xref: /openbmc/linux/arch/arm64/kvm/arm.c (revision 9cf0666f)
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 		r = num_online_cpus();
227 		break;
228 	case KVM_CAP_MAX_VCPUS:
229 	case KVM_CAP_MAX_VCPU_ID:
230 		if (kvm)
231 			r = kvm->arch.max_vcpus;
232 		else
233 			r = kvm_arm_default_max_vcpus();
234 		break;
235 	case KVM_CAP_MSI_DEVID:
236 		if (!kvm)
237 			r = -EINVAL;
238 		else
239 			r = kvm->arch.vgic.msis_require_devid;
240 		break;
241 	case KVM_CAP_ARM_USER_IRQ:
242 		/*
243 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
244 		 * (bump this number if adding more devices)
245 		 */
246 		r = 1;
247 		break;
248 	case KVM_CAP_ARM_MTE:
249 		r = system_supports_mte();
250 		break;
251 	case KVM_CAP_STEAL_TIME:
252 		r = kvm_arm_pvtime_supported();
253 		break;
254 	case KVM_CAP_ARM_EL1_32BIT:
255 		r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
256 		break;
257 	case KVM_CAP_GUEST_DEBUG_HW_BPS:
258 		r = get_num_brps();
259 		break;
260 	case KVM_CAP_GUEST_DEBUG_HW_WPS:
261 		r = get_num_wrps();
262 		break;
263 	case KVM_CAP_ARM_PMU_V3:
264 		r = kvm_arm_support_pmu_v3();
265 		break;
266 	case KVM_CAP_ARM_INJECT_SERROR_ESR:
267 		r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
268 		break;
269 	case KVM_CAP_ARM_VM_IPA_SIZE:
270 		r = get_kvm_ipa_limit();
271 		break;
272 	case KVM_CAP_ARM_SVE:
273 		r = system_supports_sve();
274 		break;
275 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
276 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
277 		r = system_has_full_ptr_auth();
278 		break;
279 	default:
280 		r = 0;
281 	}
282 
283 	return r;
284 }
285 
286 long kvm_arch_dev_ioctl(struct file *filp,
287 			unsigned int ioctl, unsigned long arg)
288 {
289 	return -EINVAL;
290 }
291 
292 struct kvm *kvm_arch_alloc_vm(void)
293 {
294 	size_t sz = sizeof(struct kvm);
295 
296 	if (!has_vhe())
297 		return kzalloc(sz, GFP_KERNEL_ACCOUNT);
298 
299 	return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
300 }
301 
302 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
303 {
304 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
305 		return -EBUSY;
306 
307 	if (id >= kvm->arch.max_vcpus)
308 		return -EINVAL;
309 
310 	return 0;
311 }
312 
313 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
314 {
315 	int err;
316 
317 	/* Force users to call KVM_ARM_VCPU_INIT */
318 	vcpu->arch.target = -1;
319 	bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
320 
321 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
322 
323 	/* Set up the timer */
324 	kvm_timer_vcpu_init(vcpu);
325 
326 	kvm_pmu_vcpu_init(vcpu);
327 
328 	kvm_arm_reset_debug_ptr(vcpu);
329 
330 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
331 
332 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
333 
334 	err = kvm_vgic_vcpu_init(vcpu);
335 	if (err)
336 		return err;
337 
338 	return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
339 }
340 
341 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
342 {
343 }
344 
345 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
346 {
347 	if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
348 		static_branch_dec(&userspace_irqchip_in_use);
349 
350 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
351 	kvm_timer_vcpu_terminate(vcpu);
352 	kvm_pmu_vcpu_destroy(vcpu);
353 
354 	kvm_arm_vcpu_destroy(vcpu);
355 }
356 
357 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
358 {
359 	return kvm_timer_is_pending(vcpu);
360 }
361 
362 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
363 {
364 	/*
365 	 * If we're about to block (most likely because we've just hit a
366 	 * WFI), we need to sync back the state of the GIC CPU interface
367 	 * so that we have the latest PMR and group enables. This ensures
368 	 * that kvm_arch_vcpu_runnable has up-to-date data to decide
369 	 * whether we have pending interrupts.
370 	 *
371 	 * For the same reason, we want to tell GICv4 that we need
372 	 * doorbells to be signalled, should an interrupt become pending.
373 	 */
374 	preempt_disable();
375 	kvm_vgic_vmcr_sync(vcpu);
376 	vgic_v4_put(vcpu, true);
377 	preempt_enable();
378 }
379 
380 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
381 {
382 	preempt_disable();
383 	vgic_v4_load(vcpu);
384 	preempt_enable();
385 }
386 
387 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
388 {
389 	struct kvm_s2_mmu *mmu;
390 	int *last_ran;
391 
392 	mmu = vcpu->arch.hw_mmu;
393 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
394 
395 	/*
396 	 * We guarantee that both TLBs and I-cache are private to each
397 	 * vcpu. If detecting that a vcpu from the same VM has
398 	 * previously run on the same physical CPU, call into the
399 	 * hypervisor code to nuke the relevant contexts.
400 	 *
401 	 * We might get preempted before the vCPU actually runs, but
402 	 * over-invalidation doesn't affect correctness.
403 	 */
404 	if (*last_ran != vcpu->vcpu_id) {
405 		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
406 		*last_ran = vcpu->vcpu_id;
407 	}
408 
409 	vcpu->cpu = cpu;
410 
411 	kvm_vgic_load(vcpu);
412 	kvm_timer_vcpu_load(vcpu);
413 	if (has_vhe())
414 		kvm_vcpu_load_sysregs_vhe(vcpu);
415 	kvm_arch_vcpu_load_fp(vcpu);
416 	kvm_vcpu_pmu_restore_guest(vcpu);
417 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
418 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
419 
420 	if (single_task_running())
421 		vcpu_clear_wfx_traps(vcpu);
422 	else
423 		vcpu_set_wfx_traps(vcpu);
424 
425 	if (vcpu_has_ptrauth(vcpu))
426 		vcpu_ptrauth_disable(vcpu);
427 	kvm_arch_vcpu_load_debug_state_flags(vcpu);
428 }
429 
430 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
431 {
432 	kvm_arch_vcpu_put_debug_state_flags(vcpu);
433 	kvm_arch_vcpu_put_fp(vcpu);
434 	if (has_vhe())
435 		kvm_vcpu_put_sysregs_vhe(vcpu);
436 	kvm_timer_vcpu_put(vcpu);
437 	kvm_vgic_put(vcpu);
438 	kvm_vcpu_pmu_restore_host(vcpu);
439 
440 	vcpu->cpu = -1;
441 }
442 
443 static void vcpu_power_off(struct kvm_vcpu *vcpu)
444 {
445 	vcpu->arch.power_off = true;
446 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
447 	kvm_vcpu_kick(vcpu);
448 }
449 
450 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
451 				    struct kvm_mp_state *mp_state)
452 {
453 	if (vcpu->arch.power_off)
454 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
455 	else
456 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
457 
458 	return 0;
459 }
460 
461 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
462 				    struct kvm_mp_state *mp_state)
463 {
464 	int ret = 0;
465 
466 	switch (mp_state->mp_state) {
467 	case KVM_MP_STATE_RUNNABLE:
468 		vcpu->arch.power_off = false;
469 		break;
470 	case KVM_MP_STATE_STOPPED:
471 		vcpu_power_off(vcpu);
472 		break;
473 	default:
474 		ret = -EINVAL;
475 	}
476 
477 	return ret;
478 }
479 
480 /**
481  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
482  * @v:		The VCPU pointer
483  *
484  * If the guest CPU is not waiting for interrupts or an interrupt line is
485  * asserted, the CPU is by definition runnable.
486  */
487 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
488 {
489 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
490 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
491 		&& !v->arch.power_off && !v->arch.pause);
492 }
493 
494 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
495 {
496 	return vcpu_mode_priv(vcpu);
497 }
498 
499 /* Just ensure a guest exit from a particular CPU */
500 static void exit_vm_noop(void *info)
501 {
502 }
503 
504 void force_vm_exit(const cpumask_t *mask)
505 {
506 	preempt_disable();
507 	smp_call_function_many(mask, exit_vm_noop, NULL, true);
508 	preempt_enable();
509 }
510 
511 /**
512  * need_new_vmid_gen - check that the VMID is still valid
513  * @vmid: The VMID to check
514  *
515  * return true if there is a new generation of VMIDs being used
516  *
517  * The hardware supports a limited set of values with the value zero reserved
518  * for the host, so we check if an assigned value belongs to a previous
519  * generation, which requires us to assign a new value. If we're the first to
520  * use a VMID for the new generation, we must flush necessary caches and TLBs
521  * on all CPUs.
522  */
523 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
524 {
525 	u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
526 	smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
527 	return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
528 }
529 
530 /**
531  * update_vmid - Update the vmid with a valid VMID for the current generation
532  * @vmid: The stage-2 VMID information struct
533  */
534 static void update_vmid(struct kvm_vmid *vmid)
535 {
536 	if (!need_new_vmid_gen(vmid))
537 		return;
538 
539 	spin_lock(&kvm_vmid_lock);
540 
541 	/*
542 	 * We need to re-check the vmid_gen here to ensure that if another vcpu
543 	 * already allocated a valid vmid for this vm, then this vcpu should
544 	 * use the same vmid.
545 	 */
546 	if (!need_new_vmid_gen(vmid)) {
547 		spin_unlock(&kvm_vmid_lock);
548 		return;
549 	}
550 
551 	/* First user of a new VMID generation? */
552 	if (unlikely(kvm_next_vmid == 0)) {
553 		atomic64_inc(&kvm_vmid_gen);
554 		kvm_next_vmid = 1;
555 
556 		/*
557 		 * On SMP we know no other CPUs can use this CPU's or each
558 		 * other's VMID after force_vm_exit returns since the
559 		 * kvm_vmid_lock blocks them from reentry to the guest.
560 		 */
561 		force_vm_exit(cpu_all_mask);
562 		/*
563 		 * Now broadcast TLB + ICACHE invalidation over the inner
564 		 * shareable domain to make sure all data structures are
565 		 * clean.
566 		 */
567 		kvm_call_hyp(__kvm_flush_vm_context);
568 	}
569 
570 	WRITE_ONCE(vmid->vmid, kvm_next_vmid);
571 	kvm_next_vmid++;
572 	kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
573 
574 	smp_wmb();
575 	WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
576 
577 	spin_unlock(&kvm_vmid_lock);
578 }
579 
580 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
581 {
582 	struct kvm *kvm = vcpu->kvm;
583 	int ret = 0;
584 
585 	if (likely(vcpu->arch.has_run_once))
586 		return 0;
587 
588 	if (!kvm_arm_vcpu_is_finalized(vcpu))
589 		return -EPERM;
590 
591 	vcpu->arch.has_run_once = true;
592 
593 	kvm_arm_vcpu_init_debug(vcpu);
594 
595 	if (likely(irqchip_in_kernel(kvm))) {
596 		/*
597 		 * Map the VGIC hardware resources before running a vcpu the
598 		 * first time on this VM.
599 		 */
600 		ret = kvm_vgic_map_resources(kvm);
601 		if (ret)
602 			return ret;
603 	} else {
604 		/*
605 		 * Tell the rest of the code that there are userspace irqchip
606 		 * VMs in the wild.
607 		 */
608 		static_branch_inc(&userspace_irqchip_in_use);
609 	}
610 
611 	ret = kvm_timer_enable(vcpu);
612 	if (ret)
613 		return ret;
614 
615 	ret = kvm_arm_pmu_v3_enable(vcpu);
616 
617 	/*
618 	 * Initialize traps for protected VMs.
619 	 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
620 	 * the code is in place for first run initialization at EL2.
621 	 */
622 	if (kvm_vm_is_protected(kvm))
623 		kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
624 
625 	return ret;
626 }
627 
628 bool kvm_arch_intc_initialized(struct kvm *kvm)
629 {
630 	return vgic_initialized(kvm);
631 }
632 
633 void kvm_arm_halt_guest(struct kvm *kvm)
634 {
635 	int i;
636 	struct kvm_vcpu *vcpu;
637 
638 	kvm_for_each_vcpu(i, vcpu, kvm)
639 		vcpu->arch.pause = true;
640 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
641 }
642 
643 void kvm_arm_resume_guest(struct kvm *kvm)
644 {
645 	int i;
646 	struct kvm_vcpu *vcpu;
647 
648 	kvm_for_each_vcpu(i, vcpu, kvm) {
649 		vcpu->arch.pause = false;
650 		rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
651 	}
652 }
653 
654 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
655 {
656 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
657 
658 	rcuwait_wait_event(wait,
659 			   (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
660 			   TASK_INTERRUPTIBLE);
661 
662 	if (vcpu->arch.power_off || vcpu->arch.pause) {
663 		/* Awaken to handle a signal, request we sleep again later. */
664 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
665 	}
666 
667 	/*
668 	 * Make sure we will observe a potential reset request if we've
669 	 * observed a change to the power state. Pairs with the smp_wmb() in
670 	 * kvm_psci_vcpu_on().
671 	 */
672 	smp_rmb();
673 }
674 
675 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
676 {
677 	return vcpu->arch.target >= 0;
678 }
679 
680 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
681 {
682 	if (kvm_request_pending(vcpu)) {
683 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
684 			vcpu_req_sleep(vcpu);
685 
686 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
687 			kvm_reset_vcpu(vcpu);
688 
689 		/*
690 		 * Clear IRQ_PENDING requests that were made to guarantee
691 		 * that a VCPU sees new virtual interrupts.
692 		 */
693 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
694 
695 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
696 			kvm_update_stolen_time(vcpu);
697 
698 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
699 			/* The distributor enable bits were changed */
700 			preempt_disable();
701 			vgic_v4_put(vcpu, false);
702 			vgic_v4_load(vcpu);
703 			preempt_enable();
704 		}
705 
706 		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
707 			kvm_pmu_handle_pmcr(vcpu,
708 					    __vcpu_sys_reg(vcpu, PMCR_EL0));
709 	}
710 }
711 
712 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
713 {
714 	if (likely(!vcpu_mode_is_32bit(vcpu)))
715 		return false;
716 
717 	return !system_supports_32bit_el0() ||
718 		static_branch_unlikely(&arm64_mismatched_32bit_el0);
719 }
720 
721 /**
722  * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
723  * @vcpu:	The VCPU pointer
724  * @ret:	Pointer to write optional return code
725  *
726  * Returns: true if the VCPU needs to return to a preemptible + interruptible
727  *	    and skip guest entry.
728  *
729  * This function disambiguates between two different types of exits: exits to a
730  * preemptible + interruptible kernel context and exits to userspace. For an
731  * exit to userspace, this function will write the return code to ret and return
732  * true. For an exit to preemptible + interruptible kernel context (i.e. check
733  * for pending work and re-enter), return true without writing to ret.
734  */
735 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
736 {
737 	struct kvm_run *run = vcpu->run;
738 
739 	/*
740 	 * If we're using a userspace irqchip, then check if we need
741 	 * to tell a userspace irqchip about timer or PMU level
742 	 * changes and if so, exit to userspace (the actual level
743 	 * state gets updated in kvm_timer_update_run and
744 	 * kvm_pmu_update_run below).
745 	 */
746 	if (static_branch_unlikely(&userspace_irqchip_in_use)) {
747 		if (kvm_timer_should_notify_user(vcpu) ||
748 		    kvm_pmu_should_notify_user(vcpu)) {
749 			*ret = -EINTR;
750 			run->exit_reason = KVM_EXIT_INTR;
751 			return true;
752 		}
753 	}
754 
755 	return kvm_request_pending(vcpu) ||
756 			need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
757 			xfer_to_guest_mode_work_pending();
758 }
759 
760 /**
761  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
762  * @vcpu:	The VCPU pointer
763  *
764  * This function is called through the VCPU_RUN ioctl called from user space. It
765  * will execute VM code in a loop until the time slice for the process is used
766  * or some emulation is needed from user space in which case the function will
767  * return with return value 0 and with the kvm_run structure filled in with the
768  * required data for the requested emulation.
769  */
770 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
771 {
772 	struct kvm_run *run = vcpu->run;
773 	int ret;
774 
775 	if (unlikely(!kvm_vcpu_initialized(vcpu)))
776 		return -ENOEXEC;
777 
778 	ret = kvm_vcpu_first_run_init(vcpu);
779 	if (ret)
780 		return ret;
781 
782 	if (run->exit_reason == KVM_EXIT_MMIO) {
783 		ret = kvm_handle_mmio_return(vcpu);
784 		if (ret)
785 			return ret;
786 	}
787 
788 	vcpu_load(vcpu);
789 
790 	if (run->immediate_exit) {
791 		ret = -EINTR;
792 		goto out;
793 	}
794 
795 	kvm_sigset_activate(vcpu);
796 
797 	ret = 1;
798 	run->exit_reason = KVM_EXIT_UNKNOWN;
799 	while (ret > 0) {
800 		/*
801 		 * Check conditions before entering the guest
802 		 */
803 		ret = xfer_to_guest_mode_handle_work(vcpu);
804 		if (!ret)
805 			ret = 1;
806 
807 		update_vmid(&vcpu->arch.hw_mmu->vmid);
808 
809 		check_vcpu_requests(vcpu);
810 
811 		/*
812 		 * Preparing the interrupts to be injected also
813 		 * involves poking the GIC, which must be done in a
814 		 * non-preemptible context.
815 		 */
816 		preempt_disable();
817 
818 		kvm_pmu_flush_hwstate(vcpu);
819 
820 		local_irq_disable();
821 
822 		kvm_vgic_flush_hwstate(vcpu);
823 
824 		/*
825 		 * Ensure we set mode to IN_GUEST_MODE after we disable
826 		 * interrupts and before the final VCPU requests check.
827 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
828 		 * Documentation/virt/kvm/vcpu-requests.rst
829 		 */
830 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
831 
832 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
833 			vcpu->mode = OUTSIDE_GUEST_MODE;
834 			isb(); /* Ensure work in x_flush_hwstate is committed */
835 			kvm_pmu_sync_hwstate(vcpu);
836 			if (static_branch_unlikely(&userspace_irqchip_in_use))
837 				kvm_timer_sync_user(vcpu);
838 			kvm_vgic_sync_hwstate(vcpu);
839 			local_irq_enable();
840 			preempt_enable();
841 			continue;
842 		}
843 
844 		kvm_arm_setup_debug(vcpu);
845 
846 		/**************************************************************
847 		 * Enter the guest
848 		 */
849 		trace_kvm_entry(*vcpu_pc(vcpu));
850 		guest_enter_irqoff();
851 
852 		ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
853 
854 		vcpu->mode = OUTSIDE_GUEST_MODE;
855 		vcpu->stat.exits++;
856 		/*
857 		 * Back from guest
858 		 *************************************************************/
859 
860 		kvm_arm_clear_debug(vcpu);
861 
862 		/*
863 		 * We must sync the PMU state before the vgic state so
864 		 * that the vgic can properly sample the updated state of the
865 		 * interrupt line.
866 		 */
867 		kvm_pmu_sync_hwstate(vcpu);
868 
869 		/*
870 		 * Sync the vgic state before syncing the timer state because
871 		 * the timer code needs to know if the virtual timer
872 		 * interrupts are active.
873 		 */
874 		kvm_vgic_sync_hwstate(vcpu);
875 
876 		/*
877 		 * Sync the timer hardware state before enabling interrupts as
878 		 * we don't want vtimer interrupts to race with syncing the
879 		 * timer virtual interrupt state.
880 		 */
881 		if (static_branch_unlikely(&userspace_irqchip_in_use))
882 			kvm_timer_sync_user(vcpu);
883 
884 		kvm_arch_vcpu_ctxsync_fp(vcpu);
885 
886 		/*
887 		 * We may have taken a host interrupt in HYP mode (ie
888 		 * while executing the guest). This interrupt is still
889 		 * pending, as we haven't serviced it yet!
890 		 *
891 		 * We're now back in SVC mode, with interrupts
892 		 * disabled.  Enabling the interrupts now will have
893 		 * the effect of taking the interrupt again, in SVC
894 		 * mode this time.
895 		 */
896 		local_irq_enable();
897 
898 		/*
899 		 * We do local_irq_enable() before calling guest_exit() so
900 		 * that if a timer interrupt hits while running the guest we
901 		 * account that tick as being spent in the guest.  We enable
902 		 * preemption after calling guest_exit() so that if we get
903 		 * preempted we make sure ticks after that is not counted as
904 		 * guest time.
905 		 */
906 		guest_exit();
907 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
908 
909 		/* Exit types that need handling before we can be preempted */
910 		handle_exit_early(vcpu, ret);
911 
912 		preempt_enable();
913 
914 		/*
915 		 * The ARMv8 architecture doesn't give the hypervisor
916 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
917 		 * if implemented by the CPU. If we spot the guest in such
918 		 * state and that we decided it wasn't supposed to do so (like
919 		 * with the asymmetric AArch32 case), return to userspace with
920 		 * a fatal error.
921 		 */
922 		if (vcpu_mode_is_bad_32bit(vcpu)) {
923 			/*
924 			 * As we have caught the guest red-handed, decide that
925 			 * it isn't fit for purpose anymore by making the vcpu
926 			 * invalid. The VMM can try and fix it by issuing  a
927 			 * KVM_ARM_VCPU_INIT if it really wants to.
928 			 */
929 			vcpu->arch.target = -1;
930 			ret = ARM_EXCEPTION_IL;
931 		}
932 
933 		ret = handle_exit(vcpu, ret);
934 	}
935 
936 	/* Tell userspace about in-kernel device output levels */
937 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
938 		kvm_timer_update_run(vcpu);
939 		kvm_pmu_update_run(vcpu);
940 	}
941 
942 	kvm_sigset_deactivate(vcpu);
943 
944 out:
945 	/*
946 	 * In the unlikely event that we are returning to userspace
947 	 * with pending exceptions or PC adjustment, commit these
948 	 * adjustments in order to give userspace a consistent view of
949 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
950 	 * being preempt-safe on VHE.
951 	 */
952 	if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
953 					 KVM_ARM64_INCREMENT_PC)))
954 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
955 
956 	vcpu_put(vcpu);
957 	return ret;
958 }
959 
960 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
961 {
962 	int bit_index;
963 	bool set;
964 	unsigned long *hcr;
965 
966 	if (number == KVM_ARM_IRQ_CPU_IRQ)
967 		bit_index = __ffs(HCR_VI);
968 	else /* KVM_ARM_IRQ_CPU_FIQ */
969 		bit_index = __ffs(HCR_VF);
970 
971 	hcr = vcpu_hcr(vcpu);
972 	if (level)
973 		set = test_and_set_bit(bit_index, hcr);
974 	else
975 		set = test_and_clear_bit(bit_index, hcr);
976 
977 	/*
978 	 * If we didn't change anything, no need to wake up or kick other CPUs
979 	 */
980 	if (set == level)
981 		return 0;
982 
983 	/*
984 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
985 	 * trigger a world-switch round on the running physical CPU to set the
986 	 * virtual IRQ/FIQ fields in the HCR appropriately.
987 	 */
988 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
989 	kvm_vcpu_kick(vcpu);
990 
991 	return 0;
992 }
993 
994 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
995 			  bool line_status)
996 {
997 	u32 irq = irq_level->irq;
998 	unsigned int irq_type, vcpu_idx, irq_num;
999 	int nrcpus = atomic_read(&kvm->online_vcpus);
1000 	struct kvm_vcpu *vcpu = NULL;
1001 	bool level = irq_level->level;
1002 
1003 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1004 	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1005 	vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1006 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1007 
1008 	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1009 
1010 	switch (irq_type) {
1011 	case KVM_ARM_IRQ_TYPE_CPU:
1012 		if (irqchip_in_kernel(kvm))
1013 			return -ENXIO;
1014 
1015 		if (vcpu_idx >= nrcpus)
1016 			return -EINVAL;
1017 
1018 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1019 		if (!vcpu)
1020 			return -EINVAL;
1021 
1022 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1023 			return -EINVAL;
1024 
1025 		return vcpu_interrupt_line(vcpu, irq_num, level);
1026 	case KVM_ARM_IRQ_TYPE_PPI:
1027 		if (!irqchip_in_kernel(kvm))
1028 			return -ENXIO;
1029 
1030 		if (vcpu_idx >= nrcpus)
1031 			return -EINVAL;
1032 
1033 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1034 		if (!vcpu)
1035 			return -EINVAL;
1036 
1037 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1038 			return -EINVAL;
1039 
1040 		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1041 	case KVM_ARM_IRQ_TYPE_SPI:
1042 		if (!irqchip_in_kernel(kvm))
1043 			return -ENXIO;
1044 
1045 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1046 			return -EINVAL;
1047 
1048 		return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1049 	}
1050 
1051 	return -EINVAL;
1052 }
1053 
1054 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1055 			       const struct kvm_vcpu_init *init)
1056 {
1057 	unsigned int i, ret;
1058 	u32 phys_target = kvm_target_cpu();
1059 
1060 	if (init->target != phys_target)
1061 		return -EINVAL;
1062 
1063 	/*
1064 	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1065 	 * use the same target.
1066 	 */
1067 	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1068 		return -EINVAL;
1069 
1070 	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1071 	for (i = 0; i < sizeof(init->features) * 8; i++) {
1072 		bool set = (init->features[i / 32] & (1 << (i % 32)));
1073 
1074 		if (set && i >= KVM_VCPU_MAX_FEATURES)
1075 			return -ENOENT;
1076 
1077 		/*
1078 		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1079 		 * use the same feature set.
1080 		 */
1081 		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1082 		    test_bit(i, vcpu->arch.features) != set)
1083 			return -EINVAL;
1084 
1085 		if (set)
1086 			set_bit(i, vcpu->arch.features);
1087 	}
1088 
1089 	vcpu->arch.target = phys_target;
1090 
1091 	/* Now we know what it is, we can reset it. */
1092 	ret = kvm_reset_vcpu(vcpu);
1093 	if (ret) {
1094 		vcpu->arch.target = -1;
1095 		bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1096 	}
1097 
1098 	return ret;
1099 }
1100 
1101 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1102 					 struct kvm_vcpu_init *init)
1103 {
1104 	int ret;
1105 
1106 	ret = kvm_vcpu_set_target(vcpu, init);
1107 	if (ret)
1108 		return ret;
1109 
1110 	/*
1111 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1112 	 * guest MMU is turned off and flush the caches as needed.
1113 	 *
1114 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1115 	 * ensuring that the data side is always coherent. We still
1116 	 * need to invalidate the I-cache though, as FWB does *not*
1117 	 * imply CTR_EL0.DIC.
1118 	 */
1119 	if (vcpu->arch.has_run_once) {
1120 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1121 			stage2_unmap_vm(vcpu->kvm);
1122 		else
1123 			icache_inval_all_pou();
1124 	}
1125 
1126 	vcpu_reset_hcr(vcpu);
1127 	vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1128 
1129 	/*
1130 	 * Handle the "start in power-off" case.
1131 	 */
1132 	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1133 		vcpu_power_off(vcpu);
1134 	else
1135 		vcpu->arch.power_off = false;
1136 
1137 	return 0;
1138 }
1139 
1140 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1141 				 struct kvm_device_attr *attr)
1142 {
1143 	int ret = -ENXIO;
1144 
1145 	switch (attr->group) {
1146 	default:
1147 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1148 		break;
1149 	}
1150 
1151 	return ret;
1152 }
1153 
1154 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1155 				 struct kvm_device_attr *attr)
1156 {
1157 	int ret = -ENXIO;
1158 
1159 	switch (attr->group) {
1160 	default:
1161 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1162 		break;
1163 	}
1164 
1165 	return ret;
1166 }
1167 
1168 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1169 				 struct kvm_device_attr *attr)
1170 {
1171 	int ret = -ENXIO;
1172 
1173 	switch (attr->group) {
1174 	default:
1175 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1176 		break;
1177 	}
1178 
1179 	return ret;
1180 }
1181 
1182 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1183 				   struct kvm_vcpu_events *events)
1184 {
1185 	memset(events, 0, sizeof(*events));
1186 
1187 	return __kvm_arm_vcpu_get_events(vcpu, events);
1188 }
1189 
1190 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1191 				   struct kvm_vcpu_events *events)
1192 {
1193 	int i;
1194 
1195 	/* check whether the reserved field is zero */
1196 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1197 		if (events->reserved[i])
1198 			return -EINVAL;
1199 
1200 	/* check whether the pad field is zero */
1201 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1202 		if (events->exception.pad[i])
1203 			return -EINVAL;
1204 
1205 	return __kvm_arm_vcpu_set_events(vcpu, events);
1206 }
1207 
1208 long kvm_arch_vcpu_ioctl(struct file *filp,
1209 			 unsigned int ioctl, unsigned long arg)
1210 {
1211 	struct kvm_vcpu *vcpu = filp->private_data;
1212 	void __user *argp = (void __user *)arg;
1213 	struct kvm_device_attr attr;
1214 	long r;
1215 
1216 	switch (ioctl) {
1217 	case KVM_ARM_VCPU_INIT: {
1218 		struct kvm_vcpu_init init;
1219 
1220 		r = -EFAULT;
1221 		if (copy_from_user(&init, argp, sizeof(init)))
1222 			break;
1223 
1224 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1225 		break;
1226 	}
1227 	case KVM_SET_ONE_REG:
1228 	case KVM_GET_ONE_REG: {
1229 		struct kvm_one_reg reg;
1230 
1231 		r = -ENOEXEC;
1232 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1233 			break;
1234 
1235 		r = -EFAULT;
1236 		if (copy_from_user(&reg, argp, sizeof(reg)))
1237 			break;
1238 
1239 		/*
1240 		 * We could owe a reset due to PSCI. Handle the pending reset
1241 		 * here to ensure userspace register accesses are ordered after
1242 		 * the reset.
1243 		 */
1244 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1245 			kvm_reset_vcpu(vcpu);
1246 
1247 		if (ioctl == KVM_SET_ONE_REG)
1248 			r = kvm_arm_set_reg(vcpu, &reg);
1249 		else
1250 			r = kvm_arm_get_reg(vcpu, &reg);
1251 		break;
1252 	}
1253 	case KVM_GET_REG_LIST: {
1254 		struct kvm_reg_list __user *user_list = argp;
1255 		struct kvm_reg_list reg_list;
1256 		unsigned n;
1257 
1258 		r = -ENOEXEC;
1259 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1260 			break;
1261 
1262 		r = -EPERM;
1263 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1264 			break;
1265 
1266 		r = -EFAULT;
1267 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1268 			break;
1269 		n = reg_list.n;
1270 		reg_list.n = kvm_arm_num_regs(vcpu);
1271 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1272 			break;
1273 		r = -E2BIG;
1274 		if (n < reg_list.n)
1275 			break;
1276 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1277 		break;
1278 	}
1279 	case KVM_SET_DEVICE_ATTR: {
1280 		r = -EFAULT;
1281 		if (copy_from_user(&attr, argp, sizeof(attr)))
1282 			break;
1283 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1284 		break;
1285 	}
1286 	case KVM_GET_DEVICE_ATTR: {
1287 		r = -EFAULT;
1288 		if (copy_from_user(&attr, argp, sizeof(attr)))
1289 			break;
1290 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1291 		break;
1292 	}
1293 	case KVM_HAS_DEVICE_ATTR: {
1294 		r = -EFAULT;
1295 		if (copy_from_user(&attr, argp, sizeof(attr)))
1296 			break;
1297 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1298 		break;
1299 	}
1300 	case KVM_GET_VCPU_EVENTS: {
1301 		struct kvm_vcpu_events events;
1302 
1303 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1304 			return -EINVAL;
1305 
1306 		if (copy_to_user(argp, &events, sizeof(events)))
1307 			return -EFAULT;
1308 
1309 		return 0;
1310 	}
1311 	case KVM_SET_VCPU_EVENTS: {
1312 		struct kvm_vcpu_events events;
1313 
1314 		if (copy_from_user(&events, argp, sizeof(events)))
1315 			return -EFAULT;
1316 
1317 		return kvm_arm_vcpu_set_events(vcpu, &events);
1318 	}
1319 	case KVM_ARM_VCPU_FINALIZE: {
1320 		int what;
1321 
1322 		if (!kvm_vcpu_initialized(vcpu))
1323 			return -ENOEXEC;
1324 
1325 		if (get_user(what, (const int __user *)argp))
1326 			return -EFAULT;
1327 
1328 		return kvm_arm_vcpu_finalize(vcpu, what);
1329 	}
1330 	default:
1331 		r = -EINVAL;
1332 	}
1333 
1334 	return r;
1335 }
1336 
1337 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1338 {
1339 
1340 }
1341 
1342 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1343 					const struct kvm_memory_slot *memslot)
1344 {
1345 	kvm_flush_remote_tlbs(kvm);
1346 }
1347 
1348 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1349 					struct kvm_arm_device_addr *dev_addr)
1350 {
1351 	unsigned long dev_id, type;
1352 
1353 	dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1354 		KVM_ARM_DEVICE_ID_SHIFT;
1355 	type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1356 		KVM_ARM_DEVICE_TYPE_SHIFT;
1357 
1358 	switch (dev_id) {
1359 	case KVM_ARM_DEVICE_VGIC_V2:
1360 		if (!vgic_present)
1361 			return -ENXIO;
1362 		return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1363 	default:
1364 		return -ENODEV;
1365 	}
1366 }
1367 
1368 long kvm_arch_vm_ioctl(struct file *filp,
1369 		       unsigned int ioctl, unsigned long arg)
1370 {
1371 	struct kvm *kvm = filp->private_data;
1372 	void __user *argp = (void __user *)arg;
1373 
1374 	switch (ioctl) {
1375 	case KVM_CREATE_IRQCHIP: {
1376 		int ret;
1377 		if (!vgic_present)
1378 			return -ENXIO;
1379 		mutex_lock(&kvm->lock);
1380 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1381 		mutex_unlock(&kvm->lock);
1382 		return ret;
1383 	}
1384 	case KVM_ARM_SET_DEVICE_ADDR: {
1385 		struct kvm_arm_device_addr dev_addr;
1386 
1387 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1388 			return -EFAULT;
1389 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1390 	}
1391 	case KVM_ARM_PREFERRED_TARGET: {
1392 		struct kvm_vcpu_init init;
1393 
1394 		kvm_vcpu_preferred_target(&init);
1395 
1396 		if (copy_to_user(argp, &init, sizeof(init)))
1397 			return -EFAULT;
1398 
1399 		return 0;
1400 	}
1401 	case KVM_ARM_MTE_COPY_TAGS: {
1402 		struct kvm_arm_copy_mte_tags copy_tags;
1403 
1404 		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1405 			return -EFAULT;
1406 		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1407 	}
1408 	default:
1409 		return -EINVAL;
1410 	}
1411 }
1412 
1413 static unsigned long nvhe_percpu_size(void)
1414 {
1415 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1416 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1417 }
1418 
1419 static unsigned long nvhe_percpu_order(void)
1420 {
1421 	unsigned long size = nvhe_percpu_size();
1422 
1423 	return size ? get_order(size) : 0;
1424 }
1425 
1426 /* A lookup table holding the hypervisor VA for each vector slot */
1427 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1428 
1429 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1430 {
1431 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1432 }
1433 
1434 static int kvm_init_vector_slots(void)
1435 {
1436 	int err;
1437 	void *base;
1438 
1439 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1440 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1441 
1442 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1443 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1444 
1445 	if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1446 		return 0;
1447 
1448 	if (!has_vhe()) {
1449 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1450 					       __BP_HARDEN_HYP_VECS_SZ, &base);
1451 		if (err)
1452 			return err;
1453 	}
1454 
1455 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1456 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1457 	return 0;
1458 }
1459 
1460 static void cpu_prepare_hyp_mode(int cpu)
1461 {
1462 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1463 	unsigned long tcr;
1464 
1465 	/*
1466 	 * Calculate the raw per-cpu offset without a translation from the
1467 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1468 	 * so that we can use adr_l to access per-cpu variables in EL2.
1469 	 * Also drop the KASAN tag which gets in the way...
1470 	 */
1471 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1472 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1473 
1474 	params->mair_el2 = read_sysreg(mair_el1);
1475 
1476 	/*
1477 	 * The ID map may be configured to use an extended virtual address
1478 	 * range. This is only the case if system RAM is out of range for the
1479 	 * currently configured page size and VA_BITS, in which case we will
1480 	 * also need the extended virtual range for the HYP ID map, or we won't
1481 	 * be able to enable the EL2 MMU.
1482 	 *
1483 	 * However, at EL2, there is only one TTBR register, and we can't switch
1484 	 * between translation tables *and* update TCR_EL2.T0SZ at the same
1485 	 * time. Bottom line: we need to use the extended range with *both* our
1486 	 * translation tables.
1487 	 *
1488 	 * So use the same T0SZ value we use for the ID map.
1489 	 */
1490 	tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1491 	tcr &= ~TCR_T0SZ_MASK;
1492 	tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1493 	params->tcr_el2 = tcr;
1494 
1495 	params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1496 	params->pgd_pa = kvm_mmu_get_httbr();
1497 	if (is_protected_kvm_enabled())
1498 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1499 	else
1500 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1501 	params->vttbr = params->vtcr = 0;
1502 
1503 	/*
1504 	 * Flush the init params from the data cache because the struct will
1505 	 * be read while the MMU is off.
1506 	 */
1507 	kvm_flush_dcache_to_poc(params, sizeof(*params));
1508 }
1509 
1510 static void hyp_install_host_vector(void)
1511 {
1512 	struct kvm_nvhe_init_params *params;
1513 	struct arm_smccc_res res;
1514 
1515 	/* Switch from the HYP stub to our own HYP init vector */
1516 	__hyp_set_vectors(kvm_get_idmap_vector());
1517 
1518 	/*
1519 	 * Call initialization code, and switch to the full blown HYP code.
1520 	 * If the cpucaps haven't been finalized yet, something has gone very
1521 	 * wrong, and hyp will crash and burn when it uses any
1522 	 * cpus_have_const_cap() wrapper.
1523 	 */
1524 	BUG_ON(!system_capabilities_finalized());
1525 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1526 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1527 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1528 }
1529 
1530 static void cpu_init_hyp_mode(void)
1531 {
1532 	hyp_install_host_vector();
1533 
1534 	/*
1535 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1536 	 * at EL2.
1537 	 */
1538 	if (this_cpu_has_cap(ARM64_SSBS) &&
1539 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1540 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1541 	}
1542 }
1543 
1544 static void cpu_hyp_reset(void)
1545 {
1546 	if (!is_kernel_in_hyp_mode())
1547 		__hyp_reset_vectors();
1548 }
1549 
1550 /*
1551  * EL2 vectors can be mapped and rerouted in a number of ways,
1552  * depending on the kernel configuration and CPU present:
1553  *
1554  * - If the CPU is affected by Spectre-v2, the hardening sequence is
1555  *   placed in one of the vector slots, which is executed before jumping
1556  *   to the real vectors.
1557  *
1558  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1559  *   containing the hardening sequence is mapped next to the idmap page,
1560  *   and executed before jumping to the real vectors.
1561  *
1562  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1563  *   empty slot is selected, mapped next to the idmap page, and
1564  *   executed before jumping to the real vectors.
1565  *
1566  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1567  * VHE, as we don't have hypervisor-specific mappings. If the system
1568  * is VHE and yet selects this capability, it will be ignored.
1569  */
1570 static void cpu_set_hyp_vector(void)
1571 {
1572 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1573 	void *vector = hyp_spectre_vector_selector[data->slot];
1574 
1575 	if (!is_protected_kvm_enabled())
1576 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1577 	else
1578 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1579 }
1580 
1581 static void cpu_hyp_init_context(void)
1582 {
1583 	kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1584 
1585 	if (!is_kernel_in_hyp_mode())
1586 		cpu_init_hyp_mode();
1587 }
1588 
1589 static void cpu_hyp_init_features(void)
1590 {
1591 	cpu_set_hyp_vector();
1592 	kvm_arm_init_debug();
1593 
1594 	if (is_kernel_in_hyp_mode())
1595 		kvm_timer_init_vhe();
1596 
1597 	if (vgic_present)
1598 		kvm_vgic_init_cpu_hardware();
1599 }
1600 
1601 static void cpu_hyp_reinit(void)
1602 {
1603 	cpu_hyp_reset();
1604 	cpu_hyp_init_context();
1605 	cpu_hyp_init_features();
1606 }
1607 
1608 static void _kvm_arch_hardware_enable(void *discard)
1609 {
1610 	if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1611 		cpu_hyp_reinit();
1612 		__this_cpu_write(kvm_arm_hardware_enabled, 1);
1613 	}
1614 }
1615 
1616 int kvm_arch_hardware_enable(void)
1617 {
1618 	_kvm_arch_hardware_enable(NULL);
1619 	return 0;
1620 }
1621 
1622 static void _kvm_arch_hardware_disable(void *discard)
1623 {
1624 	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1625 		cpu_hyp_reset();
1626 		__this_cpu_write(kvm_arm_hardware_enabled, 0);
1627 	}
1628 }
1629 
1630 void kvm_arch_hardware_disable(void)
1631 {
1632 	if (!is_protected_kvm_enabled())
1633 		_kvm_arch_hardware_disable(NULL);
1634 }
1635 
1636 #ifdef CONFIG_CPU_PM
1637 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1638 				    unsigned long cmd,
1639 				    void *v)
1640 {
1641 	/*
1642 	 * kvm_arm_hardware_enabled is left with its old value over
1643 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1644 	 * re-enable hyp.
1645 	 */
1646 	switch (cmd) {
1647 	case CPU_PM_ENTER:
1648 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1649 			/*
1650 			 * don't update kvm_arm_hardware_enabled here
1651 			 * so that the hardware will be re-enabled
1652 			 * when we resume. See below.
1653 			 */
1654 			cpu_hyp_reset();
1655 
1656 		return NOTIFY_OK;
1657 	case CPU_PM_ENTER_FAILED:
1658 	case CPU_PM_EXIT:
1659 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1660 			/* The hardware was enabled before suspend. */
1661 			cpu_hyp_reinit();
1662 
1663 		return NOTIFY_OK;
1664 
1665 	default:
1666 		return NOTIFY_DONE;
1667 	}
1668 }
1669 
1670 static struct notifier_block hyp_init_cpu_pm_nb = {
1671 	.notifier_call = hyp_init_cpu_pm_notifier,
1672 };
1673 
1674 static void hyp_cpu_pm_init(void)
1675 {
1676 	if (!is_protected_kvm_enabled())
1677 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1678 }
1679 static void hyp_cpu_pm_exit(void)
1680 {
1681 	if (!is_protected_kvm_enabled())
1682 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1683 }
1684 #else
1685 static inline void hyp_cpu_pm_init(void)
1686 {
1687 }
1688 static inline void hyp_cpu_pm_exit(void)
1689 {
1690 }
1691 #endif
1692 
1693 static void init_cpu_logical_map(void)
1694 {
1695 	unsigned int cpu;
1696 
1697 	/*
1698 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1699 	 * Only copy the set of online CPUs whose features have been chacked
1700 	 * against the finalized system capabilities. The hypervisor will not
1701 	 * allow any other CPUs from the `possible` set to boot.
1702 	 */
1703 	for_each_online_cpu(cpu)
1704 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1705 }
1706 
1707 #define init_psci_0_1_impl_state(config, what)	\
1708 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
1709 
1710 static bool init_psci_relay(void)
1711 {
1712 	/*
1713 	 * If PSCI has not been initialized, protected KVM cannot install
1714 	 * itself on newly booted CPUs.
1715 	 */
1716 	if (!psci_ops.get_version) {
1717 		kvm_err("Cannot initialize protected mode without PSCI\n");
1718 		return false;
1719 	}
1720 
1721 	kvm_host_psci_config.version = psci_ops.get_version();
1722 
1723 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1724 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1725 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1726 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1727 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1728 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1729 	}
1730 	return true;
1731 }
1732 
1733 static int init_subsystems(void)
1734 {
1735 	int err = 0;
1736 
1737 	/*
1738 	 * Enable hardware so that subsystem initialisation can access EL2.
1739 	 */
1740 	on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1741 
1742 	/*
1743 	 * Register CPU lower-power notifier
1744 	 */
1745 	hyp_cpu_pm_init();
1746 
1747 	/*
1748 	 * Init HYP view of VGIC
1749 	 */
1750 	err = kvm_vgic_hyp_init();
1751 	switch (err) {
1752 	case 0:
1753 		vgic_present = true;
1754 		break;
1755 	case -ENODEV:
1756 	case -ENXIO:
1757 		vgic_present = false;
1758 		err = 0;
1759 		break;
1760 	default:
1761 		goto out;
1762 	}
1763 
1764 	/*
1765 	 * Init HYP architected timer support
1766 	 */
1767 	err = kvm_timer_hyp_init(vgic_present);
1768 	if (err)
1769 		goto out;
1770 
1771 	kvm_perf_init();
1772 	kvm_sys_reg_table_init();
1773 
1774 out:
1775 	if (err || !is_protected_kvm_enabled())
1776 		on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1777 
1778 	return err;
1779 }
1780 
1781 static void teardown_hyp_mode(void)
1782 {
1783 	int cpu;
1784 
1785 	free_hyp_pgds();
1786 	for_each_possible_cpu(cpu) {
1787 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1788 		free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1789 	}
1790 }
1791 
1792 static int do_pkvm_init(u32 hyp_va_bits)
1793 {
1794 	void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1795 	int ret;
1796 
1797 	preempt_disable();
1798 	cpu_hyp_init_context();
1799 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1800 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
1801 				hyp_va_bits);
1802 	cpu_hyp_init_features();
1803 
1804 	/*
1805 	 * The stub hypercalls are now disabled, so set our local flag to
1806 	 * prevent a later re-init attempt in kvm_arch_hardware_enable().
1807 	 */
1808 	__this_cpu_write(kvm_arm_hardware_enabled, 1);
1809 	preempt_enable();
1810 
1811 	return ret;
1812 }
1813 
1814 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1815 {
1816 	void *addr = phys_to_virt(hyp_mem_base);
1817 	int ret;
1818 
1819 	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1820 	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1821 	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
1822 	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
1823 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1824 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1825 	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
1826 
1827 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1828 	if (ret)
1829 		return ret;
1830 
1831 	ret = do_pkvm_init(hyp_va_bits);
1832 	if (ret)
1833 		return ret;
1834 
1835 	free_hyp_pgds();
1836 
1837 	return 0;
1838 }
1839 
1840 /**
1841  * Inits Hyp-mode on all online CPUs
1842  */
1843 static int init_hyp_mode(void)
1844 {
1845 	u32 hyp_va_bits;
1846 	int cpu;
1847 	int err = -ENOMEM;
1848 
1849 	/*
1850 	 * The protected Hyp-mode cannot be initialized if the memory pool
1851 	 * allocation has failed.
1852 	 */
1853 	if (is_protected_kvm_enabled() && !hyp_mem_base)
1854 		goto out_err;
1855 
1856 	/*
1857 	 * Allocate Hyp PGD and setup Hyp identity mapping
1858 	 */
1859 	err = kvm_mmu_init(&hyp_va_bits);
1860 	if (err)
1861 		goto out_err;
1862 
1863 	/*
1864 	 * Allocate stack pages for Hypervisor-mode
1865 	 */
1866 	for_each_possible_cpu(cpu) {
1867 		unsigned long stack_page;
1868 
1869 		stack_page = __get_free_page(GFP_KERNEL);
1870 		if (!stack_page) {
1871 			err = -ENOMEM;
1872 			goto out_err;
1873 		}
1874 
1875 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1876 	}
1877 
1878 	/*
1879 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1880 	 */
1881 	for_each_possible_cpu(cpu) {
1882 		struct page *page;
1883 		void *page_addr;
1884 
1885 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1886 		if (!page) {
1887 			err = -ENOMEM;
1888 			goto out_err;
1889 		}
1890 
1891 		page_addr = page_address(page);
1892 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1893 		kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1894 	}
1895 
1896 	/*
1897 	 * Map the Hyp-code called directly from the host
1898 	 */
1899 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1900 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1901 	if (err) {
1902 		kvm_err("Cannot map world-switch code\n");
1903 		goto out_err;
1904 	}
1905 
1906 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1907 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1908 	if (err) {
1909 		kvm_err("Cannot map .hyp.rodata section\n");
1910 		goto out_err;
1911 	}
1912 
1913 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1914 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1915 	if (err) {
1916 		kvm_err("Cannot map rodata section\n");
1917 		goto out_err;
1918 	}
1919 
1920 	/*
1921 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1922 	 * section thanks to an assertion in the linker script. Map it RW and
1923 	 * the rest of .bss RO.
1924 	 */
1925 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1926 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1927 	if (err) {
1928 		kvm_err("Cannot map hyp bss section: %d\n", err);
1929 		goto out_err;
1930 	}
1931 
1932 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1933 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1934 	if (err) {
1935 		kvm_err("Cannot map bss section\n");
1936 		goto out_err;
1937 	}
1938 
1939 	/*
1940 	 * Map the Hyp stack pages
1941 	 */
1942 	for_each_possible_cpu(cpu) {
1943 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1944 		err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1945 					  PAGE_HYP);
1946 
1947 		if (err) {
1948 			kvm_err("Cannot map hyp stack\n");
1949 			goto out_err;
1950 		}
1951 	}
1952 
1953 	for_each_possible_cpu(cpu) {
1954 		char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1955 		char *percpu_end = percpu_begin + nvhe_percpu_size();
1956 
1957 		/* Map Hyp percpu pages */
1958 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1959 		if (err) {
1960 			kvm_err("Cannot map hyp percpu region\n");
1961 			goto out_err;
1962 		}
1963 
1964 		/* Prepare the CPU initialization parameters */
1965 		cpu_prepare_hyp_mode(cpu);
1966 	}
1967 
1968 	if (is_protected_kvm_enabled()) {
1969 		init_cpu_logical_map();
1970 
1971 		if (!init_psci_relay()) {
1972 			err = -ENODEV;
1973 			goto out_err;
1974 		}
1975 	}
1976 
1977 	if (is_protected_kvm_enabled()) {
1978 		err = kvm_hyp_init_protection(hyp_va_bits);
1979 		if (err) {
1980 			kvm_err("Failed to init hyp memory protection\n");
1981 			goto out_err;
1982 		}
1983 	}
1984 
1985 	return 0;
1986 
1987 out_err:
1988 	teardown_hyp_mode();
1989 	kvm_err("error initializing Hyp mode: %d\n", err);
1990 	return err;
1991 }
1992 
1993 static void _kvm_host_prot_finalize(void *arg)
1994 {
1995 	int *err = arg;
1996 
1997 	if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)))
1998 		WRITE_ONCE(*err, -EINVAL);
1999 }
2000 
2001 static int pkvm_drop_host_privileges(void)
2002 {
2003 	int ret = 0;
2004 
2005 	/*
2006 	 * Flip the static key upfront as that may no longer be possible
2007 	 * once the host stage 2 is installed.
2008 	 */
2009 	static_branch_enable(&kvm_protected_mode_initialized);
2010 	on_each_cpu(_kvm_host_prot_finalize, &ret, 1);
2011 	return ret;
2012 }
2013 
2014 static int finalize_hyp_mode(void)
2015 {
2016 	if (!is_protected_kvm_enabled())
2017 		return 0;
2018 
2019 	/*
2020 	 * Exclude HYP BSS from kmemleak so that it doesn't get peeked
2021 	 * at, which would end badly once the section is inaccessible.
2022 	 * None of other sections should ever be introspected.
2023 	 */
2024 	kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
2025 	return pkvm_drop_host_privileges();
2026 }
2027 
2028 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2029 {
2030 	struct kvm_vcpu *vcpu;
2031 	int i;
2032 
2033 	mpidr &= MPIDR_HWID_BITMASK;
2034 	kvm_for_each_vcpu(i, vcpu, kvm) {
2035 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2036 			return vcpu;
2037 	}
2038 	return NULL;
2039 }
2040 
2041 bool kvm_arch_has_irq_bypass(void)
2042 {
2043 	return true;
2044 }
2045 
2046 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2047 				      struct irq_bypass_producer *prod)
2048 {
2049 	struct kvm_kernel_irqfd *irqfd =
2050 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2051 
2052 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2053 					  &irqfd->irq_entry);
2054 }
2055 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2056 				      struct irq_bypass_producer *prod)
2057 {
2058 	struct kvm_kernel_irqfd *irqfd =
2059 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2060 
2061 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2062 				     &irqfd->irq_entry);
2063 }
2064 
2065 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2066 {
2067 	struct kvm_kernel_irqfd *irqfd =
2068 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2069 
2070 	kvm_arm_halt_guest(irqfd->kvm);
2071 }
2072 
2073 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2074 {
2075 	struct kvm_kernel_irqfd *irqfd =
2076 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2077 
2078 	kvm_arm_resume_guest(irqfd->kvm);
2079 }
2080 
2081 /**
2082  * Initialize Hyp-mode and memory mappings on all CPUs.
2083  */
2084 int kvm_arch_init(void *opaque)
2085 {
2086 	int err;
2087 	bool in_hyp_mode;
2088 
2089 	if (!is_hyp_mode_available()) {
2090 		kvm_info("HYP mode not available\n");
2091 		return -ENODEV;
2092 	}
2093 
2094 	if (kvm_get_mode() == KVM_MODE_NONE) {
2095 		kvm_info("KVM disabled from command line\n");
2096 		return -ENODEV;
2097 	}
2098 
2099 	in_hyp_mode = is_kernel_in_hyp_mode();
2100 
2101 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2102 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2103 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2104 			 "Only trusted guests should be used on this system.\n");
2105 
2106 	err = kvm_set_ipa_limit();
2107 	if (err)
2108 		return err;
2109 
2110 	err = kvm_arm_init_sve();
2111 	if (err)
2112 		return err;
2113 
2114 	if (!in_hyp_mode) {
2115 		err = init_hyp_mode();
2116 		if (err)
2117 			goto out_err;
2118 	}
2119 
2120 	err = kvm_init_vector_slots();
2121 	if (err) {
2122 		kvm_err("Cannot initialise vector slots\n");
2123 		goto out_err;
2124 	}
2125 
2126 	err = init_subsystems();
2127 	if (err)
2128 		goto out_hyp;
2129 
2130 	if (!in_hyp_mode) {
2131 		err = finalize_hyp_mode();
2132 		if (err) {
2133 			kvm_err("Failed to finalize Hyp protection\n");
2134 			goto out_hyp;
2135 		}
2136 	}
2137 
2138 	if (is_protected_kvm_enabled()) {
2139 		kvm_info("Protected nVHE mode initialized successfully\n");
2140 	} else if (in_hyp_mode) {
2141 		kvm_info("VHE mode initialized successfully\n");
2142 	} else {
2143 		kvm_info("Hyp mode initialized successfully\n");
2144 	}
2145 
2146 	return 0;
2147 
2148 out_hyp:
2149 	hyp_cpu_pm_exit();
2150 	if (!in_hyp_mode)
2151 		teardown_hyp_mode();
2152 out_err:
2153 	return err;
2154 }
2155 
2156 /* NOP: Compiling as a module not supported */
2157 void kvm_arch_exit(void)
2158 {
2159 	kvm_perf_teardown();
2160 }
2161 
2162 static int __init early_kvm_mode_cfg(char *arg)
2163 {
2164 	if (!arg)
2165 		return -EINVAL;
2166 
2167 	if (strcmp(arg, "protected") == 0) {
2168 		kvm_mode = KVM_MODE_PROTECTED;
2169 		return 0;
2170 	}
2171 
2172 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2173 		kvm_mode = KVM_MODE_DEFAULT;
2174 		return 0;
2175 	}
2176 
2177 	if (strcmp(arg, "none") == 0) {
2178 		kvm_mode = KVM_MODE_NONE;
2179 		return 0;
2180 	}
2181 
2182 	return -EINVAL;
2183 }
2184 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2185 
2186 enum kvm_mode kvm_get_mode(void)
2187 {
2188 	return kvm_mode;
2189 }
2190 
2191 static int arm_init(void)
2192 {
2193 	int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2194 	return rc;
2195 }
2196 
2197 module_init(arm_init);
2198