xref: /openbmc/linux/arch/arm64/kvm/arm.c (revision fbb6b31a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
4  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
5  */
6 
7 #include <linux/bug.h>
8 #include <linux/cpu_pm.h>
9 #include <linux/entry-kvm.h>
10 #include <linux/errno.h>
11 #include <linux/err.h>
12 #include <linux/kvm_host.h>
13 #include <linux/list.h>
14 #include <linux/module.h>
15 #include <linux/vmalloc.h>
16 #include <linux/fs.h>
17 #include <linux/mman.h>
18 #include <linux/sched.h>
19 #include <linux/kmemleak.h>
20 #include <linux/kvm.h>
21 #include <linux/kvm_irqfd.h>
22 #include <linux/irqbypass.h>
23 #include <linux/sched/stat.h>
24 #include <linux/psci.h>
25 #include <trace/events/kvm.h>
26 
27 #define CREATE_TRACE_POINTS
28 #include "trace_arm.h"
29 
30 #include <linux/uaccess.h>
31 #include <asm/ptrace.h>
32 #include <asm/mman.h>
33 #include <asm/tlbflush.h>
34 #include <asm/cacheflush.h>
35 #include <asm/cpufeature.h>
36 #include <asm/virt.h>
37 #include <asm/kvm_arm.h>
38 #include <asm/kvm_asm.h>
39 #include <asm/kvm_mmu.h>
40 #include <asm/kvm_emulate.h>
41 #include <asm/sections.h>
42 
43 #include <kvm/arm_hypercalls.h>
44 #include <kvm/arm_pmu.h>
45 #include <kvm/arm_psci.h>
46 
47 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
48 DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized);
49 
50 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
51 
52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
53 unsigned long kvm_arm_hyp_percpu_base[NR_CPUS];
54 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
55 
56 static bool vgic_present;
57 
58 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
59 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
60 
61 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
62 {
63 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
64 }
65 
66 int kvm_arch_hardware_setup(void *opaque)
67 {
68 	return 0;
69 }
70 
71 int kvm_arch_check_processor_compat(void *opaque)
72 {
73 	return 0;
74 }
75 
76 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
77 			    struct kvm_enable_cap *cap)
78 {
79 	int r;
80 
81 	if (cap->flags)
82 		return -EINVAL;
83 
84 	switch (cap->cap) {
85 	case KVM_CAP_ARM_NISV_TO_USER:
86 		r = 0;
87 		set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
88 			&kvm->arch.flags);
89 		break;
90 	case KVM_CAP_ARM_MTE:
91 		mutex_lock(&kvm->lock);
92 		if (!system_supports_mte() || kvm->created_vcpus) {
93 			r = -EINVAL;
94 		} else {
95 			r = 0;
96 			set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
97 		}
98 		mutex_unlock(&kvm->lock);
99 		break;
100 	default:
101 		r = -EINVAL;
102 		break;
103 	}
104 
105 	return r;
106 }
107 
108 static int kvm_arm_default_max_vcpus(void)
109 {
110 	return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
111 }
112 
113 static void set_default_spectre(struct kvm *kvm)
114 {
115 	/*
116 	 * The default is to expose CSV2 == 1 if the HW isn't affected.
117 	 * Although this is a per-CPU feature, we make it global because
118 	 * asymmetric systems are just a nuisance.
119 	 *
120 	 * Userspace can override this as long as it doesn't promise
121 	 * the impossible.
122 	 */
123 	if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
124 		kvm->arch.pfr0_csv2 = 1;
125 	if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
126 		kvm->arch.pfr0_csv3 = 1;
127 }
128 
129 /**
130  * kvm_arch_init_vm - initializes a VM data structure
131  * @kvm:	pointer to the KVM struct
132  */
133 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
134 {
135 	int ret;
136 
137 	ret = kvm_arm_setup_stage2(kvm, type);
138 	if (ret)
139 		return ret;
140 
141 	ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
142 	if (ret)
143 		return ret;
144 
145 	ret = kvm_share_hyp(kvm, kvm + 1);
146 	if (ret)
147 		goto out_free_stage2_pgd;
148 
149 	if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL))
150 		goto out_free_stage2_pgd;
151 	cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
152 
153 	kvm_vgic_early_init(kvm);
154 
155 	/* The maximum number of VCPUs is limited by the host's GIC model */
156 	kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
157 
158 	set_default_spectre(kvm);
159 
160 	return ret;
161 out_free_stage2_pgd:
162 	kvm_free_stage2_pgd(&kvm->arch.mmu);
163 	return ret;
164 }
165 
166 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
167 {
168 	return VM_FAULT_SIGBUS;
169 }
170 
171 
172 /**
173  * kvm_arch_destroy_vm - destroy the VM data structure
174  * @kvm:	pointer to the KVM struct
175  */
176 void kvm_arch_destroy_vm(struct kvm *kvm)
177 {
178 	bitmap_free(kvm->arch.pmu_filter);
179 	free_cpumask_var(kvm->arch.supported_cpus);
180 
181 	kvm_vgic_destroy(kvm);
182 
183 	kvm_destroy_vcpus(kvm);
184 
185 	kvm_unshare_hyp(kvm, kvm + 1);
186 }
187 
188 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
189 {
190 	int r;
191 	switch (ext) {
192 	case KVM_CAP_IRQCHIP:
193 		r = vgic_present;
194 		break;
195 	case KVM_CAP_IOEVENTFD:
196 	case KVM_CAP_DEVICE_CTRL:
197 	case KVM_CAP_USER_MEMORY:
198 	case KVM_CAP_SYNC_MMU:
199 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
200 	case KVM_CAP_ONE_REG:
201 	case KVM_CAP_ARM_PSCI:
202 	case KVM_CAP_ARM_PSCI_0_2:
203 	case KVM_CAP_READONLY_MEM:
204 	case KVM_CAP_MP_STATE:
205 	case KVM_CAP_IMMEDIATE_EXIT:
206 	case KVM_CAP_VCPU_EVENTS:
207 	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
208 	case KVM_CAP_ARM_NISV_TO_USER:
209 	case KVM_CAP_ARM_INJECT_EXT_DABT:
210 	case KVM_CAP_SET_GUEST_DEBUG:
211 	case KVM_CAP_VCPU_ATTRIBUTES:
212 	case KVM_CAP_PTP_KVM:
213 		r = 1;
214 		break;
215 	case KVM_CAP_SET_GUEST_DEBUG2:
216 		return KVM_GUESTDBG_VALID_MASK;
217 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
218 		r = 1;
219 		break;
220 	case KVM_CAP_NR_VCPUS:
221 		/*
222 		 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
223 		 * architectures, as it does not always bound it to
224 		 * KVM_CAP_MAX_VCPUS. It should not matter much because
225 		 * this is just an advisory value.
226 		 */
227 		r = min_t(unsigned int, num_online_cpus(),
228 			  kvm_arm_default_max_vcpus());
229 		break;
230 	case KVM_CAP_MAX_VCPUS:
231 	case KVM_CAP_MAX_VCPU_ID:
232 		if (kvm)
233 			r = kvm->arch.max_vcpus;
234 		else
235 			r = kvm_arm_default_max_vcpus();
236 		break;
237 	case KVM_CAP_MSI_DEVID:
238 		if (!kvm)
239 			r = -EINVAL;
240 		else
241 			r = kvm->arch.vgic.msis_require_devid;
242 		break;
243 	case KVM_CAP_ARM_USER_IRQ:
244 		/*
245 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
246 		 * (bump this number if adding more devices)
247 		 */
248 		r = 1;
249 		break;
250 	case KVM_CAP_ARM_MTE:
251 		r = system_supports_mte();
252 		break;
253 	case KVM_CAP_STEAL_TIME:
254 		r = kvm_arm_pvtime_supported();
255 		break;
256 	case KVM_CAP_ARM_EL1_32BIT:
257 		r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
258 		break;
259 	case KVM_CAP_GUEST_DEBUG_HW_BPS:
260 		r = get_num_brps();
261 		break;
262 	case KVM_CAP_GUEST_DEBUG_HW_WPS:
263 		r = get_num_wrps();
264 		break;
265 	case KVM_CAP_ARM_PMU_V3:
266 		r = kvm_arm_support_pmu_v3();
267 		break;
268 	case KVM_CAP_ARM_INJECT_SERROR_ESR:
269 		r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
270 		break;
271 	case KVM_CAP_ARM_VM_IPA_SIZE:
272 		r = get_kvm_ipa_limit();
273 		break;
274 	case KVM_CAP_ARM_SVE:
275 		r = system_supports_sve();
276 		break;
277 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
278 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
279 		r = system_has_full_ptr_auth();
280 		break;
281 	default:
282 		r = 0;
283 	}
284 
285 	return r;
286 }
287 
288 long kvm_arch_dev_ioctl(struct file *filp,
289 			unsigned int ioctl, unsigned long arg)
290 {
291 	return -EINVAL;
292 }
293 
294 struct kvm *kvm_arch_alloc_vm(void)
295 {
296 	size_t sz = sizeof(struct kvm);
297 
298 	if (!has_vhe())
299 		return kzalloc(sz, GFP_KERNEL_ACCOUNT);
300 
301 	return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
302 }
303 
304 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
305 {
306 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
307 		return -EBUSY;
308 
309 	if (id >= kvm->arch.max_vcpus)
310 		return -EINVAL;
311 
312 	return 0;
313 }
314 
315 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
316 {
317 	int err;
318 
319 	/* Force users to call KVM_ARM_VCPU_INIT */
320 	vcpu->arch.target = -1;
321 	bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
322 
323 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
324 
325 	/* Set up the timer */
326 	kvm_timer_vcpu_init(vcpu);
327 
328 	kvm_pmu_vcpu_init(vcpu);
329 
330 	kvm_arm_reset_debug_ptr(vcpu);
331 
332 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
333 
334 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
335 
336 	err = kvm_vgic_vcpu_init(vcpu);
337 	if (err)
338 		return err;
339 
340 	return kvm_share_hyp(vcpu, vcpu + 1);
341 }
342 
343 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
344 {
345 }
346 
347 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
348 {
349 	if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
350 		static_branch_dec(&userspace_irqchip_in_use);
351 
352 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
353 	kvm_timer_vcpu_terminate(vcpu);
354 	kvm_pmu_vcpu_destroy(vcpu);
355 
356 	kvm_arm_vcpu_destroy(vcpu);
357 }
358 
359 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
360 {
361 	return kvm_timer_is_pending(vcpu);
362 }
363 
364 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
365 {
366 
367 }
368 
369 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
370 {
371 
372 }
373 
374 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
375 {
376 	struct kvm_s2_mmu *mmu;
377 	int *last_ran;
378 
379 	mmu = vcpu->arch.hw_mmu;
380 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
381 
382 	/*
383 	 * We guarantee that both TLBs and I-cache are private to each
384 	 * vcpu. If detecting that a vcpu from the same VM has
385 	 * previously run on the same physical CPU, call into the
386 	 * hypervisor code to nuke the relevant contexts.
387 	 *
388 	 * We might get preempted before the vCPU actually runs, but
389 	 * over-invalidation doesn't affect correctness.
390 	 */
391 	if (*last_ran != vcpu->vcpu_id) {
392 		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
393 		*last_ran = vcpu->vcpu_id;
394 	}
395 
396 	vcpu->cpu = cpu;
397 
398 	kvm_vgic_load(vcpu);
399 	kvm_timer_vcpu_load(vcpu);
400 	if (has_vhe())
401 		kvm_vcpu_load_sysregs_vhe(vcpu);
402 	kvm_arch_vcpu_load_fp(vcpu);
403 	kvm_vcpu_pmu_restore_guest(vcpu);
404 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
405 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
406 
407 	if (single_task_running())
408 		vcpu_clear_wfx_traps(vcpu);
409 	else
410 		vcpu_set_wfx_traps(vcpu);
411 
412 	if (vcpu_has_ptrauth(vcpu))
413 		vcpu_ptrauth_disable(vcpu);
414 	kvm_arch_vcpu_load_debug_state_flags(vcpu);
415 
416 	if (!cpumask_test_cpu(smp_processor_id(), vcpu->kvm->arch.supported_cpus))
417 		vcpu_set_on_unsupported_cpu(vcpu);
418 }
419 
420 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
421 {
422 	kvm_arch_vcpu_put_debug_state_flags(vcpu);
423 	kvm_arch_vcpu_put_fp(vcpu);
424 	if (has_vhe())
425 		kvm_vcpu_put_sysregs_vhe(vcpu);
426 	kvm_timer_vcpu_put(vcpu);
427 	kvm_vgic_put(vcpu);
428 	kvm_vcpu_pmu_restore_host(vcpu);
429 	kvm_arm_vmid_clear_active();
430 
431 	vcpu_clear_on_unsupported_cpu(vcpu);
432 	vcpu->cpu = -1;
433 }
434 
435 static void vcpu_power_off(struct kvm_vcpu *vcpu)
436 {
437 	vcpu->arch.power_off = true;
438 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
439 	kvm_vcpu_kick(vcpu);
440 }
441 
442 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
443 				    struct kvm_mp_state *mp_state)
444 {
445 	if (vcpu->arch.power_off)
446 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
447 	else
448 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
449 
450 	return 0;
451 }
452 
453 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
454 				    struct kvm_mp_state *mp_state)
455 {
456 	int ret = 0;
457 
458 	switch (mp_state->mp_state) {
459 	case KVM_MP_STATE_RUNNABLE:
460 		vcpu->arch.power_off = false;
461 		break;
462 	case KVM_MP_STATE_STOPPED:
463 		vcpu_power_off(vcpu);
464 		break;
465 	default:
466 		ret = -EINVAL;
467 	}
468 
469 	return ret;
470 }
471 
472 /**
473  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
474  * @v:		The VCPU pointer
475  *
476  * If the guest CPU is not waiting for interrupts or an interrupt line is
477  * asserted, the CPU is by definition runnable.
478  */
479 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
480 {
481 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
482 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
483 		&& !v->arch.power_off && !v->arch.pause);
484 }
485 
486 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
487 {
488 	return vcpu_mode_priv(vcpu);
489 }
490 
491 #ifdef CONFIG_GUEST_PERF_EVENTS
492 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
493 {
494 	return *vcpu_pc(vcpu);
495 }
496 #endif
497 
498 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
499 {
500 	return vcpu->arch.target >= 0;
501 }
502 
503 /*
504  * Handle both the initialisation that is being done when the vcpu is
505  * run for the first time, as well as the updates that must be
506  * performed each time we get a new thread dealing with this vcpu.
507  */
508 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
509 {
510 	struct kvm *kvm = vcpu->kvm;
511 	int ret;
512 
513 	if (!kvm_vcpu_initialized(vcpu))
514 		return -ENOEXEC;
515 
516 	if (!kvm_arm_vcpu_is_finalized(vcpu))
517 		return -EPERM;
518 
519 	ret = kvm_arch_vcpu_run_map_fp(vcpu);
520 	if (ret)
521 		return ret;
522 
523 	if (likely(vcpu_has_run_once(vcpu)))
524 		return 0;
525 
526 	kvm_arm_vcpu_init_debug(vcpu);
527 
528 	if (likely(irqchip_in_kernel(kvm))) {
529 		/*
530 		 * Map the VGIC hardware resources before running a vcpu the
531 		 * first time on this VM.
532 		 */
533 		ret = kvm_vgic_map_resources(kvm);
534 		if (ret)
535 			return ret;
536 	}
537 
538 	ret = kvm_timer_enable(vcpu);
539 	if (ret)
540 		return ret;
541 
542 	ret = kvm_arm_pmu_v3_enable(vcpu);
543 	if (ret)
544 		return ret;
545 
546 	if (!irqchip_in_kernel(kvm)) {
547 		/*
548 		 * Tell the rest of the code that there are userspace irqchip
549 		 * VMs in the wild.
550 		 */
551 		static_branch_inc(&userspace_irqchip_in_use);
552 	}
553 
554 	/*
555 	 * Initialize traps for protected VMs.
556 	 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
557 	 * the code is in place for first run initialization at EL2.
558 	 */
559 	if (kvm_vm_is_protected(kvm))
560 		kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
561 
562 	mutex_lock(&kvm->lock);
563 	set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
564 	mutex_unlock(&kvm->lock);
565 
566 	return ret;
567 }
568 
569 bool kvm_arch_intc_initialized(struct kvm *kvm)
570 {
571 	return vgic_initialized(kvm);
572 }
573 
574 void kvm_arm_halt_guest(struct kvm *kvm)
575 {
576 	unsigned long i;
577 	struct kvm_vcpu *vcpu;
578 
579 	kvm_for_each_vcpu(i, vcpu, kvm)
580 		vcpu->arch.pause = true;
581 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
582 }
583 
584 void kvm_arm_resume_guest(struct kvm *kvm)
585 {
586 	unsigned long i;
587 	struct kvm_vcpu *vcpu;
588 
589 	kvm_for_each_vcpu(i, vcpu, kvm) {
590 		vcpu->arch.pause = false;
591 		__kvm_vcpu_wake_up(vcpu);
592 	}
593 }
594 
595 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
596 {
597 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
598 
599 	rcuwait_wait_event(wait,
600 			   (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
601 			   TASK_INTERRUPTIBLE);
602 
603 	if (vcpu->arch.power_off || vcpu->arch.pause) {
604 		/* Awaken to handle a signal, request we sleep again later. */
605 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
606 	}
607 
608 	/*
609 	 * Make sure we will observe a potential reset request if we've
610 	 * observed a change to the power state. Pairs with the smp_wmb() in
611 	 * kvm_psci_vcpu_on().
612 	 */
613 	smp_rmb();
614 }
615 
616 /**
617  * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
618  * @vcpu:	The VCPU pointer
619  *
620  * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
621  * the vCPU is runnable.  The vCPU may or may not be scheduled out, depending
622  * on when a wake event arrives, e.g. there may already be a pending wake event.
623  */
624 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
625 {
626 	/*
627 	 * Sync back the state of the GIC CPU interface so that we have
628 	 * the latest PMR and group enables. This ensures that
629 	 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
630 	 * we have pending interrupts, e.g. when determining if the
631 	 * vCPU should block.
632 	 *
633 	 * For the same reason, we want to tell GICv4 that we need
634 	 * doorbells to be signalled, should an interrupt become pending.
635 	 */
636 	preempt_disable();
637 	kvm_vgic_vmcr_sync(vcpu);
638 	vgic_v4_put(vcpu, true);
639 	preempt_enable();
640 
641 	kvm_vcpu_halt(vcpu);
642 	kvm_clear_request(KVM_REQ_UNHALT, vcpu);
643 
644 	preempt_disable();
645 	vgic_v4_load(vcpu);
646 	preempt_enable();
647 }
648 
649 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
650 {
651 	if (kvm_request_pending(vcpu)) {
652 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
653 			vcpu_req_sleep(vcpu);
654 
655 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
656 			kvm_reset_vcpu(vcpu);
657 
658 		/*
659 		 * Clear IRQ_PENDING requests that were made to guarantee
660 		 * that a VCPU sees new virtual interrupts.
661 		 */
662 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
663 
664 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
665 			kvm_update_stolen_time(vcpu);
666 
667 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
668 			/* The distributor enable bits were changed */
669 			preempt_disable();
670 			vgic_v4_put(vcpu, false);
671 			vgic_v4_load(vcpu);
672 			preempt_enable();
673 		}
674 
675 		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
676 			kvm_pmu_handle_pmcr(vcpu,
677 					    __vcpu_sys_reg(vcpu, PMCR_EL0));
678 	}
679 }
680 
681 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
682 {
683 	if (likely(!vcpu_mode_is_32bit(vcpu)))
684 		return false;
685 
686 	return !system_supports_32bit_el0() ||
687 		static_branch_unlikely(&arm64_mismatched_32bit_el0);
688 }
689 
690 /**
691  * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
692  * @vcpu:	The VCPU pointer
693  * @ret:	Pointer to write optional return code
694  *
695  * Returns: true if the VCPU needs to return to a preemptible + interruptible
696  *	    and skip guest entry.
697  *
698  * This function disambiguates between two different types of exits: exits to a
699  * preemptible + interruptible kernel context and exits to userspace. For an
700  * exit to userspace, this function will write the return code to ret and return
701  * true. For an exit to preemptible + interruptible kernel context (i.e. check
702  * for pending work and re-enter), return true without writing to ret.
703  */
704 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
705 {
706 	struct kvm_run *run = vcpu->run;
707 
708 	/*
709 	 * If we're using a userspace irqchip, then check if we need
710 	 * to tell a userspace irqchip about timer or PMU level
711 	 * changes and if so, exit to userspace (the actual level
712 	 * state gets updated in kvm_timer_update_run and
713 	 * kvm_pmu_update_run below).
714 	 */
715 	if (static_branch_unlikely(&userspace_irqchip_in_use)) {
716 		if (kvm_timer_should_notify_user(vcpu) ||
717 		    kvm_pmu_should_notify_user(vcpu)) {
718 			*ret = -EINTR;
719 			run->exit_reason = KVM_EXIT_INTR;
720 			return true;
721 		}
722 	}
723 
724 	if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
725 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
726 		run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
727 		run->fail_entry.cpu = smp_processor_id();
728 		*ret = 0;
729 		return true;
730 	}
731 
732 	return kvm_request_pending(vcpu) ||
733 			xfer_to_guest_mode_work_pending();
734 }
735 
736 /*
737  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
738  * the vCPU is running.
739  *
740  * This must be noinstr as instrumentation may make use of RCU, and this is not
741  * safe during the EQS.
742  */
743 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
744 {
745 	int ret;
746 
747 	guest_state_enter_irqoff();
748 	ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
749 	guest_state_exit_irqoff();
750 
751 	return ret;
752 }
753 
754 /**
755  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
756  * @vcpu:	The VCPU pointer
757  *
758  * This function is called through the VCPU_RUN ioctl called from user space. It
759  * will execute VM code in a loop until the time slice for the process is used
760  * or some emulation is needed from user space in which case the function will
761  * return with return value 0 and with the kvm_run structure filled in with the
762  * required data for the requested emulation.
763  */
764 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
765 {
766 	struct kvm_run *run = vcpu->run;
767 	int ret;
768 
769 	if (run->exit_reason == KVM_EXIT_MMIO) {
770 		ret = kvm_handle_mmio_return(vcpu);
771 		if (ret)
772 			return ret;
773 	}
774 
775 	vcpu_load(vcpu);
776 
777 	if (run->immediate_exit) {
778 		ret = -EINTR;
779 		goto out;
780 	}
781 
782 	kvm_sigset_activate(vcpu);
783 
784 	ret = 1;
785 	run->exit_reason = KVM_EXIT_UNKNOWN;
786 	while (ret > 0) {
787 		/*
788 		 * Check conditions before entering the guest
789 		 */
790 		ret = xfer_to_guest_mode_handle_work(vcpu);
791 		if (!ret)
792 			ret = 1;
793 
794 		check_vcpu_requests(vcpu);
795 
796 		/*
797 		 * Preparing the interrupts to be injected also
798 		 * involves poking the GIC, which must be done in a
799 		 * non-preemptible context.
800 		 */
801 		preempt_disable();
802 
803 		/*
804 		 * The VMID allocator only tracks active VMIDs per
805 		 * physical CPU, and therefore the VMID allocated may not be
806 		 * preserved on VMID roll-over if the task was preempted,
807 		 * making a thread's VMID inactive. So we need to call
808 		 * kvm_arm_vmid_update() in non-premptible context.
809 		 */
810 		kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid);
811 
812 		kvm_pmu_flush_hwstate(vcpu);
813 
814 		local_irq_disable();
815 
816 		kvm_vgic_flush_hwstate(vcpu);
817 
818 		/*
819 		 * Ensure we set mode to IN_GUEST_MODE after we disable
820 		 * interrupts and before the final VCPU requests check.
821 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
822 		 * Documentation/virt/kvm/vcpu-requests.rst
823 		 */
824 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
825 
826 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
827 			vcpu->mode = OUTSIDE_GUEST_MODE;
828 			isb(); /* Ensure work in x_flush_hwstate is committed */
829 			kvm_pmu_sync_hwstate(vcpu);
830 			if (static_branch_unlikely(&userspace_irqchip_in_use))
831 				kvm_timer_sync_user(vcpu);
832 			kvm_vgic_sync_hwstate(vcpu);
833 			local_irq_enable();
834 			preempt_enable();
835 			continue;
836 		}
837 
838 		kvm_arm_setup_debug(vcpu);
839 		kvm_arch_vcpu_ctxflush_fp(vcpu);
840 
841 		/**************************************************************
842 		 * Enter the guest
843 		 */
844 		trace_kvm_entry(*vcpu_pc(vcpu));
845 		guest_timing_enter_irqoff();
846 
847 		ret = kvm_arm_vcpu_enter_exit(vcpu);
848 
849 		vcpu->mode = OUTSIDE_GUEST_MODE;
850 		vcpu->stat.exits++;
851 		/*
852 		 * Back from guest
853 		 *************************************************************/
854 
855 		kvm_arm_clear_debug(vcpu);
856 
857 		/*
858 		 * We must sync the PMU state before the vgic state so
859 		 * that the vgic can properly sample the updated state of the
860 		 * interrupt line.
861 		 */
862 		kvm_pmu_sync_hwstate(vcpu);
863 
864 		/*
865 		 * Sync the vgic state before syncing the timer state because
866 		 * the timer code needs to know if the virtual timer
867 		 * interrupts are active.
868 		 */
869 		kvm_vgic_sync_hwstate(vcpu);
870 
871 		/*
872 		 * Sync the timer hardware state before enabling interrupts as
873 		 * we don't want vtimer interrupts to race with syncing the
874 		 * timer virtual interrupt state.
875 		 */
876 		if (static_branch_unlikely(&userspace_irqchip_in_use))
877 			kvm_timer_sync_user(vcpu);
878 
879 		kvm_arch_vcpu_ctxsync_fp(vcpu);
880 
881 		/*
882 		 * We must ensure that any pending interrupts are taken before
883 		 * we exit guest timing so that timer ticks are accounted as
884 		 * guest time. Transiently unmask interrupts so that any
885 		 * pending interrupts are taken.
886 		 *
887 		 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
888 		 * context synchronization event) is necessary to ensure that
889 		 * pending interrupts are taken.
890 		 */
891 		if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
892 			local_irq_enable();
893 			isb();
894 			local_irq_disable();
895 		}
896 
897 		guest_timing_exit_irqoff();
898 
899 		local_irq_enable();
900 
901 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
902 
903 		/* Exit types that need handling before we can be preempted */
904 		handle_exit_early(vcpu, ret);
905 
906 		preempt_enable();
907 
908 		/*
909 		 * The ARMv8 architecture doesn't give the hypervisor
910 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
911 		 * if implemented by the CPU. If we spot the guest in such
912 		 * state and that we decided it wasn't supposed to do so (like
913 		 * with the asymmetric AArch32 case), return to userspace with
914 		 * a fatal error.
915 		 */
916 		if (vcpu_mode_is_bad_32bit(vcpu)) {
917 			/*
918 			 * As we have caught the guest red-handed, decide that
919 			 * it isn't fit for purpose anymore by making the vcpu
920 			 * invalid. The VMM can try and fix it by issuing  a
921 			 * KVM_ARM_VCPU_INIT if it really wants to.
922 			 */
923 			vcpu->arch.target = -1;
924 			ret = ARM_EXCEPTION_IL;
925 		}
926 
927 		ret = handle_exit(vcpu, ret);
928 	}
929 
930 	/* Tell userspace about in-kernel device output levels */
931 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
932 		kvm_timer_update_run(vcpu);
933 		kvm_pmu_update_run(vcpu);
934 	}
935 
936 	kvm_sigset_deactivate(vcpu);
937 
938 out:
939 	/*
940 	 * In the unlikely event that we are returning to userspace
941 	 * with pending exceptions or PC adjustment, commit these
942 	 * adjustments in order to give userspace a consistent view of
943 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
944 	 * being preempt-safe on VHE.
945 	 */
946 	if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
947 					 KVM_ARM64_INCREMENT_PC)))
948 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
949 
950 	vcpu_put(vcpu);
951 	return ret;
952 }
953 
954 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
955 {
956 	int bit_index;
957 	bool set;
958 	unsigned long *hcr;
959 
960 	if (number == KVM_ARM_IRQ_CPU_IRQ)
961 		bit_index = __ffs(HCR_VI);
962 	else /* KVM_ARM_IRQ_CPU_FIQ */
963 		bit_index = __ffs(HCR_VF);
964 
965 	hcr = vcpu_hcr(vcpu);
966 	if (level)
967 		set = test_and_set_bit(bit_index, hcr);
968 	else
969 		set = test_and_clear_bit(bit_index, hcr);
970 
971 	/*
972 	 * If we didn't change anything, no need to wake up or kick other CPUs
973 	 */
974 	if (set == level)
975 		return 0;
976 
977 	/*
978 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
979 	 * trigger a world-switch round on the running physical CPU to set the
980 	 * virtual IRQ/FIQ fields in the HCR appropriately.
981 	 */
982 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
983 	kvm_vcpu_kick(vcpu);
984 
985 	return 0;
986 }
987 
988 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
989 			  bool line_status)
990 {
991 	u32 irq = irq_level->irq;
992 	unsigned int irq_type, vcpu_idx, irq_num;
993 	int nrcpus = atomic_read(&kvm->online_vcpus);
994 	struct kvm_vcpu *vcpu = NULL;
995 	bool level = irq_level->level;
996 
997 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
998 	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
999 	vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1000 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1001 
1002 	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1003 
1004 	switch (irq_type) {
1005 	case KVM_ARM_IRQ_TYPE_CPU:
1006 		if (irqchip_in_kernel(kvm))
1007 			return -ENXIO;
1008 
1009 		if (vcpu_idx >= nrcpus)
1010 			return -EINVAL;
1011 
1012 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1013 		if (!vcpu)
1014 			return -EINVAL;
1015 
1016 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1017 			return -EINVAL;
1018 
1019 		return vcpu_interrupt_line(vcpu, irq_num, level);
1020 	case KVM_ARM_IRQ_TYPE_PPI:
1021 		if (!irqchip_in_kernel(kvm))
1022 			return -ENXIO;
1023 
1024 		if (vcpu_idx >= nrcpus)
1025 			return -EINVAL;
1026 
1027 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1028 		if (!vcpu)
1029 			return -EINVAL;
1030 
1031 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1032 			return -EINVAL;
1033 
1034 		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1035 	case KVM_ARM_IRQ_TYPE_SPI:
1036 		if (!irqchip_in_kernel(kvm))
1037 			return -ENXIO;
1038 
1039 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1040 			return -EINVAL;
1041 
1042 		return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1043 	}
1044 
1045 	return -EINVAL;
1046 }
1047 
1048 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1049 			       const struct kvm_vcpu_init *init)
1050 {
1051 	unsigned int i, ret;
1052 	u32 phys_target = kvm_target_cpu();
1053 
1054 	if (init->target != phys_target)
1055 		return -EINVAL;
1056 
1057 	/*
1058 	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1059 	 * use the same target.
1060 	 */
1061 	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1062 		return -EINVAL;
1063 
1064 	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1065 	for (i = 0; i < sizeof(init->features) * 8; i++) {
1066 		bool set = (init->features[i / 32] & (1 << (i % 32)));
1067 
1068 		if (set && i >= KVM_VCPU_MAX_FEATURES)
1069 			return -ENOENT;
1070 
1071 		/*
1072 		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1073 		 * use the same feature set.
1074 		 */
1075 		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1076 		    test_bit(i, vcpu->arch.features) != set)
1077 			return -EINVAL;
1078 
1079 		if (set)
1080 			set_bit(i, vcpu->arch.features);
1081 	}
1082 
1083 	vcpu->arch.target = phys_target;
1084 
1085 	/* Now we know what it is, we can reset it. */
1086 	ret = kvm_reset_vcpu(vcpu);
1087 	if (ret) {
1088 		vcpu->arch.target = -1;
1089 		bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1090 	}
1091 
1092 	return ret;
1093 }
1094 
1095 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1096 					 struct kvm_vcpu_init *init)
1097 {
1098 	int ret;
1099 
1100 	ret = kvm_vcpu_set_target(vcpu, init);
1101 	if (ret)
1102 		return ret;
1103 
1104 	/*
1105 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1106 	 * guest MMU is turned off and flush the caches as needed.
1107 	 *
1108 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1109 	 * ensuring that the data side is always coherent. We still
1110 	 * need to invalidate the I-cache though, as FWB does *not*
1111 	 * imply CTR_EL0.DIC.
1112 	 */
1113 	if (vcpu_has_run_once(vcpu)) {
1114 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1115 			stage2_unmap_vm(vcpu->kvm);
1116 		else
1117 			icache_inval_all_pou();
1118 	}
1119 
1120 	vcpu_reset_hcr(vcpu);
1121 	vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1122 
1123 	/*
1124 	 * Handle the "start in power-off" case.
1125 	 */
1126 	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1127 		vcpu_power_off(vcpu);
1128 	else
1129 		vcpu->arch.power_off = false;
1130 
1131 	return 0;
1132 }
1133 
1134 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1135 				 struct kvm_device_attr *attr)
1136 {
1137 	int ret = -ENXIO;
1138 
1139 	switch (attr->group) {
1140 	default:
1141 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1142 		break;
1143 	}
1144 
1145 	return ret;
1146 }
1147 
1148 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1149 				 struct kvm_device_attr *attr)
1150 {
1151 	int ret = -ENXIO;
1152 
1153 	switch (attr->group) {
1154 	default:
1155 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1156 		break;
1157 	}
1158 
1159 	return ret;
1160 }
1161 
1162 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1163 				 struct kvm_device_attr *attr)
1164 {
1165 	int ret = -ENXIO;
1166 
1167 	switch (attr->group) {
1168 	default:
1169 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1170 		break;
1171 	}
1172 
1173 	return ret;
1174 }
1175 
1176 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1177 				   struct kvm_vcpu_events *events)
1178 {
1179 	memset(events, 0, sizeof(*events));
1180 
1181 	return __kvm_arm_vcpu_get_events(vcpu, events);
1182 }
1183 
1184 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1185 				   struct kvm_vcpu_events *events)
1186 {
1187 	int i;
1188 
1189 	/* check whether the reserved field is zero */
1190 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1191 		if (events->reserved[i])
1192 			return -EINVAL;
1193 
1194 	/* check whether the pad field is zero */
1195 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1196 		if (events->exception.pad[i])
1197 			return -EINVAL;
1198 
1199 	return __kvm_arm_vcpu_set_events(vcpu, events);
1200 }
1201 
1202 long kvm_arch_vcpu_ioctl(struct file *filp,
1203 			 unsigned int ioctl, unsigned long arg)
1204 {
1205 	struct kvm_vcpu *vcpu = filp->private_data;
1206 	void __user *argp = (void __user *)arg;
1207 	struct kvm_device_attr attr;
1208 	long r;
1209 
1210 	switch (ioctl) {
1211 	case KVM_ARM_VCPU_INIT: {
1212 		struct kvm_vcpu_init init;
1213 
1214 		r = -EFAULT;
1215 		if (copy_from_user(&init, argp, sizeof(init)))
1216 			break;
1217 
1218 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1219 		break;
1220 	}
1221 	case KVM_SET_ONE_REG:
1222 	case KVM_GET_ONE_REG: {
1223 		struct kvm_one_reg reg;
1224 
1225 		r = -ENOEXEC;
1226 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1227 			break;
1228 
1229 		r = -EFAULT;
1230 		if (copy_from_user(&reg, argp, sizeof(reg)))
1231 			break;
1232 
1233 		/*
1234 		 * We could owe a reset due to PSCI. Handle the pending reset
1235 		 * here to ensure userspace register accesses are ordered after
1236 		 * the reset.
1237 		 */
1238 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1239 			kvm_reset_vcpu(vcpu);
1240 
1241 		if (ioctl == KVM_SET_ONE_REG)
1242 			r = kvm_arm_set_reg(vcpu, &reg);
1243 		else
1244 			r = kvm_arm_get_reg(vcpu, &reg);
1245 		break;
1246 	}
1247 	case KVM_GET_REG_LIST: {
1248 		struct kvm_reg_list __user *user_list = argp;
1249 		struct kvm_reg_list reg_list;
1250 		unsigned n;
1251 
1252 		r = -ENOEXEC;
1253 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1254 			break;
1255 
1256 		r = -EPERM;
1257 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1258 			break;
1259 
1260 		r = -EFAULT;
1261 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1262 			break;
1263 		n = reg_list.n;
1264 		reg_list.n = kvm_arm_num_regs(vcpu);
1265 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1266 			break;
1267 		r = -E2BIG;
1268 		if (n < reg_list.n)
1269 			break;
1270 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1271 		break;
1272 	}
1273 	case KVM_SET_DEVICE_ATTR: {
1274 		r = -EFAULT;
1275 		if (copy_from_user(&attr, argp, sizeof(attr)))
1276 			break;
1277 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1278 		break;
1279 	}
1280 	case KVM_GET_DEVICE_ATTR: {
1281 		r = -EFAULT;
1282 		if (copy_from_user(&attr, argp, sizeof(attr)))
1283 			break;
1284 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1285 		break;
1286 	}
1287 	case KVM_HAS_DEVICE_ATTR: {
1288 		r = -EFAULT;
1289 		if (copy_from_user(&attr, argp, sizeof(attr)))
1290 			break;
1291 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1292 		break;
1293 	}
1294 	case KVM_GET_VCPU_EVENTS: {
1295 		struct kvm_vcpu_events events;
1296 
1297 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1298 			return -EINVAL;
1299 
1300 		if (copy_to_user(argp, &events, sizeof(events)))
1301 			return -EFAULT;
1302 
1303 		return 0;
1304 	}
1305 	case KVM_SET_VCPU_EVENTS: {
1306 		struct kvm_vcpu_events events;
1307 
1308 		if (copy_from_user(&events, argp, sizeof(events)))
1309 			return -EFAULT;
1310 
1311 		return kvm_arm_vcpu_set_events(vcpu, &events);
1312 	}
1313 	case KVM_ARM_VCPU_FINALIZE: {
1314 		int what;
1315 
1316 		if (!kvm_vcpu_initialized(vcpu))
1317 			return -ENOEXEC;
1318 
1319 		if (get_user(what, (const int __user *)argp))
1320 			return -EFAULT;
1321 
1322 		return kvm_arm_vcpu_finalize(vcpu, what);
1323 	}
1324 	default:
1325 		r = -EINVAL;
1326 	}
1327 
1328 	return r;
1329 }
1330 
1331 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1332 {
1333 
1334 }
1335 
1336 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1337 					const struct kvm_memory_slot *memslot)
1338 {
1339 	kvm_flush_remote_tlbs(kvm);
1340 }
1341 
1342 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1343 					struct kvm_arm_device_addr *dev_addr)
1344 {
1345 	unsigned long dev_id, type;
1346 
1347 	dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1348 		KVM_ARM_DEVICE_ID_SHIFT;
1349 	type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1350 		KVM_ARM_DEVICE_TYPE_SHIFT;
1351 
1352 	switch (dev_id) {
1353 	case KVM_ARM_DEVICE_VGIC_V2:
1354 		if (!vgic_present)
1355 			return -ENXIO;
1356 		return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1357 	default:
1358 		return -ENODEV;
1359 	}
1360 }
1361 
1362 long kvm_arch_vm_ioctl(struct file *filp,
1363 		       unsigned int ioctl, unsigned long arg)
1364 {
1365 	struct kvm *kvm = filp->private_data;
1366 	void __user *argp = (void __user *)arg;
1367 
1368 	switch (ioctl) {
1369 	case KVM_CREATE_IRQCHIP: {
1370 		int ret;
1371 		if (!vgic_present)
1372 			return -ENXIO;
1373 		mutex_lock(&kvm->lock);
1374 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1375 		mutex_unlock(&kvm->lock);
1376 		return ret;
1377 	}
1378 	case KVM_ARM_SET_DEVICE_ADDR: {
1379 		struct kvm_arm_device_addr dev_addr;
1380 
1381 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1382 			return -EFAULT;
1383 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1384 	}
1385 	case KVM_ARM_PREFERRED_TARGET: {
1386 		struct kvm_vcpu_init init;
1387 
1388 		kvm_vcpu_preferred_target(&init);
1389 
1390 		if (copy_to_user(argp, &init, sizeof(init)))
1391 			return -EFAULT;
1392 
1393 		return 0;
1394 	}
1395 	case KVM_ARM_MTE_COPY_TAGS: {
1396 		struct kvm_arm_copy_mte_tags copy_tags;
1397 
1398 		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1399 			return -EFAULT;
1400 		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1401 	}
1402 	default:
1403 		return -EINVAL;
1404 	}
1405 }
1406 
1407 static unsigned long nvhe_percpu_size(void)
1408 {
1409 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1410 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1411 }
1412 
1413 static unsigned long nvhe_percpu_order(void)
1414 {
1415 	unsigned long size = nvhe_percpu_size();
1416 
1417 	return size ? get_order(size) : 0;
1418 }
1419 
1420 /* A lookup table holding the hypervisor VA for each vector slot */
1421 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1422 
1423 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1424 {
1425 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1426 }
1427 
1428 static int kvm_init_vector_slots(void)
1429 {
1430 	int err;
1431 	void *base;
1432 
1433 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1434 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1435 
1436 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1437 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1438 
1439 	if (kvm_system_needs_idmapped_vectors() &&
1440 	    !is_protected_kvm_enabled()) {
1441 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1442 					       __BP_HARDEN_HYP_VECS_SZ, &base);
1443 		if (err)
1444 			return err;
1445 	}
1446 
1447 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1448 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1449 	return 0;
1450 }
1451 
1452 static void cpu_prepare_hyp_mode(int cpu)
1453 {
1454 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1455 	unsigned long tcr;
1456 
1457 	/*
1458 	 * Calculate the raw per-cpu offset without a translation from the
1459 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1460 	 * so that we can use adr_l to access per-cpu variables in EL2.
1461 	 * Also drop the KASAN tag which gets in the way...
1462 	 */
1463 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1464 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1465 
1466 	params->mair_el2 = read_sysreg(mair_el1);
1467 
1468 	/*
1469 	 * The ID map may be configured to use an extended virtual address
1470 	 * range. This is only the case if system RAM is out of range for the
1471 	 * currently configured page size and VA_BITS, in which case we will
1472 	 * also need the extended virtual range for the HYP ID map, or we won't
1473 	 * be able to enable the EL2 MMU.
1474 	 *
1475 	 * However, at EL2, there is only one TTBR register, and we can't switch
1476 	 * between translation tables *and* update TCR_EL2.T0SZ at the same
1477 	 * time. Bottom line: we need to use the extended range with *both* our
1478 	 * translation tables.
1479 	 *
1480 	 * So use the same T0SZ value we use for the ID map.
1481 	 */
1482 	tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1483 	tcr &= ~TCR_T0SZ_MASK;
1484 	tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1485 	params->tcr_el2 = tcr;
1486 
1487 	params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1488 	params->pgd_pa = kvm_mmu_get_httbr();
1489 	if (is_protected_kvm_enabled())
1490 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1491 	else
1492 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1493 	params->vttbr = params->vtcr = 0;
1494 
1495 	/*
1496 	 * Flush the init params from the data cache because the struct will
1497 	 * be read while the MMU is off.
1498 	 */
1499 	kvm_flush_dcache_to_poc(params, sizeof(*params));
1500 }
1501 
1502 static void hyp_install_host_vector(void)
1503 {
1504 	struct kvm_nvhe_init_params *params;
1505 	struct arm_smccc_res res;
1506 
1507 	/* Switch from the HYP stub to our own HYP init vector */
1508 	__hyp_set_vectors(kvm_get_idmap_vector());
1509 
1510 	/*
1511 	 * Call initialization code, and switch to the full blown HYP code.
1512 	 * If the cpucaps haven't been finalized yet, something has gone very
1513 	 * wrong, and hyp will crash and burn when it uses any
1514 	 * cpus_have_const_cap() wrapper.
1515 	 */
1516 	BUG_ON(!system_capabilities_finalized());
1517 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1518 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1519 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1520 }
1521 
1522 static void cpu_init_hyp_mode(void)
1523 {
1524 	hyp_install_host_vector();
1525 
1526 	/*
1527 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1528 	 * at EL2.
1529 	 */
1530 	if (this_cpu_has_cap(ARM64_SSBS) &&
1531 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1532 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1533 	}
1534 }
1535 
1536 static void cpu_hyp_reset(void)
1537 {
1538 	if (!is_kernel_in_hyp_mode())
1539 		__hyp_reset_vectors();
1540 }
1541 
1542 /*
1543  * EL2 vectors can be mapped and rerouted in a number of ways,
1544  * depending on the kernel configuration and CPU present:
1545  *
1546  * - If the CPU is affected by Spectre-v2, the hardening sequence is
1547  *   placed in one of the vector slots, which is executed before jumping
1548  *   to the real vectors.
1549  *
1550  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1551  *   containing the hardening sequence is mapped next to the idmap page,
1552  *   and executed before jumping to the real vectors.
1553  *
1554  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1555  *   empty slot is selected, mapped next to the idmap page, and
1556  *   executed before jumping to the real vectors.
1557  *
1558  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1559  * VHE, as we don't have hypervisor-specific mappings. If the system
1560  * is VHE and yet selects this capability, it will be ignored.
1561  */
1562 static void cpu_set_hyp_vector(void)
1563 {
1564 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1565 	void *vector = hyp_spectre_vector_selector[data->slot];
1566 
1567 	if (!is_protected_kvm_enabled())
1568 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1569 	else
1570 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1571 }
1572 
1573 static void cpu_hyp_init_context(void)
1574 {
1575 	kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1576 
1577 	if (!is_kernel_in_hyp_mode())
1578 		cpu_init_hyp_mode();
1579 }
1580 
1581 static void cpu_hyp_init_features(void)
1582 {
1583 	cpu_set_hyp_vector();
1584 	kvm_arm_init_debug();
1585 
1586 	if (is_kernel_in_hyp_mode())
1587 		kvm_timer_init_vhe();
1588 
1589 	if (vgic_present)
1590 		kvm_vgic_init_cpu_hardware();
1591 }
1592 
1593 static void cpu_hyp_reinit(void)
1594 {
1595 	cpu_hyp_reset();
1596 	cpu_hyp_init_context();
1597 	cpu_hyp_init_features();
1598 }
1599 
1600 static void _kvm_arch_hardware_enable(void *discard)
1601 {
1602 	if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1603 		cpu_hyp_reinit();
1604 		__this_cpu_write(kvm_arm_hardware_enabled, 1);
1605 	}
1606 }
1607 
1608 int kvm_arch_hardware_enable(void)
1609 {
1610 	_kvm_arch_hardware_enable(NULL);
1611 	return 0;
1612 }
1613 
1614 static void _kvm_arch_hardware_disable(void *discard)
1615 {
1616 	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1617 		cpu_hyp_reset();
1618 		__this_cpu_write(kvm_arm_hardware_enabled, 0);
1619 	}
1620 }
1621 
1622 void kvm_arch_hardware_disable(void)
1623 {
1624 	if (!is_protected_kvm_enabled())
1625 		_kvm_arch_hardware_disable(NULL);
1626 }
1627 
1628 #ifdef CONFIG_CPU_PM
1629 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1630 				    unsigned long cmd,
1631 				    void *v)
1632 {
1633 	/*
1634 	 * kvm_arm_hardware_enabled is left with its old value over
1635 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1636 	 * re-enable hyp.
1637 	 */
1638 	switch (cmd) {
1639 	case CPU_PM_ENTER:
1640 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1641 			/*
1642 			 * don't update kvm_arm_hardware_enabled here
1643 			 * so that the hardware will be re-enabled
1644 			 * when we resume. See below.
1645 			 */
1646 			cpu_hyp_reset();
1647 
1648 		return NOTIFY_OK;
1649 	case CPU_PM_ENTER_FAILED:
1650 	case CPU_PM_EXIT:
1651 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1652 			/* The hardware was enabled before suspend. */
1653 			cpu_hyp_reinit();
1654 
1655 		return NOTIFY_OK;
1656 
1657 	default:
1658 		return NOTIFY_DONE;
1659 	}
1660 }
1661 
1662 static struct notifier_block hyp_init_cpu_pm_nb = {
1663 	.notifier_call = hyp_init_cpu_pm_notifier,
1664 };
1665 
1666 static void hyp_cpu_pm_init(void)
1667 {
1668 	if (!is_protected_kvm_enabled())
1669 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1670 }
1671 static void hyp_cpu_pm_exit(void)
1672 {
1673 	if (!is_protected_kvm_enabled())
1674 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1675 }
1676 #else
1677 static inline void hyp_cpu_pm_init(void)
1678 {
1679 }
1680 static inline void hyp_cpu_pm_exit(void)
1681 {
1682 }
1683 #endif
1684 
1685 static void init_cpu_logical_map(void)
1686 {
1687 	unsigned int cpu;
1688 
1689 	/*
1690 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1691 	 * Only copy the set of online CPUs whose features have been checked
1692 	 * against the finalized system capabilities. The hypervisor will not
1693 	 * allow any other CPUs from the `possible` set to boot.
1694 	 */
1695 	for_each_online_cpu(cpu)
1696 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1697 }
1698 
1699 #define init_psci_0_1_impl_state(config, what)	\
1700 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
1701 
1702 static bool init_psci_relay(void)
1703 {
1704 	/*
1705 	 * If PSCI has not been initialized, protected KVM cannot install
1706 	 * itself on newly booted CPUs.
1707 	 */
1708 	if (!psci_ops.get_version) {
1709 		kvm_err("Cannot initialize protected mode without PSCI\n");
1710 		return false;
1711 	}
1712 
1713 	kvm_host_psci_config.version = psci_ops.get_version();
1714 
1715 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1716 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1717 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1718 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1719 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1720 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1721 	}
1722 	return true;
1723 }
1724 
1725 static int init_subsystems(void)
1726 {
1727 	int err = 0;
1728 
1729 	/*
1730 	 * Enable hardware so that subsystem initialisation can access EL2.
1731 	 */
1732 	on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1733 
1734 	/*
1735 	 * Register CPU lower-power notifier
1736 	 */
1737 	hyp_cpu_pm_init();
1738 
1739 	/*
1740 	 * Init HYP view of VGIC
1741 	 */
1742 	err = kvm_vgic_hyp_init();
1743 	switch (err) {
1744 	case 0:
1745 		vgic_present = true;
1746 		break;
1747 	case -ENODEV:
1748 	case -ENXIO:
1749 		vgic_present = false;
1750 		err = 0;
1751 		break;
1752 	default:
1753 		goto out;
1754 	}
1755 
1756 	/*
1757 	 * Init HYP architected timer support
1758 	 */
1759 	err = kvm_timer_hyp_init(vgic_present);
1760 	if (err)
1761 		goto out;
1762 
1763 	kvm_register_perf_callbacks(NULL);
1764 
1765 	kvm_sys_reg_table_init();
1766 
1767 out:
1768 	if (err || !is_protected_kvm_enabled())
1769 		on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1770 
1771 	return err;
1772 }
1773 
1774 static void teardown_hyp_mode(void)
1775 {
1776 	int cpu;
1777 
1778 	free_hyp_pgds();
1779 	for_each_possible_cpu(cpu) {
1780 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1781 		free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1782 	}
1783 }
1784 
1785 static int do_pkvm_init(u32 hyp_va_bits)
1786 {
1787 	void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1788 	int ret;
1789 
1790 	preempt_disable();
1791 	cpu_hyp_init_context();
1792 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1793 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
1794 				hyp_va_bits);
1795 	cpu_hyp_init_features();
1796 
1797 	/*
1798 	 * The stub hypercalls are now disabled, so set our local flag to
1799 	 * prevent a later re-init attempt in kvm_arch_hardware_enable().
1800 	 */
1801 	__this_cpu_write(kvm_arm_hardware_enabled, 1);
1802 	preempt_enable();
1803 
1804 	return ret;
1805 }
1806 
1807 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1808 {
1809 	void *addr = phys_to_virt(hyp_mem_base);
1810 	int ret;
1811 
1812 	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1813 	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1814 	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
1815 	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
1816 	kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
1817 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1818 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1819 	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
1820 
1821 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1822 	if (ret)
1823 		return ret;
1824 
1825 	ret = do_pkvm_init(hyp_va_bits);
1826 	if (ret)
1827 		return ret;
1828 
1829 	free_hyp_pgds();
1830 
1831 	return 0;
1832 }
1833 
1834 /**
1835  * Inits Hyp-mode on all online CPUs
1836  */
1837 static int init_hyp_mode(void)
1838 {
1839 	u32 hyp_va_bits;
1840 	int cpu;
1841 	int err = -ENOMEM;
1842 
1843 	/*
1844 	 * The protected Hyp-mode cannot be initialized if the memory pool
1845 	 * allocation has failed.
1846 	 */
1847 	if (is_protected_kvm_enabled() && !hyp_mem_base)
1848 		goto out_err;
1849 
1850 	/*
1851 	 * Allocate Hyp PGD and setup Hyp identity mapping
1852 	 */
1853 	err = kvm_mmu_init(&hyp_va_bits);
1854 	if (err)
1855 		goto out_err;
1856 
1857 	/*
1858 	 * Allocate stack pages for Hypervisor-mode
1859 	 */
1860 	for_each_possible_cpu(cpu) {
1861 		unsigned long stack_page;
1862 
1863 		stack_page = __get_free_page(GFP_KERNEL);
1864 		if (!stack_page) {
1865 			err = -ENOMEM;
1866 			goto out_err;
1867 		}
1868 
1869 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1870 	}
1871 
1872 	/*
1873 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1874 	 */
1875 	for_each_possible_cpu(cpu) {
1876 		struct page *page;
1877 		void *page_addr;
1878 
1879 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1880 		if (!page) {
1881 			err = -ENOMEM;
1882 			goto out_err;
1883 		}
1884 
1885 		page_addr = page_address(page);
1886 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1887 		kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1888 	}
1889 
1890 	/*
1891 	 * Map the Hyp-code called directly from the host
1892 	 */
1893 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1894 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1895 	if (err) {
1896 		kvm_err("Cannot map world-switch code\n");
1897 		goto out_err;
1898 	}
1899 
1900 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1901 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1902 	if (err) {
1903 		kvm_err("Cannot map .hyp.rodata section\n");
1904 		goto out_err;
1905 	}
1906 
1907 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1908 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1909 	if (err) {
1910 		kvm_err("Cannot map rodata section\n");
1911 		goto out_err;
1912 	}
1913 
1914 	/*
1915 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1916 	 * section thanks to an assertion in the linker script. Map it RW and
1917 	 * the rest of .bss RO.
1918 	 */
1919 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1920 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1921 	if (err) {
1922 		kvm_err("Cannot map hyp bss section: %d\n", err);
1923 		goto out_err;
1924 	}
1925 
1926 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1927 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1928 	if (err) {
1929 		kvm_err("Cannot map bss section\n");
1930 		goto out_err;
1931 	}
1932 
1933 	/*
1934 	 * Map the Hyp stack pages
1935 	 */
1936 	for_each_possible_cpu(cpu) {
1937 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1938 		err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1939 					  PAGE_HYP);
1940 
1941 		if (err) {
1942 			kvm_err("Cannot map hyp stack\n");
1943 			goto out_err;
1944 		}
1945 	}
1946 
1947 	for_each_possible_cpu(cpu) {
1948 		char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1949 		char *percpu_end = percpu_begin + nvhe_percpu_size();
1950 
1951 		/* Map Hyp percpu pages */
1952 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1953 		if (err) {
1954 			kvm_err("Cannot map hyp percpu region\n");
1955 			goto out_err;
1956 		}
1957 
1958 		/* Prepare the CPU initialization parameters */
1959 		cpu_prepare_hyp_mode(cpu);
1960 	}
1961 
1962 	if (is_protected_kvm_enabled()) {
1963 		init_cpu_logical_map();
1964 
1965 		if (!init_psci_relay()) {
1966 			err = -ENODEV;
1967 			goto out_err;
1968 		}
1969 	}
1970 
1971 	if (is_protected_kvm_enabled()) {
1972 		err = kvm_hyp_init_protection(hyp_va_bits);
1973 		if (err) {
1974 			kvm_err("Failed to init hyp memory protection\n");
1975 			goto out_err;
1976 		}
1977 	}
1978 
1979 	return 0;
1980 
1981 out_err:
1982 	teardown_hyp_mode();
1983 	kvm_err("error initializing Hyp mode: %d\n", err);
1984 	return err;
1985 }
1986 
1987 static void _kvm_host_prot_finalize(void *arg)
1988 {
1989 	int *err = arg;
1990 
1991 	if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)))
1992 		WRITE_ONCE(*err, -EINVAL);
1993 }
1994 
1995 static int pkvm_drop_host_privileges(void)
1996 {
1997 	int ret = 0;
1998 
1999 	/*
2000 	 * Flip the static key upfront as that may no longer be possible
2001 	 * once the host stage 2 is installed.
2002 	 */
2003 	static_branch_enable(&kvm_protected_mode_initialized);
2004 	on_each_cpu(_kvm_host_prot_finalize, &ret, 1);
2005 	return ret;
2006 }
2007 
2008 static int finalize_hyp_mode(void)
2009 {
2010 	if (!is_protected_kvm_enabled())
2011 		return 0;
2012 
2013 	/*
2014 	 * Exclude HYP BSS from kmemleak so that it doesn't get peeked
2015 	 * at, which would end badly once the section is inaccessible.
2016 	 * None of other sections should ever be introspected.
2017 	 */
2018 	kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
2019 	return pkvm_drop_host_privileges();
2020 }
2021 
2022 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2023 {
2024 	struct kvm_vcpu *vcpu;
2025 	unsigned long i;
2026 
2027 	mpidr &= MPIDR_HWID_BITMASK;
2028 	kvm_for_each_vcpu(i, vcpu, kvm) {
2029 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2030 			return vcpu;
2031 	}
2032 	return NULL;
2033 }
2034 
2035 bool kvm_arch_has_irq_bypass(void)
2036 {
2037 	return true;
2038 }
2039 
2040 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2041 				      struct irq_bypass_producer *prod)
2042 {
2043 	struct kvm_kernel_irqfd *irqfd =
2044 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2045 
2046 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2047 					  &irqfd->irq_entry);
2048 }
2049 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2050 				      struct irq_bypass_producer *prod)
2051 {
2052 	struct kvm_kernel_irqfd *irqfd =
2053 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2054 
2055 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2056 				     &irqfd->irq_entry);
2057 }
2058 
2059 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2060 {
2061 	struct kvm_kernel_irqfd *irqfd =
2062 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2063 
2064 	kvm_arm_halt_guest(irqfd->kvm);
2065 }
2066 
2067 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2068 {
2069 	struct kvm_kernel_irqfd *irqfd =
2070 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2071 
2072 	kvm_arm_resume_guest(irqfd->kvm);
2073 }
2074 
2075 /**
2076  * Initialize Hyp-mode and memory mappings on all CPUs.
2077  */
2078 int kvm_arch_init(void *opaque)
2079 {
2080 	int err;
2081 	bool in_hyp_mode;
2082 
2083 	if (!is_hyp_mode_available()) {
2084 		kvm_info("HYP mode not available\n");
2085 		return -ENODEV;
2086 	}
2087 
2088 	if (kvm_get_mode() == KVM_MODE_NONE) {
2089 		kvm_info("KVM disabled from command line\n");
2090 		return -ENODEV;
2091 	}
2092 
2093 	in_hyp_mode = is_kernel_in_hyp_mode();
2094 
2095 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2096 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2097 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2098 			 "Only trusted guests should be used on this system.\n");
2099 
2100 	err = kvm_set_ipa_limit();
2101 	if (err)
2102 		return err;
2103 
2104 	err = kvm_arm_init_sve();
2105 	if (err)
2106 		return err;
2107 
2108 	err = kvm_arm_vmid_alloc_init();
2109 	if (err) {
2110 		kvm_err("Failed to initialize VMID allocator.\n");
2111 		return err;
2112 	}
2113 
2114 	if (!in_hyp_mode) {
2115 		err = init_hyp_mode();
2116 		if (err)
2117 			goto out_err;
2118 	}
2119 
2120 	err = kvm_init_vector_slots();
2121 	if (err) {
2122 		kvm_err("Cannot initialise vector slots\n");
2123 		goto out_err;
2124 	}
2125 
2126 	err = init_subsystems();
2127 	if (err)
2128 		goto out_hyp;
2129 
2130 	if (!in_hyp_mode) {
2131 		err = finalize_hyp_mode();
2132 		if (err) {
2133 			kvm_err("Failed to finalize Hyp protection\n");
2134 			goto out_hyp;
2135 		}
2136 	}
2137 
2138 	if (is_protected_kvm_enabled()) {
2139 		kvm_info("Protected nVHE mode initialized successfully\n");
2140 	} else if (in_hyp_mode) {
2141 		kvm_info("VHE mode initialized successfully\n");
2142 	} else {
2143 		kvm_info("Hyp mode initialized successfully\n");
2144 	}
2145 
2146 	return 0;
2147 
2148 out_hyp:
2149 	hyp_cpu_pm_exit();
2150 	if (!in_hyp_mode)
2151 		teardown_hyp_mode();
2152 out_err:
2153 	kvm_arm_vmid_alloc_free();
2154 	return err;
2155 }
2156 
2157 /* NOP: Compiling as a module not supported */
2158 void kvm_arch_exit(void)
2159 {
2160 	kvm_unregister_perf_callbacks();
2161 }
2162 
2163 static int __init early_kvm_mode_cfg(char *arg)
2164 {
2165 	if (!arg)
2166 		return -EINVAL;
2167 
2168 	if (strcmp(arg, "protected") == 0) {
2169 		kvm_mode = KVM_MODE_PROTECTED;
2170 		return 0;
2171 	}
2172 
2173 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2174 		kvm_mode = KVM_MODE_DEFAULT;
2175 		return 0;
2176 	}
2177 
2178 	if (strcmp(arg, "none") == 0) {
2179 		kvm_mode = KVM_MODE_NONE;
2180 		return 0;
2181 	}
2182 
2183 	return -EINVAL;
2184 }
2185 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2186 
2187 enum kvm_mode kvm_get_mode(void)
2188 {
2189 	return kvm_mode;
2190 }
2191 
2192 static int arm_init(void)
2193 {
2194 	int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2195 	return rc;
2196 }
2197 
2198 module_init(arm_init);
2199