xref: /linux/arch/arm64/kvm/arm.c (revision 4359a011e259a4608afc7fb3635370c9d4ba5943)
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 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 		ret = -ENOMEM;
155 		goto out_free_stage2_pgd;
156 	}
157 	cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
158 
159 	kvm_vgic_early_init(kvm);
160 
161 	/* The maximum number of VCPUs is limited by the host's GIC model */
162 	kvm->max_vcpus = kvm_arm_default_max_vcpus();
163 
164 	set_default_spectre(kvm);
165 	kvm_arm_init_hypercalls(kvm);
166 
167 	return ret;
168 out_free_stage2_pgd:
169 	kvm_free_stage2_pgd(&kvm->arch.mmu);
170 	return ret;
171 }
172 
173 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
174 {
175 	return VM_FAULT_SIGBUS;
176 }
177 
178 
179 /**
180  * kvm_arch_destroy_vm - destroy the VM data structure
181  * @kvm:	pointer to the KVM struct
182  */
183 void kvm_arch_destroy_vm(struct kvm *kvm)
184 {
185 	bitmap_free(kvm->arch.pmu_filter);
186 	free_cpumask_var(kvm->arch.supported_cpus);
187 
188 	kvm_vgic_destroy(kvm);
189 
190 	kvm_destroy_vcpus(kvm);
191 
192 	kvm_unshare_hyp(kvm, kvm + 1);
193 }
194 
195 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
196 {
197 	int r;
198 	switch (ext) {
199 	case KVM_CAP_IRQCHIP:
200 		r = vgic_present;
201 		break;
202 	case KVM_CAP_IOEVENTFD:
203 	case KVM_CAP_DEVICE_CTRL:
204 	case KVM_CAP_USER_MEMORY:
205 	case KVM_CAP_SYNC_MMU:
206 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
207 	case KVM_CAP_ONE_REG:
208 	case KVM_CAP_ARM_PSCI:
209 	case KVM_CAP_ARM_PSCI_0_2:
210 	case KVM_CAP_READONLY_MEM:
211 	case KVM_CAP_MP_STATE:
212 	case KVM_CAP_IMMEDIATE_EXIT:
213 	case KVM_CAP_VCPU_EVENTS:
214 	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
215 	case KVM_CAP_ARM_NISV_TO_USER:
216 	case KVM_CAP_ARM_INJECT_EXT_DABT:
217 	case KVM_CAP_SET_GUEST_DEBUG:
218 	case KVM_CAP_VCPU_ATTRIBUTES:
219 	case KVM_CAP_PTP_KVM:
220 	case KVM_CAP_ARM_SYSTEM_SUSPEND:
221 		r = 1;
222 		break;
223 	case KVM_CAP_SET_GUEST_DEBUG2:
224 		return KVM_GUESTDBG_VALID_MASK;
225 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
226 		r = 1;
227 		break;
228 	case KVM_CAP_NR_VCPUS:
229 		/*
230 		 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
231 		 * architectures, as it does not always bound it to
232 		 * KVM_CAP_MAX_VCPUS. It should not matter much because
233 		 * this is just an advisory value.
234 		 */
235 		r = min_t(unsigned int, num_online_cpus(),
236 			  kvm_arm_default_max_vcpus());
237 		break;
238 	case KVM_CAP_MAX_VCPUS:
239 	case KVM_CAP_MAX_VCPU_ID:
240 		if (kvm)
241 			r = kvm->max_vcpus;
242 		else
243 			r = kvm_arm_default_max_vcpus();
244 		break;
245 	case KVM_CAP_MSI_DEVID:
246 		if (!kvm)
247 			r = -EINVAL;
248 		else
249 			r = kvm->arch.vgic.msis_require_devid;
250 		break;
251 	case KVM_CAP_ARM_USER_IRQ:
252 		/*
253 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
254 		 * (bump this number if adding more devices)
255 		 */
256 		r = 1;
257 		break;
258 	case KVM_CAP_ARM_MTE:
259 		r = system_supports_mte();
260 		break;
261 	case KVM_CAP_STEAL_TIME:
262 		r = kvm_arm_pvtime_supported();
263 		break;
264 	case KVM_CAP_ARM_EL1_32BIT:
265 		r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
266 		break;
267 	case KVM_CAP_GUEST_DEBUG_HW_BPS:
268 		r = get_num_brps();
269 		break;
270 	case KVM_CAP_GUEST_DEBUG_HW_WPS:
271 		r = get_num_wrps();
272 		break;
273 	case KVM_CAP_ARM_PMU_V3:
274 		r = kvm_arm_support_pmu_v3();
275 		break;
276 	case KVM_CAP_ARM_INJECT_SERROR_ESR:
277 		r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
278 		break;
279 	case KVM_CAP_ARM_VM_IPA_SIZE:
280 		r = get_kvm_ipa_limit();
281 		break;
282 	case KVM_CAP_ARM_SVE:
283 		r = system_supports_sve();
284 		break;
285 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
286 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
287 		r = system_has_full_ptr_auth();
288 		break;
289 	default:
290 		r = 0;
291 	}
292 
293 	return r;
294 }
295 
296 long kvm_arch_dev_ioctl(struct file *filp,
297 			unsigned int ioctl, unsigned long arg)
298 {
299 	return -EINVAL;
300 }
301 
302 struct kvm *kvm_arch_alloc_vm(void)
303 {
304 	size_t sz = sizeof(struct kvm);
305 
306 	if (!has_vhe())
307 		return kzalloc(sz, GFP_KERNEL_ACCOUNT);
308 
309 	return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
310 }
311 
312 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
313 {
314 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
315 		return -EBUSY;
316 
317 	if (id >= kvm->max_vcpus)
318 		return -EINVAL;
319 
320 	return 0;
321 }
322 
323 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
324 {
325 	int err;
326 
327 	/* Force users to call KVM_ARM_VCPU_INIT */
328 	vcpu->arch.target = -1;
329 	bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
330 
331 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
332 
333 	/*
334 	 * Default value for the FP state, will be overloaded at load
335 	 * time if we support FP (pretty likely)
336 	 */
337 	vcpu->arch.fp_state = FP_STATE_FREE;
338 
339 	/* Set up the timer */
340 	kvm_timer_vcpu_init(vcpu);
341 
342 	kvm_pmu_vcpu_init(vcpu);
343 
344 	kvm_arm_reset_debug_ptr(vcpu);
345 
346 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
347 
348 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
349 
350 	err = kvm_vgic_vcpu_init(vcpu);
351 	if (err)
352 		return err;
353 
354 	return kvm_share_hyp(vcpu, vcpu + 1);
355 }
356 
357 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
358 {
359 }
360 
361 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
362 {
363 	if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
364 		static_branch_dec(&userspace_irqchip_in_use);
365 
366 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
367 	kvm_timer_vcpu_terminate(vcpu);
368 	kvm_pmu_vcpu_destroy(vcpu);
369 
370 	kvm_arm_vcpu_destroy(vcpu);
371 }
372 
373 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
374 {
375 
376 }
377 
378 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
379 {
380 
381 }
382 
383 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
384 {
385 	struct kvm_s2_mmu *mmu;
386 	int *last_ran;
387 
388 	mmu = vcpu->arch.hw_mmu;
389 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
390 
391 	/*
392 	 * We guarantee that both TLBs and I-cache are private to each
393 	 * vcpu. If detecting that a vcpu from the same VM has
394 	 * previously run on the same physical CPU, call into the
395 	 * hypervisor code to nuke the relevant contexts.
396 	 *
397 	 * We might get preempted before the vCPU actually runs, but
398 	 * over-invalidation doesn't affect correctness.
399 	 */
400 	if (*last_ran != vcpu->vcpu_id) {
401 		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
402 		*last_ran = vcpu->vcpu_id;
403 	}
404 
405 	vcpu->cpu = cpu;
406 
407 	kvm_vgic_load(vcpu);
408 	kvm_timer_vcpu_load(vcpu);
409 	if (has_vhe())
410 		kvm_vcpu_load_sysregs_vhe(vcpu);
411 	kvm_arch_vcpu_load_fp(vcpu);
412 	kvm_vcpu_pmu_restore_guest(vcpu);
413 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
414 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
415 
416 	if (single_task_running())
417 		vcpu_clear_wfx_traps(vcpu);
418 	else
419 		vcpu_set_wfx_traps(vcpu);
420 
421 	if (vcpu_has_ptrauth(vcpu))
422 		vcpu_ptrauth_disable(vcpu);
423 	kvm_arch_vcpu_load_debug_state_flags(vcpu);
424 
425 	if (!cpumask_test_cpu(smp_processor_id(), vcpu->kvm->arch.supported_cpus))
426 		vcpu_set_on_unsupported_cpu(vcpu);
427 }
428 
429 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
430 {
431 	kvm_arch_vcpu_put_debug_state_flags(vcpu);
432 	kvm_arch_vcpu_put_fp(vcpu);
433 	if (has_vhe())
434 		kvm_vcpu_put_sysregs_vhe(vcpu);
435 	kvm_timer_vcpu_put(vcpu);
436 	kvm_vgic_put(vcpu);
437 	kvm_vcpu_pmu_restore_host(vcpu);
438 	kvm_arm_vmid_clear_active();
439 
440 	vcpu_clear_on_unsupported_cpu(vcpu);
441 	vcpu->cpu = -1;
442 }
443 
444 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
445 {
446 	vcpu->arch.mp_state.mp_state = KVM_MP_STATE_STOPPED;
447 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
448 	kvm_vcpu_kick(vcpu);
449 }
450 
451 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
452 {
453 	return vcpu->arch.mp_state.mp_state == KVM_MP_STATE_STOPPED;
454 }
455 
456 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
457 {
458 	vcpu->arch.mp_state.mp_state = KVM_MP_STATE_SUSPENDED;
459 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
460 	kvm_vcpu_kick(vcpu);
461 }
462 
463 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
464 {
465 	return vcpu->arch.mp_state.mp_state == KVM_MP_STATE_SUSPENDED;
466 }
467 
468 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
469 				    struct kvm_mp_state *mp_state)
470 {
471 	*mp_state = vcpu->arch.mp_state;
472 
473 	return 0;
474 }
475 
476 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
477 				    struct kvm_mp_state *mp_state)
478 {
479 	int ret = 0;
480 
481 	switch (mp_state->mp_state) {
482 	case KVM_MP_STATE_RUNNABLE:
483 		vcpu->arch.mp_state = *mp_state;
484 		break;
485 	case KVM_MP_STATE_STOPPED:
486 		kvm_arm_vcpu_power_off(vcpu);
487 		break;
488 	case KVM_MP_STATE_SUSPENDED:
489 		kvm_arm_vcpu_suspend(vcpu);
490 		break;
491 	default:
492 		ret = -EINVAL;
493 	}
494 
495 	return ret;
496 }
497 
498 /**
499  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
500  * @v:		The VCPU pointer
501  *
502  * If the guest CPU is not waiting for interrupts or an interrupt line is
503  * asserted, the CPU is by definition runnable.
504  */
505 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
506 {
507 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
508 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
509 		&& !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
510 }
511 
512 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
513 {
514 	return vcpu_mode_priv(vcpu);
515 }
516 
517 #ifdef CONFIG_GUEST_PERF_EVENTS
518 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
519 {
520 	return *vcpu_pc(vcpu);
521 }
522 #endif
523 
524 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
525 {
526 	return vcpu->arch.target >= 0;
527 }
528 
529 /*
530  * Handle both the initialisation that is being done when the vcpu is
531  * run for the first time, as well as the updates that must be
532  * performed each time we get a new thread dealing with this vcpu.
533  */
534 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
535 {
536 	struct kvm *kvm = vcpu->kvm;
537 	int ret;
538 
539 	if (!kvm_vcpu_initialized(vcpu))
540 		return -ENOEXEC;
541 
542 	if (!kvm_arm_vcpu_is_finalized(vcpu))
543 		return -EPERM;
544 
545 	ret = kvm_arch_vcpu_run_map_fp(vcpu);
546 	if (ret)
547 		return ret;
548 
549 	if (likely(vcpu_has_run_once(vcpu)))
550 		return 0;
551 
552 	kvm_arm_vcpu_init_debug(vcpu);
553 
554 	if (likely(irqchip_in_kernel(kvm))) {
555 		/*
556 		 * Map the VGIC hardware resources before running a vcpu the
557 		 * first time on this VM.
558 		 */
559 		ret = kvm_vgic_map_resources(kvm);
560 		if (ret)
561 			return ret;
562 	}
563 
564 	ret = kvm_timer_enable(vcpu);
565 	if (ret)
566 		return ret;
567 
568 	ret = kvm_arm_pmu_v3_enable(vcpu);
569 	if (ret)
570 		return ret;
571 
572 	if (!irqchip_in_kernel(kvm)) {
573 		/*
574 		 * Tell the rest of the code that there are userspace irqchip
575 		 * VMs in the wild.
576 		 */
577 		static_branch_inc(&userspace_irqchip_in_use);
578 	}
579 
580 	/*
581 	 * Initialize traps for protected VMs.
582 	 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
583 	 * the code is in place for first run initialization at EL2.
584 	 */
585 	if (kvm_vm_is_protected(kvm))
586 		kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
587 
588 	mutex_lock(&kvm->lock);
589 	set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
590 	mutex_unlock(&kvm->lock);
591 
592 	return ret;
593 }
594 
595 bool kvm_arch_intc_initialized(struct kvm *kvm)
596 {
597 	return vgic_initialized(kvm);
598 }
599 
600 void kvm_arm_halt_guest(struct kvm *kvm)
601 {
602 	unsigned long i;
603 	struct kvm_vcpu *vcpu;
604 
605 	kvm_for_each_vcpu(i, vcpu, kvm)
606 		vcpu->arch.pause = true;
607 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
608 }
609 
610 void kvm_arm_resume_guest(struct kvm *kvm)
611 {
612 	unsigned long i;
613 	struct kvm_vcpu *vcpu;
614 
615 	kvm_for_each_vcpu(i, vcpu, kvm) {
616 		vcpu->arch.pause = false;
617 		__kvm_vcpu_wake_up(vcpu);
618 	}
619 }
620 
621 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
622 {
623 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
624 
625 	rcuwait_wait_event(wait,
626 			   (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
627 			   TASK_INTERRUPTIBLE);
628 
629 	if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
630 		/* Awaken to handle a signal, request we sleep again later. */
631 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
632 	}
633 
634 	/*
635 	 * Make sure we will observe a potential reset request if we've
636 	 * observed a change to the power state. Pairs with the smp_wmb() in
637 	 * kvm_psci_vcpu_on().
638 	 */
639 	smp_rmb();
640 }
641 
642 /**
643  * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
644  * @vcpu:	The VCPU pointer
645  *
646  * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
647  * the vCPU is runnable.  The vCPU may or may not be scheduled out, depending
648  * on when a wake event arrives, e.g. there may already be a pending wake event.
649  */
650 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
651 {
652 	/*
653 	 * Sync back the state of the GIC CPU interface so that we have
654 	 * the latest PMR and group enables. This ensures that
655 	 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
656 	 * we have pending interrupts, e.g. when determining if the
657 	 * vCPU should block.
658 	 *
659 	 * For the same reason, we want to tell GICv4 that we need
660 	 * doorbells to be signalled, should an interrupt become pending.
661 	 */
662 	preempt_disable();
663 	kvm_vgic_vmcr_sync(vcpu);
664 	vgic_v4_put(vcpu, true);
665 	preempt_enable();
666 
667 	kvm_vcpu_halt(vcpu);
668 	vcpu_clear_flag(vcpu, IN_WFIT);
669 	kvm_clear_request(KVM_REQ_UNHALT, vcpu);
670 
671 	preempt_disable();
672 	vgic_v4_load(vcpu);
673 	preempt_enable();
674 }
675 
676 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
677 {
678 	if (!kvm_arm_vcpu_suspended(vcpu))
679 		return 1;
680 
681 	kvm_vcpu_wfi(vcpu);
682 
683 	/*
684 	 * The suspend state is sticky; we do not leave it until userspace
685 	 * explicitly marks the vCPU as runnable. Request that we suspend again
686 	 * later.
687 	 */
688 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
689 
690 	/*
691 	 * Check to make sure the vCPU is actually runnable. If so, exit to
692 	 * userspace informing it of the wakeup condition.
693 	 */
694 	if (kvm_arch_vcpu_runnable(vcpu)) {
695 		memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
696 		vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
697 		vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
698 		return 0;
699 	}
700 
701 	/*
702 	 * Otherwise, we were unblocked to process a different event, such as a
703 	 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
704 	 * process the event.
705 	 */
706 	return 1;
707 }
708 
709 /**
710  * check_vcpu_requests - check and handle pending vCPU requests
711  * @vcpu:	the VCPU pointer
712  *
713  * Return: 1 if we should enter the guest
714  *	   0 if we should exit to userspace
715  *	   < 0 if we should exit to userspace, where the return value indicates
716  *	   an error
717  */
718 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
719 {
720 	if (kvm_request_pending(vcpu)) {
721 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
722 			kvm_vcpu_sleep(vcpu);
723 
724 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
725 			kvm_reset_vcpu(vcpu);
726 
727 		/*
728 		 * Clear IRQ_PENDING requests that were made to guarantee
729 		 * that a VCPU sees new virtual interrupts.
730 		 */
731 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
732 
733 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
734 			kvm_update_stolen_time(vcpu);
735 
736 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
737 			/* The distributor enable bits were changed */
738 			preempt_disable();
739 			vgic_v4_put(vcpu, false);
740 			vgic_v4_load(vcpu);
741 			preempt_enable();
742 		}
743 
744 		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
745 			kvm_pmu_handle_pmcr(vcpu,
746 					    __vcpu_sys_reg(vcpu, PMCR_EL0));
747 
748 		if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
749 			return kvm_vcpu_suspend(vcpu);
750 	}
751 
752 	return 1;
753 }
754 
755 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
756 {
757 	if (likely(!vcpu_mode_is_32bit(vcpu)))
758 		return false;
759 
760 	return !kvm_supports_32bit_el0();
761 }
762 
763 /**
764  * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
765  * @vcpu:	The VCPU pointer
766  * @ret:	Pointer to write optional return code
767  *
768  * Returns: true if the VCPU needs to return to a preemptible + interruptible
769  *	    and skip guest entry.
770  *
771  * This function disambiguates between two different types of exits: exits to a
772  * preemptible + interruptible kernel context and exits to userspace. For an
773  * exit to userspace, this function will write the return code to ret and return
774  * true. For an exit to preemptible + interruptible kernel context (i.e. check
775  * for pending work and re-enter), return true without writing to ret.
776  */
777 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
778 {
779 	struct kvm_run *run = vcpu->run;
780 
781 	/*
782 	 * If we're using a userspace irqchip, then check if we need
783 	 * to tell a userspace irqchip about timer or PMU level
784 	 * changes and if so, exit to userspace (the actual level
785 	 * state gets updated in kvm_timer_update_run and
786 	 * kvm_pmu_update_run below).
787 	 */
788 	if (static_branch_unlikely(&userspace_irqchip_in_use)) {
789 		if (kvm_timer_should_notify_user(vcpu) ||
790 		    kvm_pmu_should_notify_user(vcpu)) {
791 			*ret = -EINTR;
792 			run->exit_reason = KVM_EXIT_INTR;
793 			return true;
794 		}
795 	}
796 
797 	if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
798 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
799 		run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
800 		run->fail_entry.cpu = smp_processor_id();
801 		*ret = 0;
802 		return true;
803 	}
804 
805 	return kvm_request_pending(vcpu) ||
806 			xfer_to_guest_mode_work_pending();
807 }
808 
809 /*
810  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
811  * the vCPU is running.
812  *
813  * This must be noinstr as instrumentation may make use of RCU, and this is not
814  * safe during the EQS.
815  */
816 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
817 {
818 	int ret;
819 
820 	guest_state_enter_irqoff();
821 	ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
822 	guest_state_exit_irqoff();
823 
824 	return ret;
825 }
826 
827 /**
828  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
829  * @vcpu:	The VCPU pointer
830  *
831  * This function is called through the VCPU_RUN ioctl called from user space. It
832  * will execute VM code in a loop until the time slice for the process is used
833  * or some emulation is needed from user space in which case the function will
834  * return with return value 0 and with the kvm_run structure filled in with the
835  * required data for the requested emulation.
836  */
837 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
838 {
839 	struct kvm_run *run = vcpu->run;
840 	int ret;
841 
842 	if (run->exit_reason == KVM_EXIT_MMIO) {
843 		ret = kvm_handle_mmio_return(vcpu);
844 		if (ret)
845 			return ret;
846 	}
847 
848 	vcpu_load(vcpu);
849 
850 	if (run->immediate_exit) {
851 		ret = -EINTR;
852 		goto out;
853 	}
854 
855 	kvm_sigset_activate(vcpu);
856 
857 	ret = 1;
858 	run->exit_reason = KVM_EXIT_UNKNOWN;
859 	run->flags = 0;
860 	while (ret > 0) {
861 		/*
862 		 * Check conditions before entering the guest
863 		 */
864 		ret = xfer_to_guest_mode_handle_work(vcpu);
865 		if (!ret)
866 			ret = 1;
867 
868 		if (ret > 0)
869 			ret = check_vcpu_requests(vcpu);
870 
871 		/*
872 		 * Preparing the interrupts to be injected also
873 		 * involves poking the GIC, which must be done in a
874 		 * non-preemptible context.
875 		 */
876 		preempt_disable();
877 
878 		/*
879 		 * The VMID allocator only tracks active VMIDs per
880 		 * physical CPU, and therefore the VMID allocated may not be
881 		 * preserved on VMID roll-over if the task was preempted,
882 		 * making a thread's VMID inactive. So we need to call
883 		 * kvm_arm_vmid_update() in non-premptible context.
884 		 */
885 		kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid);
886 
887 		kvm_pmu_flush_hwstate(vcpu);
888 
889 		local_irq_disable();
890 
891 		kvm_vgic_flush_hwstate(vcpu);
892 
893 		kvm_pmu_update_vcpu_events(vcpu);
894 
895 		/*
896 		 * Ensure we set mode to IN_GUEST_MODE after we disable
897 		 * interrupts and before the final VCPU requests check.
898 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
899 		 * Documentation/virt/kvm/vcpu-requests.rst
900 		 */
901 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
902 
903 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
904 			vcpu->mode = OUTSIDE_GUEST_MODE;
905 			isb(); /* Ensure work in x_flush_hwstate is committed */
906 			kvm_pmu_sync_hwstate(vcpu);
907 			if (static_branch_unlikely(&userspace_irqchip_in_use))
908 				kvm_timer_sync_user(vcpu);
909 			kvm_vgic_sync_hwstate(vcpu);
910 			local_irq_enable();
911 			preempt_enable();
912 			continue;
913 		}
914 
915 		kvm_arm_setup_debug(vcpu);
916 		kvm_arch_vcpu_ctxflush_fp(vcpu);
917 
918 		/**************************************************************
919 		 * Enter the guest
920 		 */
921 		trace_kvm_entry(*vcpu_pc(vcpu));
922 		guest_timing_enter_irqoff();
923 
924 		ret = kvm_arm_vcpu_enter_exit(vcpu);
925 
926 		vcpu->mode = OUTSIDE_GUEST_MODE;
927 		vcpu->stat.exits++;
928 		/*
929 		 * Back from guest
930 		 *************************************************************/
931 
932 		kvm_arm_clear_debug(vcpu);
933 
934 		/*
935 		 * We must sync the PMU state before the vgic state so
936 		 * that the vgic can properly sample the updated state of the
937 		 * interrupt line.
938 		 */
939 		kvm_pmu_sync_hwstate(vcpu);
940 
941 		/*
942 		 * Sync the vgic state before syncing the timer state because
943 		 * the timer code needs to know if the virtual timer
944 		 * interrupts are active.
945 		 */
946 		kvm_vgic_sync_hwstate(vcpu);
947 
948 		/*
949 		 * Sync the timer hardware state before enabling interrupts as
950 		 * we don't want vtimer interrupts to race with syncing the
951 		 * timer virtual interrupt state.
952 		 */
953 		if (static_branch_unlikely(&userspace_irqchip_in_use))
954 			kvm_timer_sync_user(vcpu);
955 
956 		kvm_arch_vcpu_ctxsync_fp(vcpu);
957 
958 		/*
959 		 * We must ensure that any pending interrupts are taken before
960 		 * we exit guest timing so that timer ticks are accounted as
961 		 * guest time. Transiently unmask interrupts so that any
962 		 * pending interrupts are taken.
963 		 *
964 		 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
965 		 * context synchronization event) is necessary to ensure that
966 		 * pending interrupts are taken.
967 		 */
968 		if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
969 			local_irq_enable();
970 			isb();
971 			local_irq_disable();
972 		}
973 
974 		guest_timing_exit_irqoff();
975 
976 		local_irq_enable();
977 
978 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
979 
980 		/* Exit types that need handling before we can be preempted */
981 		handle_exit_early(vcpu, ret);
982 
983 		preempt_enable();
984 
985 		/*
986 		 * The ARMv8 architecture doesn't give the hypervisor
987 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
988 		 * if implemented by the CPU. If we spot the guest in such
989 		 * state and that we decided it wasn't supposed to do so (like
990 		 * with the asymmetric AArch32 case), return to userspace with
991 		 * a fatal error.
992 		 */
993 		if (vcpu_mode_is_bad_32bit(vcpu)) {
994 			/*
995 			 * As we have caught the guest red-handed, decide that
996 			 * it isn't fit for purpose anymore by making the vcpu
997 			 * invalid. The VMM can try and fix it by issuing  a
998 			 * KVM_ARM_VCPU_INIT if it really wants to.
999 			 */
1000 			vcpu->arch.target = -1;
1001 			ret = ARM_EXCEPTION_IL;
1002 		}
1003 
1004 		ret = handle_exit(vcpu, ret);
1005 	}
1006 
1007 	/* Tell userspace about in-kernel device output levels */
1008 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1009 		kvm_timer_update_run(vcpu);
1010 		kvm_pmu_update_run(vcpu);
1011 	}
1012 
1013 	kvm_sigset_deactivate(vcpu);
1014 
1015 out:
1016 	/*
1017 	 * In the unlikely event that we are returning to userspace
1018 	 * with pending exceptions or PC adjustment, commit these
1019 	 * adjustments in order to give userspace a consistent view of
1020 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1021 	 * being preempt-safe on VHE.
1022 	 */
1023 	if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1024 		     vcpu_get_flag(vcpu, INCREMENT_PC)))
1025 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
1026 
1027 	vcpu_put(vcpu);
1028 	return ret;
1029 }
1030 
1031 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1032 {
1033 	int bit_index;
1034 	bool set;
1035 	unsigned long *hcr;
1036 
1037 	if (number == KVM_ARM_IRQ_CPU_IRQ)
1038 		bit_index = __ffs(HCR_VI);
1039 	else /* KVM_ARM_IRQ_CPU_FIQ */
1040 		bit_index = __ffs(HCR_VF);
1041 
1042 	hcr = vcpu_hcr(vcpu);
1043 	if (level)
1044 		set = test_and_set_bit(bit_index, hcr);
1045 	else
1046 		set = test_and_clear_bit(bit_index, hcr);
1047 
1048 	/*
1049 	 * If we didn't change anything, no need to wake up or kick other CPUs
1050 	 */
1051 	if (set == level)
1052 		return 0;
1053 
1054 	/*
1055 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1056 	 * trigger a world-switch round on the running physical CPU to set the
1057 	 * virtual IRQ/FIQ fields in the HCR appropriately.
1058 	 */
1059 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1060 	kvm_vcpu_kick(vcpu);
1061 
1062 	return 0;
1063 }
1064 
1065 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1066 			  bool line_status)
1067 {
1068 	u32 irq = irq_level->irq;
1069 	unsigned int irq_type, vcpu_idx, irq_num;
1070 	int nrcpus = atomic_read(&kvm->online_vcpus);
1071 	struct kvm_vcpu *vcpu = NULL;
1072 	bool level = irq_level->level;
1073 
1074 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1075 	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1076 	vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1077 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1078 
1079 	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1080 
1081 	switch (irq_type) {
1082 	case KVM_ARM_IRQ_TYPE_CPU:
1083 		if (irqchip_in_kernel(kvm))
1084 			return -ENXIO;
1085 
1086 		if (vcpu_idx >= nrcpus)
1087 			return -EINVAL;
1088 
1089 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1090 		if (!vcpu)
1091 			return -EINVAL;
1092 
1093 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1094 			return -EINVAL;
1095 
1096 		return vcpu_interrupt_line(vcpu, irq_num, level);
1097 	case KVM_ARM_IRQ_TYPE_PPI:
1098 		if (!irqchip_in_kernel(kvm))
1099 			return -ENXIO;
1100 
1101 		if (vcpu_idx >= nrcpus)
1102 			return -EINVAL;
1103 
1104 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1105 		if (!vcpu)
1106 			return -EINVAL;
1107 
1108 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1109 			return -EINVAL;
1110 
1111 		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1112 	case KVM_ARM_IRQ_TYPE_SPI:
1113 		if (!irqchip_in_kernel(kvm))
1114 			return -ENXIO;
1115 
1116 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1117 			return -EINVAL;
1118 
1119 		return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1120 	}
1121 
1122 	return -EINVAL;
1123 }
1124 
1125 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1126 			       const struct kvm_vcpu_init *init)
1127 {
1128 	unsigned int i, ret;
1129 	u32 phys_target = kvm_target_cpu();
1130 
1131 	if (init->target != phys_target)
1132 		return -EINVAL;
1133 
1134 	/*
1135 	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1136 	 * use the same target.
1137 	 */
1138 	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1139 		return -EINVAL;
1140 
1141 	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1142 	for (i = 0; i < sizeof(init->features) * 8; i++) {
1143 		bool set = (init->features[i / 32] & (1 << (i % 32)));
1144 
1145 		if (set && i >= KVM_VCPU_MAX_FEATURES)
1146 			return -ENOENT;
1147 
1148 		/*
1149 		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1150 		 * use the same feature set.
1151 		 */
1152 		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1153 		    test_bit(i, vcpu->arch.features) != set)
1154 			return -EINVAL;
1155 
1156 		if (set)
1157 			set_bit(i, vcpu->arch.features);
1158 	}
1159 
1160 	vcpu->arch.target = phys_target;
1161 
1162 	/* Now we know what it is, we can reset it. */
1163 	ret = kvm_reset_vcpu(vcpu);
1164 	if (ret) {
1165 		vcpu->arch.target = -1;
1166 		bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1167 	}
1168 
1169 	return ret;
1170 }
1171 
1172 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1173 					 struct kvm_vcpu_init *init)
1174 {
1175 	int ret;
1176 
1177 	ret = kvm_vcpu_set_target(vcpu, init);
1178 	if (ret)
1179 		return ret;
1180 
1181 	/*
1182 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1183 	 * guest MMU is turned off and flush the caches as needed.
1184 	 *
1185 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1186 	 * ensuring that the data side is always coherent. We still
1187 	 * need to invalidate the I-cache though, as FWB does *not*
1188 	 * imply CTR_EL0.DIC.
1189 	 */
1190 	if (vcpu_has_run_once(vcpu)) {
1191 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1192 			stage2_unmap_vm(vcpu->kvm);
1193 		else
1194 			icache_inval_all_pou();
1195 	}
1196 
1197 	vcpu_reset_hcr(vcpu);
1198 	vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1199 
1200 	/*
1201 	 * Handle the "start in power-off" case.
1202 	 */
1203 	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1204 		kvm_arm_vcpu_power_off(vcpu);
1205 	else
1206 		vcpu->arch.mp_state.mp_state = KVM_MP_STATE_RUNNABLE;
1207 
1208 	return 0;
1209 }
1210 
1211 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1212 				 struct kvm_device_attr *attr)
1213 {
1214 	int ret = -ENXIO;
1215 
1216 	switch (attr->group) {
1217 	default:
1218 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1219 		break;
1220 	}
1221 
1222 	return ret;
1223 }
1224 
1225 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1226 				 struct kvm_device_attr *attr)
1227 {
1228 	int ret = -ENXIO;
1229 
1230 	switch (attr->group) {
1231 	default:
1232 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1233 		break;
1234 	}
1235 
1236 	return ret;
1237 }
1238 
1239 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1240 				 struct kvm_device_attr *attr)
1241 {
1242 	int ret = -ENXIO;
1243 
1244 	switch (attr->group) {
1245 	default:
1246 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1247 		break;
1248 	}
1249 
1250 	return ret;
1251 }
1252 
1253 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1254 				   struct kvm_vcpu_events *events)
1255 {
1256 	memset(events, 0, sizeof(*events));
1257 
1258 	return __kvm_arm_vcpu_get_events(vcpu, events);
1259 }
1260 
1261 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1262 				   struct kvm_vcpu_events *events)
1263 {
1264 	int i;
1265 
1266 	/* check whether the reserved field is zero */
1267 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1268 		if (events->reserved[i])
1269 			return -EINVAL;
1270 
1271 	/* check whether the pad field is zero */
1272 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1273 		if (events->exception.pad[i])
1274 			return -EINVAL;
1275 
1276 	return __kvm_arm_vcpu_set_events(vcpu, events);
1277 }
1278 
1279 long kvm_arch_vcpu_ioctl(struct file *filp,
1280 			 unsigned int ioctl, unsigned long arg)
1281 {
1282 	struct kvm_vcpu *vcpu = filp->private_data;
1283 	void __user *argp = (void __user *)arg;
1284 	struct kvm_device_attr attr;
1285 	long r;
1286 
1287 	switch (ioctl) {
1288 	case KVM_ARM_VCPU_INIT: {
1289 		struct kvm_vcpu_init init;
1290 
1291 		r = -EFAULT;
1292 		if (copy_from_user(&init, argp, sizeof(init)))
1293 			break;
1294 
1295 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1296 		break;
1297 	}
1298 	case KVM_SET_ONE_REG:
1299 	case KVM_GET_ONE_REG: {
1300 		struct kvm_one_reg reg;
1301 
1302 		r = -ENOEXEC;
1303 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1304 			break;
1305 
1306 		r = -EFAULT;
1307 		if (copy_from_user(&reg, argp, sizeof(reg)))
1308 			break;
1309 
1310 		/*
1311 		 * We could owe a reset due to PSCI. Handle the pending reset
1312 		 * here to ensure userspace register accesses are ordered after
1313 		 * the reset.
1314 		 */
1315 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1316 			kvm_reset_vcpu(vcpu);
1317 
1318 		if (ioctl == KVM_SET_ONE_REG)
1319 			r = kvm_arm_set_reg(vcpu, &reg);
1320 		else
1321 			r = kvm_arm_get_reg(vcpu, &reg);
1322 		break;
1323 	}
1324 	case KVM_GET_REG_LIST: {
1325 		struct kvm_reg_list __user *user_list = argp;
1326 		struct kvm_reg_list reg_list;
1327 		unsigned n;
1328 
1329 		r = -ENOEXEC;
1330 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1331 			break;
1332 
1333 		r = -EPERM;
1334 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1335 			break;
1336 
1337 		r = -EFAULT;
1338 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1339 			break;
1340 		n = reg_list.n;
1341 		reg_list.n = kvm_arm_num_regs(vcpu);
1342 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1343 			break;
1344 		r = -E2BIG;
1345 		if (n < reg_list.n)
1346 			break;
1347 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1348 		break;
1349 	}
1350 	case KVM_SET_DEVICE_ATTR: {
1351 		r = -EFAULT;
1352 		if (copy_from_user(&attr, argp, sizeof(attr)))
1353 			break;
1354 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1355 		break;
1356 	}
1357 	case KVM_GET_DEVICE_ATTR: {
1358 		r = -EFAULT;
1359 		if (copy_from_user(&attr, argp, sizeof(attr)))
1360 			break;
1361 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1362 		break;
1363 	}
1364 	case KVM_HAS_DEVICE_ATTR: {
1365 		r = -EFAULT;
1366 		if (copy_from_user(&attr, argp, sizeof(attr)))
1367 			break;
1368 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1369 		break;
1370 	}
1371 	case KVM_GET_VCPU_EVENTS: {
1372 		struct kvm_vcpu_events events;
1373 
1374 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1375 			return -EINVAL;
1376 
1377 		if (copy_to_user(argp, &events, sizeof(events)))
1378 			return -EFAULT;
1379 
1380 		return 0;
1381 	}
1382 	case KVM_SET_VCPU_EVENTS: {
1383 		struct kvm_vcpu_events events;
1384 
1385 		if (copy_from_user(&events, argp, sizeof(events)))
1386 			return -EFAULT;
1387 
1388 		return kvm_arm_vcpu_set_events(vcpu, &events);
1389 	}
1390 	case KVM_ARM_VCPU_FINALIZE: {
1391 		int what;
1392 
1393 		if (!kvm_vcpu_initialized(vcpu))
1394 			return -ENOEXEC;
1395 
1396 		if (get_user(what, (const int __user *)argp))
1397 			return -EFAULT;
1398 
1399 		return kvm_arm_vcpu_finalize(vcpu, what);
1400 	}
1401 	default:
1402 		r = -EINVAL;
1403 	}
1404 
1405 	return r;
1406 }
1407 
1408 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1409 {
1410 
1411 }
1412 
1413 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1414 					const struct kvm_memory_slot *memslot)
1415 {
1416 	kvm_flush_remote_tlbs(kvm);
1417 }
1418 
1419 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1420 					struct kvm_arm_device_addr *dev_addr)
1421 {
1422 	switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1423 	case KVM_ARM_DEVICE_VGIC_V2:
1424 		if (!vgic_present)
1425 			return -ENXIO;
1426 		return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
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 sections from kmemleak so that they don't get peeked
2114 	 * at, which would end badly once inaccessible.
2115 	 */
2116 	kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
2117 	kmemleak_free_part_phys(hyp_mem_base, hyp_mem_size);
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 		if (!is_kernel_in_hyp_mode())
2275 			kvm_mode = KVM_MODE_PROTECTED;
2276 		else
2277 			pr_warn_once("Protected KVM not available with VHE\n");
2278 
2279 		return 0;
2280 	}
2281 
2282 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2283 		kvm_mode = KVM_MODE_DEFAULT;
2284 		return 0;
2285 	}
2286 
2287 	if (strcmp(arg, "none") == 0) {
2288 		kvm_mode = KVM_MODE_NONE;
2289 		return 0;
2290 	}
2291 
2292 	return -EINVAL;
2293 }
2294 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2295 
2296 enum kvm_mode kvm_get_mode(void)
2297 {
2298 	return kvm_mode;
2299 }
2300 
2301 static int arm_init(void)
2302 {
2303 	int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2304 	return rc;
2305 }
2306 
2307 module_init(arm_init);
2308