xref: /openbmc/linux/arch/arm64/kvm/arm.c (revision 9ad685db)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
4  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
5  */
6 
7 #include <linux/bug.h>
8 #include <linux/cpu_pm.h>
9 #include <linux/entry-kvm.h>
10 #include <linux/errno.h>
11 #include <linux/err.h>
12 #include <linux/kvm_host.h>
13 #include <linux/list.h>
14 #include <linux/module.h>
15 #include <linux/vmalloc.h>
16 #include <linux/fs.h>
17 #include <linux/mman.h>
18 #include <linux/sched.h>
19 #include <linux/kvm.h>
20 #include <linux/kvm_irqfd.h>
21 #include <linux/irqbypass.h>
22 #include <linux/sched/stat.h>
23 #include <linux/psci.h>
24 #include <trace/events/kvm.h>
25 
26 #define CREATE_TRACE_POINTS
27 #include "trace_arm.h"
28 
29 #include <linux/uaccess.h>
30 #include <asm/ptrace.h>
31 #include <asm/mman.h>
32 #include <asm/tlbflush.h>
33 #include <asm/cacheflush.h>
34 #include <asm/cpufeature.h>
35 #include <asm/virt.h>
36 #include <asm/kvm_arm.h>
37 #include <asm/kvm_asm.h>
38 #include <asm/kvm_mmu.h>
39 #include <asm/kvm_pkvm.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 
49 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
50 
51 DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
52 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
53 
54 static bool vgic_present;
55 
56 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
57 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
58 
59 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
60 {
61 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
62 }
63 
64 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
65 			    struct kvm_enable_cap *cap)
66 {
67 	int r;
68 
69 	if (cap->flags)
70 		return -EINVAL;
71 
72 	switch (cap->cap) {
73 	case KVM_CAP_ARM_NISV_TO_USER:
74 		r = 0;
75 		set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
76 			&kvm->arch.flags);
77 		break;
78 	case KVM_CAP_ARM_MTE:
79 		mutex_lock(&kvm->lock);
80 		if (!system_supports_mte() || kvm->created_vcpus) {
81 			r = -EINVAL;
82 		} else {
83 			r = 0;
84 			set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
85 		}
86 		mutex_unlock(&kvm->lock);
87 		break;
88 	case KVM_CAP_ARM_SYSTEM_SUSPEND:
89 		r = 0;
90 		set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags);
91 		break;
92 	default:
93 		r = -EINVAL;
94 		break;
95 	}
96 
97 	return r;
98 }
99 
100 static int kvm_arm_default_max_vcpus(void)
101 {
102 	return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
103 }
104 
105 static void set_default_spectre(struct kvm *kvm)
106 {
107 	/*
108 	 * The default is to expose CSV2 == 1 if the HW isn't affected.
109 	 * Although this is a per-CPU feature, we make it global because
110 	 * asymmetric systems are just a nuisance.
111 	 *
112 	 * Userspace can override this as long as it doesn't promise
113 	 * the impossible.
114 	 */
115 	if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
116 		kvm->arch.pfr0_csv2 = 1;
117 	if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
118 		kvm->arch.pfr0_csv3 = 1;
119 }
120 
121 /**
122  * kvm_arch_init_vm - initializes a VM data structure
123  * @kvm:	pointer to the KVM struct
124  */
125 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
126 {
127 	int ret;
128 
129 	ret = kvm_share_hyp(kvm, kvm + 1);
130 	if (ret)
131 		return ret;
132 
133 	ret = pkvm_init_host_vm(kvm);
134 	if (ret)
135 		goto err_unshare_kvm;
136 
137 	if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL_ACCOUNT)) {
138 		ret = -ENOMEM;
139 		goto err_unshare_kvm;
140 	}
141 	cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
142 
143 	ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type);
144 	if (ret)
145 		goto err_free_cpumask;
146 
147 	kvm_vgic_early_init(kvm);
148 
149 	/* The maximum number of VCPUs is limited by the host's GIC model */
150 	kvm->max_vcpus = kvm_arm_default_max_vcpus();
151 
152 	set_default_spectre(kvm);
153 	kvm_arm_init_hypercalls(kvm);
154 
155 	/*
156 	 * Initialise the default PMUver before there is a chance to
157 	 * create an actual PMU.
158 	 */
159 	kvm->arch.dfr0_pmuver.imp = kvm_arm_pmu_get_pmuver_limit();
160 
161 	return 0;
162 
163 err_free_cpumask:
164 	free_cpumask_var(kvm->arch.supported_cpus);
165 err_unshare_kvm:
166 	kvm_unshare_hyp(kvm, kvm + 1);
167 	return ret;
168 }
169 
170 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
171 {
172 	return VM_FAULT_SIGBUS;
173 }
174 
175 
176 /**
177  * kvm_arch_destroy_vm - destroy the VM data structure
178  * @kvm:	pointer to the KVM struct
179  */
180 void kvm_arch_destroy_vm(struct kvm *kvm)
181 {
182 	bitmap_free(kvm->arch.pmu_filter);
183 	free_cpumask_var(kvm->arch.supported_cpus);
184 
185 	kvm_vgic_destroy(kvm);
186 
187 	if (is_protected_kvm_enabled())
188 		pkvm_destroy_hyp_vm(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 	case KVM_CAP_IRQFD_RESAMPLE:
222 		r = 1;
223 		break;
224 	case KVM_CAP_SET_GUEST_DEBUG2:
225 		return KVM_GUESTDBG_VALID_MASK;
226 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
227 		r = 1;
228 		break;
229 	case KVM_CAP_NR_VCPUS:
230 		/*
231 		 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
232 		 * architectures, as it does not always bound it to
233 		 * KVM_CAP_MAX_VCPUS. It should not matter much because
234 		 * this is just an advisory value.
235 		 */
236 		r = min_t(unsigned int, num_online_cpus(),
237 			  kvm_arm_default_max_vcpus());
238 		break;
239 	case KVM_CAP_MAX_VCPUS:
240 	case KVM_CAP_MAX_VCPU_ID:
241 		if (kvm)
242 			r = kvm->max_vcpus;
243 		else
244 			r = kvm_arm_default_max_vcpus();
245 		break;
246 	case KVM_CAP_MSI_DEVID:
247 		if (!kvm)
248 			r = -EINVAL;
249 		else
250 			r = kvm->arch.vgic.msis_require_devid;
251 		break;
252 	case KVM_CAP_ARM_USER_IRQ:
253 		/*
254 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
255 		 * (bump this number if adding more devices)
256 		 */
257 		r = 1;
258 		break;
259 	case KVM_CAP_ARM_MTE:
260 		r = system_supports_mte();
261 		break;
262 	case KVM_CAP_STEAL_TIME:
263 		r = kvm_arm_pvtime_supported();
264 		break;
265 	case KVM_CAP_ARM_EL1_32BIT:
266 		r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
267 		break;
268 	case KVM_CAP_GUEST_DEBUG_HW_BPS:
269 		r = get_num_brps();
270 		break;
271 	case KVM_CAP_GUEST_DEBUG_HW_WPS:
272 		r = get_num_wrps();
273 		break;
274 	case KVM_CAP_ARM_PMU_V3:
275 		r = kvm_arm_support_pmu_v3();
276 		break;
277 	case KVM_CAP_ARM_INJECT_SERROR_ESR:
278 		r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
279 		break;
280 	case KVM_CAP_ARM_VM_IPA_SIZE:
281 		r = get_kvm_ipa_limit();
282 		break;
283 	case KVM_CAP_ARM_SVE:
284 		r = system_supports_sve();
285 		break;
286 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
287 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
288 		r = system_has_full_ptr_auth();
289 		break;
290 	default:
291 		r = 0;
292 	}
293 
294 	return r;
295 }
296 
297 long kvm_arch_dev_ioctl(struct file *filp,
298 			unsigned int ioctl, unsigned long arg)
299 {
300 	return -EINVAL;
301 }
302 
303 struct kvm *kvm_arch_alloc_vm(void)
304 {
305 	size_t sz = sizeof(struct kvm);
306 
307 	if (!has_vhe())
308 		return kzalloc(sz, GFP_KERNEL_ACCOUNT);
309 
310 	return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
311 }
312 
313 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
314 {
315 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
316 		return -EBUSY;
317 
318 	if (id >= kvm->max_vcpus)
319 		return -EINVAL;
320 
321 	return 0;
322 }
323 
324 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
325 {
326 	int err;
327 
328 	/* Force users to call KVM_ARM_VCPU_INIT */
329 	vcpu->arch.target = -1;
330 	bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
331 
332 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
333 
334 	/*
335 	 * Default value for the FP state, will be overloaded at load
336 	 * time if we support FP (pretty likely)
337 	 */
338 	vcpu->arch.fp_state = FP_STATE_FREE;
339 
340 	/* Set up the timer */
341 	kvm_timer_vcpu_init(vcpu);
342 
343 	kvm_pmu_vcpu_init(vcpu);
344 
345 	kvm_arm_reset_debug_ptr(vcpu);
346 
347 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
348 
349 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
350 
351 	err = kvm_vgic_vcpu_init(vcpu);
352 	if (err)
353 		return err;
354 
355 	return kvm_share_hyp(vcpu, vcpu + 1);
356 }
357 
358 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
359 {
360 }
361 
362 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
363 {
364 	if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
365 		static_branch_dec(&userspace_irqchip_in_use);
366 
367 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
368 	kvm_timer_vcpu_terminate(vcpu);
369 	kvm_pmu_vcpu_destroy(vcpu);
370 
371 	kvm_arm_vcpu_destroy(vcpu);
372 }
373 
374 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
375 {
376 
377 }
378 
379 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
380 {
381 
382 }
383 
384 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
385 {
386 	struct kvm_s2_mmu *mmu;
387 	int *last_ran;
388 
389 	mmu = vcpu->arch.hw_mmu;
390 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
391 
392 	/*
393 	 * We guarantee that both TLBs and I-cache are private to each
394 	 * vcpu. If detecting that a vcpu from the same VM has
395 	 * previously run on the same physical CPU, call into the
396 	 * hypervisor code to nuke the relevant contexts.
397 	 *
398 	 * We might get preempted before the vCPU actually runs, but
399 	 * over-invalidation doesn't affect correctness.
400 	 */
401 	if (*last_ran != vcpu->vcpu_id) {
402 		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
403 		*last_ran = vcpu->vcpu_id;
404 	}
405 
406 	vcpu->cpu = cpu;
407 
408 	kvm_vgic_load(vcpu);
409 	kvm_timer_vcpu_load(vcpu);
410 	if (has_vhe())
411 		kvm_vcpu_load_sysregs_vhe(vcpu);
412 	kvm_arch_vcpu_load_fp(vcpu);
413 	kvm_vcpu_pmu_restore_guest(vcpu);
414 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
415 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
416 
417 	if (single_task_running())
418 		vcpu_clear_wfx_traps(vcpu);
419 	else
420 		vcpu_set_wfx_traps(vcpu);
421 
422 	if (vcpu_has_ptrauth(vcpu))
423 		vcpu_ptrauth_disable(vcpu);
424 	kvm_arch_vcpu_load_debug_state_flags(vcpu);
425 
426 	if (!cpumask_test_cpu(smp_processor_id(), vcpu->kvm->arch.supported_cpus))
427 		vcpu_set_on_unsupported_cpu(vcpu);
428 }
429 
430 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
431 {
432 	kvm_arch_vcpu_put_debug_state_flags(vcpu);
433 	kvm_arch_vcpu_put_fp(vcpu);
434 	if (has_vhe())
435 		kvm_vcpu_put_sysregs_vhe(vcpu);
436 	kvm_timer_vcpu_put(vcpu);
437 	kvm_vgic_put(vcpu);
438 	kvm_vcpu_pmu_restore_host(vcpu);
439 	kvm_arm_vmid_clear_active();
440 
441 	vcpu_clear_on_unsupported_cpu(vcpu);
442 	vcpu->cpu = -1;
443 }
444 
445 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
446 {
447 	vcpu->arch.mp_state.mp_state = KVM_MP_STATE_STOPPED;
448 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
449 	kvm_vcpu_kick(vcpu);
450 }
451 
452 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
453 {
454 	return vcpu->arch.mp_state.mp_state == KVM_MP_STATE_STOPPED;
455 }
456 
457 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
458 {
459 	vcpu->arch.mp_state.mp_state = KVM_MP_STATE_SUSPENDED;
460 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
461 	kvm_vcpu_kick(vcpu);
462 }
463 
464 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
465 {
466 	return vcpu->arch.mp_state.mp_state == KVM_MP_STATE_SUSPENDED;
467 }
468 
469 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
470 				    struct kvm_mp_state *mp_state)
471 {
472 	*mp_state = vcpu->arch.mp_state;
473 
474 	return 0;
475 }
476 
477 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
478 				    struct kvm_mp_state *mp_state)
479 {
480 	int ret = 0;
481 
482 	switch (mp_state->mp_state) {
483 	case KVM_MP_STATE_RUNNABLE:
484 		vcpu->arch.mp_state = *mp_state;
485 		break;
486 	case KVM_MP_STATE_STOPPED:
487 		kvm_arm_vcpu_power_off(vcpu);
488 		break;
489 	case KVM_MP_STATE_SUSPENDED:
490 		kvm_arm_vcpu_suspend(vcpu);
491 		break;
492 	default:
493 		ret = -EINVAL;
494 	}
495 
496 	return ret;
497 }
498 
499 /**
500  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
501  * @v:		The VCPU pointer
502  *
503  * If the guest CPU is not waiting for interrupts or an interrupt line is
504  * asserted, the CPU is by definition runnable.
505  */
506 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
507 {
508 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
509 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
510 		&& !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
511 }
512 
513 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
514 {
515 	return vcpu_mode_priv(vcpu);
516 }
517 
518 #ifdef CONFIG_GUEST_PERF_EVENTS
519 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
520 {
521 	return *vcpu_pc(vcpu);
522 }
523 #endif
524 
525 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
526 {
527 	return vcpu->arch.target >= 0;
528 }
529 
530 /*
531  * Handle both the initialisation that is being done when the vcpu is
532  * run for the first time, as well as the updates that must be
533  * performed each time we get a new thread dealing with this vcpu.
534  */
535 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
536 {
537 	struct kvm *kvm = vcpu->kvm;
538 	int ret;
539 
540 	if (!kvm_vcpu_initialized(vcpu))
541 		return -ENOEXEC;
542 
543 	if (!kvm_arm_vcpu_is_finalized(vcpu))
544 		return -EPERM;
545 
546 	ret = kvm_arch_vcpu_run_map_fp(vcpu);
547 	if (ret)
548 		return ret;
549 
550 	if (likely(vcpu_has_run_once(vcpu)))
551 		return 0;
552 
553 	kvm_arm_vcpu_init_debug(vcpu);
554 
555 	if (likely(irqchip_in_kernel(kvm))) {
556 		/*
557 		 * Map the VGIC hardware resources before running a vcpu the
558 		 * first time on this VM.
559 		 */
560 		ret = kvm_vgic_map_resources(kvm);
561 		if (ret)
562 			return ret;
563 	}
564 
565 	ret = kvm_timer_enable(vcpu);
566 	if (ret)
567 		return ret;
568 
569 	ret = kvm_arm_pmu_v3_enable(vcpu);
570 	if (ret)
571 		return ret;
572 
573 	if (is_protected_kvm_enabled()) {
574 		ret = pkvm_create_hyp_vm(kvm);
575 		if (ret)
576 			return ret;
577 	}
578 
579 	if (!irqchip_in_kernel(kvm)) {
580 		/*
581 		 * Tell the rest of the code that there are userspace irqchip
582 		 * VMs in the wild.
583 		 */
584 		static_branch_inc(&userspace_irqchip_in_use);
585 	}
586 
587 	/*
588 	 * Initialize traps for protected VMs.
589 	 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
590 	 * the code is in place for first run initialization at EL2.
591 	 */
592 	if (kvm_vm_is_protected(kvm))
593 		kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
594 
595 	mutex_lock(&kvm->lock);
596 	set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
597 	mutex_unlock(&kvm->lock);
598 
599 	return ret;
600 }
601 
602 bool kvm_arch_intc_initialized(struct kvm *kvm)
603 {
604 	return vgic_initialized(kvm);
605 }
606 
607 void kvm_arm_halt_guest(struct kvm *kvm)
608 {
609 	unsigned long i;
610 	struct kvm_vcpu *vcpu;
611 
612 	kvm_for_each_vcpu(i, vcpu, kvm)
613 		vcpu->arch.pause = true;
614 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
615 }
616 
617 void kvm_arm_resume_guest(struct kvm *kvm)
618 {
619 	unsigned long i;
620 	struct kvm_vcpu *vcpu;
621 
622 	kvm_for_each_vcpu(i, vcpu, kvm) {
623 		vcpu->arch.pause = false;
624 		__kvm_vcpu_wake_up(vcpu);
625 	}
626 }
627 
628 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
629 {
630 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
631 
632 	rcuwait_wait_event(wait,
633 			   (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
634 			   TASK_INTERRUPTIBLE);
635 
636 	if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
637 		/* Awaken to handle a signal, request we sleep again later. */
638 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
639 	}
640 
641 	/*
642 	 * Make sure we will observe a potential reset request if we've
643 	 * observed a change to the power state. Pairs with the smp_wmb() in
644 	 * kvm_psci_vcpu_on().
645 	 */
646 	smp_rmb();
647 }
648 
649 /**
650  * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
651  * @vcpu:	The VCPU pointer
652  *
653  * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
654  * the vCPU is runnable.  The vCPU may or may not be scheduled out, depending
655  * on when a wake event arrives, e.g. there may already be a pending wake event.
656  */
657 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
658 {
659 	/*
660 	 * Sync back the state of the GIC CPU interface so that we have
661 	 * the latest PMR and group enables. This ensures that
662 	 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
663 	 * we have pending interrupts, e.g. when determining if the
664 	 * vCPU should block.
665 	 *
666 	 * For the same reason, we want to tell GICv4 that we need
667 	 * doorbells to be signalled, should an interrupt become pending.
668 	 */
669 	preempt_disable();
670 	kvm_vgic_vmcr_sync(vcpu);
671 	vgic_v4_put(vcpu, true);
672 	preempt_enable();
673 
674 	kvm_vcpu_halt(vcpu);
675 	vcpu_clear_flag(vcpu, IN_WFIT);
676 
677 	preempt_disable();
678 	vgic_v4_load(vcpu);
679 	preempt_enable();
680 }
681 
682 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
683 {
684 	if (!kvm_arm_vcpu_suspended(vcpu))
685 		return 1;
686 
687 	kvm_vcpu_wfi(vcpu);
688 
689 	/*
690 	 * The suspend state is sticky; we do not leave it until userspace
691 	 * explicitly marks the vCPU as runnable. Request that we suspend again
692 	 * later.
693 	 */
694 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
695 
696 	/*
697 	 * Check to make sure the vCPU is actually runnable. If so, exit to
698 	 * userspace informing it of the wakeup condition.
699 	 */
700 	if (kvm_arch_vcpu_runnable(vcpu)) {
701 		memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
702 		vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
703 		vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
704 		return 0;
705 	}
706 
707 	/*
708 	 * Otherwise, we were unblocked to process a different event, such as a
709 	 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
710 	 * process the event.
711 	 */
712 	return 1;
713 }
714 
715 /**
716  * check_vcpu_requests - check and handle pending vCPU requests
717  * @vcpu:	the VCPU pointer
718  *
719  * Return: 1 if we should enter the guest
720  *	   0 if we should exit to userspace
721  *	   < 0 if we should exit to userspace, where the return value indicates
722  *	   an error
723  */
724 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
725 {
726 	if (kvm_request_pending(vcpu)) {
727 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
728 			kvm_vcpu_sleep(vcpu);
729 
730 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
731 			kvm_reset_vcpu(vcpu);
732 
733 		/*
734 		 * Clear IRQ_PENDING requests that were made to guarantee
735 		 * that a VCPU sees new virtual interrupts.
736 		 */
737 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
738 
739 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
740 			kvm_update_stolen_time(vcpu);
741 
742 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
743 			/* The distributor enable bits were changed */
744 			preempt_disable();
745 			vgic_v4_put(vcpu, false);
746 			vgic_v4_load(vcpu);
747 			preempt_enable();
748 		}
749 
750 		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
751 			kvm_pmu_handle_pmcr(vcpu,
752 					    __vcpu_sys_reg(vcpu, PMCR_EL0));
753 
754 		if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
755 			return kvm_vcpu_suspend(vcpu);
756 
757 		if (kvm_dirty_ring_check_request(vcpu))
758 			return 0;
759 	}
760 
761 	return 1;
762 }
763 
764 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
765 {
766 	if (likely(!vcpu_mode_is_32bit(vcpu)))
767 		return false;
768 
769 	return !kvm_supports_32bit_el0();
770 }
771 
772 /**
773  * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
774  * @vcpu:	The VCPU pointer
775  * @ret:	Pointer to write optional return code
776  *
777  * Returns: true if the VCPU needs to return to a preemptible + interruptible
778  *	    and skip guest entry.
779  *
780  * This function disambiguates between two different types of exits: exits to a
781  * preemptible + interruptible kernel context and exits to userspace. For an
782  * exit to userspace, this function will write the return code to ret and return
783  * true. For an exit to preemptible + interruptible kernel context (i.e. check
784  * for pending work and re-enter), return true without writing to ret.
785  */
786 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
787 {
788 	struct kvm_run *run = vcpu->run;
789 
790 	/*
791 	 * If we're using a userspace irqchip, then check if we need
792 	 * to tell a userspace irqchip about timer or PMU level
793 	 * changes and if so, exit to userspace (the actual level
794 	 * state gets updated in kvm_timer_update_run and
795 	 * kvm_pmu_update_run below).
796 	 */
797 	if (static_branch_unlikely(&userspace_irqchip_in_use)) {
798 		if (kvm_timer_should_notify_user(vcpu) ||
799 		    kvm_pmu_should_notify_user(vcpu)) {
800 			*ret = -EINTR;
801 			run->exit_reason = KVM_EXIT_INTR;
802 			return true;
803 		}
804 	}
805 
806 	if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
807 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
808 		run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
809 		run->fail_entry.cpu = smp_processor_id();
810 		*ret = 0;
811 		return true;
812 	}
813 
814 	return kvm_request_pending(vcpu) ||
815 			xfer_to_guest_mode_work_pending();
816 }
817 
818 /*
819  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
820  * the vCPU is running.
821  *
822  * This must be noinstr as instrumentation may make use of RCU, and this is not
823  * safe during the EQS.
824  */
825 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
826 {
827 	int ret;
828 
829 	guest_state_enter_irqoff();
830 	ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
831 	guest_state_exit_irqoff();
832 
833 	return ret;
834 }
835 
836 /**
837  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
838  * @vcpu:	The VCPU pointer
839  *
840  * This function is called through the VCPU_RUN ioctl called from user space. It
841  * will execute VM code in a loop until the time slice for the process is used
842  * or some emulation is needed from user space in which case the function will
843  * return with return value 0 and with the kvm_run structure filled in with the
844  * required data for the requested emulation.
845  */
846 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
847 {
848 	struct kvm_run *run = vcpu->run;
849 	int ret;
850 
851 	if (run->exit_reason == KVM_EXIT_MMIO) {
852 		ret = kvm_handle_mmio_return(vcpu);
853 		if (ret)
854 			return ret;
855 	}
856 
857 	vcpu_load(vcpu);
858 
859 	if (run->immediate_exit) {
860 		ret = -EINTR;
861 		goto out;
862 	}
863 
864 	kvm_sigset_activate(vcpu);
865 
866 	ret = 1;
867 	run->exit_reason = KVM_EXIT_UNKNOWN;
868 	run->flags = 0;
869 	while (ret > 0) {
870 		/*
871 		 * Check conditions before entering the guest
872 		 */
873 		ret = xfer_to_guest_mode_handle_work(vcpu);
874 		if (!ret)
875 			ret = 1;
876 
877 		if (ret > 0)
878 			ret = check_vcpu_requests(vcpu);
879 
880 		/*
881 		 * Preparing the interrupts to be injected also
882 		 * involves poking the GIC, which must be done in a
883 		 * non-preemptible context.
884 		 */
885 		preempt_disable();
886 
887 		/*
888 		 * The VMID allocator only tracks active VMIDs per
889 		 * physical CPU, and therefore the VMID allocated may not be
890 		 * preserved on VMID roll-over if the task was preempted,
891 		 * making a thread's VMID inactive. So we need to call
892 		 * kvm_arm_vmid_update() in non-premptible context.
893 		 */
894 		kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid);
895 
896 		kvm_pmu_flush_hwstate(vcpu);
897 
898 		local_irq_disable();
899 
900 		kvm_vgic_flush_hwstate(vcpu);
901 
902 		kvm_pmu_update_vcpu_events(vcpu);
903 
904 		/*
905 		 * Ensure we set mode to IN_GUEST_MODE after we disable
906 		 * interrupts and before the final VCPU requests check.
907 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
908 		 * Documentation/virt/kvm/vcpu-requests.rst
909 		 */
910 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
911 
912 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
913 			vcpu->mode = OUTSIDE_GUEST_MODE;
914 			isb(); /* Ensure work in x_flush_hwstate is committed */
915 			kvm_pmu_sync_hwstate(vcpu);
916 			if (static_branch_unlikely(&userspace_irqchip_in_use))
917 				kvm_timer_sync_user(vcpu);
918 			kvm_vgic_sync_hwstate(vcpu);
919 			local_irq_enable();
920 			preempt_enable();
921 			continue;
922 		}
923 
924 		kvm_arm_setup_debug(vcpu);
925 		kvm_arch_vcpu_ctxflush_fp(vcpu);
926 
927 		/**************************************************************
928 		 * Enter the guest
929 		 */
930 		trace_kvm_entry(*vcpu_pc(vcpu));
931 		guest_timing_enter_irqoff();
932 
933 		ret = kvm_arm_vcpu_enter_exit(vcpu);
934 
935 		vcpu->mode = OUTSIDE_GUEST_MODE;
936 		vcpu->stat.exits++;
937 		/*
938 		 * Back from guest
939 		 *************************************************************/
940 
941 		kvm_arm_clear_debug(vcpu);
942 
943 		/*
944 		 * We must sync the PMU state before the vgic state so
945 		 * that the vgic can properly sample the updated state of the
946 		 * interrupt line.
947 		 */
948 		kvm_pmu_sync_hwstate(vcpu);
949 
950 		/*
951 		 * Sync the vgic state before syncing the timer state because
952 		 * the timer code needs to know if the virtual timer
953 		 * interrupts are active.
954 		 */
955 		kvm_vgic_sync_hwstate(vcpu);
956 
957 		/*
958 		 * Sync the timer hardware state before enabling interrupts as
959 		 * we don't want vtimer interrupts to race with syncing the
960 		 * timer virtual interrupt state.
961 		 */
962 		if (static_branch_unlikely(&userspace_irqchip_in_use))
963 			kvm_timer_sync_user(vcpu);
964 
965 		kvm_arch_vcpu_ctxsync_fp(vcpu);
966 
967 		/*
968 		 * We must ensure that any pending interrupts are taken before
969 		 * we exit guest timing so that timer ticks are accounted as
970 		 * guest time. Transiently unmask interrupts so that any
971 		 * pending interrupts are taken.
972 		 *
973 		 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
974 		 * context synchronization event) is necessary to ensure that
975 		 * pending interrupts are taken.
976 		 */
977 		if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
978 			local_irq_enable();
979 			isb();
980 			local_irq_disable();
981 		}
982 
983 		guest_timing_exit_irqoff();
984 
985 		local_irq_enable();
986 
987 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
988 
989 		/* Exit types that need handling before we can be preempted */
990 		handle_exit_early(vcpu, ret);
991 
992 		preempt_enable();
993 
994 		/*
995 		 * The ARMv8 architecture doesn't give the hypervisor
996 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
997 		 * if implemented by the CPU. If we spot the guest in such
998 		 * state and that we decided it wasn't supposed to do so (like
999 		 * with the asymmetric AArch32 case), return to userspace with
1000 		 * a fatal error.
1001 		 */
1002 		if (vcpu_mode_is_bad_32bit(vcpu)) {
1003 			/*
1004 			 * As we have caught the guest red-handed, decide that
1005 			 * it isn't fit for purpose anymore by making the vcpu
1006 			 * invalid. The VMM can try and fix it by issuing  a
1007 			 * KVM_ARM_VCPU_INIT if it really wants to.
1008 			 */
1009 			vcpu->arch.target = -1;
1010 			ret = ARM_EXCEPTION_IL;
1011 		}
1012 
1013 		ret = handle_exit(vcpu, ret);
1014 	}
1015 
1016 	/* Tell userspace about in-kernel device output levels */
1017 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1018 		kvm_timer_update_run(vcpu);
1019 		kvm_pmu_update_run(vcpu);
1020 	}
1021 
1022 	kvm_sigset_deactivate(vcpu);
1023 
1024 out:
1025 	/*
1026 	 * In the unlikely event that we are returning to userspace
1027 	 * with pending exceptions or PC adjustment, commit these
1028 	 * adjustments in order to give userspace a consistent view of
1029 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1030 	 * being preempt-safe on VHE.
1031 	 */
1032 	if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1033 		     vcpu_get_flag(vcpu, INCREMENT_PC)))
1034 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
1035 
1036 	vcpu_put(vcpu);
1037 	return ret;
1038 }
1039 
1040 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1041 {
1042 	int bit_index;
1043 	bool set;
1044 	unsigned long *hcr;
1045 
1046 	if (number == KVM_ARM_IRQ_CPU_IRQ)
1047 		bit_index = __ffs(HCR_VI);
1048 	else /* KVM_ARM_IRQ_CPU_FIQ */
1049 		bit_index = __ffs(HCR_VF);
1050 
1051 	hcr = vcpu_hcr(vcpu);
1052 	if (level)
1053 		set = test_and_set_bit(bit_index, hcr);
1054 	else
1055 		set = test_and_clear_bit(bit_index, hcr);
1056 
1057 	/*
1058 	 * If we didn't change anything, no need to wake up or kick other CPUs
1059 	 */
1060 	if (set == level)
1061 		return 0;
1062 
1063 	/*
1064 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1065 	 * trigger a world-switch round on the running physical CPU to set the
1066 	 * virtual IRQ/FIQ fields in the HCR appropriately.
1067 	 */
1068 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1069 	kvm_vcpu_kick(vcpu);
1070 
1071 	return 0;
1072 }
1073 
1074 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1075 			  bool line_status)
1076 {
1077 	u32 irq = irq_level->irq;
1078 	unsigned int irq_type, vcpu_idx, irq_num;
1079 	int nrcpus = atomic_read(&kvm->online_vcpus);
1080 	struct kvm_vcpu *vcpu = NULL;
1081 	bool level = irq_level->level;
1082 
1083 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1084 	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1085 	vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1086 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1087 
1088 	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1089 
1090 	switch (irq_type) {
1091 	case KVM_ARM_IRQ_TYPE_CPU:
1092 		if (irqchip_in_kernel(kvm))
1093 			return -ENXIO;
1094 
1095 		if (vcpu_idx >= nrcpus)
1096 			return -EINVAL;
1097 
1098 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1099 		if (!vcpu)
1100 			return -EINVAL;
1101 
1102 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1103 			return -EINVAL;
1104 
1105 		return vcpu_interrupt_line(vcpu, irq_num, level);
1106 	case KVM_ARM_IRQ_TYPE_PPI:
1107 		if (!irqchip_in_kernel(kvm))
1108 			return -ENXIO;
1109 
1110 		if (vcpu_idx >= nrcpus)
1111 			return -EINVAL;
1112 
1113 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1114 		if (!vcpu)
1115 			return -EINVAL;
1116 
1117 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1118 			return -EINVAL;
1119 
1120 		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1121 	case KVM_ARM_IRQ_TYPE_SPI:
1122 		if (!irqchip_in_kernel(kvm))
1123 			return -ENXIO;
1124 
1125 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1126 			return -EINVAL;
1127 
1128 		return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1129 	}
1130 
1131 	return -EINVAL;
1132 }
1133 
1134 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1135 			       const struct kvm_vcpu_init *init)
1136 {
1137 	unsigned int i, ret;
1138 	u32 phys_target = kvm_target_cpu();
1139 
1140 	if (init->target != phys_target)
1141 		return -EINVAL;
1142 
1143 	/*
1144 	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1145 	 * use the same target.
1146 	 */
1147 	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1148 		return -EINVAL;
1149 
1150 	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1151 	for (i = 0; i < sizeof(init->features) * 8; i++) {
1152 		bool set = (init->features[i / 32] & (1 << (i % 32)));
1153 
1154 		if (set && i >= KVM_VCPU_MAX_FEATURES)
1155 			return -ENOENT;
1156 
1157 		/*
1158 		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1159 		 * use the same feature set.
1160 		 */
1161 		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1162 		    test_bit(i, vcpu->arch.features) != set)
1163 			return -EINVAL;
1164 
1165 		if (set)
1166 			set_bit(i, vcpu->arch.features);
1167 	}
1168 
1169 	vcpu->arch.target = phys_target;
1170 
1171 	/* Now we know what it is, we can reset it. */
1172 	ret = kvm_reset_vcpu(vcpu);
1173 	if (ret) {
1174 		vcpu->arch.target = -1;
1175 		bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1176 	}
1177 
1178 	return ret;
1179 }
1180 
1181 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1182 					 struct kvm_vcpu_init *init)
1183 {
1184 	int ret;
1185 
1186 	ret = kvm_vcpu_set_target(vcpu, init);
1187 	if (ret)
1188 		return ret;
1189 
1190 	/*
1191 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1192 	 * guest MMU is turned off and flush the caches as needed.
1193 	 *
1194 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1195 	 * ensuring that the data side is always coherent. We still
1196 	 * need to invalidate the I-cache though, as FWB does *not*
1197 	 * imply CTR_EL0.DIC.
1198 	 */
1199 	if (vcpu_has_run_once(vcpu)) {
1200 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1201 			stage2_unmap_vm(vcpu->kvm);
1202 		else
1203 			icache_inval_all_pou();
1204 	}
1205 
1206 	vcpu_reset_hcr(vcpu);
1207 	vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1208 
1209 	/*
1210 	 * Handle the "start in power-off" case.
1211 	 */
1212 	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1213 		kvm_arm_vcpu_power_off(vcpu);
1214 	else
1215 		vcpu->arch.mp_state.mp_state = KVM_MP_STATE_RUNNABLE;
1216 
1217 	return 0;
1218 }
1219 
1220 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1221 				 struct kvm_device_attr *attr)
1222 {
1223 	int ret = -ENXIO;
1224 
1225 	switch (attr->group) {
1226 	default:
1227 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1228 		break;
1229 	}
1230 
1231 	return ret;
1232 }
1233 
1234 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1235 				 struct kvm_device_attr *attr)
1236 {
1237 	int ret = -ENXIO;
1238 
1239 	switch (attr->group) {
1240 	default:
1241 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1242 		break;
1243 	}
1244 
1245 	return ret;
1246 }
1247 
1248 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1249 				 struct kvm_device_attr *attr)
1250 {
1251 	int ret = -ENXIO;
1252 
1253 	switch (attr->group) {
1254 	default:
1255 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1256 		break;
1257 	}
1258 
1259 	return ret;
1260 }
1261 
1262 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1263 				   struct kvm_vcpu_events *events)
1264 {
1265 	memset(events, 0, sizeof(*events));
1266 
1267 	return __kvm_arm_vcpu_get_events(vcpu, events);
1268 }
1269 
1270 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1271 				   struct kvm_vcpu_events *events)
1272 {
1273 	int i;
1274 
1275 	/* check whether the reserved field is zero */
1276 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1277 		if (events->reserved[i])
1278 			return -EINVAL;
1279 
1280 	/* check whether the pad field is zero */
1281 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1282 		if (events->exception.pad[i])
1283 			return -EINVAL;
1284 
1285 	return __kvm_arm_vcpu_set_events(vcpu, events);
1286 }
1287 
1288 long kvm_arch_vcpu_ioctl(struct file *filp,
1289 			 unsigned int ioctl, unsigned long arg)
1290 {
1291 	struct kvm_vcpu *vcpu = filp->private_data;
1292 	void __user *argp = (void __user *)arg;
1293 	struct kvm_device_attr attr;
1294 	long r;
1295 
1296 	switch (ioctl) {
1297 	case KVM_ARM_VCPU_INIT: {
1298 		struct kvm_vcpu_init init;
1299 
1300 		r = -EFAULT;
1301 		if (copy_from_user(&init, argp, sizeof(init)))
1302 			break;
1303 
1304 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1305 		break;
1306 	}
1307 	case KVM_SET_ONE_REG:
1308 	case KVM_GET_ONE_REG: {
1309 		struct kvm_one_reg reg;
1310 
1311 		r = -ENOEXEC;
1312 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1313 			break;
1314 
1315 		r = -EFAULT;
1316 		if (copy_from_user(&reg, argp, sizeof(reg)))
1317 			break;
1318 
1319 		/*
1320 		 * We could owe a reset due to PSCI. Handle the pending reset
1321 		 * here to ensure userspace register accesses are ordered after
1322 		 * the reset.
1323 		 */
1324 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1325 			kvm_reset_vcpu(vcpu);
1326 
1327 		if (ioctl == KVM_SET_ONE_REG)
1328 			r = kvm_arm_set_reg(vcpu, &reg);
1329 		else
1330 			r = kvm_arm_get_reg(vcpu, &reg);
1331 		break;
1332 	}
1333 	case KVM_GET_REG_LIST: {
1334 		struct kvm_reg_list __user *user_list = argp;
1335 		struct kvm_reg_list reg_list;
1336 		unsigned n;
1337 
1338 		r = -ENOEXEC;
1339 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1340 			break;
1341 
1342 		r = -EPERM;
1343 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1344 			break;
1345 
1346 		r = -EFAULT;
1347 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1348 			break;
1349 		n = reg_list.n;
1350 		reg_list.n = kvm_arm_num_regs(vcpu);
1351 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1352 			break;
1353 		r = -E2BIG;
1354 		if (n < reg_list.n)
1355 			break;
1356 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1357 		break;
1358 	}
1359 	case KVM_SET_DEVICE_ATTR: {
1360 		r = -EFAULT;
1361 		if (copy_from_user(&attr, argp, sizeof(attr)))
1362 			break;
1363 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1364 		break;
1365 	}
1366 	case KVM_GET_DEVICE_ATTR: {
1367 		r = -EFAULT;
1368 		if (copy_from_user(&attr, argp, sizeof(attr)))
1369 			break;
1370 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1371 		break;
1372 	}
1373 	case KVM_HAS_DEVICE_ATTR: {
1374 		r = -EFAULT;
1375 		if (copy_from_user(&attr, argp, sizeof(attr)))
1376 			break;
1377 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1378 		break;
1379 	}
1380 	case KVM_GET_VCPU_EVENTS: {
1381 		struct kvm_vcpu_events events;
1382 
1383 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1384 			return -EINVAL;
1385 
1386 		if (copy_to_user(argp, &events, sizeof(events)))
1387 			return -EFAULT;
1388 
1389 		return 0;
1390 	}
1391 	case KVM_SET_VCPU_EVENTS: {
1392 		struct kvm_vcpu_events events;
1393 
1394 		if (copy_from_user(&events, argp, sizeof(events)))
1395 			return -EFAULT;
1396 
1397 		return kvm_arm_vcpu_set_events(vcpu, &events);
1398 	}
1399 	case KVM_ARM_VCPU_FINALIZE: {
1400 		int what;
1401 
1402 		if (!kvm_vcpu_initialized(vcpu))
1403 			return -ENOEXEC;
1404 
1405 		if (get_user(what, (const int __user *)argp))
1406 			return -EFAULT;
1407 
1408 		return kvm_arm_vcpu_finalize(vcpu, what);
1409 	}
1410 	default:
1411 		r = -EINVAL;
1412 	}
1413 
1414 	return r;
1415 }
1416 
1417 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1418 {
1419 
1420 }
1421 
1422 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1423 					const struct kvm_memory_slot *memslot)
1424 {
1425 	kvm_flush_remote_tlbs(kvm);
1426 }
1427 
1428 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1429 					struct kvm_arm_device_addr *dev_addr)
1430 {
1431 	switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1432 	case KVM_ARM_DEVICE_VGIC_V2:
1433 		if (!vgic_present)
1434 			return -ENXIO;
1435 		return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1436 	default:
1437 		return -ENODEV;
1438 	}
1439 }
1440 
1441 long kvm_arch_vm_ioctl(struct file *filp,
1442 		       unsigned int ioctl, unsigned long arg)
1443 {
1444 	struct kvm *kvm = filp->private_data;
1445 	void __user *argp = (void __user *)arg;
1446 
1447 	switch (ioctl) {
1448 	case KVM_CREATE_IRQCHIP: {
1449 		int ret;
1450 		if (!vgic_present)
1451 			return -ENXIO;
1452 		mutex_lock(&kvm->lock);
1453 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1454 		mutex_unlock(&kvm->lock);
1455 		return ret;
1456 	}
1457 	case KVM_ARM_SET_DEVICE_ADDR: {
1458 		struct kvm_arm_device_addr dev_addr;
1459 
1460 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1461 			return -EFAULT;
1462 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1463 	}
1464 	case KVM_ARM_PREFERRED_TARGET: {
1465 		struct kvm_vcpu_init init;
1466 
1467 		kvm_vcpu_preferred_target(&init);
1468 
1469 		if (copy_to_user(argp, &init, sizeof(init)))
1470 			return -EFAULT;
1471 
1472 		return 0;
1473 	}
1474 	case KVM_ARM_MTE_COPY_TAGS: {
1475 		struct kvm_arm_copy_mte_tags copy_tags;
1476 
1477 		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1478 			return -EFAULT;
1479 		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1480 	}
1481 	default:
1482 		return -EINVAL;
1483 	}
1484 }
1485 
1486 static unsigned long nvhe_percpu_size(void)
1487 {
1488 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1489 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1490 }
1491 
1492 static unsigned long nvhe_percpu_order(void)
1493 {
1494 	unsigned long size = nvhe_percpu_size();
1495 
1496 	return size ? get_order(size) : 0;
1497 }
1498 
1499 /* A lookup table holding the hypervisor VA for each vector slot */
1500 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1501 
1502 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1503 {
1504 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1505 }
1506 
1507 static int kvm_init_vector_slots(void)
1508 {
1509 	int err;
1510 	void *base;
1511 
1512 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1513 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1514 
1515 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1516 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1517 
1518 	if (kvm_system_needs_idmapped_vectors() &&
1519 	    !is_protected_kvm_enabled()) {
1520 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1521 					       __BP_HARDEN_HYP_VECS_SZ, &base);
1522 		if (err)
1523 			return err;
1524 	}
1525 
1526 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1527 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1528 	return 0;
1529 }
1530 
1531 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
1532 {
1533 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1534 	unsigned long tcr;
1535 
1536 	/*
1537 	 * Calculate the raw per-cpu offset without a translation from the
1538 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1539 	 * so that we can use adr_l to access per-cpu variables in EL2.
1540 	 * Also drop the KASAN tag which gets in the way...
1541 	 */
1542 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1543 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1544 
1545 	params->mair_el2 = read_sysreg(mair_el1);
1546 
1547 	tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1548 	tcr &= ~TCR_T0SZ_MASK;
1549 	tcr |= TCR_T0SZ(hyp_va_bits);
1550 	params->tcr_el2 = tcr;
1551 
1552 	params->pgd_pa = kvm_mmu_get_httbr();
1553 	if (is_protected_kvm_enabled())
1554 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1555 	else
1556 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1557 	params->vttbr = params->vtcr = 0;
1558 
1559 	/*
1560 	 * Flush the init params from the data cache because the struct will
1561 	 * be read while the MMU is off.
1562 	 */
1563 	kvm_flush_dcache_to_poc(params, sizeof(*params));
1564 }
1565 
1566 static void hyp_install_host_vector(void)
1567 {
1568 	struct kvm_nvhe_init_params *params;
1569 	struct arm_smccc_res res;
1570 
1571 	/* Switch from the HYP stub to our own HYP init vector */
1572 	__hyp_set_vectors(kvm_get_idmap_vector());
1573 
1574 	/*
1575 	 * Call initialization code, and switch to the full blown HYP code.
1576 	 * If the cpucaps haven't been finalized yet, something has gone very
1577 	 * wrong, and hyp will crash and burn when it uses any
1578 	 * cpus_have_const_cap() wrapper.
1579 	 */
1580 	BUG_ON(!system_capabilities_finalized());
1581 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1582 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1583 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1584 }
1585 
1586 static void cpu_init_hyp_mode(void)
1587 {
1588 	hyp_install_host_vector();
1589 
1590 	/*
1591 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1592 	 * at EL2.
1593 	 */
1594 	if (this_cpu_has_cap(ARM64_SSBS) &&
1595 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1596 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1597 	}
1598 }
1599 
1600 static void cpu_hyp_reset(void)
1601 {
1602 	if (!is_kernel_in_hyp_mode())
1603 		__hyp_reset_vectors();
1604 }
1605 
1606 /*
1607  * EL2 vectors can be mapped and rerouted in a number of ways,
1608  * depending on the kernel configuration and CPU present:
1609  *
1610  * - If the CPU is affected by Spectre-v2, the hardening sequence is
1611  *   placed in one of the vector slots, which is executed before jumping
1612  *   to the real vectors.
1613  *
1614  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1615  *   containing the hardening sequence is mapped next to the idmap page,
1616  *   and executed before jumping to the real vectors.
1617  *
1618  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1619  *   empty slot is selected, mapped next to the idmap page, and
1620  *   executed before jumping to the real vectors.
1621  *
1622  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1623  * VHE, as we don't have hypervisor-specific mappings. If the system
1624  * is VHE and yet selects this capability, it will be ignored.
1625  */
1626 static void cpu_set_hyp_vector(void)
1627 {
1628 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1629 	void *vector = hyp_spectre_vector_selector[data->slot];
1630 
1631 	if (!is_protected_kvm_enabled())
1632 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1633 	else
1634 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1635 }
1636 
1637 static void cpu_hyp_init_context(void)
1638 {
1639 	kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1640 
1641 	if (!is_kernel_in_hyp_mode())
1642 		cpu_init_hyp_mode();
1643 }
1644 
1645 static void cpu_hyp_init_features(void)
1646 {
1647 	cpu_set_hyp_vector();
1648 	kvm_arm_init_debug();
1649 
1650 	if (is_kernel_in_hyp_mode())
1651 		kvm_timer_init_vhe();
1652 
1653 	if (vgic_present)
1654 		kvm_vgic_init_cpu_hardware();
1655 }
1656 
1657 static void cpu_hyp_reinit(void)
1658 {
1659 	cpu_hyp_reset();
1660 	cpu_hyp_init_context();
1661 	cpu_hyp_init_features();
1662 }
1663 
1664 static void _kvm_arch_hardware_enable(void *discard)
1665 {
1666 	if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1667 		cpu_hyp_reinit();
1668 		__this_cpu_write(kvm_arm_hardware_enabled, 1);
1669 	}
1670 }
1671 
1672 int kvm_arch_hardware_enable(void)
1673 {
1674 	int was_enabled = __this_cpu_read(kvm_arm_hardware_enabled);
1675 
1676 	_kvm_arch_hardware_enable(NULL);
1677 
1678 	if (!was_enabled) {
1679 		kvm_vgic_cpu_up();
1680 		kvm_timer_cpu_up();
1681 	}
1682 
1683 	return 0;
1684 }
1685 
1686 static void _kvm_arch_hardware_disable(void *discard)
1687 {
1688 	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1689 		cpu_hyp_reset();
1690 		__this_cpu_write(kvm_arm_hardware_enabled, 0);
1691 	}
1692 }
1693 
1694 void kvm_arch_hardware_disable(void)
1695 {
1696 	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1697 		kvm_timer_cpu_down();
1698 		kvm_vgic_cpu_down();
1699 	}
1700 
1701 	if (!is_protected_kvm_enabled())
1702 		_kvm_arch_hardware_disable(NULL);
1703 }
1704 
1705 #ifdef CONFIG_CPU_PM
1706 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1707 				    unsigned long cmd,
1708 				    void *v)
1709 {
1710 	/*
1711 	 * kvm_arm_hardware_enabled is left with its old value over
1712 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1713 	 * re-enable hyp.
1714 	 */
1715 	switch (cmd) {
1716 	case CPU_PM_ENTER:
1717 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1718 			/*
1719 			 * don't update kvm_arm_hardware_enabled here
1720 			 * so that the hardware will be re-enabled
1721 			 * when we resume. See below.
1722 			 */
1723 			cpu_hyp_reset();
1724 
1725 		return NOTIFY_OK;
1726 	case CPU_PM_ENTER_FAILED:
1727 	case CPU_PM_EXIT:
1728 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1729 			/* The hardware was enabled before suspend. */
1730 			cpu_hyp_reinit();
1731 
1732 		return NOTIFY_OK;
1733 
1734 	default:
1735 		return NOTIFY_DONE;
1736 	}
1737 }
1738 
1739 static struct notifier_block hyp_init_cpu_pm_nb = {
1740 	.notifier_call = hyp_init_cpu_pm_notifier,
1741 };
1742 
1743 static void __init hyp_cpu_pm_init(void)
1744 {
1745 	if (!is_protected_kvm_enabled())
1746 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1747 }
1748 static void __init hyp_cpu_pm_exit(void)
1749 {
1750 	if (!is_protected_kvm_enabled())
1751 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1752 }
1753 #else
1754 static inline void __init hyp_cpu_pm_init(void)
1755 {
1756 }
1757 static inline void __init hyp_cpu_pm_exit(void)
1758 {
1759 }
1760 #endif
1761 
1762 static void __init init_cpu_logical_map(void)
1763 {
1764 	unsigned int cpu;
1765 
1766 	/*
1767 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1768 	 * Only copy the set of online CPUs whose features have been checked
1769 	 * against the finalized system capabilities. The hypervisor will not
1770 	 * allow any other CPUs from the `possible` set to boot.
1771 	 */
1772 	for_each_online_cpu(cpu)
1773 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1774 }
1775 
1776 #define init_psci_0_1_impl_state(config, what)	\
1777 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
1778 
1779 static bool __init init_psci_relay(void)
1780 {
1781 	/*
1782 	 * If PSCI has not been initialized, protected KVM cannot install
1783 	 * itself on newly booted CPUs.
1784 	 */
1785 	if (!psci_ops.get_version) {
1786 		kvm_err("Cannot initialize protected mode without PSCI\n");
1787 		return false;
1788 	}
1789 
1790 	kvm_host_psci_config.version = psci_ops.get_version();
1791 
1792 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1793 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1794 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1795 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1796 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1797 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1798 	}
1799 	return true;
1800 }
1801 
1802 static int __init init_subsystems(void)
1803 {
1804 	int err = 0;
1805 
1806 	/*
1807 	 * Enable hardware so that subsystem initialisation can access EL2.
1808 	 */
1809 	on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1810 
1811 	/*
1812 	 * Register CPU lower-power notifier
1813 	 */
1814 	hyp_cpu_pm_init();
1815 
1816 	/*
1817 	 * Init HYP view of VGIC
1818 	 */
1819 	err = kvm_vgic_hyp_init();
1820 	switch (err) {
1821 	case 0:
1822 		vgic_present = true;
1823 		break;
1824 	case -ENODEV:
1825 	case -ENXIO:
1826 		vgic_present = false;
1827 		err = 0;
1828 		break;
1829 	default:
1830 		goto out;
1831 	}
1832 
1833 	/*
1834 	 * Init HYP architected timer support
1835 	 */
1836 	err = kvm_timer_hyp_init(vgic_present);
1837 	if (err)
1838 		goto out;
1839 
1840 	kvm_register_perf_callbacks(NULL);
1841 
1842 out:
1843 	if (err)
1844 		hyp_cpu_pm_exit();
1845 
1846 	if (err || !is_protected_kvm_enabled())
1847 		on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1848 
1849 	return err;
1850 }
1851 
1852 static void __init teardown_subsystems(void)
1853 {
1854 	kvm_unregister_perf_callbacks();
1855 	hyp_cpu_pm_exit();
1856 }
1857 
1858 static void __init teardown_hyp_mode(void)
1859 {
1860 	int cpu;
1861 
1862 	free_hyp_pgds();
1863 	for_each_possible_cpu(cpu) {
1864 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1865 		free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
1866 	}
1867 }
1868 
1869 static int __init do_pkvm_init(u32 hyp_va_bits)
1870 {
1871 	void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
1872 	int ret;
1873 
1874 	preempt_disable();
1875 	cpu_hyp_init_context();
1876 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1877 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
1878 				hyp_va_bits);
1879 	cpu_hyp_init_features();
1880 
1881 	/*
1882 	 * The stub hypercalls are now disabled, so set our local flag to
1883 	 * prevent a later re-init attempt in kvm_arch_hardware_enable().
1884 	 */
1885 	__this_cpu_write(kvm_arm_hardware_enabled, 1);
1886 	preempt_enable();
1887 
1888 	return ret;
1889 }
1890 
1891 static u64 get_hyp_id_aa64pfr0_el1(void)
1892 {
1893 	/*
1894 	 * Track whether the system isn't affected by spectre/meltdown in the
1895 	 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
1896 	 * Although this is per-CPU, we make it global for simplicity, e.g., not
1897 	 * to have to worry about vcpu migration.
1898 	 *
1899 	 * Unlike for non-protected VMs, userspace cannot override this for
1900 	 * protected VMs.
1901 	 */
1902 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1903 
1904 	val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
1905 		 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
1906 
1907 	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
1908 			  arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
1909 	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
1910 			  arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
1911 
1912 	return val;
1913 }
1914 
1915 static void kvm_hyp_init_symbols(void)
1916 {
1917 	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
1918 	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1919 	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
1920 	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
1921 	kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
1922 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1923 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1924 	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
1925 	kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
1926 	kvm_nvhe_sym(__icache_flags) = __icache_flags;
1927 	kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
1928 }
1929 
1930 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
1931 {
1932 	void *addr = phys_to_virt(hyp_mem_base);
1933 	int ret;
1934 
1935 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1936 	if (ret)
1937 		return ret;
1938 
1939 	ret = do_pkvm_init(hyp_va_bits);
1940 	if (ret)
1941 		return ret;
1942 
1943 	free_hyp_pgds();
1944 
1945 	return 0;
1946 }
1947 
1948 /* Inits Hyp-mode on all online CPUs */
1949 static int __init init_hyp_mode(void)
1950 {
1951 	u32 hyp_va_bits;
1952 	int cpu;
1953 	int err = -ENOMEM;
1954 
1955 	/*
1956 	 * The protected Hyp-mode cannot be initialized if the memory pool
1957 	 * allocation has failed.
1958 	 */
1959 	if (is_protected_kvm_enabled() && !hyp_mem_base)
1960 		goto out_err;
1961 
1962 	/*
1963 	 * Allocate Hyp PGD and setup Hyp identity mapping
1964 	 */
1965 	err = kvm_mmu_init(&hyp_va_bits);
1966 	if (err)
1967 		goto out_err;
1968 
1969 	/*
1970 	 * Allocate stack pages for Hypervisor-mode
1971 	 */
1972 	for_each_possible_cpu(cpu) {
1973 		unsigned long stack_page;
1974 
1975 		stack_page = __get_free_page(GFP_KERNEL);
1976 		if (!stack_page) {
1977 			err = -ENOMEM;
1978 			goto out_err;
1979 		}
1980 
1981 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1982 	}
1983 
1984 	/*
1985 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1986 	 */
1987 	for_each_possible_cpu(cpu) {
1988 		struct page *page;
1989 		void *page_addr;
1990 
1991 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1992 		if (!page) {
1993 			err = -ENOMEM;
1994 			goto out_err;
1995 		}
1996 
1997 		page_addr = page_address(page);
1998 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1999 		kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2000 	}
2001 
2002 	/*
2003 	 * Map the Hyp-code called directly from the host
2004 	 */
2005 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2006 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2007 	if (err) {
2008 		kvm_err("Cannot map world-switch code\n");
2009 		goto out_err;
2010 	}
2011 
2012 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2013 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2014 	if (err) {
2015 		kvm_err("Cannot map .hyp.rodata section\n");
2016 		goto out_err;
2017 	}
2018 
2019 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2020 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2021 	if (err) {
2022 		kvm_err("Cannot map rodata section\n");
2023 		goto out_err;
2024 	}
2025 
2026 	/*
2027 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2028 	 * section thanks to an assertion in the linker script. Map it RW and
2029 	 * the rest of .bss RO.
2030 	 */
2031 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2032 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2033 	if (err) {
2034 		kvm_err("Cannot map hyp bss section: %d\n", err);
2035 		goto out_err;
2036 	}
2037 
2038 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2039 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2040 	if (err) {
2041 		kvm_err("Cannot map bss section\n");
2042 		goto out_err;
2043 	}
2044 
2045 	/*
2046 	 * Map the Hyp stack pages
2047 	 */
2048 	for_each_possible_cpu(cpu) {
2049 		struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2050 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2051 		unsigned long hyp_addr;
2052 
2053 		/*
2054 		 * Allocate a contiguous HYP private VA range for the stack
2055 		 * and guard page. The allocation is also aligned based on
2056 		 * the order of its size.
2057 		 */
2058 		err = hyp_alloc_private_va_range(PAGE_SIZE * 2, &hyp_addr);
2059 		if (err) {
2060 			kvm_err("Cannot allocate hyp stack guard page\n");
2061 			goto out_err;
2062 		}
2063 
2064 		/*
2065 		 * Since the stack grows downwards, map the stack to the page
2066 		 * at the higher address and leave the lower guard page
2067 		 * unbacked.
2068 		 *
2069 		 * Any valid stack address now has the PAGE_SHIFT bit as 1
2070 		 * and addresses corresponding to the guard page have the
2071 		 * PAGE_SHIFT bit as 0 - this is used for overflow detection.
2072 		 */
2073 		err = __create_hyp_mappings(hyp_addr + PAGE_SIZE, PAGE_SIZE,
2074 					    __pa(stack_page), PAGE_HYP);
2075 		if (err) {
2076 			kvm_err("Cannot map hyp stack\n");
2077 			goto out_err;
2078 		}
2079 
2080 		/*
2081 		 * Save the stack PA in nvhe_init_params. This will be needed
2082 		 * to recreate the stack mapping in protected nVHE mode.
2083 		 * __hyp_pa() won't do the right thing there, since the stack
2084 		 * has been mapped in the flexible private VA space.
2085 		 */
2086 		params->stack_pa = __pa(stack_page);
2087 
2088 		params->stack_hyp_va = hyp_addr + (2 * PAGE_SIZE);
2089 	}
2090 
2091 	for_each_possible_cpu(cpu) {
2092 		char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2093 		char *percpu_end = percpu_begin + nvhe_percpu_size();
2094 
2095 		/* Map Hyp percpu pages */
2096 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2097 		if (err) {
2098 			kvm_err("Cannot map hyp percpu region\n");
2099 			goto out_err;
2100 		}
2101 
2102 		/* Prepare the CPU initialization parameters */
2103 		cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2104 	}
2105 
2106 	kvm_hyp_init_symbols();
2107 
2108 	if (is_protected_kvm_enabled()) {
2109 		init_cpu_logical_map();
2110 
2111 		if (!init_psci_relay()) {
2112 			err = -ENODEV;
2113 			goto out_err;
2114 		}
2115 
2116 		err = kvm_hyp_init_protection(hyp_va_bits);
2117 		if (err) {
2118 			kvm_err("Failed to init hyp memory protection\n");
2119 			goto out_err;
2120 		}
2121 	}
2122 
2123 	return 0;
2124 
2125 out_err:
2126 	teardown_hyp_mode();
2127 	kvm_err("error initializing Hyp mode: %d\n", err);
2128 	return err;
2129 }
2130 
2131 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2132 {
2133 	struct kvm_vcpu *vcpu;
2134 	unsigned long i;
2135 
2136 	mpidr &= MPIDR_HWID_BITMASK;
2137 	kvm_for_each_vcpu(i, vcpu, kvm) {
2138 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2139 			return vcpu;
2140 	}
2141 	return NULL;
2142 }
2143 
2144 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2145 {
2146 	return irqchip_in_kernel(kvm);
2147 }
2148 
2149 bool kvm_arch_has_irq_bypass(void)
2150 {
2151 	return true;
2152 }
2153 
2154 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2155 				      struct irq_bypass_producer *prod)
2156 {
2157 	struct kvm_kernel_irqfd *irqfd =
2158 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2159 
2160 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2161 					  &irqfd->irq_entry);
2162 }
2163 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2164 				      struct irq_bypass_producer *prod)
2165 {
2166 	struct kvm_kernel_irqfd *irqfd =
2167 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2168 
2169 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2170 				     &irqfd->irq_entry);
2171 }
2172 
2173 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2174 {
2175 	struct kvm_kernel_irqfd *irqfd =
2176 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2177 
2178 	kvm_arm_halt_guest(irqfd->kvm);
2179 }
2180 
2181 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2182 {
2183 	struct kvm_kernel_irqfd *irqfd =
2184 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2185 
2186 	kvm_arm_resume_guest(irqfd->kvm);
2187 }
2188 
2189 /* Initialize Hyp-mode and memory mappings on all CPUs */
2190 static __init int kvm_arm_init(void)
2191 {
2192 	int err;
2193 	bool in_hyp_mode;
2194 
2195 	if (!is_hyp_mode_available()) {
2196 		kvm_info("HYP mode not available\n");
2197 		return -ENODEV;
2198 	}
2199 
2200 	if (kvm_get_mode() == KVM_MODE_NONE) {
2201 		kvm_info("KVM disabled from command line\n");
2202 		return -ENODEV;
2203 	}
2204 
2205 	err = kvm_sys_reg_table_init();
2206 	if (err) {
2207 		kvm_info("Error initializing system register tables");
2208 		return err;
2209 	}
2210 
2211 	in_hyp_mode = is_kernel_in_hyp_mode();
2212 
2213 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2214 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2215 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2216 			 "Only trusted guests should be used on this system.\n");
2217 
2218 	err = kvm_set_ipa_limit();
2219 	if (err)
2220 		return err;
2221 
2222 	err = kvm_arm_init_sve();
2223 	if (err)
2224 		return err;
2225 
2226 	err = kvm_arm_vmid_alloc_init();
2227 	if (err) {
2228 		kvm_err("Failed to initialize VMID allocator.\n");
2229 		return err;
2230 	}
2231 
2232 	if (!in_hyp_mode) {
2233 		err = init_hyp_mode();
2234 		if (err)
2235 			goto out_err;
2236 	}
2237 
2238 	err = kvm_init_vector_slots();
2239 	if (err) {
2240 		kvm_err("Cannot initialise vector slots\n");
2241 		goto out_hyp;
2242 	}
2243 
2244 	err = init_subsystems();
2245 	if (err)
2246 		goto out_hyp;
2247 
2248 	if (is_protected_kvm_enabled()) {
2249 		kvm_info("Protected nVHE mode initialized successfully\n");
2250 	} else if (in_hyp_mode) {
2251 		kvm_info("VHE mode initialized successfully\n");
2252 	} else {
2253 		kvm_info("Hyp mode initialized successfully\n");
2254 	}
2255 
2256 	/*
2257 	 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2258 	 * hypervisor protection is finalized.
2259 	 */
2260 	err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2261 	if (err)
2262 		goto out_subs;
2263 
2264 	return 0;
2265 
2266 out_subs:
2267 	teardown_subsystems();
2268 out_hyp:
2269 	if (!in_hyp_mode)
2270 		teardown_hyp_mode();
2271 out_err:
2272 	kvm_arm_vmid_alloc_free();
2273 	return err;
2274 }
2275 
2276 static int __init early_kvm_mode_cfg(char *arg)
2277 {
2278 	if (!arg)
2279 		return -EINVAL;
2280 
2281 	if (strcmp(arg, "none") == 0) {
2282 		kvm_mode = KVM_MODE_NONE;
2283 		return 0;
2284 	}
2285 
2286 	if (!is_hyp_mode_available()) {
2287 		pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2288 		return 0;
2289 	}
2290 
2291 	if (strcmp(arg, "protected") == 0) {
2292 		if (!is_kernel_in_hyp_mode())
2293 			kvm_mode = KVM_MODE_PROTECTED;
2294 		else
2295 			pr_warn_once("Protected KVM not available with VHE\n");
2296 
2297 		return 0;
2298 	}
2299 
2300 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2301 		kvm_mode = KVM_MODE_DEFAULT;
2302 		return 0;
2303 	}
2304 
2305 	if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2306 		kvm_mode = KVM_MODE_NV;
2307 		return 0;
2308 	}
2309 
2310 	return -EINVAL;
2311 }
2312 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2313 
2314 enum kvm_mode kvm_get_mode(void)
2315 {
2316 	return kvm_mode;
2317 }
2318 
2319 module_init(kvm_arm_init);
2320