xref: /linux/arch/arm64/kvm/arm.c (revision 79e03f29fc84db76f78834b1f73ed329b377d753)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
4  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
5  */
6 
7 #include <linux/bug.h>
8 #include <linux/cpu_pm.h>
9 #include <linux/entry-kvm.h>
10 #include <linux/errno.h>
11 #include <linux/err.h>
12 #include <linux/kvm_host.h>
13 #include <linux/list.h>
14 #include <linux/module.h>
15 #include <linux/vmalloc.h>
16 #include <linux/fs.h>
17 #include <linux/mman.h>
18 #include <linux/sched.h>
19 #include <linux/kmemleak.h>
20 #include <linux/kvm.h>
21 #include <linux/kvm_irqfd.h>
22 #include <linux/irqbypass.h>
23 #include <linux/sched/stat.h>
24 #include <linux/psci.h>
25 #include <trace/events/kvm.h>
26 
27 #define CREATE_TRACE_POINTS
28 #include "trace_arm.h"
29 
30 #include <linux/uaccess.h>
31 #include <asm/ptrace.h>
32 #include <asm/mman.h>
33 #include <asm/tlbflush.h>
34 #include <asm/cacheflush.h>
35 #include <asm/cpufeature.h>
36 #include <asm/virt.h>
37 #include <asm/kvm_arm.h>
38 #include <asm/kvm_asm.h>
39 #include <asm/kvm_mmu.h>
40 #include <asm/kvm_emulate.h>
41 #include <asm/sections.h>
42 
43 #include <kvm/arm_hypercalls.h>
44 #include <kvm/arm_pmu.h>
45 #include <kvm/arm_psci.h>
46 
47 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
48 DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized);
49 
50 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
51 
52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
53 unsigned long kvm_arm_hyp_percpu_base[NR_CPUS];
54 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
55 
56 /* The VMID used in the VTTBR */
57 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
58 static u32 kvm_next_vmid;
59 static DEFINE_SPINLOCK(kvm_vmid_lock);
60 
61 static bool vgic_present;
62 
63 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
64 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
65 
66 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
67 {
68 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
69 }
70 
71 int kvm_arch_hardware_setup(void *opaque)
72 {
73 	return 0;
74 }
75 
76 int kvm_arch_check_processor_compat(void *opaque)
77 {
78 	return 0;
79 }
80 
81 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
82 			    struct kvm_enable_cap *cap)
83 {
84 	int r;
85 
86 	if (cap->flags)
87 		return -EINVAL;
88 
89 	switch (cap->cap) {
90 	case KVM_CAP_ARM_NISV_TO_USER:
91 		r = 0;
92 		kvm->arch.return_nisv_io_abort_to_user = true;
93 		break;
94 	case KVM_CAP_ARM_MTE:
95 		mutex_lock(&kvm->lock);
96 		if (!system_supports_mte() || kvm->created_vcpus) {
97 			r = -EINVAL;
98 		} else {
99 			r = 0;
100 			kvm->arch.mte_enabled = true;
101 		}
102 		mutex_unlock(&kvm->lock);
103 		break;
104 	default:
105 		r = -EINVAL;
106 		break;
107 	}
108 
109 	return r;
110 }
111 
112 static int kvm_arm_default_max_vcpus(void)
113 {
114 	return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
115 }
116 
117 static void set_default_spectre(struct kvm *kvm)
118 {
119 	/*
120 	 * The default is to expose CSV2 == 1 if the HW isn't affected.
121 	 * Although this is a per-CPU feature, we make it global because
122 	 * asymmetric systems are just a nuisance.
123 	 *
124 	 * Userspace can override this as long as it doesn't promise
125 	 * the impossible.
126 	 */
127 	if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
128 		kvm->arch.pfr0_csv2 = 1;
129 	if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
130 		kvm->arch.pfr0_csv3 = 1;
131 }
132 
133 /**
134  * kvm_arch_init_vm - initializes a VM data structure
135  * @kvm:	pointer to the KVM struct
136  */
137 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
138 {
139 	int ret;
140 
141 	ret = kvm_arm_setup_stage2(kvm, type);
142 	if (ret)
143 		return ret;
144 
145 	ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
146 	if (ret)
147 		return ret;
148 
149 	ret = kvm_share_hyp(kvm, kvm + 1);
150 	if (ret)
151 		goto out_free_stage2_pgd;
152 
153 	kvm_vgic_early_init(kvm);
154 
155 	/* The maximum number of VCPUs is limited by the host's GIC model */
156 	kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
157 
158 	set_default_spectre(kvm);
159 
160 	return ret;
161 out_free_stage2_pgd:
162 	kvm_free_stage2_pgd(&kvm->arch.mmu);
163 	return ret;
164 }
165 
166 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
167 {
168 	return VM_FAULT_SIGBUS;
169 }
170 
171 
172 /**
173  * kvm_arch_destroy_vm - destroy the VM data structure
174  * @kvm:	pointer to the KVM struct
175  */
176 void kvm_arch_destroy_vm(struct kvm *kvm)
177 {
178 	bitmap_free(kvm->arch.pmu_filter);
179 
180 	kvm_vgic_destroy(kvm);
181 
182 	kvm_destroy_vcpus(kvm);
183 
184 	kvm_unshare_hyp(kvm, kvm + 1);
185 }
186 
187 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
188 {
189 	int r;
190 	switch (ext) {
191 	case KVM_CAP_IRQCHIP:
192 		r = vgic_present;
193 		break;
194 	case KVM_CAP_IOEVENTFD:
195 	case KVM_CAP_DEVICE_CTRL:
196 	case KVM_CAP_USER_MEMORY:
197 	case KVM_CAP_SYNC_MMU:
198 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
199 	case KVM_CAP_ONE_REG:
200 	case KVM_CAP_ARM_PSCI:
201 	case KVM_CAP_ARM_PSCI_0_2:
202 	case KVM_CAP_READONLY_MEM:
203 	case KVM_CAP_MP_STATE:
204 	case KVM_CAP_IMMEDIATE_EXIT:
205 	case KVM_CAP_VCPU_EVENTS:
206 	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
207 	case KVM_CAP_ARM_NISV_TO_USER:
208 	case KVM_CAP_ARM_INJECT_EXT_DABT:
209 	case KVM_CAP_SET_GUEST_DEBUG:
210 	case KVM_CAP_VCPU_ATTRIBUTES:
211 	case KVM_CAP_PTP_KVM:
212 		r = 1;
213 		break;
214 	case KVM_CAP_SET_GUEST_DEBUG2:
215 		return KVM_GUESTDBG_VALID_MASK;
216 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
217 		r = 1;
218 		break;
219 	case KVM_CAP_NR_VCPUS:
220 		/*
221 		 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
222 		 * architectures, as it does not always bound it to
223 		 * KVM_CAP_MAX_VCPUS. It should not matter much because
224 		 * this is just an advisory value.
225 		 */
226 		r = min_t(unsigned int, num_online_cpus(),
227 			  kvm_arm_default_max_vcpus());
228 		break;
229 	case KVM_CAP_MAX_VCPUS:
230 	case KVM_CAP_MAX_VCPU_ID:
231 		if (kvm)
232 			r = kvm->arch.max_vcpus;
233 		else
234 			r = kvm_arm_default_max_vcpus();
235 		break;
236 	case KVM_CAP_MSI_DEVID:
237 		if (!kvm)
238 			r = -EINVAL;
239 		else
240 			r = kvm->arch.vgic.msis_require_devid;
241 		break;
242 	case KVM_CAP_ARM_USER_IRQ:
243 		/*
244 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
245 		 * (bump this number if adding more devices)
246 		 */
247 		r = 1;
248 		break;
249 	case KVM_CAP_ARM_MTE:
250 		r = system_supports_mte();
251 		break;
252 	case KVM_CAP_STEAL_TIME:
253 		r = kvm_arm_pvtime_supported();
254 		break;
255 	case KVM_CAP_ARM_EL1_32BIT:
256 		r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
257 		break;
258 	case KVM_CAP_GUEST_DEBUG_HW_BPS:
259 		r = get_num_brps();
260 		break;
261 	case KVM_CAP_GUEST_DEBUG_HW_WPS:
262 		r = get_num_wrps();
263 		break;
264 	case KVM_CAP_ARM_PMU_V3:
265 		r = kvm_arm_support_pmu_v3();
266 		break;
267 	case KVM_CAP_ARM_INJECT_SERROR_ESR:
268 		r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
269 		break;
270 	case KVM_CAP_ARM_VM_IPA_SIZE:
271 		r = get_kvm_ipa_limit();
272 		break;
273 	case KVM_CAP_ARM_SVE:
274 		r = system_supports_sve();
275 		break;
276 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
277 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
278 		r = system_has_full_ptr_auth();
279 		break;
280 	default:
281 		r = 0;
282 	}
283 
284 	return r;
285 }
286 
287 long kvm_arch_dev_ioctl(struct file *filp,
288 			unsigned int ioctl, unsigned long arg)
289 {
290 	return -EINVAL;
291 }
292 
293 struct kvm *kvm_arch_alloc_vm(void)
294 {
295 	size_t sz = sizeof(struct kvm);
296 
297 	if (!has_vhe())
298 		return kzalloc(sz, GFP_KERNEL_ACCOUNT);
299 
300 	return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
301 }
302 
303 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
304 {
305 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
306 		return -EBUSY;
307 
308 	if (id >= kvm->arch.max_vcpus)
309 		return -EINVAL;
310 
311 	return 0;
312 }
313 
314 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
315 {
316 	int err;
317 
318 	/* Force users to call KVM_ARM_VCPU_INIT */
319 	vcpu->arch.target = -1;
320 	bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
321 
322 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
323 
324 	/* Set up the timer */
325 	kvm_timer_vcpu_init(vcpu);
326 
327 	kvm_pmu_vcpu_init(vcpu);
328 
329 	kvm_arm_reset_debug_ptr(vcpu);
330 
331 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
332 
333 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
334 
335 	err = kvm_vgic_vcpu_init(vcpu);
336 	if (err)
337 		return err;
338 
339 	return kvm_share_hyp(vcpu, vcpu + 1);
340 }
341 
342 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
343 {
344 }
345 
346 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
347 {
348 	if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
349 		static_branch_dec(&userspace_irqchip_in_use);
350 
351 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
352 	kvm_timer_vcpu_terminate(vcpu);
353 	kvm_pmu_vcpu_destroy(vcpu);
354 
355 	kvm_arm_vcpu_destroy(vcpu);
356 }
357 
358 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
359 {
360 	return kvm_timer_is_pending(vcpu);
361 }
362 
363 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
364 {
365 
366 }
367 
368 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
369 {
370 
371 }
372 
373 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
374 {
375 	struct kvm_s2_mmu *mmu;
376 	int *last_ran;
377 
378 	mmu = vcpu->arch.hw_mmu;
379 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
380 
381 	/*
382 	 * We guarantee that both TLBs and I-cache are private to each
383 	 * vcpu. If detecting that a vcpu from the same VM has
384 	 * previously run on the same physical CPU, call into the
385 	 * hypervisor code to nuke the relevant contexts.
386 	 *
387 	 * We might get preempted before the vCPU actually runs, but
388 	 * over-invalidation doesn't affect correctness.
389 	 */
390 	if (*last_ran != vcpu->vcpu_id) {
391 		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
392 		*last_ran = vcpu->vcpu_id;
393 	}
394 
395 	vcpu->cpu = cpu;
396 
397 	kvm_vgic_load(vcpu);
398 	kvm_timer_vcpu_load(vcpu);
399 	if (has_vhe())
400 		kvm_vcpu_load_sysregs_vhe(vcpu);
401 	kvm_arch_vcpu_load_fp(vcpu);
402 	kvm_vcpu_pmu_restore_guest(vcpu);
403 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
404 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
405 
406 	if (single_task_running())
407 		vcpu_clear_wfx_traps(vcpu);
408 	else
409 		vcpu_set_wfx_traps(vcpu);
410 
411 	if (vcpu_has_ptrauth(vcpu))
412 		vcpu_ptrauth_disable(vcpu);
413 	kvm_arch_vcpu_load_debug_state_flags(vcpu);
414 }
415 
416 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
417 {
418 	kvm_arch_vcpu_put_debug_state_flags(vcpu);
419 	kvm_arch_vcpu_put_fp(vcpu);
420 	if (has_vhe())
421 		kvm_vcpu_put_sysregs_vhe(vcpu);
422 	kvm_timer_vcpu_put(vcpu);
423 	kvm_vgic_put(vcpu);
424 	kvm_vcpu_pmu_restore_host(vcpu);
425 
426 	vcpu->cpu = -1;
427 }
428 
429 static void vcpu_power_off(struct kvm_vcpu *vcpu)
430 {
431 	vcpu->arch.power_off = true;
432 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
433 	kvm_vcpu_kick(vcpu);
434 }
435 
436 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
437 				    struct kvm_mp_state *mp_state)
438 {
439 	if (vcpu->arch.power_off)
440 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
441 	else
442 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
443 
444 	return 0;
445 }
446 
447 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
448 				    struct kvm_mp_state *mp_state)
449 {
450 	int ret = 0;
451 
452 	switch (mp_state->mp_state) {
453 	case KVM_MP_STATE_RUNNABLE:
454 		vcpu->arch.power_off = false;
455 		break;
456 	case KVM_MP_STATE_STOPPED:
457 		vcpu_power_off(vcpu);
458 		break;
459 	default:
460 		ret = -EINVAL;
461 	}
462 
463 	return ret;
464 }
465 
466 /**
467  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
468  * @v:		The VCPU pointer
469  *
470  * If the guest CPU is not waiting for interrupts or an interrupt line is
471  * asserted, the CPU is by definition runnable.
472  */
473 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
474 {
475 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
476 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
477 		&& !v->arch.power_off && !v->arch.pause);
478 }
479 
480 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
481 {
482 	return vcpu_mode_priv(vcpu);
483 }
484 
485 #ifdef CONFIG_GUEST_PERF_EVENTS
486 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
487 {
488 	return *vcpu_pc(vcpu);
489 }
490 #endif
491 
492 /* Just ensure a guest exit from a particular CPU */
493 static void exit_vm_noop(void *info)
494 {
495 }
496 
497 void force_vm_exit(const cpumask_t *mask)
498 {
499 	preempt_disable();
500 	smp_call_function_many(mask, exit_vm_noop, NULL, true);
501 	preempt_enable();
502 }
503 
504 /**
505  * need_new_vmid_gen - check that the VMID is still valid
506  * @vmid: The VMID to check
507  *
508  * return true if there is a new generation of VMIDs being used
509  *
510  * The hardware supports a limited set of values with the value zero reserved
511  * for the host, so we check if an assigned value belongs to a previous
512  * generation, which requires us to assign a new value. If we're the first to
513  * use a VMID for the new generation, we must flush necessary caches and TLBs
514  * on all CPUs.
515  */
516 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
517 {
518 	u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
519 	smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
520 	return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
521 }
522 
523 /**
524  * update_vmid - Update the vmid with a valid VMID for the current generation
525  * @vmid: The stage-2 VMID information struct
526  */
527 static void update_vmid(struct kvm_vmid *vmid)
528 {
529 	if (!need_new_vmid_gen(vmid))
530 		return;
531 
532 	spin_lock(&kvm_vmid_lock);
533 
534 	/*
535 	 * We need to re-check the vmid_gen here to ensure that if another vcpu
536 	 * already allocated a valid vmid for this vm, then this vcpu should
537 	 * use the same vmid.
538 	 */
539 	if (!need_new_vmid_gen(vmid)) {
540 		spin_unlock(&kvm_vmid_lock);
541 		return;
542 	}
543 
544 	/* First user of a new VMID generation? */
545 	if (unlikely(kvm_next_vmid == 0)) {
546 		atomic64_inc(&kvm_vmid_gen);
547 		kvm_next_vmid = 1;
548 
549 		/*
550 		 * On SMP we know no other CPUs can use this CPU's or each
551 		 * other's VMID after force_vm_exit returns since the
552 		 * kvm_vmid_lock blocks them from reentry to the guest.
553 		 */
554 		force_vm_exit(cpu_all_mask);
555 		/*
556 		 * Now broadcast TLB + ICACHE invalidation over the inner
557 		 * shareable domain to make sure all data structures are
558 		 * clean.
559 		 */
560 		kvm_call_hyp(__kvm_flush_vm_context);
561 	}
562 
563 	WRITE_ONCE(vmid->vmid, kvm_next_vmid);
564 	kvm_next_vmid++;
565 	kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
566 
567 	smp_wmb();
568 	WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
569 
570 	spin_unlock(&kvm_vmid_lock);
571 }
572 
573 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
574 {
575 	return vcpu->arch.target >= 0;
576 }
577 
578 /*
579  * Handle both the initialisation that is being done when the vcpu is
580  * run for the first time, as well as the updates that must be
581  * performed each time we get a new thread dealing with this vcpu.
582  */
583 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
584 {
585 	struct kvm *kvm = vcpu->kvm;
586 	int ret;
587 
588 	if (!kvm_vcpu_initialized(vcpu))
589 		return -ENOEXEC;
590 
591 	if (!kvm_arm_vcpu_is_finalized(vcpu))
592 		return -EPERM;
593 
594 	ret = kvm_arch_vcpu_run_map_fp(vcpu);
595 	if (ret)
596 		return ret;
597 
598 	if (likely(vcpu_has_run_once(vcpu)))
599 		return 0;
600 
601 	kvm_arm_vcpu_init_debug(vcpu);
602 
603 	if (likely(irqchip_in_kernel(kvm))) {
604 		/*
605 		 * Map the VGIC hardware resources before running a vcpu the
606 		 * first time on this VM.
607 		 */
608 		ret = kvm_vgic_map_resources(kvm);
609 		if (ret)
610 			return ret;
611 	}
612 
613 	ret = kvm_timer_enable(vcpu);
614 	if (ret)
615 		return ret;
616 
617 	ret = kvm_arm_pmu_v3_enable(vcpu);
618 	if (ret)
619 		return ret;
620 
621 	if (!irqchip_in_kernel(kvm)) {
622 		/*
623 		 * Tell the rest of the code that there are userspace irqchip
624 		 * VMs in the wild.
625 		 */
626 		static_branch_inc(&userspace_irqchip_in_use);
627 	}
628 
629 	/*
630 	 * Initialize traps for protected VMs.
631 	 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
632 	 * the code is in place for first run initialization at EL2.
633 	 */
634 	if (kvm_vm_is_protected(kvm))
635 		kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
636 
637 	return ret;
638 }
639 
640 bool kvm_arch_intc_initialized(struct kvm *kvm)
641 {
642 	return vgic_initialized(kvm);
643 }
644 
645 void kvm_arm_halt_guest(struct kvm *kvm)
646 {
647 	unsigned long i;
648 	struct kvm_vcpu *vcpu;
649 
650 	kvm_for_each_vcpu(i, vcpu, kvm)
651 		vcpu->arch.pause = true;
652 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
653 }
654 
655 void kvm_arm_resume_guest(struct kvm *kvm)
656 {
657 	unsigned long i;
658 	struct kvm_vcpu *vcpu;
659 
660 	kvm_for_each_vcpu(i, vcpu, kvm) {
661 		vcpu->arch.pause = false;
662 		__kvm_vcpu_wake_up(vcpu);
663 	}
664 }
665 
666 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
667 {
668 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
669 
670 	rcuwait_wait_event(wait,
671 			   (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
672 			   TASK_INTERRUPTIBLE);
673 
674 	if (vcpu->arch.power_off || vcpu->arch.pause) {
675 		/* Awaken to handle a signal, request we sleep again later. */
676 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
677 	}
678 
679 	/*
680 	 * Make sure we will observe a potential reset request if we've
681 	 * observed a change to the power state. Pairs with the smp_wmb() in
682 	 * kvm_psci_vcpu_on().
683 	 */
684 	smp_rmb();
685 }
686 
687 /**
688  * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
689  * @vcpu:	The VCPU pointer
690  *
691  * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
692  * the vCPU is runnable.  The vCPU may or may not be scheduled out, depending
693  * on when a wake event arrives, e.g. there may already be a pending wake event.
694  */
695 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
696 {
697 	/*
698 	 * Sync back the state of the GIC CPU interface so that we have
699 	 * the latest PMR and group enables. This ensures that
700 	 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
701 	 * we have pending interrupts, e.g. when determining if the
702 	 * vCPU should block.
703 	 *
704 	 * For the same reason, we want to tell GICv4 that we need
705 	 * doorbells to be signalled, should an interrupt become pending.
706 	 */
707 	preempt_disable();
708 	kvm_vgic_vmcr_sync(vcpu);
709 	vgic_v4_put(vcpu, true);
710 	preempt_enable();
711 
712 	kvm_vcpu_halt(vcpu);
713 	kvm_clear_request(KVM_REQ_UNHALT, vcpu);
714 
715 	preempt_disable();
716 	vgic_v4_load(vcpu);
717 	preempt_enable();
718 }
719 
720 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
721 {
722 	if (kvm_request_pending(vcpu)) {
723 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
724 			vcpu_req_sleep(vcpu);
725 
726 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
727 			kvm_reset_vcpu(vcpu);
728 
729 		/*
730 		 * Clear IRQ_PENDING requests that were made to guarantee
731 		 * that a VCPU sees new virtual interrupts.
732 		 */
733 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
734 
735 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
736 			kvm_update_stolen_time(vcpu);
737 
738 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
739 			/* The distributor enable bits were changed */
740 			preempt_disable();
741 			vgic_v4_put(vcpu, false);
742 			vgic_v4_load(vcpu);
743 			preempt_enable();
744 		}
745 
746 		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
747 			kvm_pmu_handle_pmcr(vcpu,
748 					    __vcpu_sys_reg(vcpu, PMCR_EL0));
749 	}
750 }
751 
752 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
753 {
754 	if (likely(!vcpu_mode_is_32bit(vcpu)))
755 		return false;
756 
757 	return !system_supports_32bit_el0() ||
758 		static_branch_unlikely(&arm64_mismatched_32bit_el0);
759 }
760 
761 /**
762  * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
763  * @vcpu:	The VCPU pointer
764  * @ret:	Pointer to write optional return code
765  *
766  * Returns: true if the VCPU needs to return to a preemptible + interruptible
767  *	    and skip guest entry.
768  *
769  * This function disambiguates between two different types of exits: exits to a
770  * preemptible + interruptible kernel context and exits to userspace. For an
771  * exit to userspace, this function will write the return code to ret and return
772  * true. For an exit to preemptible + interruptible kernel context (i.e. check
773  * for pending work and re-enter), return true without writing to ret.
774  */
775 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
776 {
777 	struct kvm_run *run = vcpu->run;
778 
779 	/*
780 	 * If we're using a userspace irqchip, then check if we need
781 	 * to tell a userspace irqchip about timer or PMU level
782 	 * changes and if so, exit to userspace (the actual level
783 	 * state gets updated in kvm_timer_update_run and
784 	 * kvm_pmu_update_run below).
785 	 */
786 	if (static_branch_unlikely(&userspace_irqchip_in_use)) {
787 		if (kvm_timer_should_notify_user(vcpu) ||
788 		    kvm_pmu_should_notify_user(vcpu)) {
789 			*ret = -EINTR;
790 			run->exit_reason = KVM_EXIT_INTR;
791 			return true;
792 		}
793 	}
794 
795 	return kvm_request_pending(vcpu) ||
796 			need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
797 			xfer_to_guest_mode_work_pending();
798 }
799 
800 /*
801  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
802  * the vCPU is running.
803  *
804  * This must be noinstr as instrumentation may make use of RCU, and this is not
805  * safe during the EQS.
806  */
807 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
808 {
809 	int ret;
810 
811 	guest_state_enter_irqoff();
812 	ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
813 	guest_state_exit_irqoff();
814 
815 	return ret;
816 }
817 
818 /**
819  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
820  * @vcpu:	The VCPU pointer
821  *
822  * This function is called through the VCPU_RUN ioctl called from user space. It
823  * will execute VM code in a loop until the time slice for the process is used
824  * or some emulation is needed from user space in which case the function will
825  * return with return value 0 and with the kvm_run structure filled in with the
826  * required data for the requested emulation.
827  */
828 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
829 {
830 	struct kvm_run *run = vcpu->run;
831 	int ret;
832 
833 	if (run->exit_reason == KVM_EXIT_MMIO) {
834 		ret = kvm_handle_mmio_return(vcpu);
835 		if (ret)
836 			return ret;
837 	}
838 
839 	vcpu_load(vcpu);
840 
841 	if (run->immediate_exit) {
842 		ret = -EINTR;
843 		goto out;
844 	}
845 
846 	kvm_sigset_activate(vcpu);
847 
848 	ret = 1;
849 	run->exit_reason = KVM_EXIT_UNKNOWN;
850 	while (ret > 0) {
851 		/*
852 		 * Check conditions before entering the guest
853 		 */
854 		ret = xfer_to_guest_mode_handle_work(vcpu);
855 		if (!ret)
856 			ret = 1;
857 
858 		update_vmid(&vcpu->arch.hw_mmu->vmid);
859 
860 		check_vcpu_requests(vcpu);
861 
862 		/*
863 		 * Preparing the interrupts to be injected also
864 		 * involves poking the GIC, which must be done in a
865 		 * non-preemptible context.
866 		 */
867 		preempt_disable();
868 
869 		kvm_pmu_flush_hwstate(vcpu);
870 
871 		local_irq_disable();
872 
873 		kvm_vgic_flush_hwstate(vcpu);
874 
875 		/*
876 		 * Ensure we set mode to IN_GUEST_MODE after we disable
877 		 * interrupts and before the final VCPU requests check.
878 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
879 		 * Documentation/virt/kvm/vcpu-requests.rst
880 		 */
881 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
882 
883 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
884 			vcpu->mode = OUTSIDE_GUEST_MODE;
885 			isb(); /* Ensure work in x_flush_hwstate is committed */
886 			kvm_pmu_sync_hwstate(vcpu);
887 			if (static_branch_unlikely(&userspace_irqchip_in_use))
888 				kvm_timer_sync_user(vcpu);
889 			kvm_vgic_sync_hwstate(vcpu);
890 			local_irq_enable();
891 			preempt_enable();
892 			continue;
893 		}
894 
895 		kvm_arm_setup_debug(vcpu);
896 		kvm_arch_vcpu_ctxflush_fp(vcpu);
897 
898 		/**************************************************************
899 		 * Enter the guest
900 		 */
901 		trace_kvm_entry(*vcpu_pc(vcpu));
902 		guest_timing_enter_irqoff();
903 
904 		ret = kvm_arm_vcpu_enter_exit(vcpu);
905 
906 		vcpu->mode = OUTSIDE_GUEST_MODE;
907 		vcpu->stat.exits++;
908 		/*
909 		 * Back from guest
910 		 *************************************************************/
911 
912 		kvm_arm_clear_debug(vcpu);
913 
914 		/*
915 		 * We must sync the PMU state before the vgic state so
916 		 * that the vgic can properly sample the updated state of the
917 		 * interrupt line.
918 		 */
919 		kvm_pmu_sync_hwstate(vcpu);
920 
921 		/*
922 		 * Sync the vgic state before syncing the timer state because
923 		 * the timer code needs to know if the virtual timer
924 		 * interrupts are active.
925 		 */
926 		kvm_vgic_sync_hwstate(vcpu);
927 
928 		/*
929 		 * Sync the timer hardware state before enabling interrupts as
930 		 * we don't want vtimer interrupts to race with syncing the
931 		 * timer virtual interrupt state.
932 		 */
933 		if (static_branch_unlikely(&userspace_irqchip_in_use))
934 			kvm_timer_sync_user(vcpu);
935 
936 		kvm_arch_vcpu_ctxsync_fp(vcpu);
937 
938 		/*
939 		 * We must ensure that any pending interrupts are taken before
940 		 * we exit guest timing so that timer ticks are accounted as
941 		 * guest time. Transiently unmask interrupts so that any
942 		 * pending interrupts are taken.
943 		 *
944 		 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
945 		 * context synchronization event) is necessary to ensure that
946 		 * pending interrupts are taken.
947 		 */
948 		local_irq_enable();
949 		isb();
950 		local_irq_disable();
951 
952 		guest_timing_exit_irqoff();
953 
954 		local_irq_enable();
955 
956 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
957 
958 		/* Exit types that need handling before we can be preempted */
959 		handle_exit_early(vcpu, ret);
960 
961 		preempt_enable();
962 
963 		/*
964 		 * The ARMv8 architecture doesn't give the hypervisor
965 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
966 		 * if implemented by the CPU. If we spot the guest in such
967 		 * state and that we decided it wasn't supposed to do so (like
968 		 * with the asymmetric AArch32 case), return to userspace with
969 		 * a fatal error.
970 		 */
971 		if (vcpu_mode_is_bad_32bit(vcpu)) {
972 			/*
973 			 * As we have caught the guest red-handed, decide that
974 			 * it isn't fit for purpose anymore by making the vcpu
975 			 * invalid. The VMM can try and fix it by issuing  a
976 			 * KVM_ARM_VCPU_INIT if it really wants to.
977 			 */
978 			vcpu->arch.target = -1;
979 			ret = ARM_EXCEPTION_IL;
980 		}
981 
982 		ret = handle_exit(vcpu, ret);
983 	}
984 
985 	/* Tell userspace about in-kernel device output levels */
986 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
987 		kvm_timer_update_run(vcpu);
988 		kvm_pmu_update_run(vcpu);
989 	}
990 
991 	kvm_sigset_deactivate(vcpu);
992 
993 out:
994 	/*
995 	 * In the unlikely event that we are returning to userspace
996 	 * with pending exceptions or PC adjustment, commit these
997 	 * adjustments in order to give userspace a consistent view of
998 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
999 	 * being preempt-safe on VHE.
1000 	 */
1001 	if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
1002 					 KVM_ARM64_INCREMENT_PC)))
1003 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
1004 
1005 	vcpu_put(vcpu);
1006 	return ret;
1007 }
1008 
1009 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1010 {
1011 	int bit_index;
1012 	bool set;
1013 	unsigned long *hcr;
1014 
1015 	if (number == KVM_ARM_IRQ_CPU_IRQ)
1016 		bit_index = __ffs(HCR_VI);
1017 	else /* KVM_ARM_IRQ_CPU_FIQ */
1018 		bit_index = __ffs(HCR_VF);
1019 
1020 	hcr = vcpu_hcr(vcpu);
1021 	if (level)
1022 		set = test_and_set_bit(bit_index, hcr);
1023 	else
1024 		set = test_and_clear_bit(bit_index, hcr);
1025 
1026 	/*
1027 	 * If we didn't change anything, no need to wake up or kick other CPUs
1028 	 */
1029 	if (set == level)
1030 		return 0;
1031 
1032 	/*
1033 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1034 	 * trigger a world-switch round on the running physical CPU to set the
1035 	 * virtual IRQ/FIQ fields in the HCR appropriately.
1036 	 */
1037 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1038 	kvm_vcpu_kick(vcpu);
1039 
1040 	return 0;
1041 }
1042 
1043 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1044 			  bool line_status)
1045 {
1046 	u32 irq = irq_level->irq;
1047 	unsigned int irq_type, vcpu_idx, irq_num;
1048 	int nrcpus = atomic_read(&kvm->online_vcpus);
1049 	struct kvm_vcpu *vcpu = NULL;
1050 	bool level = irq_level->level;
1051 
1052 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1053 	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1054 	vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1055 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1056 
1057 	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1058 
1059 	switch (irq_type) {
1060 	case KVM_ARM_IRQ_TYPE_CPU:
1061 		if (irqchip_in_kernel(kvm))
1062 			return -ENXIO;
1063 
1064 		if (vcpu_idx >= nrcpus)
1065 			return -EINVAL;
1066 
1067 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1068 		if (!vcpu)
1069 			return -EINVAL;
1070 
1071 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1072 			return -EINVAL;
1073 
1074 		return vcpu_interrupt_line(vcpu, irq_num, level);
1075 	case KVM_ARM_IRQ_TYPE_PPI:
1076 		if (!irqchip_in_kernel(kvm))
1077 			return -ENXIO;
1078 
1079 		if (vcpu_idx >= nrcpus)
1080 			return -EINVAL;
1081 
1082 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1083 		if (!vcpu)
1084 			return -EINVAL;
1085 
1086 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1087 			return -EINVAL;
1088 
1089 		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1090 	case KVM_ARM_IRQ_TYPE_SPI:
1091 		if (!irqchip_in_kernel(kvm))
1092 			return -ENXIO;
1093 
1094 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1095 			return -EINVAL;
1096 
1097 		return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1098 	}
1099 
1100 	return -EINVAL;
1101 }
1102 
1103 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1104 			       const struct kvm_vcpu_init *init)
1105 {
1106 	unsigned int i, ret;
1107 	u32 phys_target = kvm_target_cpu();
1108 
1109 	if (init->target != phys_target)
1110 		return -EINVAL;
1111 
1112 	/*
1113 	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1114 	 * use the same target.
1115 	 */
1116 	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1117 		return -EINVAL;
1118 
1119 	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1120 	for (i = 0; i < sizeof(init->features) * 8; i++) {
1121 		bool set = (init->features[i / 32] & (1 << (i % 32)));
1122 
1123 		if (set && i >= KVM_VCPU_MAX_FEATURES)
1124 			return -ENOENT;
1125 
1126 		/*
1127 		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1128 		 * use the same feature set.
1129 		 */
1130 		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1131 		    test_bit(i, vcpu->arch.features) != set)
1132 			return -EINVAL;
1133 
1134 		if (set)
1135 			set_bit(i, vcpu->arch.features);
1136 	}
1137 
1138 	vcpu->arch.target = phys_target;
1139 
1140 	/* Now we know what it is, we can reset it. */
1141 	ret = kvm_reset_vcpu(vcpu);
1142 	if (ret) {
1143 		vcpu->arch.target = -1;
1144 		bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1145 	}
1146 
1147 	return ret;
1148 }
1149 
1150 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1151 					 struct kvm_vcpu_init *init)
1152 {
1153 	int ret;
1154 
1155 	ret = kvm_vcpu_set_target(vcpu, init);
1156 	if (ret)
1157 		return ret;
1158 
1159 	/*
1160 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1161 	 * guest MMU is turned off and flush the caches as needed.
1162 	 *
1163 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1164 	 * ensuring that the data side is always coherent. We still
1165 	 * need to invalidate the I-cache though, as FWB does *not*
1166 	 * imply CTR_EL0.DIC.
1167 	 */
1168 	if (vcpu_has_run_once(vcpu)) {
1169 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1170 			stage2_unmap_vm(vcpu->kvm);
1171 		else
1172 			icache_inval_all_pou();
1173 	}
1174 
1175 	vcpu_reset_hcr(vcpu);
1176 	vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1177 
1178 	/*
1179 	 * Handle the "start in power-off" case.
1180 	 */
1181 	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1182 		vcpu_power_off(vcpu);
1183 	else
1184 		vcpu->arch.power_off = false;
1185 
1186 	return 0;
1187 }
1188 
1189 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1190 				 struct kvm_device_attr *attr)
1191 {
1192 	int ret = -ENXIO;
1193 
1194 	switch (attr->group) {
1195 	default:
1196 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1197 		break;
1198 	}
1199 
1200 	return ret;
1201 }
1202 
1203 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1204 				 struct kvm_device_attr *attr)
1205 {
1206 	int ret = -ENXIO;
1207 
1208 	switch (attr->group) {
1209 	default:
1210 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1211 		break;
1212 	}
1213 
1214 	return ret;
1215 }
1216 
1217 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1218 				 struct kvm_device_attr *attr)
1219 {
1220 	int ret = -ENXIO;
1221 
1222 	switch (attr->group) {
1223 	default:
1224 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1225 		break;
1226 	}
1227 
1228 	return ret;
1229 }
1230 
1231 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1232 				   struct kvm_vcpu_events *events)
1233 {
1234 	memset(events, 0, sizeof(*events));
1235 
1236 	return __kvm_arm_vcpu_get_events(vcpu, events);
1237 }
1238 
1239 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1240 				   struct kvm_vcpu_events *events)
1241 {
1242 	int i;
1243 
1244 	/* check whether the reserved field is zero */
1245 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1246 		if (events->reserved[i])
1247 			return -EINVAL;
1248 
1249 	/* check whether the pad field is zero */
1250 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1251 		if (events->exception.pad[i])
1252 			return -EINVAL;
1253 
1254 	return __kvm_arm_vcpu_set_events(vcpu, events);
1255 }
1256 
1257 long kvm_arch_vcpu_ioctl(struct file *filp,
1258 			 unsigned int ioctl, unsigned long arg)
1259 {
1260 	struct kvm_vcpu *vcpu = filp->private_data;
1261 	void __user *argp = (void __user *)arg;
1262 	struct kvm_device_attr attr;
1263 	long r;
1264 
1265 	switch (ioctl) {
1266 	case KVM_ARM_VCPU_INIT: {
1267 		struct kvm_vcpu_init init;
1268 
1269 		r = -EFAULT;
1270 		if (copy_from_user(&init, argp, sizeof(init)))
1271 			break;
1272 
1273 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1274 		break;
1275 	}
1276 	case KVM_SET_ONE_REG:
1277 	case KVM_GET_ONE_REG: {
1278 		struct kvm_one_reg reg;
1279 
1280 		r = -ENOEXEC;
1281 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1282 			break;
1283 
1284 		r = -EFAULT;
1285 		if (copy_from_user(&reg, argp, sizeof(reg)))
1286 			break;
1287 
1288 		/*
1289 		 * We could owe a reset due to PSCI. Handle the pending reset
1290 		 * here to ensure userspace register accesses are ordered after
1291 		 * the reset.
1292 		 */
1293 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1294 			kvm_reset_vcpu(vcpu);
1295 
1296 		if (ioctl == KVM_SET_ONE_REG)
1297 			r = kvm_arm_set_reg(vcpu, &reg);
1298 		else
1299 			r = kvm_arm_get_reg(vcpu, &reg);
1300 		break;
1301 	}
1302 	case KVM_GET_REG_LIST: {
1303 		struct kvm_reg_list __user *user_list = argp;
1304 		struct kvm_reg_list reg_list;
1305 		unsigned n;
1306 
1307 		r = -ENOEXEC;
1308 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1309 			break;
1310 
1311 		r = -EPERM;
1312 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1313 			break;
1314 
1315 		r = -EFAULT;
1316 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1317 			break;
1318 		n = reg_list.n;
1319 		reg_list.n = kvm_arm_num_regs(vcpu);
1320 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1321 			break;
1322 		r = -E2BIG;
1323 		if (n < reg_list.n)
1324 			break;
1325 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1326 		break;
1327 	}
1328 	case KVM_SET_DEVICE_ATTR: {
1329 		r = -EFAULT;
1330 		if (copy_from_user(&attr, argp, sizeof(attr)))
1331 			break;
1332 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1333 		break;
1334 	}
1335 	case KVM_GET_DEVICE_ATTR: {
1336 		r = -EFAULT;
1337 		if (copy_from_user(&attr, argp, sizeof(attr)))
1338 			break;
1339 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1340 		break;
1341 	}
1342 	case KVM_HAS_DEVICE_ATTR: {
1343 		r = -EFAULT;
1344 		if (copy_from_user(&attr, argp, sizeof(attr)))
1345 			break;
1346 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1347 		break;
1348 	}
1349 	case KVM_GET_VCPU_EVENTS: {
1350 		struct kvm_vcpu_events events;
1351 
1352 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1353 			return -EINVAL;
1354 
1355 		if (copy_to_user(argp, &events, sizeof(events)))
1356 			return -EFAULT;
1357 
1358 		return 0;
1359 	}
1360 	case KVM_SET_VCPU_EVENTS: {
1361 		struct kvm_vcpu_events events;
1362 
1363 		if (copy_from_user(&events, argp, sizeof(events)))
1364 			return -EFAULT;
1365 
1366 		return kvm_arm_vcpu_set_events(vcpu, &events);
1367 	}
1368 	case KVM_ARM_VCPU_FINALIZE: {
1369 		int what;
1370 
1371 		if (!kvm_vcpu_initialized(vcpu))
1372 			return -ENOEXEC;
1373 
1374 		if (get_user(what, (const int __user *)argp))
1375 			return -EFAULT;
1376 
1377 		return kvm_arm_vcpu_finalize(vcpu, what);
1378 	}
1379 	default:
1380 		r = -EINVAL;
1381 	}
1382 
1383 	return r;
1384 }
1385 
1386 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1387 {
1388 
1389 }
1390 
1391 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1392 					const struct kvm_memory_slot *memslot)
1393 {
1394 	kvm_flush_remote_tlbs(kvm);
1395 }
1396 
1397 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1398 					struct kvm_arm_device_addr *dev_addr)
1399 {
1400 	unsigned long dev_id, type;
1401 
1402 	dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1403 		KVM_ARM_DEVICE_ID_SHIFT;
1404 	type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1405 		KVM_ARM_DEVICE_TYPE_SHIFT;
1406 
1407 	switch (dev_id) {
1408 	case KVM_ARM_DEVICE_VGIC_V2:
1409 		if (!vgic_present)
1410 			return -ENXIO;
1411 		return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1412 	default:
1413 		return -ENODEV;
1414 	}
1415 }
1416 
1417 long kvm_arch_vm_ioctl(struct file *filp,
1418 		       unsigned int ioctl, unsigned long arg)
1419 {
1420 	struct kvm *kvm = filp->private_data;
1421 	void __user *argp = (void __user *)arg;
1422 
1423 	switch (ioctl) {
1424 	case KVM_CREATE_IRQCHIP: {
1425 		int ret;
1426 		if (!vgic_present)
1427 			return -ENXIO;
1428 		mutex_lock(&kvm->lock);
1429 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1430 		mutex_unlock(&kvm->lock);
1431 		return ret;
1432 	}
1433 	case KVM_ARM_SET_DEVICE_ADDR: {
1434 		struct kvm_arm_device_addr dev_addr;
1435 
1436 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1437 			return -EFAULT;
1438 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1439 	}
1440 	case KVM_ARM_PREFERRED_TARGET: {
1441 		struct kvm_vcpu_init init;
1442 
1443 		kvm_vcpu_preferred_target(&init);
1444 
1445 		if (copy_to_user(argp, &init, sizeof(init)))
1446 			return -EFAULT;
1447 
1448 		return 0;
1449 	}
1450 	case KVM_ARM_MTE_COPY_TAGS: {
1451 		struct kvm_arm_copy_mte_tags copy_tags;
1452 
1453 		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1454 			return -EFAULT;
1455 		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1456 	}
1457 	default:
1458 		return -EINVAL;
1459 	}
1460 }
1461 
1462 static unsigned long nvhe_percpu_size(void)
1463 {
1464 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1465 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1466 }
1467 
1468 static unsigned long nvhe_percpu_order(void)
1469 {
1470 	unsigned long size = nvhe_percpu_size();
1471 
1472 	return size ? get_order(size) : 0;
1473 }
1474 
1475 /* A lookup table holding the hypervisor VA for each vector slot */
1476 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1477 
1478 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1479 {
1480 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1481 }
1482 
1483 static int kvm_init_vector_slots(void)
1484 {
1485 	int err;
1486 	void *base;
1487 
1488 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1489 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1490 
1491 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1492 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1493 
1494 	if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1495 		return 0;
1496 
1497 	if (!has_vhe()) {
1498 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1499 					       __BP_HARDEN_HYP_VECS_SZ, &base);
1500 		if (err)
1501 			return err;
1502 	}
1503 
1504 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1505 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1506 	return 0;
1507 }
1508 
1509 static void cpu_prepare_hyp_mode(int cpu)
1510 {
1511 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1512 	unsigned long tcr;
1513 
1514 	/*
1515 	 * Calculate the raw per-cpu offset without a translation from the
1516 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1517 	 * so that we can use adr_l to access per-cpu variables in EL2.
1518 	 * Also drop the KASAN tag which gets in the way...
1519 	 */
1520 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1521 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1522 
1523 	params->mair_el2 = read_sysreg(mair_el1);
1524 
1525 	/*
1526 	 * The ID map may be configured to use an extended virtual address
1527 	 * range. This is only the case if system RAM is out of range for the
1528 	 * currently configured page size and VA_BITS, in which case we will
1529 	 * also need the extended virtual range for the HYP ID map, or we won't
1530 	 * be able to enable the EL2 MMU.
1531 	 *
1532 	 * However, at EL2, there is only one TTBR register, and we can't switch
1533 	 * between translation tables *and* update TCR_EL2.T0SZ at the same
1534 	 * time. Bottom line: we need to use the extended range with *both* our
1535 	 * translation tables.
1536 	 *
1537 	 * So use the same T0SZ value we use for the ID map.
1538 	 */
1539 	tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1540 	tcr &= ~TCR_T0SZ_MASK;
1541 	tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1542 	params->tcr_el2 = tcr;
1543 
1544 	params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1545 	params->pgd_pa = kvm_mmu_get_httbr();
1546 	if (is_protected_kvm_enabled())
1547 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1548 	else
1549 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1550 	params->vttbr = params->vtcr = 0;
1551 
1552 	/*
1553 	 * Flush the init params from the data cache because the struct will
1554 	 * be read while the MMU is off.
1555 	 */
1556 	kvm_flush_dcache_to_poc(params, sizeof(*params));
1557 }
1558 
1559 static void hyp_install_host_vector(void)
1560 {
1561 	struct kvm_nvhe_init_params *params;
1562 	struct arm_smccc_res res;
1563 
1564 	/* Switch from the HYP stub to our own HYP init vector */
1565 	__hyp_set_vectors(kvm_get_idmap_vector());
1566 
1567 	/*
1568 	 * Call initialization code, and switch to the full blown HYP code.
1569 	 * If the cpucaps haven't been finalized yet, something has gone very
1570 	 * wrong, and hyp will crash and burn when it uses any
1571 	 * cpus_have_const_cap() wrapper.
1572 	 */
1573 	BUG_ON(!system_capabilities_finalized());
1574 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1575 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1576 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1577 }
1578 
1579 static void cpu_init_hyp_mode(void)
1580 {
1581 	hyp_install_host_vector();
1582 
1583 	/*
1584 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1585 	 * at EL2.
1586 	 */
1587 	if (this_cpu_has_cap(ARM64_SSBS) &&
1588 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1589 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1590 	}
1591 }
1592 
1593 static void cpu_hyp_reset(void)
1594 {
1595 	if (!is_kernel_in_hyp_mode())
1596 		__hyp_reset_vectors();
1597 }
1598 
1599 /*
1600  * EL2 vectors can be mapped and rerouted in a number of ways,
1601  * depending on the kernel configuration and CPU present:
1602  *
1603  * - If the CPU is affected by Spectre-v2, the hardening sequence is
1604  *   placed in one of the vector slots, which is executed before jumping
1605  *   to the real vectors.
1606  *
1607  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1608  *   containing the hardening sequence is mapped next to the idmap page,
1609  *   and executed before jumping to the real vectors.
1610  *
1611  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1612  *   empty slot is selected, mapped next to the idmap page, and
1613  *   executed before jumping to the real vectors.
1614  *
1615  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1616  * VHE, as we don't have hypervisor-specific mappings. If the system
1617  * is VHE and yet selects this capability, it will be ignored.
1618  */
1619 static void cpu_set_hyp_vector(void)
1620 {
1621 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1622 	void *vector = hyp_spectre_vector_selector[data->slot];
1623 
1624 	if (!is_protected_kvm_enabled())
1625 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1626 	else
1627 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1628 }
1629 
1630 static void cpu_hyp_init_context(void)
1631 {
1632 	kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1633 
1634 	if (!is_kernel_in_hyp_mode())
1635 		cpu_init_hyp_mode();
1636 }
1637 
1638 static void cpu_hyp_init_features(void)
1639 {
1640 	cpu_set_hyp_vector();
1641 	kvm_arm_init_debug();
1642 
1643 	if (is_kernel_in_hyp_mode())
1644 		kvm_timer_init_vhe();
1645 
1646 	if (vgic_present)
1647 		kvm_vgic_init_cpu_hardware();
1648 }
1649 
1650 static void cpu_hyp_reinit(void)
1651 {
1652 	cpu_hyp_reset();
1653 	cpu_hyp_init_context();
1654 	cpu_hyp_init_features();
1655 }
1656 
1657 static void _kvm_arch_hardware_enable(void *discard)
1658 {
1659 	if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1660 		cpu_hyp_reinit();
1661 		__this_cpu_write(kvm_arm_hardware_enabled, 1);
1662 	}
1663 }
1664 
1665 int kvm_arch_hardware_enable(void)
1666 {
1667 	_kvm_arch_hardware_enable(NULL);
1668 	return 0;
1669 }
1670 
1671 static void _kvm_arch_hardware_disable(void *discard)
1672 {
1673 	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1674 		cpu_hyp_reset();
1675 		__this_cpu_write(kvm_arm_hardware_enabled, 0);
1676 	}
1677 }
1678 
1679 void kvm_arch_hardware_disable(void)
1680 {
1681 	if (!is_protected_kvm_enabled())
1682 		_kvm_arch_hardware_disable(NULL);
1683 }
1684 
1685 #ifdef CONFIG_CPU_PM
1686 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1687 				    unsigned long cmd,
1688 				    void *v)
1689 {
1690 	/*
1691 	 * kvm_arm_hardware_enabled is left with its old value over
1692 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1693 	 * re-enable hyp.
1694 	 */
1695 	switch (cmd) {
1696 	case CPU_PM_ENTER:
1697 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1698 			/*
1699 			 * don't update kvm_arm_hardware_enabled here
1700 			 * so that the hardware will be re-enabled
1701 			 * when we resume. See below.
1702 			 */
1703 			cpu_hyp_reset();
1704 
1705 		return NOTIFY_OK;
1706 	case CPU_PM_ENTER_FAILED:
1707 	case CPU_PM_EXIT:
1708 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1709 			/* The hardware was enabled before suspend. */
1710 			cpu_hyp_reinit();
1711 
1712 		return NOTIFY_OK;
1713 
1714 	default:
1715 		return NOTIFY_DONE;
1716 	}
1717 }
1718 
1719 static struct notifier_block hyp_init_cpu_pm_nb = {
1720 	.notifier_call = hyp_init_cpu_pm_notifier,
1721 };
1722 
1723 static void hyp_cpu_pm_init(void)
1724 {
1725 	if (!is_protected_kvm_enabled())
1726 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1727 }
1728 static void hyp_cpu_pm_exit(void)
1729 {
1730 	if (!is_protected_kvm_enabled())
1731 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1732 }
1733 #else
1734 static inline void hyp_cpu_pm_init(void)
1735 {
1736 }
1737 static inline void hyp_cpu_pm_exit(void)
1738 {
1739 }
1740 #endif
1741 
1742 static void init_cpu_logical_map(void)
1743 {
1744 	unsigned int cpu;
1745 
1746 	/*
1747 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1748 	 * Only copy the set of online CPUs whose features have been chacked
1749 	 * against the finalized system capabilities. The hypervisor will not
1750 	 * allow any other CPUs from the `possible` set to boot.
1751 	 */
1752 	for_each_online_cpu(cpu)
1753 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1754 }
1755 
1756 #define init_psci_0_1_impl_state(config, what)	\
1757 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
1758 
1759 static bool init_psci_relay(void)
1760 {
1761 	/*
1762 	 * If PSCI has not been initialized, protected KVM cannot install
1763 	 * itself on newly booted CPUs.
1764 	 */
1765 	if (!psci_ops.get_version) {
1766 		kvm_err("Cannot initialize protected mode without PSCI\n");
1767 		return false;
1768 	}
1769 
1770 	kvm_host_psci_config.version = psci_ops.get_version();
1771 
1772 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1773 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1774 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1775 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1776 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1777 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1778 	}
1779 	return true;
1780 }
1781 
1782 static int init_subsystems(void)
1783 {
1784 	int err = 0;
1785 
1786 	/*
1787 	 * Enable hardware so that subsystem initialisation can access EL2.
1788 	 */
1789 	on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1790 
1791 	/*
1792 	 * Register CPU lower-power notifier
1793 	 */
1794 	hyp_cpu_pm_init();
1795 
1796 	/*
1797 	 * Init HYP view of VGIC
1798 	 */
1799 	err = kvm_vgic_hyp_init();
1800 	switch (err) {
1801 	case 0:
1802 		vgic_present = true;
1803 		break;
1804 	case -ENODEV:
1805 	case -ENXIO:
1806 		vgic_present = false;
1807 		err = 0;
1808 		break;
1809 	default:
1810 		goto out;
1811 	}
1812 
1813 	/*
1814 	 * Init HYP architected timer support
1815 	 */
1816 	err = kvm_timer_hyp_init(vgic_present);
1817 	if (err)
1818 		goto out;
1819 
1820 	kvm_register_perf_callbacks(NULL);
1821 
1822 	kvm_sys_reg_table_init();
1823 
1824 out:
1825 	if (err || !is_protected_kvm_enabled())
1826 		on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1827 
1828 	return err;
1829 }
1830 
1831 static void teardown_hyp_mode(void)
1832 {
1833 	int cpu;
1834 
1835 	free_hyp_pgds();
1836 	for_each_possible_cpu(cpu) {
1837 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1838 		free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1839 	}
1840 }
1841 
1842 static int do_pkvm_init(u32 hyp_va_bits)
1843 {
1844 	void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1845 	int ret;
1846 
1847 	preempt_disable();
1848 	cpu_hyp_init_context();
1849 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1850 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
1851 				hyp_va_bits);
1852 	cpu_hyp_init_features();
1853 
1854 	/*
1855 	 * The stub hypercalls are now disabled, so set our local flag to
1856 	 * prevent a later re-init attempt in kvm_arch_hardware_enable().
1857 	 */
1858 	__this_cpu_write(kvm_arm_hardware_enabled, 1);
1859 	preempt_enable();
1860 
1861 	return ret;
1862 }
1863 
1864 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1865 {
1866 	void *addr = phys_to_virt(hyp_mem_base);
1867 	int ret;
1868 
1869 	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1870 	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1871 	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
1872 	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
1873 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1874 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1875 	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
1876 
1877 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1878 	if (ret)
1879 		return ret;
1880 
1881 	ret = do_pkvm_init(hyp_va_bits);
1882 	if (ret)
1883 		return ret;
1884 
1885 	free_hyp_pgds();
1886 
1887 	return 0;
1888 }
1889 
1890 /**
1891  * Inits Hyp-mode on all online CPUs
1892  */
1893 static int init_hyp_mode(void)
1894 {
1895 	u32 hyp_va_bits;
1896 	int cpu;
1897 	int err = -ENOMEM;
1898 
1899 	/*
1900 	 * The protected Hyp-mode cannot be initialized if the memory pool
1901 	 * allocation has failed.
1902 	 */
1903 	if (is_protected_kvm_enabled() && !hyp_mem_base)
1904 		goto out_err;
1905 
1906 	/*
1907 	 * Allocate Hyp PGD and setup Hyp identity mapping
1908 	 */
1909 	err = kvm_mmu_init(&hyp_va_bits);
1910 	if (err)
1911 		goto out_err;
1912 
1913 	/*
1914 	 * Allocate stack pages for Hypervisor-mode
1915 	 */
1916 	for_each_possible_cpu(cpu) {
1917 		unsigned long stack_page;
1918 
1919 		stack_page = __get_free_page(GFP_KERNEL);
1920 		if (!stack_page) {
1921 			err = -ENOMEM;
1922 			goto out_err;
1923 		}
1924 
1925 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1926 	}
1927 
1928 	/*
1929 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1930 	 */
1931 	for_each_possible_cpu(cpu) {
1932 		struct page *page;
1933 		void *page_addr;
1934 
1935 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1936 		if (!page) {
1937 			err = -ENOMEM;
1938 			goto out_err;
1939 		}
1940 
1941 		page_addr = page_address(page);
1942 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1943 		kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1944 	}
1945 
1946 	/*
1947 	 * Map the Hyp-code called directly from the host
1948 	 */
1949 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1950 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1951 	if (err) {
1952 		kvm_err("Cannot map world-switch code\n");
1953 		goto out_err;
1954 	}
1955 
1956 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1957 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1958 	if (err) {
1959 		kvm_err("Cannot map .hyp.rodata section\n");
1960 		goto out_err;
1961 	}
1962 
1963 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1964 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1965 	if (err) {
1966 		kvm_err("Cannot map rodata section\n");
1967 		goto out_err;
1968 	}
1969 
1970 	/*
1971 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1972 	 * section thanks to an assertion in the linker script. Map it RW and
1973 	 * the rest of .bss RO.
1974 	 */
1975 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1976 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1977 	if (err) {
1978 		kvm_err("Cannot map hyp bss section: %d\n", err);
1979 		goto out_err;
1980 	}
1981 
1982 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1983 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1984 	if (err) {
1985 		kvm_err("Cannot map bss section\n");
1986 		goto out_err;
1987 	}
1988 
1989 	/*
1990 	 * Map the Hyp stack pages
1991 	 */
1992 	for_each_possible_cpu(cpu) {
1993 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1994 		err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1995 					  PAGE_HYP);
1996 
1997 		if (err) {
1998 			kvm_err("Cannot map hyp stack\n");
1999 			goto out_err;
2000 		}
2001 	}
2002 
2003 	for_each_possible_cpu(cpu) {
2004 		char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
2005 		char *percpu_end = percpu_begin + nvhe_percpu_size();
2006 
2007 		/* Map Hyp percpu pages */
2008 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2009 		if (err) {
2010 			kvm_err("Cannot map hyp percpu region\n");
2011 			goto out_err;
2012 		}
2013 
2014 		/* Prepare the CPU initialization parameters */
2015 		cpu_prepare_hyp_mode(cpu);
2016 	}
2017 
2018 	if (is_protected_kvm_enabled()) {
2019 		init_cpu_logical_map();
2020 
2021 		if (!init_psci_relay()) {
2022 			err = -ENODEV;
2023 			goto out_err;
2024 		}
2025 	}
2026 
2027 	if (is_protected_kvm_enabled()) {
2028 		err = kvm_hyp_init_protection(hyp_va_bits);
2029 		if (err) {
2030 			kvm_err("Failed to init hyp memory protection\n");
2031 			goto out_err;
2032 		}
2033 	}
2034 
2035 	return 0;
2036 
2037 out_err:
2038 	teardown_hyp_mode();
2039 	kvm_err("error initializing Hyp mode: %d\n", err);
2040 	return err;
2041 }
2042 
2043 static void _kvm_host_prot_finalize(void *arg)
2044 {
2045 	int *err = arg;
2046 
2047 	if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)))
2048 		WRITE_ONCE(*err, -EINVAL);
2049 }
2050 
2051 static int pkvm_drop_host_privileges(void)
2052 {
2053 	int ret = 0;
2054 
2055 	/*
2056 	 * Flip the static key upfront as that may no longer be possible
2057 	 * once the host stage 2 is installed.
2058 	 */
2059 	static_branch_enable(&kvm_protected_mode_initialized);
2060 	on_each_cpu(_kvm_host_prot_finalize, &ret, 1);
2061 	return ret;
2062 }
2063 
2064 static int finalize_hyp_mode(void)
2065 {
2066 	if (!is_protected_kvm_enabled())
2067 		return 0;
2068 
2069 	/*
2070 	 * Exclude HYP BSS from kmemleak so that it doesn't get peeked
2071 	 * at, which would end badly once the section is inaccessible.
2072 	 * None of other sections should ever be introspected.
2073 	 */
2074 	kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
2075 	return pkvm_drop_host_privileges();
2076 }
2077 
2078 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2079 {
2080 	struct kvm_vcpu *vcpu;
2081 	unsigned long i;
2082 
2083 	mpidr &= MPIDR_HWID_BITMASK;
2084 	kvm_for_each_vcpu(i, vcpu, kvm) {
2085 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2086 			return vcpu;
2087 	}
2088 	return NULL;
2089 }
2090 
2091 bool kvm_arch_has_irq_bypass(void)
2092 {
2093 	return true;
2094 }
2095 
2096 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2097 				      struct irq_bypass_producer *prod)
2098 {
2099 	struct kvm_kernel_irqfd *irqfd =
2100 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2101 
2102 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2103 					  &irqfd->irq_entry);
2104 }
2105 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2106 				      struct irq_bypass_producer *prod)
2107 {
2108 	struct kvm_kernel_irqfd *irqfd =
2109 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2110 
2111 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2112 				     &irqfd->irq_entry);
2113 }
2114 
2115 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2116 {
2117 	struct kvm_kernel_irqfd *irqfd =
2118 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2119 
2120 	kvm_arm_halt_guest(irqfd->kvm);
2121 }
2122 
2123 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2124 {
2125 	struct kvm_kernel_irqfd *irqfd =
2126 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2127 
2128 	kvm_arm_resume_guest(irqfd->kvm);
2129 }
2130 
2131 /**
2132  * Initialize Hyp-mode and memory mappings on all CPUs.
2133  */
2134 int kvm_arch_init(void *opaque)
2135 {
2136 	int err;
2137 	bool in_hyp_mode;
2138 
2139 	if (!is_hyp_mode_available()) {
2140 		kvm_info("HYP mode not available\n");
2141 		return -ENODEV;
2142 	}
2143 
2144 	if (kvm_get_mode() == KVM_MODE_NONE) {
2145 		kvm_info("KVM disabled from command line\n");
2146 		return -ENODEV;
2147 	}
2148 
2149 	in_hyp_mode = is_kernel_in_hyp_mode();
2150 
2151 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2152 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2153 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2154 			 "Only trusted guests should be used on this system.\n");
2155 
2156 	err = kvm_set_ipa_limit();
2157 	if (err)
2158 		return err;
2159 
2160 	err = kvm_arm_init_sve();
2161 	if (err)
2162 		return err;
2163 
2164 	if (!in_hyp_mode) {
2165 		err = init_hyp_mode();
2166 		if (err)
2167 			goto out_err;
2168 	}
2169 
2170 	err = kvm_init_vector_slots();
2171 	if (err) {
2172 		kvm_err("Cannot initialise vector slots\n");
2173 		goto out_err;
2174 	}
2175 
2176 	err = init_subsystems();
2177 	if (err)
2178 		goto out_hyp;
2179 
2180 	if (!in_hyp_mode) {
2181 		err = finalize_hyp_mode();
2182 		if (err) {
2183 			kvm_err("Failed to finalize Hyp protection\n");
2184 			goto out_hyp;
2185 		}
2186 	}
2187 
2188 	if (is_protected_kvm_enabled()) {
2189 		kvm_info("Protected nVHE mode initialized successfully\n");
2190 	} else if (in_hyp_mode) {
2191 		kvm_info("VHE mode initialized successfully\n");
2192 	} else {
2193 		kvm_info("Hyp mode initialized successfully\n");
2194 	}
2195 
2196 	return 0;
2197 
2198 out_hyp:
2199 	hyp_cpu_pm_exit();
2200 	if (!in_hyp_mode)
2201 		teardown_hyp_mode();
2202 out_err:
2203 	return err;
2204 }
2205 
2206 /* NOP: Compiling as a module not supported */
2207 void kvm_arch_exit(void)
2208 {
2209 	kvm_unregister_perf_callbacks();
2210 }
2211 
2212 static int __init early_kvm_mode_cfg(char *arg)
2213 {
2214 	if (!arg)
2215 		return -EINVAL;
2216 
2217 	if (strcmp(arg, "protected") == 0) {
2218 		kvm_mode = KVM_MODE_PROTECTED;
2219 		return 0;
2220 	}
2221 
2222 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2223 		kvm_mode = KVM_MODE_DEFAULT;
2224 		return 0;
2225 	}
2226 
2227 	if (strcmp(arg, "none") == 0) {
2228 		kvm_mode = KVM_MODE_NONE;
2229 		return 0;
2230 	}
2231 
2232 	return -EINVAL;
2233 }
2234 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2235 
2236 enum kvm_mode kvm_get_mode(void)
2237 {
2238 	return kvm_mode;
2239 }
2240 
2241 static int arm_init(void)
2242 {
2243 	int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2244 	return rc;
2245 }
2246 
2247 module_init(arm_init);
2248