xref: /linux/arch/arm64/kvm/arm.c (revision 68550cbc6129159b7a6434796b721e8b66ee12f6)
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/kmemleak.h>
20 #include <linux/kvm.h>
21 #include <linux/kvm_irqfd.h>
22 #include <linux/irqbypass.h>
23 #include <linux/sched/stat.h>
24 #include <linux/psci.h>
25 #include <trace/events/kvm.h>
26 
27 #define CREATE_TRACE_POINTS
28 #include "trace_arm.h"
29 
30 #include <linux/uaccess.h>
31 #include <asm/ptrace.h>
32 #include <asm/mman.h>
33 #include <asm/tlbflush.h>
34 #include <asm/cacheflush.h>
35 #include <asm/cpufeature.h>
36 #include <asm/virt.h>
37 #include <asm/kvm_arm.h>
38 #include <asm/kvm_asm.h>
39 #include <asm/kvm_mmu.h>
40 #include <asm/kvm_emulate.h>
41 #include <asm/sections.h>
42 
43 #include <kvm/arm_hypercalls.h>
44 #include <kvm/arm_pmu.h>
45 #include <kvm/arm_psci.h>
46 
47 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
48 DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized);
49 
50 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
51 
52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
53 unsigned long kvm_arm_hyp_percpu_base[NR_CPUS];
54 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
55 
56 /* The VMID used in the VTTBR */
57 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
58 static u32 kvm_next_vmid;
59 static DEFINE_SPINLOCK(kvm_vmid_lock);
60 
61 static bool vgic_present;
62 
63 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
64 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
65 
66 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
67 {
68 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
69 }
70 
71 int kvm_arch_hardware_setup(void *opaque)
72 {
73 	return 0;
74 }
75 
76 int kvm_arch_check_processor_compat(void *opaque)
77 {
78 	return 0;
79 }
80 
81 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
82 			    struct kvm_enable_cap *cap)
83 {
84 	int r;
85 
86 	if (cap->flags)
87 		return -EINVAL;
88 
89 	switch (cap->cap) {
90 	case KVM_CAP_ARM_NISV_TO_USER:
91 		r = 0;
92 		kvm->arch.return_nisv_io_abort_to_user = true;
93 		break;
94 	case KVM_CAP_ARM_MTE:
95 		mutex_lock(&kvm->lock);
96 		if (!system_supports_mte() || kvm->created_vcpus) {
97 			r = -EINVAL;
98 		} else {
99 			r = 0;
100 			kvm->arch.mte_enabled = true;
101 		}
102 		mutex_unlock(&kvm->lock);
103 		break;
104 	default:
105 		r = -EINVAL;
106 		break;
107 	}
108 
109 	return r;
110 }
111 
112 static int kvm_arm_default_max_vcpus(void)
113 {
114 	return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
115 }
116 
117 static void set_default_spectre(struct kvm *kvm)
118 {
119 	/*
120 	 * The default is to expose CSV2 == 1 if the HW isn't affected.
121 	 * Although this is a per-CPU feature, we make it global because
122 	 * asymmetric systems are just a nuisance.
123 	 *
124 	 * Userspace can override this as long as it doesn't promise
125 	 * the impossible.
126 	 */
127 	if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
128 		kvm->arch.pfr0_csv2 = 1;
129 	if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
130 		kvm->arch.pfr0_csv3 = 1;
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 	ret = kvm_arm_setup_stage2(kvm, type);
142 	if (ret)
143 		return ret;
144 
145 	ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
146 	if (ret)
147 		return ret;
148 
149 	ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
150 	if (ret)
151 		goto out_free_stage2_pgd;
152 
153 	kvm_vgic_early_init(kvm);
154 
155 	/* The maximum number of VCPUs is limited by the host's GIC model */
156 	kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
157 
158 	set_default_spectre(kvm);
159 
160 	return ret;
161 out_free_stage2_pgd:
162 	kvm_free_stage2_pgd(&kvm->arch.mmu);
163 	return ret;
164 }
165 
166 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
167 {
168 	return VM_FAULT_SIGBUS;
169 }
170 
171 
172 /**
173  * kvm_arch_destroy_vm - destroy the VM data structure
174  * @kvm:	pointer to the KVM struct
175  */
176 void kvm_arch_destroy_vm(struct kvm *kvm)
177 {
178 	int i;
179 
180 	bitmap_free(kvm->arch.pmu_filter);
181 
182 	kvm_vgic_destroy(kvm);
183 
184 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
185 		if (kvm->vcpus[i]) {
186 			kvm_vcpu_destroy(kvm->vcpus[i]);
187 			kvm->vcpus[i] = NULL;
188 		}
189 	}
190 	atomic_set(&kvm->online_vcpus, 0);
191 }
192 
193 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
194 {
195 	int r;
196 	switch (ext) {
197 	case KVM_CAP_IRQCHIP:
198 		r = vgic_present;
199 		break;
200 	case KVM_CAP_IOEVENTFD:
201 	case KVM_CAP_DEVICE_CTRL:
202 	case KVM_CAP_USER_MEMORY:
203 	case KVM_CAP_SYNC_MMU:
204 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
205 	case KVM_CAP_ONE_REG:
206 	case KVM_CAP_ARM_PSCI:
207 	case KVM_CAP_ARM_PSCI_0_2:
208 	case KVM_CAP_READONLY_MEM:
209 	case KVM_CAP_MP_STATE:
210 	case KVM_CAP_IMMEDIATE_EXIT:
211 	case KVM_CAP_VCPU_EVENTS:
212 	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
213 	case KVM_CAP_ARM_NISV_TO_USER:
214 	case KVM_CAP_ARM_INJECT_EXT_DABT:
215 	case KVM_CAP_SET_GUEST_DEBUG:
216 	case KVM_CAP_VCPU_ATTRIBUTES:
217 	case KVM_CAP_PTP_KVM:
218 		r = 1;
219 		break;
220 	case KVM_CAP_SET_GUEST_DEBUG2:
221 		return KVM_GUESTDBG_VALID_MASK;
222 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
223 		r = 1;
224 		break;
225 	case KVM_CAP_NR_VCPUS:
226 		/*
227 		 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
228 		 * architectures, as it does not always bound it to
229 		 * KVM_CAP_MAX_VCPUS. It should not matter much because
230 		 * this is just an advisory value.
231 		 */
232 		r = min_t(unsigned int, num_online_cpus(),
233 			  kvm_arm_default_max_vcpus());
234 		break;
235 	case KVM_CAP_MAX_VCPUS:
236 	case KVM_CAP_MAX_VCPU_ID:
237 		if (kvm)
238 			r = kvm->arch.max_vcpus;
239 		else
240 			r = kvm_arm_default_max_vcpus();
241 		break;
242 	case KVM_CAP_MSI_DEVID:
243 		if (!kvm)
244 			r = -EINVAL;
245 		else
246 			r = kvm->arch.vgic.msis_require_devid;
247 		break;
248 	case KVM_CAP_ARM_USER_IRQ:
249 		/*
250 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
251 		 * (bump this number if adding more devices)
252 		 */
253 		r = 1;
254 		break;
255 	case KVM_CAP_ARM_MTE:
256 		r = system_supports_mte();
257 		break;
258 	case KVM_CAP_STEAL_TIME:
259 		r = kvm_arm_pvtime_supported();
260 		break;
261 	case KVM_CAP_ARM_EL1_32BIT:
262 		r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
263 		break;
264 	case KVM_CAP_GUEST_DEBUG_HW_BPS:
265 		r = get_num_brps();
266 		break;
267 	case KVM_CAP_GUEST_DEBUG_HW_WPS:
268 		r = get_num_wrps();
269 		break;
270 	case KVM_CAP_ARM_PMU_V3:
271 		r = kvm_arm_support_pmu_v3();
272 		break;
273 	case KVM_CAP_ARM_INJECT_SERROR_ESR:
274 		r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
275 		break;
276 	case KVM_CAP_ARM_VM_IPA_SIZE:
277 		r = get_kvm_ipa_limit();
278 		break;
279 	case KVM_CAP_ARM_SVE:
280 		r = system_supports_sve();
281 		break;
282 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
283 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
284 		r = system_has_full_ptr_auth();
285 		break;
286 	default:
287 		r = 0;
288 	}
289 
290 	return r;
291 }
292 
293 long kvm_arch_dev_ioctl(struct file *filp,
294 			unsigned int ioctl, unsigned long arg)
295 {
296 	return -EINVAL;
297 }
298 
299 struct kvm *kvm_arch_alloc_vm(void)
300 {
301 	size_t sz = sizeof(struct kvm);
302 
303 	if (!has_vhe())
304 		return kzalloc(sz, GFP_KERNEL_ACCOUNT);
305 
306 	return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
307 }
308 
309 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
310 {
311 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
312 		return -EBUSY;
313 
314 	if (id >= kvm->arch.max_vcpus)
315 		return -EINVAL;
316 
317 	return 0;
318 }
319 
320 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
321 {
322 	int err;
323 
324 	/* Force users to call KVM_ARM_VCPU_INIT */
325 	vcpu->arch.target = -1;
326 	bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
327 
328 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
329 
330 	/* Set up the timer */
331 	kvm_timer_vcpu_init(vcpu);
332 
333 	kvm_pmu_vcpu_init(vcpu);
334 
335 	kvm_arm_reset_debug_ptr(vcpu);
336 
337 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
338 
339 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
340 
341 	err = kvm_vgic_vcpu_init(vcpu);
342 	if (err)
343 		return err;
344 
345 	return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
346 }
347 
348 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
349 {
350 }
351 
352 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
353 {
354 	if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
355 		static_branch_dec(&userspace_irqchip_in_use);
356 
357 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
358 	kvm_timer_vcpu_terminate(vcpu);
359 	kvm_pmu_vcpu_destroy(vcpu);
360 
361 	kvm_arm_vcpu_destroy(vcpu);
362 }
363 
364 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
365 {
366 	return kvm_timer_is_pending(vcpu);
367 }
368 
369 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
370 {
371 	/*
372 	 * If we're about to block (most likely because we've just hit a
373 	 * WFI), we need to sync back the state of the GIC CPU interface
374 	 * so that we have the latest PMR and group enables. This ensures
375 	 * that kvm_arch_vcpu_runnable has up-to-date data to decide
376 	 * whether we have pending interrupts.
377 	 *
378 	 * For the same reason, we want to tell GICv4 that we need
379 	 * doorbells to be signalled, should an interrupt become pending.
380 	 */
381 	preempt_disable();
382 	kvm_vgic_vmcr_sync(vcpu);
383 	vgic_v4_put(vcpu, true);
384 	preempt_enable();
385 }
386 
387 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
388 {
389 	preempt_disable();
390 	vgic_v4_load(vcpu);
391 	preempt_enable();
392 }
393 
394 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
395 {
396 	struct kvm_s2_mmu *mmu;
397 	int *last_ran;
398 
399 	mmu = vcpu->arch.hw_mmu;
400 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
401 
402 	/*
403 	 * We guarantee that both TLBs and I-cache are private to each
404 	 * vcpu. If detecting that a vcpu from the same VM has
405 	 * previously run on the same physical CPU, call into the
406 	 * hypervisor code to nuke the relevant contexts.
407 	 *
408 	 * We might get preempted before the vCPU actually runs, but
409 	 * over-invalidation doesn't affect correctness.
410 	 */
411 	if (*last_ran != vcpu->vcpu_id) {
412 		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
413 		*last_ran = vcpu->vcpu_id;
414 	}
415 
416 	vcpu->cpu = cpu;
417 
418 	kvm_vgic_load(vcpu);
419 	kvm_timer_vcpu_load(vcpu);
420 	if (has_vhe())
421 		kvm_vcpu_load_sysregs_vhe(vcpu);
422 	kvm_arch_vcpu_load_fp(vcpu);
423 	kvm_vcpu_pmu_restore_guest(vcpu);
424 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
425 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
426 
427 	if (single_task_running())
428 		vcpu_clear_wfx_traps(vcpu);
429 	else
430 		vcpu_set_wfx_traps(vcpu);
431 
432 	if (vcpu_has_ptrauth(vcpu))
433 		vcpu_ptrauth_disable(vcpu);
434 	kvm_arch_vcpu_load_debug_state_flags(vcpu);
435 }
436 
437 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
438 {
439 	kvm_arch_vcpu_put_debug_state_flags(vcpu);
440 	kvm_arch_vcpu_put_fp(vcpu);
441 	if (has_vhe())
442 		kvm_vcpu_put_sysregs_vhe(vcpu);
443 	kvm_timer_vcpu_put(vcpu);
444 	kvm_vgic_put(vcpu);
445 	kvm_vcpu_pmu_restore_host(vcpu);
446 
447 	vcpu->cpu = -1;
448 }
449 
450 static void vcpu_power_off(struct kvm_vcpu *vcpu)
451 {
452 	vcpu->arch.power_off = true;
453 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
454 	kvm_vcpu_kick(vcpu);
455 }
456 
457 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
458 				    struct kvm_mp_state *mp_state)
459 {
460 	if (vcpu->arch.power_off)
461 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
462 	else
463 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
464 
465 	return 0;
466 }
467 
468 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
469 				    struct kvm_mp_state *mp_state)
470 {
471 	int ret = 0;
472 
473 	switch (mp_state->mp_state) {
474 	case KVM_MP_STATE_RUNNABLE:
475 		vcpu->arch.power_off = false;
476 		break;
477 	case KVM_MP_STATE_STOPPED:
478 		vcpu_power_off(vcpu);
479 		break;
480 	default:
481 		ret = -EINVAL;
482 	}
483 
484 	return ret;
485 }
486 
487 /**
488  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
489  * @v:		The VCPU pointer
490  *
491  * If the guest CPU is not waiting for interrupts or an interrupt line is
492  * asserted, the CPU is by definition runnable.
493  */
494 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
495 {
496 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
497 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
498 		&& !v->arch.power_off && !v->arch.pause);
499 }
500 
501 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
502 {
503 	return vcpu_mode_priv(vcpu);
504 }
505 
506 /* Just ensure a guest exit from a particular CPU */
507 static void exit_vm_noop(void *info)
508 {
509 }
510 
511 void force_vm_exit(const cpumask_t *mask)
512 {
513 	preempt_disable();
514 	smp_call_function_many(mask, exit_vm_noop, NULL, true);
515 	preempt_enable();
516 }
517 
518 /**
519  * need_new_vmid_gen - check that the VMID is still valid
520  * @vmid: The VMID to check
521  *
522  * return true if there is a new generation of VMIDs being used
523  *
524  * The hardware supports a limited set of values with the value zero reserved
525  * for the host, so we check if an assigned value belongs to a previous
526  * generation, which requires us to assign a new value. If we're the first to
527  * use a VMID for the new generation, we must flush necessary caches and TLBs
528  * on all CPUs.
529  */
530 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
531 {
532 	u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
533 	smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
534 	return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
535 }
536 
537 /**
538  * update_vmid - Update the vmid with a valid VMID for the current generation
539  * @vmid: The stage-2 VMID information struct
540  */
541 static void update_vmid(struct kvm_vmid *vmid)
542 {
543 	if (!need_new_vmid_gen(vmid))
544 		return;
545 
546 	spin_lock(&kvm_vmid_lock);
547 
548 	/*
549 	 * We need to re-check the vmid_gen here to ensure that if another vcpu
550 	 * already allocated a valid vmid for this vm, then this vcpu should
551 	 * use the same vmid.
552 	 */
553 	if (!need_new_vmid_gen(vmid)) {
554 		spin_unlock(&kvm_vmid_lock);
555 		return;
556 	}
557 
558 	/* First user of a new VMID generation? */
559 	if (unlikely(kvm_next_vmid == 0)) {
560 		atomic64_inc(&kvm_vmid_gen);
561 		kvm_next_vmid = 1;
562 
563 		/*
564 		 * On SMP we know no other CPUs can use this CPU's or each
565 		 * other's VMID after force_vm_exit returns since the
566 		 * kvm_vmid_lock blocks them from reentry to the guest.
567 		 */
568 		force_vm_exit(cpu_all_mask);
569 		/*
570 		 * Now broadcast TLB + ICACHE invalidation over the inner
571 		 * shareable domain to make sure all data structures are
572 		 * clean.
573 		 */
574 		kvm_call_hyp(__kvm_flush_vm_context);
575 	}
576 
577 	WRITE_ONCE(vmid->vmid, kvm_next_vmid);
578 	kvm_next_vmid++;
579 	kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
580 
581 	smp_wmb();
582 	WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
583 
584 	spin_unlock(&kvm_vmid_lock);
585 }
586 
587 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
588 {
589 	struct kvm *kvm = vcpu->kvm;
590 	int ret = 0;
591 
592 	if (likely(vcpu->arch.has_run_once))
593 		return 0;
594 
595 	if (!kvm_arm_vcpu_is_finalized(vcpu))
596 		return -EPERM;
597 
598 	vcpu->arch.has_run_once = true;
599 
600 	kvm_arm_vcpu_init_debug(vcpu);
601 
602 	if (likely(irqchip_in_kernel(kvm))) {
603 		/*
604 		 * Map the VGIC hardware resources before running a vcpu the
605 		 * first time on this VM.
606 		 */
607 		ret = kvm_vgic_map_resources(kvm);
608 		if (ret)
609 			return ret;
610 	} else {
611 		/*
612 		 * Tell the rest of the code that there are userspace irqchip
613 		 * VMs in the wild.
614 		 */
615 		static_branch_inc(&userspace_irqchip_in_use);
616 	}
617 
618 	ret = kvm_timer_enable(vcpu);
619 	if (ret)
620 		return ret;
621 
622 	ret = kvm_arm_pmu_v3_enable(vcpu);
623 
624 	/*
625 	 * Initialize traps for protected VMs.
626 	 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
627 	 * the code is in place for first run initialization at EL2.
628 	 */
629 	if (kvm_vm_is_protected(kvm))
630 		kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
631 
632 	return ret;
633 }
634 
635 bool kvm_arch_intc_initialized(struct kvm *kvm)
636 {
637 	return vgic_initialized(kvm);
638 }
639 
640 void kvm_arm_halt_guest(struct kvm *kvm)
641 {
642 	int i;
643 	struct kvm_vcpu *vcpu;
644 
645 	kvm_for_each_vcpu(i, vcpu, kvm)
646 		vcpu->arch.pause = true;
647 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
648 }
649 
650 void kvm_arm_resume_guest(struct kvm *kvm)
651 {
652 	int i;
653 	struct kvm_vcpu *vcpu;
654 
655 	kvm_for_each_vcpu(i, vcpu, kvm) {
656 		vcpu->arch.pause = false;
657 		rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
658 	}
659 }
660 
661 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
662 {
663 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
664 
665 	rcuwait_wait_event(wait,
666 			   (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
667 			   TASK_INTERRUPTIBLE);
668 
669 	if (vcpu->arch.power_off || vcpu->arch.pause) {
670 		/* Awaken to handle a signal, request we sleep again later. */
671 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
672 	}
673 
674 	/*
675 	 * Make sure we will observe a potential reset request if we've
676 	 * observed a change to the power state. Pairs with the smp_wmb() in
677 	 * kvm_psci_vcpu_on().
678 	 */
679 	smp_rmb();
680 }
681 
682 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
683 {
684 	return vcpu->arch.target >= 0;
685 }
686 
687 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
688 {
689 	if (kvm_request_pending(vcpu)) {
690 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
691 			vcpu_req_sleep(vcpu);
692 
693 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
694 			kvm_reset_vcpu(vcpu);
695 
696 		/*
697 		 * Clear IRQ_PENDING requests that were made to guarantee
698 		 * that a VCPU sees new virtual interrupts.
699 		 */
700 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
701 
702 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
703 			kvm_update_stolen_time(vcpu);
704 
705 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
706 			/* The distributor enable bits were changed */
707 			preempt_disable();
708 			vgic_v4_put(vcpu, false);
709 			vgic_v4_load(vcpu);
710 			preempt_enable();
711 		}
712 
713 		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
714 			kvm_pmu_handle_pmcr(vcpu,
715 					    __vcpu_sys_reg(vcpu, PMCR_EL0));
716 	}
717 }
718 
719 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
720 {
721 	if (likely(!vcpu_mode_is_32bit(vcpu)))
722 		return false;
723 
724 	return !system_supports_32bit_el0() ||
725 		static_branch_unlikely(&arm64_mismatched_32bit_el0);
726 }
727 
728 /**
729  * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
730  * @vcpu:	The VCPU pointer
731  * @ret:	Pointer to write optional return code
732  *
733  * Returns: true if the VCPU needs to return to a preemptible + interruptible
734  *	    and skip guest entry.
735  *
736  * This function disambiguates between two different types of exits: exits to a
737  * preemptible + interruptible kernel context and exits to userspace. For an
738  * exit to userspace, this function will write the return code to ret and return
739  * true. For an exit to preemptible + interruptible kernel context (i.e. check
740  * for pending work and re-enter), return true without writing to ret.
741  */
742 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
743 {
744 	struct kvm_run *run = vcpu->run;
745 
746 	/*
747 	 * If we're using a userspace irqchip, then check if we need
748 	 * to tell a userspace irqchip about timer or PMU level
749 	 * changes and if so, exit to userspace (the actual level
750 	 * state gets updated in kvm_timer_update_run and
751 	 * kvm_pmu_update_run below).
752 	 */
753 	if (static_branch_unlikely(&userspace_irqchip_in_use)) {
754 		if (kvm_timer_should_notify_user(vcpu) ||
755 		    kvm_pmu_should_notify_user(vcpu)) {
756 			*ret = -EINTR;
757 			run->exit_reason = KVM_EXIT_INTR;
758 			return true;
759 		}
760 	}
761 
762 	return kvm_request_pending(vcpu) ||
763 			need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
764 			xfer_to_guest_mode_work_pending();
765 }
766 
767 /**
768  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
769  * @vcpu:	The VCPU pointer
770  *
771  * This function is called through the VCPU_RUN ioctl called from user space. It
772  * will execute VM code in a loop until the time slice for the process is used
773  * or some emulation is needed from user space in which case the function will
774  * return with return value 0 and with the kvm_run structure filled in with the
775  * required data for the requested emulation.
776  */
777 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
778 {
779 	struct kvm_run *run = vcpu->run;
780 	int ret;
781 
782 	if (unlikely(!kvm_vcpu_initialized(vcpu)))
783 		return -ENOEXEC;
784 
785 	ret = kvm_vcpu_first_run_init(vcpu);
786 	if (ret)
787 		return ret;
788 
789 	if (run->exit_reason == KVM_EXIT_MMIO) {
790 		ret = kvm_handle_mmio_return(vcpu);
791 		if (ret)
792 			return ret;
793 	}
794 
795 	vcpu_load(vcpu);
796 
797 	if (run->immediate_exit) {
798 		ret = -EINTR;
799 		goto out;
800 	}
801 
802 	kvm_sigset_activate(vcpu);
803 
804 	ret = 1;
805 	run->exit_reason = KVM_EXIT_UNKNOWN;
806 	while (ret > 0) {
807 		/*
808 		 * Check conditions before entering the guest
809 		 */
810 		ret = xfer_to_guest_mode_handle_work(vcpu);
811 		if (!ret)
812 			ret = 1;
813 
814 		update_vmid(&vcpu->arch.hw_mmu->vmid);
815 
816 		check_vcpu_requests(vcpu);
817 
818 		/*
819 		 * Preparing the interrupts to be injected also
820 		 * involves poking the GIC, which must be done in a
821 		 * non-preemptible context.
822 		 */
823 		preempt_disable();
824 
825 		kvm_pmu_flush_hwstate(vcpu);
826 
827 		local_irq_disable();
828 
829 		kvm_vgic_flush_hwstate(vcpu);
830 
831 		/*
832 		 * Ensure we set mode to IN_GUEST_MODE after we disable
833 		 * interrupts and before the final VCPU requests check.
834 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
835 		 * Documentation/virt/kvm/vcpu-requests.rst
836 		 */
837 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
838 
839 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
840 			vcpu->mode = OUTSIDE_GUEST_MODE;
841 			isb(); /* Ensure work in x_flush_hwstate is committed */
842 			kvm_pmu_sync_hwstate(vcpu);
843 			if (static_branch_unlikely(&userspace_irqchip_in_use))
844 				kvm_timer_sync_user(vcpu);
845 			kvm_vgic_sync_hwstate(vcpu);
846 			local_irq_enable();
847 			preempt_enable();
848 			continue;
849 		}
850 
851 		kvm_arm_setup_debug(vcpu);
852 
853 		/**************************************************************
854 		 * Enter the guest
855 		 */
856 		trace_kvm_entry(*vcpu_pc(vcpu));
857 		guest_enter_irqoff();
858 
859 		ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
860 
861 		vcpu->mode = OUTSIDE_GUEST_MODE;
862 		vcpu->stat.exits++;
863 		/*
864 		 * Back from guest
865 		 *************************************************************/
866 
867 		kvm_arm_clear_debug(vcpu);
868 
869 		/*
870 		 * We must sync the PMU state before the vgic state so
871 		 * that the vgic can properly sample the updated state of the
872 		 * interrupt line.
873 		 */
874 		kvm_pmu_sync_hwstate(vcpu);
875 
876 		/*
877 		 * Sync the vgic state before syncing the timer state because
878 		 * the timer code needs to know if the virtual timer
879 		 * interrupts are active.
880 		 */
881 		kvm_vgic_sync_hwstate(vcpu);
882 
883 		/*
884 		 * Sync the timer hardware state before enabling interrupts as
885 		 * we don't want vtimer interrupts to race with syncing the
886 		 * timer virtual interrupt state.
887 		 */
888 		if (static_branch_unlikely(&userspace_irqchip_in_use))
889 			kvm_timer_sync_user(vcpu);
890 
891 		kvm_arch_vcpu_ctxsync_fp(vcpu);
892 
893 		/*
894 		 * We may have taken a host interrupt in HYP mode (ie
895 		 * while executing the guest). This interrupt is still
896 		 * pending, as we haven't serviced it yet!
897 		 *
898 		 * We're now back in SVC mode, with interrupts
899 		 * disabled.  Enabling the interrupts now will have
900 		 * the effect of taking the interrupt again, in SVC
901 		 * mode this time.
902 		 */
903 		local_irq_enable();
904 
905 		/*
906 		 * We do local_irq_enable() before calling guest_exit() so
907 		 * that if a timer interrupt hits while running the guest we
908 		 * account that tick as being spent in the guest.  We enable
909 		 * preemption after calling guest_exit() so that if we get
910 		 * preempted we make sure ticks after that is not counted as
911 		 * guest time.
912 		 */
913 		guest_exit();
914 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
915 
916 		/* Exit types that need handling before we can be preempted */
917 		handle_exit_early(vcpu, ret);
918 
919 		preempt_enable();
920 
921 		/*
922 		 * The ARMv8 architecture doesn't give the hypervisor
923 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
924 		 * if implemented by the CPU. If we spot the guest in such
925 		 * state and that we decided it wasn't supposed to do so (like
926 		 * with the asymmetric AArch32 case), return to userspace with
927 		 * a fatal error.
928 		 */
929 		if (vcpu_mode_is_bad_32bit(vcpu)) {
930 			/*
931 			 * As we have caught the guest red-handed, decide that
932 			 * it isn't fit for purpose anymore by making the vcpu
933 			 * invalid. The VMM can try and fix it by issuing  a
934 			 * KVM_ARM_VCPU_INIT if it really wants to.
935 			 */
936 			vcpu->arch.target = -1;
937 			ret = ARM_EXCEPTION_IL;
938 		}
939 
940 		ret = handle_exit(vcpu, ret);
941 	}
942 
943 	/* Tell userspace about in-kernel device output levels */
944 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
945 		kvm_timer_update_run(vcpu);
946 		kvm_pmu_update_run(vcpu);
947 	}
948 
949 	kvm_sigset_deactivate(vcpu);
950 
951 out:
952 	/*
953 	 * In the unlikely event that we are returning to userspace
954 	 * with pending exceptions or PC adjustment, commit these
955 	 * adjustments in order to give userspace a consistent view of
956 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
957 	 * being preempt-safe on VHE.
958 	 */
959 	if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
960 					 KVM_ARM64_INCREMENT_PC)))
961 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
962 
963 	vcpu_put(vcpu);
964 	return ret;
965 }
966 
967 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
968 {
969 	int bit_index;
970 	bool set;
971 	unsigned long *hcr;
972 
973 	if (number == KVM_ARM_IRQ_CPU_IRQ)
974 		bit_index = __ffs(HCR_VI);
975 	else /* KVM_ARM_IRQ_CPU_FIQ */
976 		bit_index = __ffs(HCR_VF);
977 
978 	hcr = vcpu_hcr(vcpu);
979 	if (level)
980 		set = test_and_set_bit(bit_index, hcr);
981 	else
982 		set = test_and_clear_bit(bit_index, hcr);
983 
984 	/*
985 	 * If we didn't change anything, no need to wake up or kick other CPUs
986 	 */
987 	if (set == level)
988 		return 0;
989 
990 	/*
991 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
992 	 * trigger a world-switch round on the running physical CPU to set the
993 	 * virtual IRQ/FIQ fields in the HCR appropriately.
994 	 */
995 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
996 	kvm_vcpu_kick(vcpu);
997 
998 	return 0;
999 }
1000 
1001 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1002 			  bool line_status)
1003 {
1004 	u32 irq = irq_level->irq;
1005 	unsigned int irq_type, vcpu_idx, irq_num;
1006 	int nrcpus = atomic_read(&kvm->online_vcpus);
1007 	struct kvm_vcpu *vcpu = NULL;
1008 	bool level = irq_level->level;
1009 
1010 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1011 	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1012 	vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1013 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1014 
1015 	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1016 
1017 	switch (irq_type) {
1018 	case KVM_ARM_IRQ_TYPE_CPU:
1019 		if (irqchip_in_kernel(kvm))
1020 			return -ENXIO;
1021 
1022 		if (vcpu_idx >= nrcpus)
1023 			return -EINVAL;
1024 
1025 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1026 		if (!vcpu)
1027 			return -EINVAL;
1028 
1029 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1030 			return -EINVAL;
1031 
1032 		return vcpu_interrupt_line(vcpu, irq_num, level);
1033 	case KVM_ARM_IRQ_TYPE_PPI:
1034 		if (!irqchip_in_kernel(kvm))
1035 			return -ENXIO;
1036 
1037 		if (vcpu_idx >= nrcpus)
1038 			return -EINVAL;
1039 
1040 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1041 		if (!vcpu)
1042 			return -EINVAL;
1043 
1044 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1045 			return -EINVAL;
1046 
1047 		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1048 	case KVM_ARM_IRQ_TYPE_SPI:
1049 		if (!irqchip_in_kernel(kvm))
1050 			return -ENXIO;
1051 
1052 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1053 			return -EINVAL;
1054 
1055 		return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1056 	}
1057 
1058 	return -EINVAL;
1059 }
1060 
1061 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1062 			       const struct kvm_vcpu_init *init)
1063 {
1064 	unsigned int i, ret;
1065 	u32 phys_target = kvm_target_cpu();
1066 
1067 	if (init->target != phys_target)
1068 		return -EINVAL;
1069 
1070 	/*
1071 	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1072 	 * use the same target.
1073 	 */
1074 	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1075 		return -EINVAL;
1076 
1077 	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1078 	for (i = 0; i < sizeof(init->features) * 8; i++) {
1079 		bool set = (init->features[i / 32] & (1 << (i % 32)));
1080 
1081 		if (set && i >= KVM_VCPU_MAX_FEATURES)
1082 			return -ENOENT;
1083 
1084 		/*
1085 		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1086 		 * use the same feature set.
1087 		 */
1088 		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1089 		    test_bit(i, vcpu->arch.features) != set)
1090 			return -EINVAL;
1091 
1092 		if (set)
1093 			set_bit(i, vcpu->arch.features);
1094 	}
1095 
1096 	vcpu->arch.target = phys_target;
1097 
1098 	/* Now we know what it is, we can reset it. */
1099 	ret = kvm_reset_vcpu(vcpu);
1100 	if (ret) {
1101 		vcpu->arch.target = -1;
1102 		bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1103 	}
1104 
1105 	return ret;
1106 }
1107 
1108 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1109 					 struct kvm_vcpu_init *init)
1110 {
1111 	int ret;
1112 
1113 	ret = kvm_vcpu_set_target(vcpu, init);
1114 	if (ret)
1115 		return ret;
1116 
1117 	/*
1118 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1119 	 * guest MMU is turned off and flush the caches as needed.
1120 	 *
1121 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1122 	 * ensuring that the data side is always coherent. We still
1123 	 * need to invalidate the I-cache though, as FWB does *not*
1124 	 * imply CTR_EL0.DIC.
1125 	 */
1126 	if (vcpu->arch.has_run_once) {
1127 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1128 			stage2_unmap_vm(vcpu->kvm);
1129 		else
1130 			icache_inval_all_pou();
1131 	}
1132 
1133 	vcpu_reset_hcr(vcpu);
1134 	vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1135 
1136 	/*
1137 	 * Handle the "start in power-off" case.
1138 	 */
1139 	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1140 		vcpu_power_off(vcpu);
1141 	else
1142 		vcpu->arch.power_off = false;
1143 
1144 	return 0;
1145 }
1146 
1147 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1148 				 struct kvm_device_attr *attr)
1149 {
1150 	int ret = -ENXIO;
1151 
1152 	switch (attr->group) {
1153 	default:
1154 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1155 		break;
1156 	}
1157 
1158 	return ret;
1159 }
1160 
1161 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1162 				 struct kvm_device_attr *attr)
1163 {
1164 	int ret = -ENXIO;
1165 
1166 	switch (attr->group) {
1167 	default:
1168 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1169 		break;
1170 	}
1171 
1172 	return ret;
1173 }
1174 
1175 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1176 				 struct kvm_device_attr *attr)
1177 {
1178 	int ret = -ENXIO;
1179 
1180 	switch (attr->group) {
1181 	default:
1182 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1183 		break;
1184 	}
1185 
1186 	return ret;
1187 }
1188 
1189 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1190 				   struct kvm_vcpu_events *events)
1191 {
1192 	memset(events, 0, sizeof(*events));
1193 
1194 	return __kvm_arm_vcpu_get_events(vcpu, events);
1195 }
1196 
1197 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1198 				   struct kvm_vcpu_events *events)
1199 {
1200 	int i;
1201 
1202 	/* check whether the reserved field is zero */
1203 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1204 		if (events->reserved[i])
1205 			return -EINVAL;
1206 
1207 	/* check whether the pad field is zero */
1208 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1209 		if (events->exception.pad[i])
1210 			return -EINVAL;
1211 
1212 	return __kvm_arm_vcpu_set_events(vcpu, events);
1213 }
1214 
1215 long kvm_arch_vcpu_ioctl(struct file *filp,
1216 			 unsigned int ioctl, unsigned long arg)
1217 {
1218 	struct kvm_vcpu *vcpu = filp->private_data;
1219 	void __user *argp = (void __user *)arg;
1220 	struct kvm_device_attr attr;
1221 	long r;
1222 
1223 	switch (ioctl) {
1224 	case KVM_ARM_VCPU_INIT: {
1225 		struct kvm_vcpu_init init;
1226 
1227 		r = -EFAULT;
1228 		if (copy_from_user(&init, argp, sizeof(init)))
1229 			break;
1230 
1231 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1232 		break;
1233 	}
1234 	case KVM_SET_ONE_REG:
1235 	case KVM_GET_ONE_REG: {
1236 		struct kvm_one_reg reg;
1237 
1238 		r = -ENOEXEC;
1239 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1240 			break;
1241 
1242 		r = -EFAULT;
1243 		if (copy_from_user(&reg, argp, sizeof(reg)))
1244 			break;
1245 
1246 		/*
1247 		 * We could owe a reset due to PSCI. Handle the pending reset
1248 		 * here to ensure userspace register accesses are ordered after
1249 		 * the reset.
1250 		 */
1251 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1252 			kvm_reset_vcpu(vcpu);
1253 
1254 		if (ioctl == KVM_SET_ONE_REG)
1255 			r = kvm_arm_set_reg(vcpu, &reg);
1256 		else
1257 			r = kvm_arm_get_reg(vcpu, &reg);
1258 		break;
1259 	}
1260 	case KVM_GET_REG_LIST: {
1261 		struct kvm_reg_list __user *user_list = argp;
1262 		struct kvm_reg_list reg_list;
1263 		unsigned n;
1264 
1265 		r = -ENOEXEC;
1266 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1267 			break;
1268 
1269 		r = -EPERM;
1270 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1271 			break;
1272 
1273 		r = -EFAULT;
1274 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1275 			break;
1276 		n = reg_list.n;
1277 		reg_list.n = kvm_arm_num_regs(vcpu);
1278 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1279 			break;
1280 		r = -E2BIG;
1281 		if (n < reg_list.n)
1282 			break;
1283 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1284 		break;
1285 	}
1286 	case KVM_SET_DEVICE_ATTR: {
1287 		r = -EFAULT;
1288 		if (copy_from_user(&attr, argp, sizeof(attr)))
1289 			break;
1290 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1291 		break;
1292 	}
1293 	case KVM_GET_DEVICE_ATTR: {
1294 		r = -EFAULT;
1295 		if (copy_from_user(&attr, argp, sizeof(attr)))
1296 			break;
1297 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1298 		break;
1299 	}
1300 	case KVM_HAS_DEVICE_ATTR: {
1301 		r = -EFAULT;
1302 		if (copy_from_user(&attr, argp, sizeof(attr)))
1303 			break;
1304 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1305 		break;
1306 	}
1307 	case KVM_GET_VCPU_EVENTS: {
1308 		struct kvm_vcpu_events events;
1309 
1310 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1311 			return -EINVAL;
1312 
1313 		if (copy_to_user(argp, &events, sizeof(events)))
1314 			return -EFAULT;
1315 
1316 		return 0;
1317 	}
1318 	case KVM_SET_VCPU_EVENTS: {
1319 		struct kvm_vcpu_events events;
1320 
1321 		if (copy_from_user(&events, argp, sizeof(events)))
1322 			return -EFAULT;
1323 
1324 		return kvm_arm_vcpu_set_events(vcpu, &events);
1325 	}
1326 	case KVM_ARM_VCPU_FINALIZE: {
1327 		int what;
1328 
1329 		if (!kvm_vcpu_initialized(vcpu))
1330 			return -ENOEXEC;
1331 
1332 		if (get_user(what, (const int __user *)argp))
1333 			return -EFAULT;
1334 
1335 		return kvm_arm_vcpu_finalize(vcpu, what);
1336 	}
1337 	default:
1338 		r = -EINVAL;
1339 	}
1340 
1341 	return r;
1342 }
1343 
1344 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1345 {
1346 
1347 }
1348 
1349 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1350 					const struct kvm_memory_slot *memslot)
1351 {
1352 	kvm_flush_remote_tlbs(kvm);
1353 }
1354 
1355 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1356 					struct kvm_arm_device_addr *dev_addr)
1357 {
1358 	unsigned long dev_id, type;
1359 
1360 	dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1361 		KVM_ARM_DEVICE_ID_SHIFT;
1362 	type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1363 		KVM_ARM_DEVICE_TYPE_SHIFT;
1364 
1365 	switch (dev_id) {
1366 	case KVM_ARM_DEVICE_VGIC_V2:
1367 		if (!vgic_present)
1368 			return -ENXIO;
1369 		return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1370 	default:
1371 		return -ENODEV;
1372 	}
1373 }
1374 
1375 long kvm_arch_vm_ioctl(struct file *filp,
1376 		       unsigned int ioctl, unsigned long arg)
1377 {
1378 	struct kvm *kvm = filp->private_data;
1379 	void __user *argp = (void __user *)arg;
1380 
1381 	switch (ioctl) {
1382 	case KVM_CREATE_IRQCHIP: {
1383 		int ret;
1384 		if (!vgic_present)
1385 			return -ENXIO;
1386 		mutex_lock(&kvm->lock);
1387 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1388 		mutex_unlock(&kvm->lock);
1389 		return ret;
1390 	}
1391 	case KVM_ARM_SET_DEVICE_ADDR: {
1392 		struct kvm_arm_device_addr dev_addr;
1393 
1394 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1395 			return -EFAULT;
1396 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1397 	}
1398 	case KVM_ARM_PREFERRED_TARGET: {
1399 		struct kvm_vcpu_init init;
1400 
1401 		kvm_vcpu_preferred_target(&init);
1402 
1403 		if (copy_to_user(argp, &init, sizeof(init)))
1404 			return -EFAULT;
1405 
1406 		return 0;
1407 	}
1408 	case KVM_ARM_MTE_COPY_TAGS: {
1409 		struct kvm_arm_copy_mte_tags copy_tags;
1410 
1411 		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1412 			return -EFAULT;
1413 		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1414 	}
1415 	default:
1416 		return -EINVAL;
1417 	}
1418 }
1419 
1420 static unsigned long nvhe_percpu_size(void)
1421 {
1422 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1423 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1424 }
1425 
1426 static unsigned long nvhe_percpu_order(void)
1427 {
1428 	unsigned long size = nvhe_percpu_size();
1429 
1430 	return size ? get_order(size) : 0;
1431 }
1432 
1433 /* A lookup table holding the hypervisor VA for each vector slot */
1434 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1435 
1436 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1437 {
1438 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1439 }
1440 
1441 static int kvm_init_vector_slots(void)
1442 {
1443 	int err;
1444 	void *base;
1445 
1446 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1447 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1448 
1449 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1450 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1451 
1452 	if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1453 		return 0;
1454 
1455 	if (!has_vhe()) {
1456 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1457 					       __BP_HARDEN_HYP_VECS_SZ, &base);
1458 		if (err)
1459 			return err;
1460 	}
1461 
1462 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1463 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1464 	return 0;
1465 }
1466 
1467 static void cpu_prepare_hyp_mode(int cpu)
1468 {
1469 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1470 	unsigned long tcr;
1471 
1472 	/*
1473 	 * Calculate the raw per-cpu offset without a translation from the
1474 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1475 	 * so that we can use adr_l to access per-cpu variables in EL2.
1476 	 * Also drop the KASAN tag which gets in the way...
1477 	 */
1478 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1479 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1480 
1481 	params->mair_el2 = read_sysreg(mair_el1);
1482 
1483 	/*
1484 	 * The ID map may be configured to use an extended virtual address
1485 	 * range. This is only the case if system RAM is out of range for the
1486 	 * currently configured page size and VA_BITS, in which case we will
1487 	 * also need the extended virtual range for the HYP ID map, or we won't
1488 	 * be able to enable the EL2 MMU.
1489 	 *
1490 	 * However, at EL2, there is only one TTBR register, and we can't switch
1491 	 * between translation tables *and* update TCR_EL2.T0SZ at the same
1492 	 * time. Bottom line: we need to use the extended range with *both* our
1493 	 * translation tables.
1494 	 *
1495 	 * So use the same T0SZ value we use for the ID map.
1496 	 */
1497 	tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1498 	tcr &= ~TCR_T0SZ_MASK;
1499 	tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1500 	params->tcr_el2 = tcr;
1501 
1502 	params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1503 	params->pgd_pa = kvm_mmu_get_httbr();
1504 	if (is_protected_kvm_enabled())
1505 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1506 	else
1507 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1508 	params->vttbr = params->vtcr = 0;
1509 
1510 	/*
1511 	 * Flush the init params from the data cache because the struct will
1512 	 * be read while the MMU is off.
1513 	 */
1514 	kvm_flush_dcache_to_poc(params, sizeof(*params));
1515 }
1516 
1517 static void hyp_install_host_vector(void)
1518 {
1519 	struct kvm_nvhe_init_params *params;
1520 	struct arm_smccc_res res;
1521 
1522 	/* Switch from the HYP stub to our own HYP init vector */
1523 	__hyp_set_vectors(kvm_get_idmap_vector());
1524 
1525 	/*
1526 	 * Call initialization code, and switch to the full blown HYP code.
1527 	 * If the cpucaps haven't been finalized yet, something has gone very
1528 	 * wrong, and hyp will crash and burn when it uses any
1529 	 * cpus_have_const_cap() wrapper.
1530 	 */
1531 	BUG_ON(!system_capabilities_finalized());
1532 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1533 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1534 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1535 }
1536 
1537 static void cpu_init_hyp_mode(void)
1538 {
1539 	hyp_install_host_vector();
1540 
1541 	/*
1542 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1543 	 * at EL2.
1544 	 */
1545 	if (this_cpu_has_cap(ARM64_SSBS) &&
1546 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1547 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1548 	}
1549 }
1550 
1551 static void cpu_hyp_reset(void)
1552 {
1553 	if (!is_kernel_in_hyp_mode())
1554 		__hyp_reset_vectors();
1555 }
1556 
1557 /*
1558  * EL2 vectors can be mapped and rerouted in a number of ways,
1559  * depending on the kernel configuration and CPU present:
1560  *
1561  * - If the CPU is affected by Spectre-v2, the hardening sequence is
1562  *   placed in one of the vector slots, which is executed before jumping
1563  *   to the real vectors.
1564  *
1565  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1566  *   containing the hardening sequence is mapped next to the idmap page,
1567  *   and executed before jumping to the real vectors.
1568  *
1569  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1570  *   empty slot is selected, mapped next to the idmap page, and
1571  *   executed before jumping to the real vectors.
1572  *
1573  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1574  * VHE, as we don't have hypervisor-specific mappings. If the system
1575  * is VHE and yet selects this capability, it will be ignored.
1576  */
1577 static void cpu_set_hyp_vector(void)
1578 {
1579 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1580 	void *vector = hyp_spectre_vector_selector[data->slot];
1581 
1582 	if (!is_protected_kvm_enabled())
1583 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1584 	else
1585 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1586 }
1587 
1588 static void cpu_hyp_init_context(void)
1589 {
1590 	kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1591 
1592 	if (!is_kernel_in_hyp_mode())
1593 		cpu_init_hyp_mode();
1594 }
1595 
1596 static void cpu_hyp_init_features(void)
1597 {
1598 	cpu_set_hyp_vector();
1599 	kvm_arm_init_debug();
1600 
1601 	if (is_kernel_in_hyp_mode())
1602 		kvm_timer_init_vhe();
1603 
1604 	if (vgic_present)
1605 		kvm_vgic_init_cpu_hardware();
1606 }
1607 
1608 static void cpu_hyp_reinit(void)
1609 {
1610 	cpu_hyp_reset();
1611 	cpu_hyp_init_context();
1612 	cpu_hyp_init_features();
1613 }
1614 
1615 static void _kvm_arch_hardware_enable(void *discard)
1616 {
1617 	if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1618 		cpu_hyp_reinit();
1619 		__this_cpu_write(kvm_arm_hardware_enabled, 1);
1620 	}
1621 }
1622 
1623 int kvm_arch_hardware_enable(void)
1624 {
1625 	_kvm_arch_hardware_enable(NULL);
1626 	return 0;
1627 }
1628 
1629 static void _kvm_arch_hardware_disable(void *discard)
1630 {
1631 	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1632 		cpu_hyp_reset();
1633 		__this_cpu_write(kvm_arm_hardware_enabled, 0);
1634 	}
1635 }
1636 
1637 void kvm_arch_hardware_disable(void)
1638 {
1639 	if (!is_protected_kvm_enabled())
1640 		_kvm_arch_hardware_disable(NULL);
1641 }
1642 
1643 #ifdef CONFIG_CPU_PM
1644 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1645 				    unsigned long cmd,
1646 				    void *v)
1647 {
1648 	/*
1649 	 * kvm_arm_hardware_enabled is left with its old value over
1650 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1651 	 * re-enable hyp.
1652 	 */
1653 	switch (cmd) {
1654 	case CPU_PM_ENTER:
1655 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1656 			/*
1657 			 * don't update kvm_arm_hardware_enabled here
1658 			 * so that the hardware will be re-enabled
1659 			 * when we resume. See below.
1660 			 */
1661 			cpu_hyp_reset();
1662 
1663 		return NOTIFY_OK;
1664 	case CPU_PM_ENTER_FAILED:
1665 	case CPU_PM_EXIT:
1666 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1667 			/* The hardware was enabled before suspend. */
1668 			cpu_hyp_reinit();
1669 
1670 		return NOTIFY_OK;
1671 
1672 	default:
1673 		return NOTIFY_DONE;
1674 	}
1675 }
1676 
1677 static struct notifier_block hyp_init_cpu_pm_nb = {
1678 	.notifier_call = hyp_init_cpu_pm_notifier,
1679 };
1680 
1681 static void hyp_cpu_pm_init(void)
1682 {
1683 	if (!is_protected_kvm_enabled())
1684 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1685 }
1686 static void hyp_cpu_pm_exit(void)
1687 {
1688 	if (!is_protected_kvm_enabled())
1689 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1690 }
1691 #else
1692 static inline void hyp_cpu_pm_init(void)
1693 {
1694 }
1695 static inline void hyp_cpu_pm_exit(void)
1696 {
1697 }
1698 #endif
1699 
1700 static void init_cpu_logical_map(void)
1701 {
1702 	unsigned int cpu;
1703 
1704 	/*
1705 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1706 	 * Only copy the set of online CPUs whose features have been chacked
1707 	 * against the finalized system capabilities. The hypervisor will not
1708 	 * allow any other CPUs from the `possible` set to boot.
1709 	 */
1710 	for_each_online_cpu(cpu)
1711 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1712 }
1713 
1714 #define init_psci_0_1_impl_state(config, what)	\
1715 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
1716 
1717 static bool init_psci_relay(void)
1718 {
1719 	/*
1720 	 * If PSCI has not been initialized, protected KVM cannot install
1721 	 * itself on newly booted CPUs.
1722 	 */
1723 	if (!psci_ops.get_version) {
1724 		kvm_err("Cannot initialize protected mode without PSCI\n");
1725 		return false;
1726 	}
1727 
1728 	kvm_host_psci_config.version = psci_ops.get_version();
1729 
1730 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1731 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1732 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1733 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1734 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1735 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1736 	}
1737 	return true;
1738 }
1739 
1740 static int init_subsystems(void)
1741 {
1742 	int err = 0;
1743 
1744 	/*
1745 	 * Enable hardware so that subsystem initialisation can access EL2.
1746 	 */
1747 	on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1748 
1749 	/*
1750 	 * Register CPU lower-power notifier
1751 	 */
1752 	hyp_cpu_pm_init();
1753 
1754 	/*
1755 	 * Init HYP view of VGIC
1756 	 */
1757 	err = kvm_vgic_hyp_init();
1758 	switch (err) {
1759 	case 0:
1760 		vgic_present = true;
1761 		break;
1762 	case -ENODEV:
1763 	case -ENXIO:
1764 		vgic_present = false;
1765 		err = 0;
1766 		break;
1767 	default:
1768 		goto out;
1769 	}
1770 
1771 	/*
1772 	 * Init HYP architected timer support
1773 	 */
1774 	err = kvm_timer_hyp_init(vgic_present);
1775 	if (err)
1776 		goto out;
1777 
1778 	kvm_perf_init();
1779 	kvm_sys_reg_table_init();
1780 
1781 out:
1782 	if (err || !is_protected_kvm_enabled())
1783 		on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1784 
1785 	return err;
1786 }
1787 
1788 static void teardown_hyp_mode(void)
1789 {
1790 	int cpu;
1791 
1792 	free_hyp_pgds();
1793 	for_each_possible_cpu(cpu) {
1794 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1795 		free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1796 	}
1797 }
1798 
1799 static int do_pkvm_init(u32 hyp_va_bits)
1800 {
1801 	void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1802 	int ret;
1803 
1804 	preempt_disable();
1805 	cpu_hyp_init_context();
1806 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1807 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
1808 				hyp_va_bits);
1809 	cpu_hyp_init_features();
1810 
1811 	/*
1812 	 * The stub hypercalls are now disabled, so set our local flag to
1813 	 * prevent a later re-init attempt in kvm_arch_hardware_enable().
1814 	 */
1815 	__this_cpu_write(kvm_arm_hardware_enabled, 1);
1816 	preempt_enable();
1817 
1818 	return ret;
1819 }
1820 
1821 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1822 {
1823 	void *addr = phys_to_virt(hyp_mem_base);
1824 	int ret;
1825 
1826 	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1827 	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1828 	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
1829 	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
1830 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1831 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1832 	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
1833 
1834 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1835 	if (ret)
1836 		return ret;
1837 
1838 	ret = do_pkvm_init(hyp_va_bits);
1839 	if (ret)
1840 		return ret;
1841 
1842 	free_hyp_pgds();
1843 
1844 	return 0;
1845 }
1846 
1847 /**
1848  * Inits Hyp-mode on all online CPUs
1849  */
1850 static int init_hyp_mode(void)
1851 {
1852 	u32 hyp_va_bits;
1853 	int cpu;
1854 	int err = -ENOMEM;
1855 
1856 	/*
1857 	 * The protected Hyp-mode cannot be initialized if the memory pool
1858 	 * allocation has failed.
1859 	 */
1860 	if (is_protected_kvm_enabled() && !hyp_mem_base)
1861 		goto out_err;
1862 
1863 	/*
1864 	 * Allocate Hyp PGD and setup Hyp identity mapping
1865 	 */
1866 	err = kvm_mmu_init(&hyp_va_bits);
1867 	if (err)
1868 		goto out_err;
1869 
1870 	/*
1871 	 * Allocate stack pages for Hypervisor-mode
1872 	 */
1873 	for_each_possible_cpu(cpu) {
1874 		unsigned long stack_page;
1875 
1876 		stack_page = __get_free_page(GFP_KERNEL);
1877 		if (!stack_page) {
1878 			err = -ENOMEM;
1879 			goto out_err;
1880 		}
1881 
1882 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1883 	}
1884 
1885 	/*
1886 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1887 	 */
1888 	for_each_possible_cpu(cpu) {
1889 		struct page *page;
1890 		void *page_addr;
1891 
1892 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1893 		if (!page) {
1894 			err = -ENOMEM;
1895 			goto out_err;
1896 		}
1897 
1898 		page_addr = page_address(page);
1899 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1900 		kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1901 	}
1902 
1903 	/*
1904 	 * Map the Hyp-code called directly from the host
1905 	 */
1906 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1907 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1908 	if (err) {
1909 		kvm_err("Cannot map world-switch code\n");
1910 		goto out_err;
1911 	}
1912 
1913 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1914 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1915 	if (err) {
1916 		kvm_err("Cannot map .hyp.rodata section\n");
1917 		goto out_err;
1918 	}
1919 
1920 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1921 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1922 	if (err) {
1923 		kvm_err("Cannot map rodata section\n");
1924 		goto out_err;
1925 	}
1926 
1927 	/*
1928 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1929 	 * section thanks to an assertion in the linker script. Map it RW and
1930 	 * the rest of .bss RO.
1931 	 */
1932 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1933 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1934 	if (err) {
1935 		kvm_err("Cannot map hyp bss section: %d\n", err);
1936 		goto out_err;
1937 	}
1938 
1939 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1940 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1941 	if (err) {
1942 		kvm_err("Cannot map bss section\n");
1943 		goto out_err;
1944 	}
1945 
1946 	/*
1947 	 * Map the Hyp stack pages
1948 	 */
1949 	for_each_possible_cpu(cpu) {
1950 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1951 		err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1952 					  PAGE_HYP);
1953 
1954 		if (err) {
1955 			kvm_err("Cannot map hyp stack\n");
1956 			goto out_err;
1957 		}
1958 	}
1959 
1960 	for_each_possible_cpu(cpu) {
1961 		char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1962 		char *percpu_end = percpu_begin + nvhe_percpu_size();
1963 
1964 		/* Map Hyp percpu pages */
1965 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1966 		if (err) {
1967 			kvm_err("Cannot map hyp percpu region\n");
1968 			goto out_err;
1969 		}
1970 
1971 		/* Prepare the CPU initialization parameters */
1972 		cpu_prepare_hyp_mode(cpu);
1973 	}
1974 
1975 	if (is_protected_kvm_enabled()) {
1976 		init_cpu_logical_map();
1977 
1978 		if (!init_psci_relay()) {
1979 			err = -ENODEV;
1980 			goto out_err;
1981 		}
1982 	}
1983 
1984 	if (is_protected_kvm_enabled()) {
1985 		err = kvm_hyp_init_protection(hyp_va_bits);
1986 		if (err) {
1987 			kvm_err("Failed to init hyp memory protection\n");
1988 			goto out_err;
1989 		}
1990 	}
1991 
1992 	return 0;
1993 
1994 out_err:
1995 	teardown_hyp_mode();
1996 	kvm_err("error initializing Hyp mode: %d\n", err);
1997 	return err;
1998 }
1999 
2000 static void _kvm_host_prot_finalize(void *arg)
2001 {
2002 	int *err = arg;
2003 
2004 	if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)))
2005 		WRITE_ONCE(*err, -EINVAL);
2006 }
2007 
2008 static int pkvm_drop_host_privileges(void)
2009 {
2010 	int ret = 0;
2011 
2012 	/*
2013 	 * Flip the static key upfront as that may no longer be possible
2014 	 * once the host stage 2 is installed.
2015 	 */
2016 	static_branch_enable(&kvm_protected_mode_initialized);
2017 	on_each_cpu(_kvm_host_prot_finalize, &ret, 1);
2018 	return ret;
2019 }
2020 
2021 static int finalize_hyp_mode(void)
2022 {
2023 	if (!is_protected_kvm_enabled())
2024 		return 0;
2025 
2026 	/*
2027 	 * Exclude HYP BSS from kmemleak so that it doesn't get peeked
2028 	 * at, which would end badly once the section is inaccessible.
2029 	 * None of other sections should ever be introspected.
2030 	 */
2031 	kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
2032 	return pkvm_drop_host_privileges();
2033 }
2034 
2035 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2036 {
2037 	struct kvm_vcpu *vcpu;
2038 	int i;
2039 
2040 	mpidr &= MPIDR_HWID_BITMASK;
2041 	kvm_for_each_vcpu(i, vcpu, kvm) {
2042 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2043 			return vcpu;
2044 	}
2045 	return NULL;
2046 }
2047 
2048 bool kvm_arch_has_irq_bypass(void)
2049 {
2050 	return true;
2051 }
2052 
2053 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2054 				      struct irq_bypass_producer *prod)
2055 {
2056 	struct kvm_kernel_irqfd *irqfd =
2057 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2058 
2059 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2060 					  &irqfd->irq_entry);
2061 }
2062 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2063 				      struct irq_bypass_producer *prod)
2064 {
2065 	struct kvm_kernel_irqfd *irqfd =
2066 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2067 
2068 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2069 				     &irqfd->irq_entry);
2070 }
2071 
2072 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2073 {
2074 	struct kvm_kernel_irqfd *irqfd =
2075 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2076 
2077 	kvm_arm_halt_guest(irqfd->kvm);
2078 }
2079 
2080 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2081 {
2082 	struct kvm_kernel_irqfd *irqfd =
2083 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2084 
2085 	kvm_arm_resume_guest(irqfd->kvm);
2086 }
2087 
2088 /**
2089  * Initialize Hyp-mode and memory mappings on all CPUs.
2090  */
2091 int kvm_arch_init(void *opaque)
2092 {
2093 	int err;
2094 	bool in_hyp_mode;
2095 
2096 	if (!is_hyp_mode_available()) {
2097 		kvm_info("HYP mode not available\n");
2098 		return -ENODEV;
2099 	}
2100 
2101 	if (kvm_get_mode() == KVM_MODE_NONE) {
2102 		kvm_info("KVM disabled from command line\n");
2103 		return -ENODEV;
2104 	}
2105 
2106 	in_hyp_mode = is_kernel_in_hyp_mode();
2107 
2108 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2109 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2110 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2111 			 "Only trusted guests should be used on this system.\n");
2112 
2113 	err = kvm_set_ipa_limit();
2114 	if (err)
2115 		return err;
2116 
2117 	err = kvm_arm_init_sve();
2118 	if (err)
2119 		return err;
2120 
2121 	if (!in_hyp_mode) {
2122 		err = init_hyp_mode();
2123 		if (err)
2124 			goto out_err;
2125 	}
2126 
2127 	err = kvm_init_vector_slots();
2128 	if (err) {
2129 		kvm_err("Cannot initialise vector slots\n");
2130 		goto out_err;
2131 	}
2132 
2133 	err = init_subsystems();
2134 	if (err)
2135 		goto out_hyp;
2136 
2137 	if (!in_hyp_mode) {
2138 		err = finalize_hyp_mode();
2139 		if (err) {
2140 			kvm_err("Failed to finalize Hyp protection\n");
2141 			goto out_hyp;
2142 		}
2143 	}
2144 
2145 	if (is_protected_kvm_enabled()) {
2146 		kvm_info("Protected nVHE mode initialized successfully\n");
2147 	} else if (in_hyp_mode) {
2148 		kvm_info("VHE mode initialized successfully\n");
2149 	} else {
2150 		kvm_info("Hyp mode initialized successfully\n");
2151 	}
2152 
2153 	return 0;
2154 
2155 out_hyp:
2156 	hyp_cpu_pm_exit();
2157 	if (!in_hyp_mode)
2158 		teardown_hyp_mode();
2159 out_err:
2160 	return err;
2161 }
2162 
2163 /* NOP: Compiling as a module not supported */
2164 void kvm_arch_exit(void)
2165 {
2166 	kvm_perf_teardown();
2167 }
2168 
2169 static int __init early_kvm_mode_cfg(char *arg)
2170 {
2171 	if (!arg)
2172 		return -EINVAL;
2173 
2174 	if (strcmp(arg, "protected") == 0) {
2175 		kvm_mode = KVM_MODE_PROTECTED;
2176 		return 0;
2177 	}
2178 
2179 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2180 		kvm_mode = KVM_MODE_DEFAULT;
2181 		return 0;
2182 	}
2183 
2184 	if (strcmp(arg, "none") == 0) {
2185 		kvm_mode = KVM_MODE_NONE;
2186 		return 0;
2187 	}
2188 
2189 	return -EINVAL;
2190 }
2191 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2192 
2193 enum kvm_mode kvm_get_mode(void)
2194 {
2195 	return kvm_mode;
2196 }
2197 
2198 static int arm_init(void)
2199 {
2200 	int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2201 	return rc;
2202 }
2203 
2204 module_init(arm_init);
2205