xref: /linux/arch/arm64/kvm/arm.c (revision dec1c62e91ba268ab2a6e339d4d7a59287d5eba1)
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 static bool vgic_present;
57 
58 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
59 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
60 
61 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
62 {
63 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
64 }
65 
66 int kvm_arch_hardware_setup(void *opaque)
67 {
68 	return 0;
69 }
70 
71 int kvm_arch_check_processor_compat(void *opaque)
72 {
73 	return 0;
74 }
75 
76 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
77 			    struct kvm_enable_cap *cap)
78 {
79 	int r;
80 
81 	if (cap->flags)
82 		return -EINVAL;
83 
84 	switch (cap->cap) {
85 	case KVM_CAP_ARM_NISV_TO_USER:
86 		r = 0;
87 		set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
88 			&kvm->arch.flags);
89 		break;
90 	case KVM_CAP_ARM_MTE:
91 		mutex_lock(&kvm->lock);
92 		if (!system_supports_mte() || kvm->created_vcpus) {
93 			r = -EINVAL;
94 		} else {
95 			r = 0;
96 			set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
97 		}
98 		mutex_unlock(&kvm->lock);
99 		break;
100 	case KVM_CAP_ARM_SYSTEM_SUSPEND:
101 		r = 0;
102 		set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags);
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 = kvm_share_hyp(kvm, kvm + 1);
150 	if (ret)
151 		goto out_free_stage2_pgd;
152 
153 	if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL))
154 		goto out_free_stage2_pgd;
155 	cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
156 
157 	kvm_vgic_early_init(kvm);
158 
159 	/* The maximum number of VCPUs is limited by the host's GIC model */
160 	kvm->max_vcpus = kvm_arm_default_max_vcpus();
161 
162 	set_default_spectre(kvm);
163 	kvm_arm_init_hypercalls(kvm);
164 
165 	return ret;
166 out_free_stage2_pgd:
167 	kvm_free_stage2_pgd(&kvm->arch.mmu);
168 	return ret;
169 }
170 
171 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
172 {
173 	return VM_FAULT_SIGBUS;
174 }
175 
176 
177 /**
178  * kvm_arch_destroy_vm - destroy the VM data structure
179  * @kvm:	pointer to the KVM struct
180  */
181 void kvm_arch_destroy_vm(struct kvm *kvm)
182 {
183 	bitmap_free(kvm->arch.pmu_filter);
184 	free_cpumask_var(kvm->arch.supported_cpus);
185 
186 	kvm_vgic_destroy(kvm);
187 
188 	kvm_destroy_vcpus(kvm);
189 
190 	kvm_unshare_hyp(kvm, kvm + 1);
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 	case KVM_CAP_ARM_SYSTEM_SUSPEND:
219 		r = 1;
220 		break;
221 	case KVM_CAP_SET_GUEST_DEBUG2:
222 		return KVM_GUESTDBG_VALID_MASK;
223 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
224 		r = 1;
225 		break;
226 	case KVM_CAP_NR_VCPUS:
227 		/*
228 		 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
229 		 * architectures, as it does not always bound it to
230 		 * KVM_CAP_MAX_VCPUS. It should not matter much because
231 		 * this is just an advisory value.
232 		 */
233 		r = min_t(unsigned int, num_online_cpus(),
234 			  kvm_arm_default_max_vcpus());
235 		break;
236 	case KVM_CAP_MAX_VCPUS:
237 	case KVM_CAP_MAX_VCPU_ID:
238 		if (kvm)
239 			r = kvm->max_vcpus;
240 		else
241 			r = kvm_arm_default_max_vcpus();
242 		break;
243 	case KVM_CAP_MSI_DEVID:
244 		if (!kvm)
245 			r = -EINVAL;
246 		else
247 			r = kvm->arch.vgic.msis_require_devid;
248 		break;
249 	case KVM_CAP_ARM_USER_IRQ:
250 		/*
251 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
252 		 * (bump this number if adding more devices)
253 		 */
254 		r = 1;
255 		break;
256 	case KVM_CAP_ARM_MTE:
257 		r = system_supports_mte();
258 		break;
259 	case KVM_CAP_STEAL_TIME:
260 		r = kvm_arm_pvtime_supported();
261 		break;
262 	case KVM_CAP_ARM_EL1_32BIT:
263 		r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
264 		break;
265 	case KVM_CAP_GUEST_DEBUG_HW_BPS:
266 		r = get_num_brps();
267 		break;
268 	case KVM_CAP_GUEST_DEBUG_HW_WPS:
269 		r = get_num_wrps();
270 		break;
271 	case KVM_CAP_ARM_PMU_V3:
272 		r = kvm_arm_support_pmu_v3();
273 		break;
274 	case KVM_CAP_ARM_INJECT_SERROR_ESR:
275 		r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
276 		break;
277 	case KVM_CAP_ARM_VM_IPA_SIZE:
278 		r = get_kvm_ipa_limit();
279 		break;
280 	case KVM_CAP_ARM_SVE:
281 		r = system_supports_sve();
282 		break;
283 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
284 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
285 		r = system_has_full_ptr_auth();
286 		break;
287 	default:
288 		r = 0;
289 	}
290 
291 	return r;
292 }
293 
294 long kvm_arch_dev_ioctl(struct file *filp,
295 			unsigned int ioctl, unsigned long arg)
296 {
297 	return -EINVAL;
298 }
299 
300 struct kvm *kvm_arch_alloc_vm(void)
301 {
302 	size_t sz = sizeof(struct kvm);
303 
304 	if (!has_vhe())
305 		return kzalloc(sz, GFP_KERNEL_ACCOUNT);
306 
307 	return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
308 }
309 
310 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
311 {
312 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
313 		return -EBUSY;
314 
315 	if (id >= kvm->max_vcpus)
316 		return -EINVAL;
317 
318 	return 0;
319 }
320 
321 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
322 {
323 	int err;
324 
325 	/* Force users to call KVM_ARM_VCPU_INIT */
326 	vcpu->arch.target = -1;
327 	bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
328 
329 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
330 
331 	/* Set up the timer */
332 	kvm_timer_vcpu_init(vcpu);
333 
334 	kvm_pmu_vcpu_init(vcpu);
335 
336 	kvm_arm_reset_debug_ptr(vcpu);
337 
338 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
339 
340 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
341 
342 	err = kvm_vgic_vcpu_init(vcpu);
343 	if (err)
344 		return err;
345 
346 	return kvm_share_hyp(vcpu, vcpu + 1);
347 }
348 
349 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
350 {
351 }
352 
353 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
354 {
355 	if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
356 		static_branch_dec(&userspace_irqchip_in_use);
357 
358 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
359 	kvm_timer_vcpu_terminate(vcpu);
360 	kvm_pmu_vcpu_destroy(vcpu);
361 
362 	kvm_arm_vcpu_destroy(vcpu);
363 }
364 
365 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
366 {
367 
368 }
369 
370 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
371 {
372 
373 }
374 
375 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
376 {
377 	struct kvm_s2_mmu *mmu;
378 	int *last_ran;
379 
380 	mmu = vcpu->arch.hw_mmu;
381 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
382 
383 	/*
384 	 * We guarantee that both TLBs and I-cache are private to each
385 	 * vcpu. If detecting that a vcpu from the same VM has
386 	 * previously run on the same physical CPU, call into the
387 	 * hypervisor code to nuke the relevant contexts.
388 	 *
389 	 * We might get preempted before the vCPU actually runs, but
390 	 * over-invalidation doesn't affect correctness.
391 	 */
392 	if (*last_ran != vcpu->vcpu_id) {
393 		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
394 		*last_ran = vcpu->vcpu_id;
395 	}
396 
397 	vcpu->cpu = cpu;
398 
399 	kvm_vgic_load(vcpu);
400 	kvm_timer_vcpu_load(vcpu);
401 	if (has_vhe())
402 		kvm_vcpu_load_sysregs_vhe(vcpu);
403 	kvm_arch_vcpu_load_fp(vcpu);
404 	kvm_vcpu_pmu_restore_guest(vcpu);
405 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
406 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
407 
408 	if (single_task_running())
409 		vcpu_clear_wfx_traps(vcpu);
410 	else
411 		vcpu_set_wfx_traps(vcpu);
412 
413 	if (vcpu_has_ptrauth(vcpu))
414 		vcpu_ptrauth_disable(vcpu);
415 	kvm_arch_vcpu_load_debug_state_flags(vcpu);
416 
417 	if (!cpumask_test_cpu(smp_processor_id(), vcpu->kvm->arch.supported_cpus))
418 		vcpu_set_on_unsupported_cpu(vcpu);
419 }
420 
421 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
422 {
423 	kvm_arch_vcpu_put_debug_state_flags(vcpu);
424 	kvm_arch_vcpu_put_fp(vcpu);
425 	if (has_vhe())
426 		kvm_vcpu_put_sysregs_vhe(vcpu);
427 	kvm_timer_vcpu_put(vcpu);
428 	kvm_vgic_put(vcpu);
429 	kvm_vcpu_pmu_restore_host(vcpu);
430 	kvm_arm_vmid_clear_active();
431 
432 	vcpu_clear_on_unsupported_cpu(vcpu);
433 	vcpu->cpu = -1;
434 }
435 
436 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
437 {
438 	vcpu->arch.mp_state.mp_state = KVM_MP_STATE_STOPPED;
439 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
440 	kvm_vcpu_kick(vcpu);
441 }
442 
443 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
444 {
445 	return vcpu->arch.mp_state.mp_state == KVM_MP_STATE_STOPPED;
446 }
447 
448 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
449 {
450 	vcpu->arch.mp_state.mp_state = KVM_MP_STATE_SUSPENDED;
451 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
452 	kvm_vcpu_kick(vcpu);
453 }
454 
455 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
456 {
457 	return vcpu->arch.mp_state.mp_state == KVM_MP_STATE_SUSPENDED;
458 }
459 
460 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
461 				    struct kvm_mp_state *mp_state)
462 {
463 	*mp_state = vcpu->arch.mp_state;
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.mp_state = *mp_state;
476 		break;
477 	case KVM_MP_STATE_STOPPED:
478 		kvm_arm_vcpu_power_off(vcpu);
479 		break;
480 	case KVM_MP_STATE_SUSPENDED:
481 		kvm_arm_vcpu_suspend(vcpu);
482 		break;
483 	default:
484 		ret = -EINVAL;
485 	}
486 
487 	return ret;
488 }
489 
490 /**
491  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
492  * @v:		The VCPU pointer
493  *
494  * If the guest CPU is not waiting for interrupts or an interrupt line is
495  * asserted, the CPU is by definition runnable.
496  */
497 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
498 {
499 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
500 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
501 		&& !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
502 }
503 
504 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
505 {
506 	return vcpu_mode_priv(vcpu);
507 }
508 
509 #ifdef CONFIG_GUEST_PERF_EVENTS
510 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
511 {
512 	return *vcpu_pc(vcpu);
513 }
514 #endif
515 
516 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
517 {
518 	return vcpu->arch.target >= 0;
519 }
520 
521 /*
522  * Handle both the initialisation that is being done when the vcpu is
523  * run for the first time, as well as the updates that must be
524  * performed each time we get a new thread dealing with this vcpu.
525  */
526 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
527 {
528 	struct kvm *kvm = vcpu->kvm;
529 	int ret;
530 
531 	if (!kvm_vcpu_initialized(vcpu))
532 		return -ENOEXEC;
533 
534 	if (!kvm_arm_vcpu_is_finalized(vcpu))
535 		return -EPERM;
536 
537 	ret = kvm_arch_vcpu_run_map_fp(vcpu);
538 	if (ret)
539 		return ret;
540 
541 	if (likely(vcpu_has_run_once(vcpu)))
542 		return 0;
543 
544 	kvm_arm_vcpu_init_debug(vcpu);
545 
546 	if (likely(irqchip_in_kernel(kvm))) {
547 		/*
548 		 * Map the VGIC hardware resources before running a vcpu the
549 		 * first time on this VM.
550 		 */
551 		ret = kvm_vgic_map_resources(kvm);
552 		if (ret)
553 			return ret;
554 	}
555 
556 	ret = kvm_timer_enable(vcpu);
557 	if (ret)
558 		return ret;
559 
560 	ret = kvm_arm_pmu_v3_enable(vcpu);
561 	if (ret)
562 		return ret;
563 
564 	if (!irqchip_in_kernel(kvm)) {
565 		/*
566 		 * Tell the rest of the code that there are userspace irqchip
567 		 * VMs in the wild.
568 		 */
569 		static_branch_inc(&userspace_irqchip_in_use);
570 	}
571 
572 	/*
573 	 * Initialize traps for protected VMs.
574 	 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
575 	 * the code is in place for first run initialization at EL2.
576 	 */
577 	if (kvm_vm_is_protected(kvm))
578 		kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
579 
580 	mutex_lock(&kvm->lock);
581 	set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
582 	mutex_unlock(&kvm->lock);
583 
584 	return ret;
585 }
586 
587 bool kvm_arch_intc_initialized(struct kvm *kvm)
588 {
589 	return vgic_initialized(kvm);
590 }
591 
592 void kvm_arm_halt_guest(struct kvm *kvm)
593 {
594 	unsigned long i;
595 	struct kvm_vcpu *vcpu;
596 
597 	kvm_for_each_vcpu(i, vcpu, kvm)
598 		vcpu->arch.pause = true;
599 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
600 }
601 
602 void kvm_arm_resume_guest(struct kvm *kvm)
603 {
604 	unsigned long i;
605 	struct kvm_vcpu *vcpu;
606 
607 	kvm_for_each_vcpu(i, vcpu, kvm) {
608 		vcpu->arch.pause = false;
609 		__kvm_vcpu_wake_up(vcpu);
610 	}
611 }
612 
613 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
614 {
615 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
616 
617 	rcuwait_wait_event(wait,
618 			   (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
619 			   TASK_INTERRUPTIBLE);
620 
621 	if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
622 		/* Awaken to handle a signal, request we sleep again later. */
623 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
624 	}
625 
626 	/*
627 	 * Make sure we will observe a potential reset request if we've
628 	 * observed a change to the power state. Pairs with the smp_wmb() in
629 	 * kvm_psci_vcpu_on().
630 	 */
631 	smp_rmb();
632 }
633 
634 /**
635  * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
636  * @vcpu:	The VCPU pointer
637  *
638  * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
639  * the vCPU is runnable.  The vCPU may or may not be scheduled out, depending
640  * on when a wake event arrives, e.g. there may already be a pending wake event.
641  */
642 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
643 {
644 	/*
645 	 * Sync back the state of the GIC CPU interface so that we have
646 	 * the latest PMR and group enables. This ensures that
647 	 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
648 	 * we have pending interrupts, e.g. when determining if the
649 	 * vCPU should block.
650 	 *
651 	 * For the same reason, we want to tell GICv4 that we need
652 	 * doorbells to be signalled, should an interrupt become pending.
653 	 */
654 	preempt_disable();
655 	kvm_vgic_vmcr_sync(vcpu);
656 	vgic_v4_put(vcpu, true);
657 	preempt_enable();
658 
659 	kvm_vcpu_halt(vcpu);
660 	vcpu->arch.flags &= ~KVM_ARM64_WFIT;
661 	kvm_clear_request(KVM_REQ_UNHALT, vcpu);
662 
663 	preempt_disable();
664 	vgic_v4_load(vcpu);
665 	preempt_enable();
666 }
667 
668 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
669 {
670 	if (!kvm_arm_vcpu_suspended(vcpu))
671 		return 1;
672 
673 	kvm_vcpu_wfi(vcpu);
674 
675 	/*
676 	 * The suspend state is sticky; we do not leave it until userspace
677 	 * explicitly marks the vCPU as runnable. Request that we suspend again
678 	 * later.
679 	 */
680 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
681 
682 	/*
683 	 * Check to make sure the vCPU is actually runnable. If so, exit to
684 	 * userspace informing it of the wakeup condition.
685 	 */
686 	if (kvm_arch_vcpu_runnable(vcpu)) {
687 		memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
688 		vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
689 		vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
690 		return 0;
691 	}
692 
693 	/*
694 	 * Otherwise, we were unblocked to process a different event, such as a
695 	 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
696 	 * process the event.
697 	 */
698 	return 1;
699 }
700 
701 /**
702  * check_vcpu_requests - check and handle pending vCPU requests
703  * @vcpu:	the VCPU pointer
704  *
705  * Return: 1 if we should enter the guest
706  *	   0 if we should exit to userspace
707  *	   < 0 if we should exit to userspace, where the return value indicates
708  *	   an error
709  */
710 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
711 {
712 	if (kvm_request_pending(vcpu)) {
713 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
714 			kvm_vcpu_sleep(vcpu);
715 
716 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
717 			kvm_reset_vcpu(vcpu);
718 
719 		/*
720 		 * Clear IRQ_PENDING requests that were made to guarantee
721 		 * that a VCPU sees new virtual interrupts.
722 		 */
723 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
724 
725 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
726 			kvm_update_stolen_time(vcpu);
727 
728 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
729 			/* The distributor enable bits were changed */
730 			preempt_disable();
731 			vgic_v4_put(vcpu, false);
732 			vgic_v4_load(vcpu);
733 			preempt_enable();
734 		}
735 
736 		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
737 			kvm_pmu_handle_pmcr(vcpu,
738 					    __vcpu_sys_reg(vcpu, PMCR_EL0));
739 
740 		if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
741 			return kvm_vcpu_suspend(vcpu);
742 	}
743 
744 	return 1;
745 }
746 
747 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
748 {
749 	if (likely(!vcpu_mode_is_32bit(vcpu)))
750 		return false;
751 
752 	return !system_supports_32bit_el0() ||
753 		static_branch_unlikely(&arm64_mismatched_32bit_el0);
754 }
755 
756 /**
757  * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
758  * @vcpu:	The VCPU pointer
759  * @ret:	Pointer to write optional return code
760  *
761  * Returns: true if the VCPU needs to return to a preemptible + interruptible
762  *	    and skip guest entry.
763  *
764  * This function disambiguates between two different types of exits: exits to a
765  * preemptible + interruptible kernel context and exits to userspace. For an
766  * exit to userspace, this function will write the return code to ret and return
767  * true. For an exit to preemptible + interruptible kernel context (i.e. check
768  * for pending work and re-enter), return true without writing to ret.
769  */
770 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
771 {
772 	struct kvm_run *run = vcpu->run;
773 
774 	/*
775 	 * If we're using a userspace irqchip, then check if we need
776 	 * to tell a userspace irqchip about timer or PMU level
777 	 * changes and if so, exit to userspace (the actual level
778 	 * state gets updated in kvm_timer_update_run and
779 	 * kvm_pmu_update_run below).
780 	 */
781 	if (static_branch_unlikely(&userspace_irqchip_in_use)) {
782 		if (kvm_timer_should_notify_user(vcpu) ||
783 		    kvm_pmu_should_notify_user(vcpu)) {
784 			*ret = -EINTR;
785 			run->exit_reason = KVM_EXIT_INTR;
786 			return true;
787 		}
788 	}
789 
790 	if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
791 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
792 		run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
793 		run->fail_entry.cpu = smp_processor_id();
794 		*ret = 0;
795 		return true;
796 	}
797 
798 	return kvm_request_pending(vcpu) ||
799 			xfer_to_guest_mode_work_pending();
800 }
801 
802 /*
803  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
804  * the vCPU is running.
805  *
806  * This must be noinstr as instrumentation may make use of RCU, and this is not
807  * safe during the EQS.
808  */
809 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
810 {
811 	int ret;
812 
813 	guest_state_enter_irqoff();
814 	ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
815 	guest_state_exit_irqoff();
816 
817 	return ret;
818 }
819 
820 /**
821  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
822  * @vcpu:	The VCPU pointer
823  *
824  * This function is called through the VCPU_RUN ioctl called from user space. It
825  * will execute VM code in a loop until the time slice for the process is used
826  * or some emulation is needed from user space in which case the function will
827  * return with return value 0 and with the kvm_run structure filled in with the
828  * required data for the requested emulation.
829  */
830 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
831 {
832 	struct kvm_run *run = vcpu->run;
833 	int ret;
834 
835 	if (run->exit_reason == KVM_EXIT_MMIO) {
836 		ret = kvm_handle_mmio_return(vcpu);
837 		if (ret)
838 			return ret;
839 	}
840 
841 	vcpu_load(vcpu);
842 
843 	if (run->immediate_exit) {
844 		ret = -EINTR;
845 		goto out;
846 	}
847 
848 	kvm_sigset_activate(vcpu);
849 
850 	ret = 1;
851 	run->exit_reason = KVM_EXIT_UNKNOWN;
852 	run->flags = 0;
853 	while (ret > 0) {
854 		/*
855 		 * Check conditions before entering the guest
856 		 */
857 		ret = xfer_to_guest_mode_handle_work(vcpu);
858 		if (!ret)
859 			ret = 1;
860 
861 		if (ret > 0)
862 			ret = check_vcpu_requests(vcpu);
863 
864 		/*
865 		 * Preparing the interrupts to be injected also
866 		 * involves poking the GIC, which must be done in a
867 		 * non-preemptible context.
868 		 */
869 		preempt_disable();
870 
871 		/*
872 		 * The VMID allocator only tracks active VMIDs per
873 		 * physical CPU, and therefore the VMID allocated may not be
874 		 * preserved on VMID roll-over if the task was preempted,
875 		 * making a thread's VMID inactive. So we need to call
876 		 * kvm_arm_vmid_update() in non-premptible context.
877 		 */
878 		kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid);
879 
880 		kvm_pmu_flush_hwstate(vcpu);
881 
882 		local_irq_disable();
883 
884 		kvm_vgic_flush_hwstate(vcpu);
885 
886 		kvm_pmu_update_vcpu_events(vcpu);
887 
888 		/*
889 		 * Ensure we set mode to IN_GUEST_MODE after we disable
890 		 * interrupts and before the final VCPU requests check.
891 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
892 		 * Documentation/virt/kvm/vcpu-requests.rst
893 		 */
894 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
895 
896 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
897 			vcpu->mode = OUTSIDE_GUEST_MODE;
898 			isb(); /* Ensure work in x_flush_hwstate is committed */
899 			kvm_pmu_sync_hwstate(vcpu);
900 			if (static_branch_unlikely(&userspace_irqchip_in_use))
901 				kvm_timer_sync_user(vcpu);
902 			kvm_vgic_sync_hwstate(vcpu);
903 			local_irq_enable();
904 			preempt_enable();
905 			continue;
906 		}
907 
908 		kvm_arm_setup_debug(vcpu);
909 		kvm_arch_vcpu_ctxflush_fp(vcpu);
910 
911 		/**************************************************************
912 		 * Enter the guest
913 		 */
914 		trace_kvm_entry(*vcpu_pc(vcpu));
915 		guest_timing_enter_irqoff();
916 
917 		ret = kvm_arm_vcpu_enter_exit(vcpu);
918 
919 		vcpu->mode = OUTSIDE_GUEST_MODE;
920 		vcpu->stat.exits++;
921 		/*
922 		 * Back from guest
923 		 *************************************************************/
924 
925 		kvm_arm_clear_debug(vcpu);
926 
927 		/*
928 		 * We must sync the PMU state before the vgic state so
929 		 * that the vgic can properly sample the updated state of the
930 		 * interrupt line.
931 		 */
932 		kvm_pmu_sync_hwstate(vcpu);
933 
934 		/*
935 		 * Sync the vgic state before syncing the timer state because
936 		 * the timer code needs to know if the virtual timer
937 		 * interrupts are active.
938 		 */
939 		kvm_vgic_sync_hwstate(vcpu);
940 
941 		/*
942 		 * Sync the timer hardware state before enabling interrupts as
943 		 * we don't want vtimer interrupts to race with syncing the
944 		 * timer virtual interrupt state.
945 		 */
946 		if (static_branch_unlikely(&userspace_irqchip_in_use))
947 			kvm_timer_sync_user(vcpu);
948 
949 		kvm_arch_vcpu_ctxsync_fp(vcpu);
950 
951 		/*
952 		 * We must ensure that any pending interrupts are taken before
953 		 * we exit guest timing so that timer ticks are accounted as
954 		 * guest time. Transiently unmask interrupts so that any
955 		 * pending interrupts are taken.
956 		 *
957 		 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
958 		 * context synchronization event) is necessary to ensure that
959 		 * pending interrupts are taken.
960 		 */
961 		if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
962 			local_irq_enable();
963 			isb();
964 			local_irq_disable();
965 		}
966 
967 		guest_timing_exit_irqoff();
968 
969 		local_irq_enable();
970 
971 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
972 
973 		/* Exit types that need handling before we can be preempted */
974 		handle_exit_early(vcpu, ret);
975 
976 		preempt_enable();
977 
978 		/*
979 		 * The ARMv8 architecture doesn't give the hypervisor
980 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
981 		 * if implemented by the CPU. If we spot the guest in such
982 		 * state and that we decided it wasn't supposed to do so (like
983 		 * with the asymmetric AArch32 case), return to userspace with
984 		 * a fatal error.
985 		 */
986 		if (vcpu_mode_is_bad_32bit(vcpu)) {
987 			/*
988 			 * As we have caught the guest red-handed, decide that
989 			 * it isn't fit for purpose anymore by making the vcpu
990 			 * invalid. The VMM can try and fix it by issuing  a
991 			 * KVM_ARM_VCPU_INIT if it really wants to.
992 			 */
993 			vcpu->arch.target = -1;
994 			ret = ARM_EXCEPTION_IL;
995 		}
996 
997 		ret = handle_exit(vcpu, ret);
998 	}
999 
1000 	/* Tell userspace about in-kernel device output levels */
1001 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1002 		kvm_timer_update_run(vcpu);
1003 		kvm_pmu_update_run(vcpu);
1004 	}
1005 
1006 	kvm_sigset_deactivate(vcpu);
1007 
1008 out:
1009 	/*
1010 	 * In the unlikely event that we are returning to userspace
1011 	 * with pending exceptions or PC adjustment, commit these
1012 	 * adjustments in order to give userspace a consistent view of
1013 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1014 	 * being preempt-safe on VHE.
1015 	 */
1016 	if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
1017 					 KVM_ARM64_INCREMENT_PC)))
1018 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
1019 
1020 	vcpu_put(vcpu);
1021 	return ret;
1022 }
1023 
1024 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1025 {
1026 	int bit_index;
1027 	bool set;
1028 	unsigned long *hcr;
1029 
1030 	if (number == KVM_ARM_IRQ_CPU_IRQ)
1031 		bit_index = __ffs(HCR_VI);
1032 	else /* KVM_ARM_IRQ_CPU_FIQ */
1033 		bit_index = __ffs(HCR_VF);
1034 
1035 	hcr = vcpu_hcr(vcpu);
1036 	if (level)
1037 		set = test_and_set_bit(bit_index, hcr);
1038 	else
1039 		set = test_and_clear_bit(bit_index, hcr);
1040 
1041 	/*
1042 	 * If we didn't change anything, no need to wake up or kick other CPUs
1043 	 */
1044 	if (set == level)
1045 		return 0;
1046 
1047 	/*
1048 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1049 	 * trigger a world-switch round on the running physical CPU to set the
1050 	 * virtual IRQ/FIQ fields in the HCR appropriately.
1051 	 */
1052 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1053 	kvm_vcpu_kick(vcpu);
1054 
1055 	return 0;
1056 }
1057 
1058 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1059 			  bool line_status)
1060 {
1061 	u32 irq = irq_level->irq;
1062 	unsigned int irq_type, vcpu_idx, irq_num;
1063 	int nrcpus = atomic_read(&kvm->online_vcpus);
1064 	struct kvm_vcpu *vcpu = NULL;
1065 	bool level = irq_level->level;
1066 
1067 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1068 	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1069 	vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1070 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1071 
1072 	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1073 
1074 	switch (irq_type) {
1075 	case KVM_ARM_IRQ_TYPE_CPU:
1076 		if (irqchip_in_kernel(kvm))
1077 			return -ENXIO;
1078 
1079 		if (vcpu_idx >= nrcpus)
1080 			return -EINVAL;
1081 
1082 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1083 		if (!vcpu)
1084 			return -EINVAL;
1085 
1086 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1087 			return -EINVAL;
1088 
1089 		return vcpu_interrupt_line(vcpu, irq_num, level);
1090 	case KVM_ARM_IRQ_TYPE_PPI:
1091 		if (!irqchip_in_kernel(kvm))
1092 			return -ENXIO;
1093 
1094 		if (vcpu_idx >= nrcpus)
1095 			return -EINVAL;
1096 
1097 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1098 		if (!vcpu)
1099 			return -EINVAL;
1100 
1101 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1102 			return -EINVAL;
1103 
1104 		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1105 	case KVM_ARM_IRQ_TYPE_SPI:
1106 		if (!irqchip_in_kernel(kvm))
1107 			return -ENXIO;
1108 
1109 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1110 			return -EINVAL;
1111 
1112 		return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1113 	}
1114 
1115 	return -EINVAL;
1116 }
1117 
1118 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1119 			       const struct kvm_vcpu_init *init)
1120 {
1121 	unsigned int i, ret;
1122 	u32 phys_target = kvm_target_cpu();
1123 
1124 	if (init->target != phys_target)
1125 		return -EINVAL;
1126 
1127 	/*
1128 	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1129 	 * use the same target.
1130 	 */
1131 	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1132 		return -EINVAL;
1133 
1134 	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1135 	for (i = 0; i < sizeof(init->features) * 8; i++) {
1136 		bool set = (init->features[i / 32] & (1 << (i % 32)));
1137 
1138 		if (set && i >= KVM_VCPU_MAX_FEATURES)
1139 			return -ENOENT;
1140 
1141 		/*
1142 		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1143 		 * use the same feature set.
1144 		 */
1145 		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1146 		    test_bit(i, vcpu->arch.features) != set)
1147 			return -EINVAL;
1148 
1149 		if (set)
1150 			set_bit(i, vcpu->arch.features);
1151 	}
1152 
1153 	vcpu->arch.target = phys_target;
1154 
1155 	/* Now we know what it is, we can reset it. */
1156 	ret = kvm_reset_vcpu(vcpu);
1157 	if (ret) {
1158 		vcpu->arch.target = -1;
1159 		bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1160 	}
1161 
1162 	return ret;
1163 }
1164 
1165 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1166 					 struct kvm_vcpu_init *init)
1167 {
1168 	int ret;
1169 
1170 	ret = kvm_vcpu_set_target(vcpu, init);
1171 	if (ret)
1172 		return ret;
1173 
1174 	/*
1175 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1176 	 * guest MMU is turned off and flush the caches as needed.
1177 	 *
1178 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1179 	 * ensuring that the data side is always coherent. We still
1180 	 * need to invalidate the I-cache though, as FWB does *not*
1181 	 * imply CTR_EL0.DIC.
1182 	 */
1183 	if (vcpu_has_run_once(vcpu)) {
1184 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1185 			stage2_unmap_vm(vcpu->kvm);
1186 		else
1187 			icache_inval_all_pou();
1188 	}
1189 
1190 	vcpu_reset_hcr(vcpu);
1191 	vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1192 
1193 	/*
1194 	 * Handle the "start in power-off" case.
1195 	 */
1196 	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1197 		kvm_arm_vcpu_power_off(vcpu);
1198 	else
1199 		vcpu->arch.mp_state.mp_state = KVM_MP_STATE_RUNNABLE;
1200 
1201 	return 0;
1202 }
1203 
1204 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1205 				 struct kvm_device_attr *attr)
1206 {
1207 	int ret = -ENXIO;
1208 
1209 	switch (attr->group) {
1210 	default:
1211 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1212 		break;
1213 	}
1214 
1215 	return ret;
1216 }
1217 
1218 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1219 				 struct kvm_device_attr *attr)
1220 {
1221 	int ret = -ENXIO;
1222 
1223 	switch (attr->group) {
1224 	default:
1225 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1226 		break;
1227 	}
1228 
1229 	return ret;
1230 }
1231 
1232 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1233 				 struct kvm_device_attr *attr)
1234 {
1235 	int ret = -ENXIO;
1236 
1237 	switch (attr->group) {
1238 	default:
1239 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1240 		break;
1241 	}
1242 
1243 	return ret;
1244 }
1245 
1246 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1247 				   struct kvm_vcpu_events *events)
1248 {
1249 	memset(events, 0, sizeof(*events));
1250 
1251 	return __kvm_arm_vcpu_get_events(vcpu, events);
1252 }
1253 
1254 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1255 				   struct kvm_vcpu_events *events)
1256 {
1257 	int i;
1258 
1259 	/* check whether the reserved field is zero */
1260 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1261 		if (events->reserved[i])
1262 			return -EINVAL;
1263 
1264 	/* check whether the pad field is zero */
1265 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1266 		if (events->exception.pad[i])
1267 			return -EINVAL;
1268 
1269 	return __kvm_arm_vcpu_set_events(vcpu, events);
1270 }
1271 
1272 long kvm_arch_vcpu_ioctl(struct file *filp,
1273 			 unsigned int ioctl, unsigned long arg)
1274 {
1275 	struct kvm_vcpu *vcpu = filp->private_data;
1276 	void __user *argp = (void __user *)arg;
1277 	struct kvm_device_attr attr;
1278 	long r;
1279 
1280 	switch (ioctl) {
1281 	case KVM_ARM_VCPU_INIT: {
1282 		struct kvm_vcpu_init init;
1283 
1284 		r = -EFAULT;
1285 		if (copy_from_user(&init, argp, sizeof(init)))
1286 			break;
1287 
1288 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1289 		break;
1290 	}
1291 	case KVM_SET_ONE_REG:
1292 	case KVM_GET_ONE_REG: {
1293 		struct kvm_one_reg reg;
1294 
1295 		r = -ENOEXEC;
1296 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1297 			break;
1298 
1299 		r = -EFAULT;
1300 		if (copy_from_user(&reg, argp, sizeof(reg)))
1301 			break;
1302 
1303 		/*
1304 		 * We could owe a reset due to PSCI. Handle the pending reset
1305 		 * here to ensure userspace register accesses are ordered after
1306 		 * the reset.
1307 		 */
1308 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1309 			kvm_reset_vcpu(vcpu);
1310 
1311 		if (ioctl == KVM_SET_ONE_REG)
1312 			r = kvm_arm_set_reg(vcpu, &reg);
1313 		else
1314 			r = kvm_arm_get_reg(vcpu, &reg);
1315 		break;
1316 	}
1317 	case KVM_GET_REG_LIST: {
1318 		struct kvm_reg_list __user *user_list = argp;
1319 		struct kvm_reg_list reg_list;
1320 		unsigned n;
1321 
1322 		r = -ENOEXEC;
1323 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1324 			break;
1325 
1326 		r = -EPERM;
1327 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1328 			break;
1329 
1330 		r = -EFAULT;
1331 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1332 			break;
1333 		n = reg_list.n;
1334 		reg_list.n = kvm_arm_num_regs(vcpu);
1335 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1336 			break;
1337 		r = -E2BIG;
1338 		if (n < reg_list.n)
1339 			break;
1340 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1341 		break;
1342 	}
1343 	case KVM_SET_DEVICE_ATTR: {
1344 		r = -EFAULT;
1345 		if (copy_from_user(&attr, argp, sizeof(attr)))
1346 			break;
1347 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1348 		break;
1349 	}
1350 	case KVM_GET_DEVICE_ATTR: {
1351 		r = -EFAULT;
1352 		if (copy_from_user(&attr, argp, sizeof(attr)))
1353 			break;
1354 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1355 		break;
1356 	}
1357 	case KVM_HAS_DEVICE_ATTR: {
1358 		r = -EFAULT;
1359 		if (copy_from_user(&attr, argp, sizeof(attr)))
1360 			break;
1361 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1362 		break;
1363 	}
1364 	case KVM_GET_VCPU_EVENTS: {
1365 		struct kvm_vcpu_events events;
1366 
1367 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1368 			return -EINVAL;
1369 
1370 		if (copy_to_user(argp, &events, sizeof(events)))
1371 			return -EFAULT;
1372 
1373 		return 0;
1374 	}
1375 	case KVM_SET_VCPU_EVENTS: {
1376 		struct kvm_vcpu_events events;
1377 
1378 		if (copy_from_user(&events, argp, sizeof(events)))
1379 			return -EFAULT;
1380 
1381 		return kvm_arm_vcpu_set_events(vcpu, &events);
1382 	}
1383 	case KVM_ARM_VCPU_FINALIZE: {
1384 		int what;
1385 
1386 		if (!kvm_vcpu_initialized(vcpu))
1387 			return -ENOEXEC;
1388 
1389 		if (get_user(what, (const int __user *)argp))
1390 			return -EFAULT;
1391 
1392 		return kvm_arm_vcpu_finalize(vcpu, what);
1393 	}
1394 	default:
1395 		r = -EINVAL;
1396 	}
1397 
1398 	return r;
1399 }
1400 
1401 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1402 {
1403 
1404 }
1405 
1406 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1407 					const struct kvm_memory_slot *memslot)
1408 {
1409 	kvm_flush_remote_tlbs(kvm);
1410 }
1411 
1412 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1413 					struct kvm_arm_device_addr *dev_addr)
1414 {
1415 	unsigned long dev_id, type;
1416 
1417 	dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1418 		KVM_ARM_DEVICE_ID_SHIFT;
1419 	type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1420 		KVM_ARM_DEVICE_TYPE_SHIFT;
1421 
1422 	switch (dev_id) {
1423 	case KVM_ARM_DEVICE_VGIC_V2:
1424 		if (!vgic_present)
1425 			return -ENXIO;
1426 		return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1427 	default:
1428 		return -ENODEV;
1429 	}
1430 }
1431 
1432 long kvm_arch_vm_ioctl(struct file *filp,
1433 		       unsigned int ioctl, unsigned long arg)
1434 {
1435 	struct kvm *kvm = filp->private_data;
1436 	void __user *argp = (void __user *)arg;
1437 
1438 	switch (ioctl) {
1439 	case KVM_CREATE_IRQCHIP: {
1440 		int ret;
1441 		if (!vgic_present)
1442 			return -ENXIO;
1443 		mutex_lock(&kvm->lock);
1444 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1445 		mutex_unlock(&kvm->lock);
1446 		return ret;
1447 	}
1448 	case KVM_ARM_SET_DEVICE_ADDR: {
1449 		struct kvm_arm_device_addr dev_addr;
1450 
1451 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1452 			return -EFAULT;
1453 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1454 	}
1455 	case KVM_ARM_PREFERRED_TARGET: {
1456 		struct kvm_vcpu_init init;
1457 
1458 		kvm_vcpu_preferred_target(&init);
1459 
1460 		if (copy_to_user(argp, &init, sizeof(init)))
1461 			return -EFAULT;
1462 
1463 		return 0;
1464 	}
1465 	case KVM_ARM_MTE_COPY_TAGS: {
1466 		struct kvm_arm_copy_mte_tags copy_tags;
1467 
1468 		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1469 			return -EFAULT;
1470 		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1471 	}
1472 	default:
1473 		return -EINVAL;
1474 	}
1475 }
1476 
1477 static unsigned long nvhe_percpu_size(void)
1478 {
1479 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1480 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1481 }
1482 
1483 static unsigned long nvhe_percpu_order(void)
1484 {
1485 	unsigned long size = nvhe_percpu_size();
1486 
1487 	return size ? get_order(size) : 0;
1488 }
1489 
1490 /* A lookup table holding the hypervisor VA for each vector slot */
1491 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1492 
1493 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1494 {
1495 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1496 }
1497 
1498 static int kvm_init_vector_slots(void)
1499 {
1500 	int err;
1501 	void *base;
1502 
1503 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1504 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1505 
1506 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1507 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1508 
1509 	if (kvm_system_needs_idmapped_vectors() &&
1510 	    !is_protected_kvm_enabled()) {
1511 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1512 					       __BP_HARDEN_HYP_VECS_SZ, &base);
1513 		if (err)
1514 			return err;
1515 	}
1516 
1517 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1518 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1519 	return 0;
1520 }
1521 
1522 static void cpu_prepare_hyp_mode(int cpu)
1523 {
1524 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1525 	unsigned long tcr;
1526 
1527 	/*
1528 	 * Calculate the raw per-cpu offset without a translation from the
1529 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1530 	 * so that we can use adr_l to access per-cpu variables in EL2.
1531 	 * Also drop the KASAN tag which gets in the way...
1532 	 */
1533 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1534 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1535 
1536 	params->mair_el2 = read_sysreg(mair_el1);
1537 
1538 	/*
1539 	 * The ID map may be configured to use an extended virtual address
1540 	 * range. This is only the case if system RAM is out of range for the
1541 	 * currently configured page size and VA_BITS, in which case we will
1542 	 * also need the extended virtual range for the HYP ID map, or we won't
1543 	 * be able to enable the EL2 MMU.
1544 	 *
1545 	 * However, at EL2, there is only one TTBR register, and we can't switch
1546 	 * between translation tables *and* update TCR_EL2.T0SZ at the same
1547 	 * time. Bottom line: we need to use the extended range with *both* our
1548 	 * translation tables.
1549 	 *
1550 	 * So use the same T0SZ value we use for the ID map.
1551 	 */
1552 	tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1553 	tcr &= ~TCR_T0SZ_MASK;
1554 	tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1555 	params->tcr_el2 = tcr;
1556 
1557 	params->pgd_pa = kvm_mmu_get_httbr();
1558 	if (is_protected_kvm_enabled())
1559 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1560 	else
1561 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1562 	params->vttbr = params->vtcr = 0;
1563 
1564 	/*
1565 	 * Flush the init params from the data cache because the struct will
1566 	 * be read while the MMU is off.
1567 	 */
1568 	kvm_flush_dcache_to_poc(params, sizeof(*params));
1569 }
1570 
1571 static void hyp_install_host_vector(void)
1572 {
1573 	struct kvm_nvhe_init_params *params;
1574 	struct arm_smccc_res res;
1575 
1576 	/* Switch from the HYP stub to our own HYP init vector */
1577 	__hyp_set_vectors(kvm_get_idmap_vector());
1578 
1579 	/*
1580 	 * Call initialization code, and switch to the full blown HYP code.
1581 	 * If the cpucaps haven't been finalized yet, something has gone very
1582 	 * wrong, and hyp will crash and burn when it uses any
1583 	 * cpus_have_const_cap() wrapper.
1584 	 */
1585 	BUG_ON(!system_capabilities_finalized());
1586 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1587 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1588 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1589 }
1590 
1591 static void cpu_init_hyp_mode(void)
1592 {
1593 	hyp_install_host_vector();
1594 
1595 	/*
1596 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1597 	 * at EL2.
1598 	 */
1599 	if (this_cpu_has_cap(ARM64_SSBS) &&
1600 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1601 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1602 	}
1603 }
1604 
1605 static void cpu_hyp_reset(void)
1606 {
1607 	if (!is_kernel_in_hyp_mode())
1608 		__hyp_reset_vectors();
1609 }
1610 
1611 /*
1612  * EL2 vectors can be mapped and rerouted in a number of ways,
1613  * depending on the kernel configuration and CPU present:
1614  *
1615  * - If the CPU is affected by Spectre-v2, the hardening sequence is
1616  *   placed in one of the vector slots, which is executed before jumping
1617  *   to the real vectors.
1618  *
1619  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1620  *   containing the hardening sequence is mapped next to the idmap page,
1621  *   and executed before jumping to the real vectors.
1622  *
1623  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1624  *   empty slot is selected, mapped next to the idmap page, and
1625  *   executed before jumping to the real vectors.
1626  *
1627  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1628  * VHE, as we don't have hypervisor-specific mappings. If the system
1629  * is VHE and yet selects this capability, it will be ignored.
1630  */
1631 static void cpu_set_hyp_vector(void)
1632 {
1633 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1634 	void *vector = hyp_spectre_vector_selector[data->slot];
1635 
1636 	if (!is_protected_kvm_enabled())
1637 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1638 	else
1639 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1640 }
1641 
1642 static void cpu_hyp_init_context(void)
1643 {
1644 	kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1645 
1646 	if (!is_kernel_in_hyp_mode())
1647 		cpu_init_hyp_mode();
1648 }
1649 
1650 static void cpu_hyp_init_features(void)
1651 {
1652 	cpu_set_hyp_vector();
1653 	kvm_arm_init_debug();
1654 
1655 	if (is_kernel_in_hyp_mode())
1656 		kvm_timer_init_vhe();
1657 
1658 	if (vgic_present)
1659 		kvm_vgic_init_cpu_hardware();
1660 }
1661 
1662 static void cpu_hyp_reinit(void)
1663 {
1664 	cpu_hyp_reset();
1665 	cpu_hyp_init_context();
1666 	cpu_hyp_init_features();
1667 }
1668 
1669 static void _kvm_arch_hardware_enable(void *discard)
1670 {
1671 	if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1672 		cpu_hyp_reinit();
1673 		__this_cpu_write(kvm_arm_hardware_enabled, 1);
1674 	}
1675 }
1676 
1677 int kvm_arch_hardware_enable(void)
1678 {
1679 	_kvm_arch_hardware_enable(NULL);
1680 	return 0;
1681 }
1682 
1683 static void _kvm_arch_hardware_disable(void *discard)
1684 {
1685 	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1686 		cpu_hyp_reset();
1687 		__this_cpu_write(kvm_arm_hardware_enabled, 0);
1688 	}
1689 }
1690 
1691 void kvm_arch_hardware_disable(void)
1692 {
1693 	if (!is_protected_kvm_enabled())
1694 		_kvm_arch_hardware_disable(NULL);
1695 }
1696 
1697 #ifdef CONFIG_CPU_PM
1698 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1699 				    unsigned long cmd,
1700 				    void *v)
1701 {
1702 	/*
1703 	 * kvm_arm_hardware_enabled is left with its old value over
1704 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1705 	 * re-enable hyp.
1706 	 */
1707 	switch (cmd) {
1708 	case CPU_PM_ENTER:
1709 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1710 			/*
1711 			 * don't update kvm_arm_hardware_enabled here
1712 			 * so that the hardware will be re-enabled
1713 			 * when we resume. See below.
1714 			 */
1715 			cpu_hyp_reset();
1716 
1717 		return NOTIFY_OK;
1718 	case CPU_PM_ENTER_FAILED:
1719 	case CPU_PM_EXIT:
1720 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1721 			/* The hardware was enabled before suspend. */
1722 			cpu_hyp_reinit();
1723 
1724 		return NOTIFY_OK;
1725 
1726 	default:
1727 		return NOTIFY_DONE;
1728 	}
1729 }
1730 
1731 static struct notifier_block hyp_init_cpu_pm_nb = {
1732 	.notifier_call = hyp_init_cpu_pm_notifier,
1733 };
1734 
1735 static void hyp_cpu_pm_init(void)
1736 {
1737 	if (!is_protected_kvm_enabled())
1738 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1739 }
1740 static void hyp_cpu_pm_exit(void)
1741 {
1742 	if (!is_protected_kvm_enabled())
1743 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1744 }
1745 #else
1746 static inline void hyp_cpu_pm_init(void)
1747 {
1748 }
1749 static inline void hyp_cpu_pm_exit(void)
1750 {
1751 }
1752 #endif
1753 
1754 static void init_cpu_logical_map(void)
1755 {
1756 	unsigned int cpu;
1757 
1758 	/*
1759 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1760 	 * Only copy the set of online CPUs whose features have been checked
1761 	 * against the finalized system capabilities. The hypervisor will not
1762 	 * allow any other CPUs from the `possible` set to boot.
1763 	 */
1764 	for_each_online_cpu(cpu)
1765 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1766 }
1767 
1768 #define init_psci_0_1_impl_state(config, what)	\
1769 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
1770 
1771 static bool init_psci_relay(void)
1772 {
1773 	/*
1774 	 * If PSCI has not been initialized, protected KVM cannot install
1775 	 * itself on newly booted CPUs.
1776 	 */
1777 	if (!psci_ops.get_version) {
1778 		kvm_err("Cannot initialize protected mode without PSCI\n");
1779 		return false;
1780 	}
1781 
1782 	kvm_host_psci_config.version = psci_ops.get_version();
1783 
1784 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1785 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1786 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1787 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1788 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1789 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1790 	}
1791 	return true;
1792 }
1793 
1794 static int init_subsystems(void)
1795 {
1796 	int err = 0;
1797 
1798 	/*
1799 	 * Enable hardware so that subsystem initialisation can access EL2.
1800 	 */
1801 	on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1802 
1803 	/*
1804 	 * Register CPU lower-power notifier
1805 	 */
1806 	hyp_cpu_pm_init();
1807 
1808 	/*
1809 	 * Init HYP view of VGIC
1810 	 */
1811 	err = kvm_vgic_hyp_init();
1812 	switch (err) {
1813 	case 0:
1814 		vgic_present = true;
1815 		break;
1816 	case -ENODEV:
1817 	case -ENXIO:
1818 		vgic_present = false;
1819 		err = 0;
1820 		break;
1821 	default:
1822 		goto out;
1823 	}
1824 
1825 	/*
1826 	 * Init HYP architected timer support
1827 	 */
1828 	err = kvm_timer_hyp_init(vgic_present);
1829 	if (err)
1830 		goto out;
1831 
1832 	kvm_register_perf_callbacks(NULL);
1833 
1834 out:
1835 	if (err || !is_protected_kvm_enabled())
1836 		on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1837 
1838 	return err;
1839 }
1840 
1841 static void teardown_hyp_mode(void)
1842 {
1843 	int cpu;
1844 
1845 	free_hyp_pgds();
1846 	for_each_possible_cpu(cpu) {
1847 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1848 		free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1849 	}
1850 }
1851 
1852 static int do_pkvm_init(u32 hyp_va_bits)
1853 {
1854 	void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1855 	int ret;
1856 
1857 	preempt_disable();
1858 	cpu_hyp_init_context();
1859 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1860 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
1861 				hyp_va_bits);
1862 	cpu_hyp_init_features();
1863 
1864 	/*
1865 	 * The stub hypercalls are now disabled, so set our local flag to
1866 	 * prevent a later re-init attempt in kvm_arch_hardware_enable().
1867 	 */
1868 	__this_cpu_write(kvm_arm_hardware_enabled, 1);
1869 	preempt_enable();
1870 
1871 	return ret;
1872 }
1873 
1874 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1875 {
1876 	void *addr = phys_to_virt(hyp_mem_base);
1877 	int ret;
1878 
1879 	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1880 	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1881 	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
1882 	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
1883 	kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
1884 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1885 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1886 	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
1887 
1888 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1889 	if (ret)
1890 		return ret;
1891 
1892 	ret = do_pkvm_init(hyp_va_bits);
1893 	if (ret)
1894 		return ret;
1895 
1896 	free_hyp_pgds();
1897 
1898 	return 0;
1899 }
1900 
1901 /**
1902  * Inits Hyp-mode on all online CPUs
1903  */
1904 static int init_hyp_mode(void)
1905 {
1906 	u32 hyp_va_bits;
1907 	int cpu;
1908 	int err = -ENOMEM;
1909 
1910 	/*
1911 	 * The protected Hyp-mode cannot be initialized if the memory pool
1912 	 * allocation has failed.
1913 	 */
1914 	if (is_protected_kvm_enabled() && !hyp_mem_base)
1915 		goto out_err;
1916 
1917 	/*
1918 	 * Allocate Hyp PGD and setup Hyp identity mapping
1919 	 */
1920 	err = kvm_mmu_init(&hyp_va_bits);
1921 	if (err)
1922 		goto out_err;
1923 
1924 	/*
1925 	 * Allocate stack pages for Hypervisor-mode
1926 	 */
1927 	for_each_possible_cpu(cpu) {
1928 		unsigned long stack_page;
1929 
1930 		stack_page = __get_free_page(GFP_KERNEL);
1931 		if (!stack_page) {
1932 			err = -ENOMEM;
1933 			goto out_err;
1934 		}
1935 
1936 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1937 	}
1938 
1939 	/*
1940 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1941 	 */
1942 	for_each_possible_cpu(cpu) {
1943 		struct page *page;
1944 		void *page_addr;
1945 
1946 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1947 		if (!page) {
1948 			err = -ENOMEM;
1949 			goto out_err;
1950 		}
1951 
1952 		page_addr = page_address(page);
1953 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1954 		kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1955 	}
1956 
1957 	/*
1958 	 * Map the Hyp-code called directly from the host
1959 	 */
1960 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1961 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1962 	if (err) {
1963 		kvm_err("Cannot map world-switch code\n");
1964 		goto out_err;
1965 	}
1966 
1967 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1968 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1969 	if (err) {
1970 		kvm_err("Cannot map .hyp.rodata section\n");
1971 		goto out_err;
1972 	}
1973 
1974 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1975 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1976 	if (err) {
1977 		kvm_err("Cannot map rodata section\n");
1978 		goto out_err;
1979 	}
1980 
1981 	/*
1982 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1983 	 * section thanks to an assertion in the linker script. Map it RW and
1984 	 * the rest of .bss RO.
1985 	 */
1986 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1987 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1988 	if (err) {
1989 		kvm_err("Cannot map hyp bss section: %d\n", err);
1990 		goto out_err;
1991 	}
1992 
1993 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1994 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1995 	if (err) {
1996 		kvm_err("Cannot map bss section\n");
1997 		goto out_err;
1998 	}
1999 
2000 	/*
2001 	 * Map the Hyp stack pages
2002 	 */
2003 	for_each_possible_cpu(cpu) {
2004 		struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2005 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2006 		unsigned long hyp_addr;
2007 
2008 		/*
2009 		 * Allocate a contiguous HYP private VA range for the stack
2010 		 * and guard page. The allocation is also aligned based on
2011 		 * the order of its size.
2012 		 */
2013 		err = hyp_alloc_private_va_range(PAGE_SIZE * 2, &hyp_addr);
2014 		if (err) {
2015 			kvm_err("Cannot allocate hyp stack guard page\n");
2016 			goto out_err;
2017 		}
2018 
2019 		/*
2020 		 * Since the stack grows downwards, map the stack to the page
2021 		 * at the higher address and leave the lower guard page
2022 		 * unbacked.
2023 		 *
2024 		 * Any valid stack address now has the PAGE_SHIFT bit as 1
2025 		 * and addresses corresponding to the guard page have the
2026 		 * PAGE_SHIFT bit as 0 - this is used for overflow detection.
2027 		 */
2028 		err = __create_hyp_mappings(hyp_addr + PAGE_SIZE, PAGE_SIZE,
2029 					    __pa(stack_page), PAGE_HYP);
2030 		if (err) {
2031 			kvm_err("Cannot map hyp stack\n");
2032 			goto out_err;
2033 		}
2034 
2035 		/*
2036 		 * Save the stack PA in nvhe_init_params. This will be needed
2037 		 * to recreate the stack mapping in protected nVHE mode.
2038 		 * __hyp_pa() won't do the right thing there, since the stack
2039 		 * has been mapped in the flexible private VA space.
2040 		 */
2041 		params->stack_pa = __pa(stack_page);
2042 
2043 		params->stack_hyp_va = hyp_addr + (2 * PAGE_SIZE);
2044 	}
2045 
2046 	for_each_possible_cpu(cpu) {
2047 		char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
2048 		char *percpu_end = percpu_begin + nvhe_percpu_size();
2049 
2050 		/* Map Hyp percpu pages */
2051 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2052 		if (err) {
2053 			kvm_err("Cannot map hyp percpu region\n");
2054 			goto out_err;
2055 		}
2056 
2057 		/* Prepare the CPU initialization parameters */
2058 		cpu_prepare_hyp_mode(cpu);
2059 	}
2060 
2061 	if (is_protected_kvm_enabled()) {
2062 		init_cpu_logical_map();
2063 
2064 		if (!init_psci_relay()) {
2065 			err = -ENODEV;
2066 			goto out_err;
2067 		}
2068 	}
2069 
2070 	if (is_protected_kvm_enabled()) {
2071 		err = kvm_hyp_init_protection(hyp_va_bits);
2072 		if (err) {
2073 			kvm_err("Failed to init hyp memory protection\n");
2074 			goto out_err;
2075 		}
2076 	}
2077 
2078 	return 0;
2079 
2080 out_err:
2081 	teardown_hyp_mode();
2082 	kvm_err("error initializing Hyp mode: %d\n", err);
2083 	return err;
2084 }
2085 
2086 static void _kvm_host_prot_finalize(void *arg)
2087 {
2088 	int *err = arg;
2089 
2090 	if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)))
2091 		WRITE_ONCE(*err, -EINVAL);
2092 }
2093 
2094 static int pkvm_drop_host_privileges(void)
2095 {
2096 	int ret = 0;
2097 
2098 	/*
2099 	 * Flip the static key upfront as that may no longer be possible
2100 	 * once the host stage 2 is installed.
2101 	 */
2102 	static_branch_enable(&kvm_protected_mode_initialized);
2103 	on_each_cpu(_kvm_host_prot_finalize, &ret, 1);
2104 	return ret;
2105 }
2106 
2107 static int finalize_hyp_mode(void)
2108 {
2109 	if (!is_protected_kvm_enabled())
2110 		return 0;
2111 
2112 	/*
2113 	 * Exclude HYP BSS from kmemleak so that it doesn't get peeked
2114 	 * at, which would end badly once the section is inaccessible.
2115 	 * None of other sections should ever be introspected.
2116 	 */
2117 	kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
2118 	return pkvm_drop_host_privileges();
2119 }
2120 
2121 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2122 {
2123 	struct kvm_vcpu *vcpu;
2124 	unsigned long i;
2125 
2126 	mpidr &= MPIDR_HWID_BITMASK;
2127 	kvm_for_each_vcpu(i, vcpu, kvm) {
2128 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2129 			return vcpu;
2130 	}
2131 	return NULL;
2132 }
2133 
2134 bool kvm_arch_has_irq_bypass(void)
2135 {
2136 	return true;
2137 }
2138 
2139 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2140 				      struct irq_bypass_producer *prod)
2141 {
2142 	struct kvm_kernel_irqfd *irqfd =
2143 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2144 
2145 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2146 					  &irqfd->irq_entry);
2147 }
2148 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2149 				      struct irq_bypass_producer *prod)
2150 {
2151 	struct kvm_kernel_irqfd *irqfd =
2152 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2153 
2154 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2155 				     &irqfd->irq_entry);
2156 }
2157 
2158 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2159 {
2160 	struct kvm_kernel_irqfd *irqfd =
2161 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2162 
2163 	kvm_arm_halt_guest(irqfd->kvm);
2164 }
2165 
2166 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2167 {
2168 	struct kvm_kernel_irqfd *irqfd =
2169 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2170 
2171 	kvm_arm_resume_guest(irqfd->kvm);
2172 }
2173 
2174 /**
2175  * Initialize Hyp-mode and memory mappings on all CPUs.
2176  */
2177 int kvm_arch_init(void *opaque)
2178 {
2179 	int err;
2180 	bool in_hyp_mode;
2181 
2182 	if (!is_hyp_mode_available()) {
2183 		kvm_info("HYP mode not available\n");
2184 		return -ENODEV;
2185 	}
2186 
2187 	if (kvm_get_mode() == KVM_MODE_NONE) {
2188 		kvm_info("KVM disabled from command line\n");
2189 		return -ENODEV;
2190 	}
2191 
2192 	err = kvm_sys_reg_table_init();
2193 	if (err) {
2194 		kvm_info("Error initializing system register tables");
2195 		return err;
2196 	}
2197 
2198 	in_hyp_mode = is_kernel_in_hyp_mode();
2199 
2200 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2201 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2202 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2203 			 "Only trusted guests should be used on this system.\n");
2204 
2205 	err = kvm_set_ipa_limit();
2206 	if (err)
2207 		return err;
2208 
2209 	err = kvm_arm_init_sve();
2210 	if (err)
2211 		return err;
2212 
2213 	err = kvm_arm_vmid_alloc_init();
2214 	if (err) {
2215 		kvm_err("Failed to initialize VMID allocator.\n");
2216 		return err;
2217 	}
2218 
2219 	if (!in_hyp_mode) {
2220 		err = init_hyp_mode();
2221 		if (err)
2222 			goto out_err;
2223 	}
2224 
2225 	err = kvm_init_vector_slots();
2226 	if (err) {
2227 		kvm_err("Cannot initialise vector slots\n");
2228 		goto out_err;
2229 	}
2230 
2231 	err = init_subsystems();
2232 	if (err)
2233 		goto out_hyp;
2234 
2235 	if (!in_hyp_mode) {
2236 		err = finalize_hyp_mode();
2237 		if (err) {
2238 			kvm_err("Failed to finalize Hyp protection\n");
2239 			goto out_hyp;
2240 		}
2241 	}
2242 
2243 	if (is_protected_kvm_enabled()) {
2244 		kvm_info("Protected nVHE mode initialized successfully\n");
2245 	} else if (in_hyp_mode) {
2246 		kvm_info("VHE mode initialized successfully\n");
2247 	} else {
2248 		kvm_info("Hyp mode initialized successfully\n");
2249 	}
2250 
2251 	return 0;
2252 
2253 out_hyp:
2254 	hyp_cpu_pm_exit();
2255 	if (!in_hyp_mode)
2256 		teardown_hyp_mode();
2257 out_err:
2258 	kvm_arm_vmid_alloc_free();
2259 	return err;
2260 }
2261 
2262 /* NOP: Compiling as a module not supported */
2263 void kvm_arch_exit(void)
2264 {
2265 	kvm_unregister_perf_callbacks();
2266 }
2267 
2268 static int __init early_kvm_mode_cfg(char *arg)
2269 {
2270 	if (!arg)
2271 		return -EINVAL;
2272 
2273 	if (strcmp(arg, "protected") == 0) {
2274 		kvm_mode = KVM_MODE_PROTECTED;
2275 		return 0;
2276 	}
2277 
2278 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2279 		kvm_mode = KVM_MODE_DEFAULT;
2280 		return 0;
2281 	}
2282 
2283 	if (strcmp(arg, "none") == 0) {
2284 		kvm_mode = KVM_MODE_NONE;
2285 		return 0;
2286 	}
2287 
2288 	return -EINVAL;
2289 }
2290 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2291 
2292 enum kvm_mode kvm_get_mode(void)
2293 {
2294 	return kvm_mode;
2295 }
2296 
2297 static int arm_init(void)
2298 {
2299 	int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2300 	return rc;
2301 }
2302 
2303 module_init(arm_init);
2304