waiting for interrupts or an interrupt line is
629  * asserted, the CPU is by definition runnable.
630  */
631 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
632 {
633 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
634 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
635 		&& !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
636 }
637 
638 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
639 {
640 	return vcpu_mode_priv(vcpu);
641 }
642 
643 #ifdef CONFIG_GUEST_PERF_EVENTS
644 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
645 {
646 	return *vcpu_pc(vcpu);
647 }
648 #endif
649 
650 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
651 {
652 	return vcpu_get_flag(vcpu, VCPU_INITIALIZED);
653 }
654 
655 static void kvm_init_mpidr_data(struct kvm *kvm)
656 {
657 	struct kvm_mpidr_data *data = NULL;
658 	unsigned long c, mask, nr_entries;
659 	u64 aff_set = 0, aff_clr = ~0UL;
660 	struct kvm_vcpu *vcpu;
661 
662 	mutex_lock(&kvm->arch.config_lock);
663 
664 	if (kvm->arch.mpidr_data || atomic_read(&kvm->online_vcpus) == 1)
665 		goto out;
666 
667 	kvm_for_each_vcpu(c, vcpu, kvm) {
668 		u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
669 		aff_set |= aff;
670 		aff_clr &= aff;
671 	}
672 
673 	/*
674 	 * A significant bit can be either 0 or 1, and will only appear in
675 	 * aff_set. Use aff_clr to weed out the useless stuff.
676 	 */
677 	mask = aff_set ^ aff_clr;
678 	nr_entries = BIT_ULL(hweight_long(mask));
679 
680 	/*
681 	 * Don't let userspace fool us. If we need more than a single page
682 	 * to describe the compressed MPIDR array, just fall back to the
683 	 * iterative method. Single vcpu VMs do not need this either.
684 	 */
685 	if (struct_size(data, cmpidr_to_idx, nr_entries) <= PAGE_SIZE)
686 		data = kzalloc(struct_size(data, cmpidr_to_idx, nr_entries),
687 			       GFP_KERNEL_ACCOUNT);
688 
689 	if (!data)
690 		goto out;
691 
692 	data->mpidr_mask = mask;
693 
694 	kvm_for_each_vcpu(c, vcpu, kvm) {
695 		u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
696 		u16 index = kvm_mpidr_index(data, aff);
697 
698 		data->cmpidr_to_idx[index] = c;
699 	}
700 
701 	kvm->arch.mpidr_data = data;
702 out:
703 	mutex_unlock(&kvm->arch.config_lock);
704 }
705 
706 /*
707  * Handle both the initialisation that is being done when the vcpu is
708  * run for the first time, as well as the updates that must be
709  * performed each time we get a new thread dealing with this vcpu.
710  */
711 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
712 {
713 	struct kvm *kvm = vcpu->kvm;
714 	int ret;
715 
716 	if (!kvm_vcpu_initialized(vcpu))
717 		return -ENOEXEC;
718 
719 	if (!kvm_arm_vcpu_is_finalized(vcpu))
720 		return -EPERM;
721 
722 	ret = kvm_arch_vcpu_run_map_fp(vcpu);
723 	if (ret)
724 		return ret;
725 
726 	if (likely(vcpu_has_run_once(vcpu)))
727 		return 0;
728 
729 	kvm_init_mpidr_data(kvm);
730 
731 	kvm_arm_vcpu_init_debug(vcpu);
732 
733 	if (likely(irqchip_in_kernel(kvm))) {
734 		/*
735 		 * Map the VGIC hardware resources before running a vcpu the
736 		 * first time on this VM.
737 		 */
738 		ret = kvm_vgic_map_resources(kvm);
739 		if (ret)
740 			return ret;
741 	}
742 
743 	if (vcpu_has_nv(vcpu)) {
744 		ret = kvm_init_nv_sysregs(vcpu->kvm);
745 		if (ret)
746 			return ret;
747 	}
748 
749 	/*
750 	 * This needs to happen after NV has imposed its own restrictions on
751 	 * the feature set
752 	 */
753 	kvm_init_sysreg(vcpu);
754 
755 	ret = kvm_timer_enable(vcpu);
756 	if (ret)
757 		return ret;
758 
759 	ret = kvm_arm_pmu_v3_enable(vcpu);
760 	if (ret)
761 		return ret;
762 
763 	if (is_protected_kvm_enabled()) {
764 		ret = pkvm_create_hyp_vm(kvm);
765 		if (ret)
766 			return ret;
767 	}
768 
769 	if (!irqchip_in_kernel(kvm)) {
770 		/*
771 		 * Tell the rest of the code that there are userspace irqchip
772 		 * VMs in the wild.
773 		 */
774 		static_branch_inc(&userspace_irqchip_in_use);
775 	}
776 
777 	/*
778 	 * Initialize traps for protected VMs.
779 	 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
780 	 * the code is in place for first run initialization at EL2.
781 	 */
782 	if (kvm_vm_is_protected(kvm))
783 		kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
784 
785 	mutex_lock(&kvm->arch.config_lock);
786 	set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
787 	mutex_unlock(&kvm->arch.config_lock);
788 
789 	return ret;
790 }
791 
792 bool kvm_arch_intc_initialized(struct kvm *kvm)
793 {
794 	return vgic_initialized(kvm);
795 }
796 
797 void kvm_arm_halt_guest(struct kvm *kvm)
798 {
799 	unsigned long i;
800 	struct kvm_vcpu *vcpu;
801 
802 	kvm_for_each_vcpu(i, vcpu, kvm)
803 		vcpu->arch.pause = true;
804 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
805 }
806 
807 void kvm_arm_resume_guest(struct kvm *kvm)
808 {
809 	unsigned long i;
810 	struct kvm_vcpu *vcpu;
811 
812 	kvm_for_each_vcpu(i, vcpu, kvm) {
813 		vcpu->arch.pause = false;
814 		__kvm_vcpu_wake_up(vcpu);
815 	}
816 }
817 
818 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
819 {
820 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
821 
822 	rcuwait_wait_event(wait,
823 			   (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
824 			   TASK_INTERRUPTIBLE);
825 
826 	if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
827 		/* Awaken to handle a signal, request we sleep again later. */
828 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
829 	}
830 
831 	/*
832 	 * Make sure we will observe a potential reset request if we've
833 	 * observed a change to the power state. Pairs with the smp_wmb() in
834 	 * kvm_psci_vcpu_on().
835 	 */
836 	smp_rmb();
837 }
838 
839 /**
840  * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
841  * @vcpu:	The VCPU pointer
842  *
843  * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
844  * the vCPU is runnable.  The vCPU may or may not be scheduled out, depending
845  * on when a wake event arrives, e.g. there may already be a pending wake event.
846  */
847 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
848 {
849 	/*
850 	 * Sync back the state of the GIC CPU interface so that we have
851 	 * the latest PMR and group enables. This ensures that
852 	 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
853 	 * we have pending interrupts, e.g. when determining if the
854 	 * vCPU should block.
855 	 *
856 	 * For the same reason, we want to tell GICv4 that we need
857 	 * doorbells to be signalled, should an interrupt become pending.
858 	 */
859 	preempt_disable();
860 	kvm_vgic_vmcr_sync(vcpu);
861 	vcpu_set_flag(vcpu, IN_WFI);
862 	vgic_v4_put(vcpu);
863 	preempt_enable();
864 
865 	kvm_vcpu_halt(vcpu);
866 	vcpu_clear_flag(vcpu, IN_WFIT);
867 
868 	preempt_disable();
869 	vcpu_clear_flag(vcpu, IN_WFI);
870 	vgic_v4_load(vcpu);
871 	preempt_enable();
872 }
873 
874 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
875 {
876 	if (!kvm_arm_vcpu_suspended(vcpu))
877 		return 1;
878 
879 	kvm_vcpu_wfi(vcpu);
880 
881 	/*
882 	 * The suspend state is sticky; we do not leave it until userspace
883 	 * explicitly marks the vCPU as runnable. Request that we suspend again
884 	 * later.
885 	 */
886 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
887 
888 	/*
889 	 * Check to make sure the vCPU is actually runnable. If so, exit to
890 	 * userspace informing it of the wakeup condition.
891 	 */
892 	if (kvm_arch_vcpu_runnable(vcpu)) {
893 		memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
894 		vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
895 		vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
896 		return 0;
897 	}
898 
899 	/*
900 	 * Otherwise, we were unblocked to process a different event, such as a
901 	 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
902 	 * process the event.
903 	 */
904 	return 1;
905 }
906 
907 /**
908  * check_vcpu_requests - check and handle pending vCPU requests
909  * @vcpu:	the VCPU pointer
910  *
911  * Return: 1 if we should enter the guest
912  *	   0 if we should exit to userspace
913  *	   < 0 if we should exit to userspace, where the return value indicates
914  *	   an error
915  */
916 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
917 {
918 	if (kvm_request_pending(vcpu)) {
919 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
920 			kvm_vcpu_sleep(vcpu);
921 
922 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
923 			kvm_reset_vcpu(vcpu);
924 
925 		/*
926 		 * Clear IRQ_PENDING requests that were made to guarantee
927 		 * that a VCPU sees new virtual interrupts.
928 		 */
929 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
930 
931 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
932 			kvm_update_stolen_time(vcpu);
933 
934 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
935 			/* The distributor enable bits were changed */
936 			preempt_disable();
937 			vgic_v4_put(vcpu);
938 			vgic_v4_load(vcpu);
939 			preempt_enable();
940 		}
941 
942 		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
943 			kvm_vcpu_reload_pmu(vcpu);
944 
945 		if (kvm_check_request(KVM_REQ_RESYNC_PMU_EL0, vcpu))
946 			kvm_vcpu_pmu_restore_guest(vcpu);
947 
948 		if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
949 			return kvm_vcpu_suspend(vcpu);
950 
951 		if (kvm_dirty_ring_check_request(vcpu))
952 			return 0;
953 	}
954 
955 	return 1;
956 }
957 
958 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
959 {
960 	if (likely(!vcpu_mode_is_32bit(vcpu)))
961 		return false;
962 
963 	if (vcpu_has_nv(vcpu))
964 		return true;
965 
966 	return !kvm_supports_32bit_el0();
967 }
968 
969 /**
970  * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
971  * @vcpu:	The VCPU pointer
972  * @ret:	Pointer to write optional return code
973  *
974  * Returns: true if the VCPU needs to return to a preemptible + interruptible
975  *	    and skip guest entry.
976  *
977  * This function disambiguates between two different types of exits: exits to a
978  * preemptible + interruptible kernel context and exits to userspace. For an
979  * exit to userspace, this function will write the return code to ret and return
980  * true. For an exit to preemptible + interruptible kernel context (i.e. check
981  * for pending work and re-enter), return true without writing to ret.
982  */
983 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
984 {
985 	struct kvm_run *run = vcpu->run;
986 
987 	/*
988 	 * If we're using a userspace irqchip, then check if we need
989 	 * to tell a userspace irqchip about timer or PMU level
990 	 * changes and if so, exit to userspace (the actual level
991 	 * state gets updated in kvm_timer_update_run and
992 	 * kvm_pmu_update_run below).
993 	 */
994 	if (static_branch_unlikely(&userspace_irqchip_in_use)) {
995 		if (kvm_timer_should_notify_user(vcpu) ||
996 		    kvm_pmu_should_notify_user(vcpu)) {
997 			*ret = -EINTR;
998 			run->exit_reason = KVM_EXIT_INTR;
999 			return true;
1000 		}
1001 	}
1002 
1003 	if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
1004 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1005 		run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
1006 		run->fail_entry.cpu = smp_processor_id();
1007 		*ret = 0;
1008 		return true;
1009 	}
1010 
1011 	return kvm_request_pending(vcpu) ||
1012 			xfer_to_guest_mode_work_pending();
1013 }
1014 
1015 /*
1016  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
1017  * the vCPU is running.
1018  *
1019  * This must be noinstr as instrumentation may make use of RCU, and this is not
1020  * safe during the EQS.
1021  */
1022 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
1023 {
1024 	int ret;
1025 
1026 	guest_state_enter_irqoff();
1027 	ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
1028 	guest_state_exit_irqoff();
1029 
1030 	return ret;
1031 }
1032 
1033 /**
1034  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
1035  * @vcpu:	The VCPU pointer
1036  *
1037  * This function is called through the VCPU_RUN ioctl called from user space. It
1038  * will execute VM code in a loop until the time slice for the process is used
1039  * or some emulation is needed from user space in which case the function will
1040  * return with return value 0 and with the kvm_run structure filled in with the
1041  * required data for the requested emulation.
1042  */
1043 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1044 {
1045 	struct kvm_run *run = vcpu->run;
1046 	int ret;
1047 
1048 	if (run->exit_reason == KVM_EXIT_MMIO) {
1049 		ret = kvm_handle_mmio_return(vcpu);
1050 		if (ret)
1051 			return ret;
1052 	}
1053 
1054 	vcpu_load(vcpu);
1055 
1056 	if (run->immediate_exit) {
1057 		ret = -EINTR;
1058 		goto out;
1059 	}
1060 
1061 	kvm_sigset_activate(vcpu);
1062 
1063 	ret = 1;
1064 	run->exit_reason = KVM_EXIT_UNKNOWN;
1065 	run->flags = 0;
1066 	while (ret > 0) {
1067 		/*
1068 		 * Check conditions before entering the guest
1069 		 */
1070 		ret = xfer_to_guest_mode_handle_work(vcpu);
1071 		if (!ret)
1072 			ret = 1;
1073 
1074 		if (ret > 0)
1075 			ret = check_vcpu_requests(vcpu);
1076 
1077 		/*
1078 		 * Preparing the interrupts to be injected also
1079 		 * involves poking the GIC, which must be done in a
1080 		 * non-preemptible context.
1081 		 */
1082 		preempt_disable();
1083 
1084 		/*
1085 		 * The VMID allocator only tracks active VMIDs per
1086 		 * physical CPU, and therefore the VMID allocated may not be
1087 		 * preserved on VMID roll-over if the task was preempted,
1088 		 * making a thread's VMID inactive. So we need to call
1089 		 * kvm_arm_vmid_update() in non-premptible context.
1090 		 */
1091 		if (kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid) &&
1092 		    has_vhe())
1093 			__load_stage2(vcpu->arch.hw_mmu,
1094 				      vcpu->arch.hw_mmu->arch);
1095 
1096 		kvm_pmu_flush_hwstate(vcpu);
1097 
1098 		local_irq_disable();
1099 
1100 		kvm_vgic_flush_hwstate(vcpu);
1101 
1102 		kvm_pmu_update_vcpu_events(vcpu);
1103 
1104 		/*
1105 		 * Ensure we set mode to IN_GUEST_MODE after we disable
1106 		 * interrupts and before the final VCPU requests check.
1107 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
1108 		 * Documentation/virt/kvm/vcpu-requests.rst
1109 		 */
1110 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1111 
1112 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
1113 			vcpu->mode = OUTSIDE_GUEST_MODE;
1114 			isb(); /* Ensure work in x_flush_hwstate is committed */
1115 			kvm_pmu_sync_hwstate(vcpu);
1116 			if (static_branch_unlikely(&userspace_irqchip_in_use))
1117 				kvm_timer_sync_user(vcpu);
1118 			kvm_vgic_sync_hwstate(vcpu);
1119 			local_irq_enable();
1120 			preempt_enable();
1121 			continue;
1122 		}
1123 
1124 		kvm_arm_setup_debug(vcpu);
1125 		kvm_arch_vcpu_ctxflush_fp(vcpu);
1126 
1127 		/**************************************************************
1128 		 * Enter the guest
1129 		 */
1130 		trace_kvm_entry(*vcpu_pc(vcpu));
1131 		guest_timing_enter_irqoff();
1132 
1133 		ret = kvm_arm_vcpu_enter_exit(vcpu);
1134 
1135 		vcpu->mode = OUTSIDE_GUEST_MODE;
1136 		vcpu->stat.exits++;
1137 		/*
1138 		 * Back from guest
1139 		 *************************************************************/
1140 
1141 		kvm_arm_clear_debug(vcpu);
1142 
1143 		/*
1144 		 * We must sync the PMU state before the vgic state so
1145 		 * that the vgic can properly sample the updated state of the
1146 		 * interrupt line.
1147 		 */
1148 		kvm_pmu_sync_hwstate(vcpu);
1149 
1150 		/*
1151 		 * Sync the vgic state before syncing the timer state because
1152 		 * the timer code needs to know if the virtual timer
1153 		 * interrupts are active.
1154 		 */
1155 		kvm_vgic_sync_hwstate(vcpu);
1156 
1157 		/*
1158 		 * Sync the timer hardware state before enabling interrupts as
1159 		 * we don't want vtimer interrupts to race with syncing the
1160 		 * timer virtual interrupt state.
1161 		 */
1162 		if (static_branch_unlikely(&userspace_irqchip_in_use))
1163 			kvm_timer_sync_user(vcpu);
1164 
1165 		kvm_arch_vcpu_ctxsync_fp(vcpu);
1166 
1167 		/*
1168 		 * We must ensure that any pending interrupts are taken before
1169 		 * we exit guest timing so that timer ticks are accounted as
1170 		 * guest time. Transiently unmask interrupts so that any
1171 		 * pending interrupts are taken.
1172 		 *
1173 		 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1174 		 * context synchronization event) is necessary to ensure that
1175 		 * pending interrupts are taken.
1176 		 */
1177 		if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1178 			local_irq_enable();
1179 			isb();
1180 			local_irq_disable();
1181 		}
1182 
1183 		guest_timing_exit_irqoff();
1184 
1185 		local_irq_enable();
1186 
1187 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1188 
1189 		/* Exit types that need handling before we can be preempted */
1190 		handle_exit_early(vcpu, ret);
1191 
1192 		preempt_enable();
1193 
1194 		/*
1195 		 * The ARMv8 architecture doesn't give the hypervisor
1196 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
1197 		 * if implemented by the CPU. If we spot the guest in such
1198 		 * state and that we decided it wasn't supposed to do so (like
1199 		 * with the asymmetric AArch32 case), return to userspace with
1200 		 * a fatal error.
1201 		 */
1202 		if (vcpu_mode_is_bad_32bit(vcpu)) {
1203 			/*
1204 			 * As we have caught the guest red-handed, decide that
1205 			 * it isn't fit for purpose anymore by making the vcpu
1206 			 * invalid. The VMM can try and fix it by issuing  a
1207 			 * KVM_ARM_VCPU_INIT if it really wants to.
1208 			 */
1209 			vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
1210 			ret = ARM_EXCEPTION_IL;
1211 		}
1212 
1213 		ret = handle_exit(vcpu, ret);
1214 	}
1215 
1216 	/* Tell userspace about in-kernel device output levels */
1217 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1218 		kvm_timer_update_run(vcpu);
1219 		kvm_pmu_update_run(vcpu);
1220 	}
1221 
1222 	kvm_sigset_deactivate(vcpu);
1223 
1224 out:
1225 	/*
1226 	 * In the unlikely event that we are returning to userspace
1227 	 * with pending exceptions or PC adjustment, commit these
1228 	 * adjustments in order to give userspace a consistent view of
1229 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1230 	 * being preempt-safe on VHE.
1231 	 */
1232 	if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1233 		     vcpu_get_flag(vcpu, INCREMENT_PC)))
1234 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
1235 
1236 	vcpu_put(vcpu);
1237 	return ret;
1238 }
1239 
1240 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1241 {
1242 	int bit_index;
1243 	bool set;
1244 	unsigned long *hcr;
1245 
1246 	if (number == KVM_ARM_IRQ_CPU_IRQ)
1247 		bit_index = __ffs(HCR_VI);
1248 	else /* KVM_ARM_IRQ_CPU_FIQ */
1249 		bit_index = __ffs(HCR_VF);
1250 
1251 	hcr = vcpu_hcr(vcpu);
1252 	if (level)
1253 		set = test_and_set_bit(bit_index, hcr);
1254 	else
1255 		set = test_and_clear_bit(bit_index, hcr);
1256 
1257 	/*
1258 	 * If we didn't change anything, no need to wake up or kick other CPUs
1259 	 */
1260 	if (set == level)
1261 		return 0;
1262 
1263 	/*
1264 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1265 	 * trigger a world-switch round on the running physical CPU to set the
1266 	 * virtual IRQ/FIQ fields in the HCR appropriately.
1267 	 */
1268 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1269 	kvm_vcpu_kick(vcpu);
1270 
1271 	return 0;
1272 }
1273 
1274 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1275 			  bool line_status)
1276 {
1277 	u32 irq = irq_level->irq;
1278 	unsigned int irq_type, vcpu_id, irq_num;
1279 	struct kvm_vcpu *vcpu = NULL;
1280 	bool level = irq_level->level;
1281 
1282 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1283 	vcpu_id = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1284 	vcpu_id += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1285 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1286 
1287 	trace_kvm_irq_line(irq_type, vcpu_id, irq_num, irq_level->level);
1288 
1289 	switch (irq_type) {
1290 	case KVM_ARM_IRQ_TYPE_CPU:
1291 		if (irqchip_in_kernel(kvm))
1292 			return -ENXIO;
1293 
1294 		vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1295 		if (!vcpu)
1296 			return -EINVAL;
1297 
1298 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1299 			return -EINVAL;
1300 
1301 		return vcpu_interrupt_line(vcpu, irq_num, level);
1302 	case KVM_ARM_IRQ_TYPE_PPI:
1303 		if (!irqchip_in_kernel(kvm))
1304 			return -ENXIO;
1305 
1306 		vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1307 		if (!vcpu)
1308 			return -EINVAL;
1309 
1310 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1311 			return -EINVAL;
1312 
1313 		return kvm_vgic_inject_irq(kvm, vcpu, irq_num, level, NULL);
1314 	case KVM_ARM_IRQ_TYPE_SPI:
1315 		if (!irqchip_in_kernel(kvm))
1316 			return -ENXIO;
1317 
1318 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1319 			return -EINVAL;
1320 
1321 		return kvm_vgic_inject_irq(kvm, NULL, irq_num, level, NULL);
1322 	}
1323 
1324 	return -EINVAL;
1325 }
1326 
1327 static unsigned long system_supported_vcpu_features(void)
1328 {
1329 	unsigned long features = KVM_VCPU_VALID_FEATURES;
1330 
1331 	if (!cpus_have_final_cap(ARM64_HAS_32BIT_EL1))
1332 		clear_bit(KVM_ARM_VCPU_EL1_32BIT, &features);
1333 
1334 	if (!kvm_arm_support_pmu_v3())
1335 		clear_bit(KVM_ARM_VCPU_PMU_V3, &features);
1336 
1337 	if (!system_supports_sve())
1338 		clear_bit(KVM_ARM_VCPU_SVE, &features);
1339 
1340 	if (!kvm_has_full_ptr_auth()) {
1341 		clear_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features);
1342 		clear_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features);
1343 	}
1344 
1345 	if (!cpus_have_final_cap(ARM64_HAS_NESTED_VIRT))
1346 		clear_bit(KVM_ARM_VCPU_HAS_EL2, &features);
1347 
1348 	return features;
1349 }
1350 
1351 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1352 					const struct kvm_vcpu_init *init)
1353 {
1354 	unsigned long features = init->features[0];
1355 	int i;
1356 
1357 	if (features & ~KVM_VCPU_VALID_FEATURES)
1358 		return -ENOENT;
1359 
1360 	for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1361 		if (init->features[i])
1362 			return -ENOENT;
1363 	}
1364 
1365 	if (features & ~system_supported_vcpu_features())
1366 		return -EINVAL;
1367 
1368 	/*
1369 	 * For now make sure that both address/generic pointer authentication
1370 	 * features are requested by the userspace together.
1371 	 */
1372 	if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features) !=
1373 	    test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features))
1374 		return -EINVAL;
1375 
1376 	/* Disallow NV+SVE for the time being */
1377 	if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features) &&
1378 	    test_bit(KVM_ARM_VCPU_SVE, &features))
1379 		return -EINVAL;
1380 
1381 	if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features))
1382 		return 0;
1383 
1384 	/* MTE is incompatible with AArch32 */
1385 	if (kvm_has_mte(vcpu->kvm))
1386 		return -EINVAL;
1387 
1388 	/* NV is incompatible with AArch32 */
1389 	if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features))
1390 		return -EINVAL;
1391 
1392 	return 0;
1393 }
1394 
1395 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1396 				  const struct kvm_vcpu_init *init)
1397 {
1398 	unsigned long features = init->features[0];
1399 
1400 	return !bitmap_equal(vcpu->kvm->arch.vcpu_features, &features,
1401 			     KVM_VCPU_MAX_FEATURES);
1402 }
1403 
1404 static int kvm_setup_vcpu(struct kvm_vcpu *vcpu)
1405 {
1406 	struct kvm *kvm = vcpu->kvm;
1407 	int ret = 0;
1408 
1409 	/*
1410 	 * When the vCPU has a PMU, but no PMU is set for the guest
1411 	 * yet, set the default one.
1412 	 */
1413 	if (kvm_vcpu_has_pmu(vcpu) && !kvm->arch.arm_pmu)
1414 		ret = kvm_arm_set_default_pmu(kvm);
1415 
1416 	return ret;
1417 }
1418 
1419 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1420 				 const struct kvm_vcpu_init *init)
1421 {
1422 	unsigned long features = init->features[0];
1423 	struct kvm *kvm = vcpu->kvm;
1424 	int ret = -EINVAL;
1425 
1426 	mutex_lock(&kvm->arch.config_lock);
1427 
1428 	if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) &&
1429 	    kvm_vcpu_init_changed(vcpu, init))
1430 		goto out_unlock;
1431 
1432 	bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
1433 
1434 	ret = kvm_setup_vcpu(vcpu);
1435 	if (ret)
1436 		goto out_unlock;
1437 
1438 	/* Now we know what it is, we can reset it. */
1439 	kvm_reset_vcpu(vcpu);
1440 
1441 	set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags);
1442 	vcpu_set_flag(vcpu, VCPU_INITIALIZED);
1443 	ret = 0;
1444 out_unlock:
1445 	mutex_unlock(&kvm->arch.config_lock);
1446 	return ret;
1447 }
1448 
1449 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1450 			       const struct kvm_vcpu_init *init)
1451 {
1452 	int ret;
1453 
1454 	if (init->target != KVM_ARM_TARGET_GENERIC_V8 &&
1455 	    init->target != kvm_target_cpu())
1456 		return -EINVAL;
1457 
1458 	ret = kvm_vcpu_init_check_features(vcpu, init);
1459 	if (ret)
1460 		return ret;
1461 
1462 	if (!kvm_vcpu_initialized(vcpu))
1463 		return __kvm_vcpu_set_target(vcpu, init);
1464 
1465 	if (kvm_vcpu_init_changed(vcpu, init))
1466 		return -EINVAL;
1467 
1468 	kvm_reset_vcpu(vcpu);
1469 	return 0;
1470 }
1471 
1472 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1473 					 struct kvm_vcpu_init *init)
1474 {
1475 	bool power_off = false;
1476 	int ret;
1477 
1478 	/*
1479 	 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid
1480 	 * reflecting it in the finalized feature set, thus limiting its scope
1481 	 * to a single KVM_ARM_VCPU_INIT call.
1482 	 */
1483 	if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) {
1484 		init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF);
1485 		power_off = true;
1486 	}
1487 
1488 	ret = kvm_vcpu_set_target(vcpu, init);
1489 	if (ret)
1490 		return ret;
1491 
1492 	/*
1493 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1494 	 * guest MMU is turned off and flush the caches as needed.
1495 	 *
1496 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1497 	 * ensuring that the data side is always coherent. We still
1498 	 * need to invalidate the I-cache though, as FWB does *not*
1499 	 * imply CTR_EL0.DIC.
1500 	 */
1501 	if (vcpu_has_run_once(vcpu)) {
1502 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1503 			stage2_unmap_vm(vcpu->kvm);
1504 		else
1505 			icache_inval_all_pou();
1506 	}
1507 
1508 	vcpu_reset_hcr(vcpu);
1509 	vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu);
1510 
1511 	/*
1512 	 * Handle the "start in power-off" case.
1513 	 */
1514 	spin_lock(&vcpu->arch.mp_state_lock);
1515 
1516 	if (power_off)
1517 		__kvm_arm_vcpu_power_off(vcpu);
1518 	else
1519 		WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1520 
1521 	spin_unlock(&vcpu->arch.mp_state_lock);
1522 
1523 	return 0;
1524 }
1525 
1526 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1527 				 struct kvm_device_attr *attr)
1528 {
1529 	int ret = -ENXIO;
1530 
1531 	switch (attr->group) {
1532 	default:
1533 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1534 		break;
1535 	}
1536 
1537 	return ret;
1538 }
1539 
1540 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1541 				 struct kvm_device_attr *attr)
1542 {
1543 	int ret = -ENXIO;
1544 
1545 	switch (attr->group) {
1546 	default:
1547 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1548 		break;
1549 	}
1550 
1551 	return ret;
1552 }
1553 
1554 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1555 				 struct kvm_device_attr *attr)
1556 {
1557 	int ret = -ENXIO;
1558 
1559 	switch (attr->group) {
1560 	default:
1561 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1562 		break;
1563 	}
1564 
1565 	return ret;
1566 }
1567 
1568 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1569 				   struct kvm_vcpu_events *events)
1570 {
1571 	memset(events, 0, sizeof(*events));
1572 
1573 	return __kvm_arm_vcpu_get_events(vcpu, events);
1574 }
1575 
1576 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1577 				   struct kvm_vcpu_events *events)
1578 {
1579 	int i;
1580 
1581 	/* check whether the reserved field is zero */
1582 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1583 		if (events->reserved[i])
1584 			return -EINVAL;
1585 
1586 	/* check whether the pad field is zero */
1587 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1588 		if (events->exception.pad[i])
1589 			return -EINVAL;
1590 
1591 	return __kvm_arm_vcpu_set_events(vcpu, events);
1592 }
1593 
1594 long kvm_arch_vcpu_ioctl(struct file *filp,
1595 			 unsigned int ioctl, unsigned long arg)
1596 {
1597 	struct kvm_vcpu *vcpu = filp->private_data;
1598 	void __user *argp = (void __user *)arg;
1599 	struct kvm_device_attr attr;
1600 	long r;
1601 
1602 	switch (ioctl) {
1603 	case KVM_ARM_VCPU_INIT: {
1604 		struct kvm_vcpu_init init;
1605 
1606 		r = -EFAULT;
1607 		if (copy_from_user(&init, argp, sizeof(init)))
1608 			break;
1609 
1610 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1611 		break;
1612 	}
1613 	case KVM_SET_ONE_REG:
1614 	case KVM_GET_ONE_REG: {
1615 		struct kvm_one_reg reg;
1616 
1617 		r = -ENOEXEC;
1618 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1619 			break;
1620 
1621 		r = -EFAULT;
1622 		if (copy_from_user(&reg, argp, sizeof(reg)))
1623 			break;
1624 
1625 		/*
1626 		 * We could owe a reset due to PSCI. Handle the pending reset
1627 		 * here to ensure userspace register accesses are ordered after
1628 		 * the reset.
1629 		 */
1630 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1631 			kvm_reset_vcpu(vcpu);
1632 
1633 		if (ioctl == KVM_SET_ONE_REG)
1634 			r = kvm_arm_set_reg(vcpu, &reg);
1635 		else
1636 			r = kvm_arm_get_reg(vcpu, &reg);
1637 		break;
1638 	}
1639 	case KVM_GET_REG_LIST: {
1640 		struct kvm_reg_list __user *user_list = argp;
1641 		struct kvm_reg_list reg_list;
1642 		unsigned n;
1643 
1644 		r = -ENOEXEC;
1645 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1646 			break;
1647 
1648 		r = -EPERM;
1649 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1650 			break;
1651 
1652 		r = -EFAULT;
1653 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1654 			break;
1655 		n = reg_list.n;
1656 		reg_list.n = kvm_arm_num_regs(vcpu);
1657 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1658 			break;
1659 		r = -E2BIG;
1660 		if (n < reg_list.n)
1661 			break;
1662 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1663 		break;
1664 	}
1665 	case KVM_SET_DEVICE_ATTR: {
1666 		r = -EFAULT;
1667 		if (copy_from_user(&attr, argp, sizeof(attr)))
1668 			break;
1669 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1670 		break;
1671 	}
1672 	case KVM_GET_DEVICE_ATTR: {
1673 		r = -EFAULT;
1674 		if (copy_from_user(&attr, argp, sizeof(attr)))
1675 			break;
1676 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1677 		break;
1678 	}
1679 	case KVM_HAS_DEVICE_ATTR: {
1680 		r = -EFAULT;
1681 		if (copy_from_user(&attr, argp, sizeof(attr)))
1682 			break;
1683 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1684 		break;
1685 	}
1686 	case KVM_GET_VCPU_EVENTS: {
1687 		struct kvm_vcpu_events events;
1688 
1689 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1690 			return -EINVAL;
1691 
1692 		if (copy_to_user(argp, &events, sizeof(events)))
1693 			return -EFAULT;
1694 
1695 		return 0;
1696 	}
1697 	case KVM_SET_VCPU_EVENTS: {
1698 		struct kvm_vcpu_events events;
1699 
1700 		if (copy_from_user(&events, argp, sizeof(events)))
1701 			return -EFAULT;
1702 
1703 		return kvm_arm_vcpu_set_events(vcpu, &events);
1704 	}
1705 	case KVM_ARM_VCPU_FINALIZE: {
1706 		int what;
1707 
1708 		if (!kvm_vcpu_initialized(vcpu))
1709 			return -ENOEXEC;
1710 
1711 		if (get_user(what, (const int __user *)argp))
1712 			return -EFAULT;
1713 
1714 		return kvm_arm_vcpu_finalize(vcpu, what);
1715 	}
1716 	default:
1717 		r = -EINVAL;
1718 	}
1719 
1720 	return r;
1721 }
1722 
1723 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1724 {
1725 
1726 }
1727 
1728 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1729 					struct kvm_arm_device_addr *dev_addr)
1730 {
1731 	switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1732 	case KVM_ARM_DEVICE_VGIC_V2:
1733 		if (!vgic_present)
1734 			return -ENXIO;
1735 		return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1736 	default:
1737 		return -ENODEV;
1738 	}
1739 }
1740 
1741 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1742 {
1743 	switch (attr->group) {
1744 	case KVM_ARM_VM_SMCCC_CTRL:
1745 		return kvm_vm_smccc_has_attr(kvm, attr);
1746 	default:
1747 		return -ENXIO;
1748 	}
1749 }
1750 
1751 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1752 {
1753 	switch (attr->group) {
1754 	case KVM_ARM_VM_SMCCC_CTRL:
1755 		return kvm_vm_smccc_set_attr(kvm, attr);
1756 	default:
1757 		return -ENXIO;
1758 	}
1759 }
1760 
1761 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1762 {
1763 	struct kvm *kvm = filp->private_data;
1764 	void __user *argp = (void __user *)arg;
1765 	struct kvm_device_attr attr;
1766 
1767 	switch (ioctl) {
1768 	case KVM_CREATE_IRQCHIP: {
1769 		int ret;
1770 		if (!vgic_present)
1771 			return -ENXIO;
1772 		mutex_lock(&kvm->lock);
1773 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1774 		mutex_unlock(&kvm->lock);
1775 		return ret;
1776 	}
1777 	case KVM_ARM_SET_DEVICE_ADDR: {
1778 		struct kvm_arm_device_addr dev_addr;
1779 
1780 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1781 			return -EFAULT;
1782 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1783 	}
1784 	case KVM_ARM_PREFERRED_TARGET: {
1785 		struct kvm_vcpu_init init = {
1786 			.target = KVM_ARM_TARGET_GENERIC_V8,
1787 		};
1788 
1789 		if (copy_to_user(argp, &init, sizeof(init)))
1790 			return -EFAULT;
1791 
1792 		return 0;
1793 	}
1794 	case KVM_ARM_MTE_COPY_TAGS: {
1795 		struct kvm_arm_copy_mte_tags copy_tags;
1796 
1797 		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1798 			return -EFAULT;
1799 		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1800 	}
1801 	case KVM_ARM_SET_COUNTER_OFFSET: {
1802 		struct kvm_arm_counter_offset offset;
1803 
1804 		if (copy_from_user(&offset, argp, sizeof(offset)))
1805 			return -EFAULT;
1806 		return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1807 	}
1808 	case KVM_HAS_DEVICE_ATTR: {
1809 		if (copy_from_user(&attr, argp, sizeof(attr)))
1810 			return -EFAULT;
1811 
1812 		return kvm_vm_has_attr(kvm, &attr);
1813 	}
1814 	case KVM_SET_DEVICE_ATTR: {
1815 		if (copy_from_user(&attr, argp, sizeof(attr)))
1816 			return -EFAULT;
1817 
1818 		return kvm_vm_set_attr(kvm, &attr);
1819 	}
1820 	case KVM_ARM_GET_REG_WRITABLE_MASKS: {
1821 		struct reg_mask_range range;
1822 
1823 		if (copy_from_user(&range, argp, sizeof(range)))
1824 			return -EFAULT;
1825 		return kvm_vm_ioctl_get_reg_writable_masks(kvm, &range);
1826 	}
1827 	default:
1828 		return -EINVAL;
1829 	}
1830 }
1831 
1832 /* unlocks vcpus from @vcpu_lock_idx and smaller */
1833 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1834 {
1835 	struct kvm_vcpu *tmp_vcpu;
1836 
1837 	for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1838 		tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1839 		mutex_unlock(&tmp_vcpu->mutex);
1840 	}
1841 }
1842 
1843 void unlock_all_vcpus(struct kvm *kvm)
1844 {
1845 	lockdep_assert_held(&kvm->lock);
1846 
1847 	unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1848 }
1849 
1850 /* Returns true if all vcpus were locked, false otherwise */
1851 bool lock_all_vcpus(struct kvm *kvm)
1852 {
1853 	struct kvm_vcpu *tmp_vcpu;
1854 	unsigned long c;
1855 
1856 	lockdep_assert_held(&kvm->lock);
1857 
1858 	/*
1859 	 * Any time a vcpu is in an ioctl (including running), the
1860 	 * core KVM code tries to grab the vcpu->mutex.
1861 	 *
1862 	 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1863 	 * other VCPUs can fiddle with the state while we access it.
1864 	 */
1865 	kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1866 		if (!mutex_trylock(&tmp_vcpu->mutex)) {
1867 			unlock_vcpus(kvm, c - 1);
1868 			return false;
1869 		}
1870 	}
1871 
1872 	return true;
1873 }
1874 
1875 static unsigned long nvhe_percpu_size(void)
1876 {
1877 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1878 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1879 }
1880 
1881 static unsigned long nvhe_percpu_order(void)
1882 {
1883 	unsigned long size = nvhe_percpu_size();
1884 
1885 	return size ? get_order(size) : 0;
1886 }
1887 
1888 /* A lookup table holding the hypervisor VA for each vector slot */
1889 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1890 
1891 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1892 {
1893 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1894 }
1895 
1896 static int kvm_init_vector_slots(void)
1897 {
1898 	int err;
1899 	void *base;
1900 
1901 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1902 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1903 
1904 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1905 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1906 
1907 	if (kvm_system_needs_idmapped_vectors() &&
1908 	    !is_protected_kvm_enabled()) {
1909 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1910 					       __BP_HARDEN_HYP_VECS_SZ, &base);
1911 		if (err)
1912 			return err;
1913 	}
1914 
1915 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1916 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1917 	return 0;
1918 }
1919 
1920 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
1921 {
1922 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1923 	u64 mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1924 	unsigned long tcr;
1925 
1926 	/*
1927 	 * Calculate the raw per-cpu offset without a translation from the
1928 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1929 	 * so that we can use adr_l to access per-cpu variables in EL2.
1930 	 * Also drop the KASAN tag which gets in the way...
1931 	 */
1932 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1933 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1934 
1935 	params->mair_el2 = read_sysreg(mair_el1);
1936 
1937 	tcr = read_sysreg(tcr_el1);
1938 	if (cpus_have_final_cap(ARM64_KVM_HVHE)) {
1939 		tcr |= TCR_EPD1_MASK;
1940 	} else {
1941 		tcr &= TCR_EL2_MASK;
1942 		tcr |= TCR_EL2_RES1;
1943 	}
1944 	tcr &= ~TCR_T0SZ_MASK;
1945 	tcr |= TCR_T0SZ(hyp_va_bits);
1946 	tcr &= ~TCR_EL2_PS_MASK;
1947 	tcr |= FIELD_PREP(TCR_EL2_PS_MASK, kvm_get_parange(mmfr0));
1948 	if (kvm_lpa2_is_enabled())
1949 		tcr |= TCR_EL2_DS;
1950 	params->tcr_el2 = tcr;
1951 
1952 	params->pgd_pa = kvm_mmu_get_httbr();
1953 	if (is_protected_kvm_enabled())
1954 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1955 	else
1956 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1957 	if (cpus_have_final_cap(ARM64_KVM_HVHE))
1958 		params->hcr_el2 |= HCR_E2H;
1959 	params->vttbr = params->vtcr = 0;
1960 
1961 	/*
1962 	 * Flush the init params from the data cache because the struct will
1963 	 * be read while the MMU is off.
1964 	 */
1965 	kvm_flush_dcache_to_poc(params, sizeof(*params));
1966 }
1967 
1968 static void hyp_install_host_vector(void)
1969 {
1970 	struct kvm_nvhe_init_params *params;
1971 	struct arm_smccc_res res;
1972 
1973 	/* Switch from the HYP stub to our own HYP init vector */
1974 	__hyp_set_vectors(kvm_get_idmap_vector());
1975 
1976 	/*
1977 	 * Call initialization code, and switch to the full blown HYP code.
1978 	 * If the cpucaps haven't been finalized yet, something has gone very
1979 	 * wrong, and hyp will crash and burn when it uses any
1980 	 * cpus_have_*_cap() wrapper.
1981 	 */
1982 	BUG_ON(!system_capabilities_finalized());
1983 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1984 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1985 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1986 }
1987 
1988 static void cpu_init_hyp_mode(void)
1989 {
1990 	hyp_install_host_vector();
1991 
1992 	/*
1993 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1994 	 * at EL2.
1995 	 */
1996 	if (this_cpu_has_cap(ARM64_SSBS) &&
1997 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1998 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1999 	}
2000 }
2001 
2002 static void cpu_hyp_reset(void)
2003 {
2004 	if (!is_kernel_in_hyp_mode())
2005 		__hyp_reset_vectors();
2006 }
2007 
2008 /*
2009  * EL2 vectors can be mapped and rerouted in a number of ways,
2010  * depending on the kernel configuration and CPU present:
2011  *
2012  * - If the CPU is affected by Spectre-v2, the hardening sequence is
2013  *   placed in one of the vector slots, which is executed before jumping
2014  *   to the real vectors.
2015  *
2016  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
2017  *   containing the hardening sequence is mapped next to the idmap page,
2018  *   and executed before jumping to the real vectors.
2019  *
2020  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
2021  *   empty slot is selected, mapped next to the idmap page, and
2022  *   executed before jumping to the real vectors.
2023  *
2024  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
2025  * VHE, as we don't have hypervisor-specific mappings. If the system
2026  * is VHE and yet selects this capability, it will be ignored.
2027  */
2028 static void cpu_set_hyp_vector(void)
2029 {
2030 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
2031 	void *vector = hyp_spectre_vector_selector[data->slot];
2032 
2033 	if (!is_protected_kvm_enabled())
2034 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
2035 	else
2036 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
2037 }
2038 
2039 static void cpu_hyp_init_context(void)
2040 {
2041 	kvm_init_host_cpu_context(host_data_ptr(host_ctxt));
2042 
2043 	if (!is_kernel_in_hyp_mode())
2044 		cpu_init_hyp_mode();
2045 }
2046 
2047 static void cpu_hyp_init_features(void)
2048 {
2049 	cpu_set_hyp_vector();
2050 	kvm_arm_init_debug();
2051 
2052 	if (is_kernel_in_hyp_mode())
2053 		kvm_timer_init_vhe();
2054 
2055 	if (vgic_present)
2056 		kvm_vgic_init_cpu_hardware();
2057 }
2058 
2059 static void cpu_hyp_reinit(void)
2060 {
2061 	cpu_hyp_reset();
2062 	cpu_hyp_init_context();
2063 	cpu_hyp_init_features();
2064 }
2065 
2066 static void cpu_hyp_init(void *discard)
2067 {
2068 	if (!__this_cpu_read(kvm_hyp_initialized)) {
2069 		cpu_hyp_reinit();
2070 		__this_cpu_write(kvm_hyp_initialized, 1);
2071 	}
2072 }
2073 
2074 static void cpu_hyp_uninit(void *discard)
2075 {
2076 	if (__this_cpu_read(kvm_hyp_initialized)) {
2077 		cpu_hyp_reset();
2078 		__this_cpu_write(kvm_hyp_initialized, 0);
2079 	}
2080 }
2081 
2082 int kvm_arch_hardware_enable(void)
2083 {
2084 	/*
2085 	 * Most calls to this function are made with migration
2086 	 * disabled, but not with preemption disabled. The former is
2087 	 * enough to ensure correctness, but most of the helpers
2088 	 * expect the later and will throw a tantrum otherwise.
2089 	 */
2090 	preempt_disable();
2091 
2092 	cpu_hyp_init(NULL);
2093 
2094 	kvm_vgic_cpu_up();
2095 	kvm_timer_cpu_up();
2096 
2097 	preempt_enable();
2098 
2099 	return 0;
2100 }
2101 
2102 void kvm_arch_hardware_disable(void)
2103 {
2104 	kvm_timer_cpu_down();
2105 	kvm_vgic_cpu_down();
2106 
2107 	if (!is_protected_kvm_enabled())
2108 		cpu_hyp_uninit(NULL);
2109 }
2110 
2111 #ifdef CONFIG_CPU_PM
2112 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
2113 				    unsigned long cmd,
2114 				    void *v)
2115 {
2116 	/*
2117 	 * kvm_hyp_initialized is left with its old value over
2118 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
2119 	 * re-enable hyp.
2120 	 */
2121 	switch (cmd) {
2122 	case CPU_PM_ENTER:
2123 		if (__this_cpu_read(kvm_hyp_initialized))
2124 			/*
2125 			 * don't update kvm_hyp_initialized here
2126 			 * so that the hyp will be re-enabled
2127 			 * when we resume. See below.
2128 			 */
2129 			cpu_hyp_reset();
2130 
2131 		return NOTIFY_OK;
2132 	case CPU_PM_ENTER_FAILED:
2133 	case CPU_PM_EXIT:
2134 		if (__this_cpu_read(kvm_hyp_initialized))
2135 			/* The hyp was enabled before suspend. */
2136 			cpu_hyp_reinit();
2137 
2138 		return NOTIFY_OK;
2139 
2140 	default:
2141 		return NOTIFY_DONE;
2142 	}
2143 }
2144 
2145 static struct notifier_block hyp_init_cpu_pm_nb = {
2146 	.notifier_call = hyp_init_cpu_pm_notifier,
2147 };
2148 
2149 static void __init hyp_cpu_pm_init(void)
2150 {
2151 	if (!is_protected_kvm_enabled())
2152 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
2153 }
2154 static void __init hyp_cpu_pm_exit(void)
2155 {
2156 	if (!is_protected_kvm_enabled())
2157 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
2158 }
2159 #else
2160 static inline void __init hyp_cpu_pm_init(void)
2161 {
2162 }
2163 static inline void __init hyp_cpu_pm_exit(void)
2164 {
2165 }
2166 #endif
2167 
2168 static void __init init_cpu_logical_map(void)
2169 {
2170 	unsigned int cpu;
2171 
2172 	/*
2173 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
2174 	 * Only copy the set of online CPUs whose features have been checked
2175 	 * against the finalized system capabilities. The hypervisor will not
2176 	 * allow any other CPUs from the `possible` set to boot.
2177 	 */
2178 	for_each_online_cpu(cpu)
2179 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
2180 }
2181 
2182 #define init_psci_0_1_impl_state(config, what)	\
2183 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
2184 
2185 static bool __init init_psci_relay(void)
2186 {
2187 	/*
2188 	 * If PSCI has not been initialized, protected KVM cannot install
2189 	 * itself on newly booted CPUs.
2190 	 */
2191 	if (!psci_ops.get_version) {
2192 		kvm_err("Cannot initialize protected mode without PSCI\n");
2193 		return false;
2194 	}
2195 
2196 	kvm_host_psci_config.version = psci_ops.get_version();
2197 	kvm_host_psci_config.smccc_version = arm_smccc_get_version();
2198 
2199 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
2200 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
2201 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
2202 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
2203 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
2204 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
2205 	}
2206 	return true;
2207 }
2208 
2209 static int __init init_subsystems(void)
2210 {
2211 	int err = 0;
2212 
2213 	/*
2214 	 * Enable hardware so that subsystem initialisation can access EL2.
2215 	 */
2216 	on_each_cpu(cpu_hyp_init, NULL, 1);
2217 
2218 	/*
2219 	 * Register CPU lower-power notifier
2220 	 */
2221 	hyp_cpu_pm_init();
2222 
2223 	/*
2224 	 * Init HYP view of VGIC
2225 	 */
2226 	err = kvm_vgic_hyp_init();
2227 	switch (err) {
2228 	case 0:
2229 		vgic_present = true;
2230 		break;
2231 	case -ENODEV:
2232 	case -ENXIO:
2233 		vgic_present = false;
2234 		err = 0;
2235 		break;
2236 	default:
2237 		goto out;
2238 	}
2239 
2240 	/*
2241 	 * Init HYP architected timer support
2242 	 */
2243 	err = kvm_timer_hyp_init(vgic_present);
2244 	if (err)
2245 		goto out;
2246 
2247 	kvm_register_perf_callbacks(NULL);
2248 
2249 out:
2250 	if (err)
2251 		hyp_cpu_pm_exit();
2252 
2253 	if (err || !is_protected_kvm_enabled())
2254 		on_each_cpu(cpu_hyp_uninit, NULL, 1);
2255 
2256 	return err;
2257 }
2258 
2259 static void __init teardown_subsystems(void)
2260 {
2261 	kvm_unregister_perf_callbacks();
2262 	hyp_cpu_pm_exit();
2263 }
2264 
2265 static void __init teardown_hyp_mode(void)
2266 {
2267 	int cpu;
2268 
2269 	free_hyp_pgds();
2270 	for_each_possible_cpu(cpu) {
2271 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2272 		free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2273 	}
2274 }
2275 
2276 static int __init do_pkvm_init(u32 hyp_va_bits)
2277 {
2278 	void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2279 	int ret;
2280 
2281 	preempt_disable();
2282 	cpu_hyp_init_context();
2283 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2284 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
2285 				hyp_va_bits);
2286 	cpu_hyp_init_features();
2287 
2288 	/*
2289 	 * The stub hypercalls are now disabled, so set our local flag to
2290 	 * prevent a later re-init attempt in kvm_arch_hardware_enable().
2291 	 */
2292 	__this_cpu_write(kvm_hyp_initialized, 1);
2293 	preempt_enable();
2294 
2295 	return ret;
2296 }
2297 
2298 static u64 get_hyp_id_aa64pfr0_el1(void)
2299 {
2300 	/*
2301 	 * Track whether the system isn't affected by spectre/meltdown in the
2302 	 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2303 	 * Although this is per-CPU, we make it global for simplicity, e.g., not
2304 	 * to have to worry about vcpu migration.
2305 	 *
2306 	 * Unlike for non-protected VMs, userspace cannot override this for
2307 	 * protected VMs.
2308 	 */
2309 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2310 
2311 	val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2312 		 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2313 
2314 	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2315 			  arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2316 	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2317 			  arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2318 
2319 	return val;
2320 }
2321 
2322 static void kvm_hyp_init_symbols(void)
2323 {
2324 	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2325 	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2326 	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2327 	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2328 	kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2329 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2330 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2331 	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2332 	kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2333 	kvm_nvhe_sym(__icache_flags) = __icache_flags;
2334 	kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2335 }
2336 
2337 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2338 {
2339 	void *addr = phys_to_virt(hyp_mem_base);
2340 	int ret;
2341 
2342 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2343 	if (ret)
2344 		return ret;
2345 
2346 	ret = do_pkvm_init(hyp_va_bits);
2347 	if (ret)
2348 		return ret;
2349 
2350 	free_hyp_pgds();
2351 
2352 	return 0;
2353 }
2354 
2355 static void pkvm_hyp_init_ptrauth(void)
2356 {
2357 	struct kvm_cpu_context *hyp_ctxt;
2358 	int cpu;
2359 
2360 	for_each_possible_cpu(cpu) {
2361 		hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu);
2362 		hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long();
2363 		hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long();
2364 		hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long();
2365 		hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long();
2366 		hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long();
2367 		hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long();
2368 		hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long();
2369 		hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long();
2370 		hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long();
2371 		hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long();
2372 	}
2373 }
2374 
2375 /* Inits Hyp-mode on all online CPUs */
2376 static int __init init_hyp_mode(void)
2377 {
2378 	u32 hyp_va_bits;
2379 	int cpu;
2380 	int err = -ENOMEM;
2381 
2382 	/*
2383 	 * The protected Hyp-mode cannot be initialized if the memory pool
2384 	 * allocation has failed.
2385 	 */
2386 	if (is_protected_kvm_enabled() && !hyp_mem_base)
2387 		goto out_err;
2388 
2389 	/*
2390 	 * Allocate Hyp PGD and setup Hyp identity mapping
2391 	 */
2392 	err = kvm_mmu_init(&hyp_va_bits);
2393 	if (err)
2394 		goto out_err;
2395 
2396 	/*
2397 	 * Allocate stack pages for Hypervisor-mode
2398 	 */
2399 	for_each_possible_cpu(cpu) {
2400 		unsigned long stack_page;
2401 
2402 		stack_page = __get_free_page(GFP_KERNEL);
2403 		if (!stack_page) {
2404 			err = -ENOMEM;
2405 			goto out_err;
2406 		}
2407 
2408 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2409 	}
2410 
2411 	/*
2412 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2413 	 */
2414 	for_each_possible_cpu(cpu) {
2415 		struct page *page;
2416 		void *page_addr;
2417 
2418 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2419 		if (!page) {
2420 			err = -ENOMEM;
2421 			goto out_err;
2422 		}
2423 
2424 		page_addr = page_address(page);
2425 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2426 		kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2427 	}
2428 
2429 	/*
2430 	 * Map the Hyp-code called directly from the host
2431 	 */
2432 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2433 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2434 	if (err) {
2435 		kvm_err("Cannot map world-switch code\n");
2436 		goto out_err;
2437 	}
2438 
2439 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2440 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2441 	if (err) {
2442 		kvm_err("Cannot map .hyp.rodata section\n");
2443 		goto out_err;
2444 	}
2445 
2446 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2447 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2448 	if (err) {
2449 		kvm_err("Cannot map rodata section\n");
2450 		goto out_err;
2451 	}
2452 
2453 	/*
2454 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2455 	 * section thanks to an assertion in the linker script. Map it RW and
2456 	 * the rest of .bss RO.
2457 	 */
2458 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2459 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2460 	if (err) {
2461 		kvm_err("Cannot map hyp bss section: %d\n", err);
2462 		goto out_err;
2463 	}
2464 
2465 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2466 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2467 	if (err) {
2468 		kvm_err("Cannot map bss section\n");
2469 		goto out_err;
2470 	}
2471 
2472 	/*
2473 	 * Map the Hyp stack pages
2474 	 */
2475 	for_each_possible_cpu(cpu) {
2476 		struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2477 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2478 
2479 		err = create_hyp_stack(__pa(stack_page), &params->stack_hyp_va);
2480 		if (err) {
2481 			kvm_err("Cannot map hyp stack\n");
2482 			goto out_err;
2483 		}
2484 
2485 		/*
2486 		 * Save the stack PA in nvhe_init_params. This will be needed
2487 		 * to recreate the stack mapping in protected nVHE mode.
2488 		 * __hyp_pa() won't do the right thing there, since the stack
2489 		 * has been mapped in the flexible private VA space.
2490 		 */
2491 		params->stack_pa = __pa(stack_page);
2492 	}
2493 
2494 	for_each_possible_cpu(cpu) {
2495 		char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2496 		char *percpu_end = percpu_begin + nvhe_percpu_size();
2497 
2498 		/* Map Hyp percpu pages */
2499 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2500 		if (err) {
2501 			kvm_err("Cannot map hyp percpu region\n");
2502 			goto out_err;
2503 		}
2504 
2505 		/* Prepare the CPU initialization parameters */
2506 		cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2507 	}
2508 
2509 	kvm_hyp_init_symbols();
2510 
2511 	if (is_protected_kvm_enabled()) {
2512 		if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) &&
2513 		    cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH))
2514 			pkvm_hyp_init_ptrauth();
2515 
2516 		init_cpu_logical_map();
2517 
2518 		if (!init_psci_relay()) {
2519 			err = -ENODEV;
2520 			goto out_err;
2521 		}
2522 
2523 		err = kvm_hyp_init_protection(hyp_va_bits);
2524 		if (err) {
2525 			kvm_err("Failed to init hyp memory protection\n");
2526 			goto out_err;
2527 		}
2528 	}
2529 
2530 	return 0;
2531 
2532 out_err:
2533 	teardown_hyp_mode();
2534 	kvm_err("error initializing Hyp mode: %d\n", err);
2535 	return err;
2536 }
2537 
2538 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2539 {
2540 	struct kvm_vcpu *vcpu;
2541 	unsigned long i;
2542 
2543 	mpidr &= MPIDR_HWID_BITMASK;
2544 
2545 	if (kvm->arch.mpidr_data) {
2546 		u16 idx = kvm_mpidr_index(kvm->arch.mpidr_data, mpidr);
2547 
2548 		vcpu = kvm_get_vcpu(kvm,
2549 				    kvm->arch.mpidr_data->cmpidr_to_idx[idx]);
2550 		if (mpidr != kvm_vcpu_get_mpidr_aff(vcpu))
2551 			vcpu = NULL;
2552 
2553 		return vcpu;
2554 	}
2555 
2556 	kvm_for_each_vcpu(i, vcpu, kvm) {
2557 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2558 			return vcpu;
2559 	}
2560 	return NULL;
2561 }
2562 
2563 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2564 {
2565 	return irqchip_in_kernel(kvm);
2566 }
2567 
2568 bool kvm_arch_has_irq_bypass(void)
2569 {
2570 	return true;
2571 }
2572 
2573 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2574 				      struct irq_bypass_producer *prod)
2575 {
2576 	struct kvm_kernel_irqfd *irqfd =
2577 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2578 
2579 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2580 					  &irqfd->irq_entry);
2581 }
2582 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2583 				      struct irq_bypass_producer *prod)
2584 {
2585 	struct kvm_kernel_irqfd *irqfd =
2586 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2587 
2588 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2589 				     &irqfd->irq_entry);
2590 }
2591 
2592 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2593 {
2594 	struct kvm_kernel_irqfd *irqfd =
2595 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2596 
2597 	kvm_arm_halt_guest(irqfd->kvm);
2598 }
2599 
2600 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2601 {
2602 	struct kvm_kernel_irqfd *irqfd =
2603 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2604 
2605 	kvm_arm_resume_guest(irqfd->kvm);
2606 }
2607 
2608 /* Initialize Hyp-mode and memory mappings on all CPUs */
2609 static __init int kvm_arm_init(void)
2610 {
2611 	int err;
2612 	bool in_hyp_mode;
2613 
2614 	if (!is_hyp_mode_available()) {
2615 		kvm_info("HYP mode not available\n");
2616 		return -ENODEV;
2617 	}
2618 
2619 	if (kvm_get_mode() == KVM_MODE_NONE) {
2620 		kvm_info("KVM disabled from command line\n");
2621 		return -ENODEV;
2622 	}
2623 
2624 	err = kvm_sys_reg_table_init();
2625 	if (err) {
2626 		kvm_info("Error initializing system register tables");
2627 		return err;
2628 	}
2629 
2630 	in_hyp_mode = is_kernel_in_hyp_mode();
2631 
2632 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2633 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2634 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2635 			 "Only trusted guests should be used on this system.\n");
2636 
2637 	err = kvm_set_ipa_limit();
2638 	if (err)
2639 		return err;
2640 
2641 	err = kvm_arm_init_sve();
2642 	if (err)
2643 		return err;
2644 
2645 	err = kvm_arm_vmid_alloc_init();
2646 	if (err) {
2647 		kvm_err("Failed to initialize VMID allocator.\n");
2648 		return err;
2649 	}
2650 
2651 	if (!in_hyp_mode) {
2652 		err = init_hyp_mode();
2653 		if (err)
2654 			goto out_err;
2655 	}
2656 
2657 	err = kvm_init_vector_slots();
2658 	if (err) {
2659 		kvm_err("Cannot initialise vector slots\n");
2660 		goto out_hyp;
2661 	}
2662 
2663 	err = init_subsystems();
2664 	if (err)
2665 		goto out_hyp;
2666 
2667 	kvm_info("%s%sVHE mode initialized successfully\n",
2668 		 in_hyp_mode ? "" : (is_protected_kvm_enabled() ?
2669 				     "Protected " : "Hyp "),
2670 		 in_hyp_mode ? "" : (cpus_have_final_cap(ARM64_KVM_HVHE) ?
2671 				     "h" : "n"));
2672 
2673 	/*
2674 	 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2675 	 * hypervisor protection is finalized.
2676 	 */
2677 	err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2678 	if (err)
2679 		goto out_subs;
2680 
2681 	kvm_arm_initialised = true;
2682 
2683 	return 0;
2684 
2685 out_subs:
2686 	teardown_subsystems();
2687 out_hyp:
2688 	if (!in_hyp_mode)
2689 		teardown_hyp_mode();
2690 out_err:
2691 	kvm_arm_vmid_alloc_free();
2692 	return err;
2693 }
2694 
2695 static int __init early_kvm_mode_cfg(char *arg)
2696 {
2697 	if (!arg)
2698 		return -EINVAL;
2699 
2700 	if (strcmp(arg, "none") == 0) {
2701 		kvm_mode = KVM_MODE_NONE;
2702 		return 0;
2703 	}
2704 
2705 	if (!is_hyp_mode_available()) {
2706 		pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2707 		return 0;
2708 	}
2709 
2710 	if (strcmp(arg, "protected") == 0) {
2711 		if (!is_kernel_in_hyp_mode())
2712 			kvm_mode = KVM_MODE_PROTECTED;
2713 		else
2714 			pr_warn_once("Protected KVM not available with VHE\n");
2715 
2716 		return 0;
2717 	}
2718 
2719 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2720 		kvm_mode = KVM_MODE_DEFAULT;
2721 		return 0;
2722 	}
2723 
2724 	if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2725 		kvm_mode = KVM_MODE_NV;
2726 		return 0;
2727 	}
2728 
2729 	return -EINVAL;
2730 }
2731 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2732 
2733 enum kvm_mode kvm_get_mode(void)
2734 {
2735 	return kvm_mode;
2736 }
2737 
2738 module_init(kvm_arm_init);
2739