xref: /openbmc/linux/arch/arm64/kvm/arm.c (revision 5b448065)
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/errno.h>
10 #include <linux/err.h>
11 #include <linux/kvm_host.h>
12 #include <linux/list.h>
13 #include <linux/module.h>
14 #include <linux/vmalloc.h>
15 #include <linux/fs.h>
16 #include <linux/mman.h>
17 #include <linux/sched.h>
18 #include <linux/kvm.h>
19 #include <linux/kvm_irqfd.h>
20 #include <linux/irqbypass.h>
21 #include <linux/sched/stat.h>
22 #include <linux/psci.h>
23 #include <trace/events/kvm.h>
24 
25 #define CREATE_TRACE_POINTS
26 #include "trace_arm.h"
27 
28 #include <linux/uaccess.h>
29 #include <asm/ptrace.h>
30 #include <asm/mman.h>
31 #include <asm/tlbflush.h>
32 #include <asm/cacheflush.h>
33 #include <asm/cpufeature.h>
34 #include <asm/virt.h>
35 #include <asm/kvm_arm.h>
36 #include <asm/kvm_asm.h>
37 #include <asm/kvm_mmu.h>
38 #include <asm/kvm_emulate.h>
39 #include <asm/sections.h>
40 
41 #include <kvm/arm_hypercalls.h>
42 #include <kvm/arm_pmu.h>
43 #include <kvm/arm_psci.h>
44 
45 #ifdef REQUIRES_VIRT
46 __asm__(".arch_extension	virt");
47 #endif
48 
49 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
50 DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized);
51 
52 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
53 
54 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
55 unsigned long kvm_arm_hyp_percpu_base[NR_CPUS];
56 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
57 
58 /* The VMID used in the VTTBR */
59 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
60 static u32 kvm_next_vmid;
61 static DEFINE_SPINLOCK(kvm_vmid_lock);
62 
63 static bool vgic_present;
64 
65 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
66 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
67 
68 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
69 {
70 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
71 }
72 
73 int kvm_arch_hardware_setup(void *opaque)
74 {
75 	return 0;
76 }
77 
78 int kvm_arch_check_processor_compat(void *opaque)
79 {
80 	return 0;
81 }
82 
83 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
84 			    struct kvm_enable_cap *cap)
85 {
86 	int r;
87 
88 	if (cap->flags)
89 		return -EINVAL;
90 
91 	switch (cap->cap) {
92 	case KVM_CAP_ARM_NISV_TO_USER:
93 		r = 0;
94 		kvm->arch.return_nisv_io_abort_to_user = true;
95 		break;
96 	default:
97 		r = -EINVAL;
98 		break;
99 	}
100 
101 	return r;
102 }
103 
104 static int kvm_arm_default_max_vcpus(void)
105 {
106 	return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
107 }
108 
109 static void set_default_spectre(struct kvm *kvm)
110 {
111 	/*
112 	 * The default is to expose CSV2 == 1 if the HW isn't affected.
113 	 * Although this is a per-CPU feature, we make it global because
114 	 * asymmetric systems are just a nuisance.
115 	 *
116 	 * Userspace can override this as long as it doesn't promise
117 	 * the impossible.
118 	 */
119 	if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
120 		kvm->arch.pfr0_csv2 = 1;
121 	if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
122 		kvm->arch.pfr0_csv3 = 1;
123 }
124 
125 /**
126  * kvm_arch_init_vm - initializes a VM data structure
127  * @kvm:	pointer to the KVM struct
128  */
129 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
130 {
131 	int ret;
132 
133 	ret = kvm_arm_setup_stage2(kvm, type);
134 	if (ret)
135 		return ret;
136 
137 	ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
138 	if (ret)
139 		return ret;
140 
141 	ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
142 	if (ret)
143 		goto out_free_stage2_pgd;
144 
145 	kvm_vgic_early_init(kvm);
146 
147 	/* The maximum number of VCPUs is limited by the host's GIC model */
148 	kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
149 
150 	set_default_spectre(kvm);
151 
152 	return ret;
153 out_free_stage2_pgd:
154 	kvm_free_stage2_pgd(&kvm->arch.mmu);
155 	return ret;
156 }
157 
158 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
159 {
160 	return VM_FAULT_SIGBUS;
161 }
162 
163 
164 /**
165  * kvm_arch_destroy_vm - destroy the VM data structure
166  * @kvm:	pointer to the KVM struct
167  */
168 void kvm_arch_destroy_vm(struct kvm *kvm)
169 {
170 	int i;
171 
172 	bitmap_free(kvm->arch.pmu_filter);
173 
174 	kvm_vgic_destroy(kvm);
175 
176 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
177 		if (kvm->vcpus[i]) {
178 			kvm_vcpu_destroy(kvm->vcpus[i]);
179 			kvm->vcpus[i] = NULL;
180 		}
181 	}
182 	atomic_set(&kvm->online_vcpus, 0);
183 }
184 
185 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
186 {
187 	int r;
188 	switch (ext) {
189 	case KVM_CAP_IRQCHIP:
190 		r = vgic_present;
191 		break;
192 	case KVM_CAP_IOEVENTFD:
193 	case KVM_CAP_DEVICE_CTRL:
194 	case KVM_CAP_USER_MEMORY:
195 	case KVM_CAP_SYNC_MMU:
196 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
197 	case KVM_CAP_ONE_REG:
198 	case KVM_CAP_ARM_PSCI:
199 	case KVM_CAP_ARM_PSCI_0_2:
200 	case KVM_CAP_READONLY_MEM:
201 	case KVM_CAP_MP_STATE:
202 	case KVM_CAP_IMMEDIATE_EXIT:
203 	case KVM_CAP_VCPU_EVENTS:
204 	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
205 	case KVM_CAP_ARM_NISV_TO_USER:
206 	case KVM_CAP_ARM_INJECT_EXT_DABT:
207 	case KVM_CAP_SET_GUEST_DEBUG:
208 	case KVM_CAP_VCPU_ATTRIBUTES:
209 	case KVM_CAP_PTP_KVM:
210 		r = 1;
211 		break;
212 	case KVM_CAP_SET_GUEST_DEBUG2:
213 		return KVM_GUESTDBG_VALID_MASK;
214 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
215 		r = 1;
216 		break;
217 	case KVM_CAP_NR_VCPUS:
218 		r = num_online_cpus();
219 		break;
220 	case KVM_CAP_MAX_VCPUS:
221 	case KVM_CAP_MAX_VCPU_ID:
222 		if (kvm)
223 			r = kvm->arch.max_vcpus;
224 		else
225 			r = kvm_arm_default_max_vcpus();
226 		break;
227 	case KVM_CAP_MSI_DEVID:
228 		if (!kvm)
229 			r = -EINVAL;
230 		else
231 			r = kvm->arch.vgic.msis_require_devid;
232 		break;
233 	case KVM_CAP_ARM_USER_IRQ:
234 		/*
235 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
236 		 * (bump this number if adding more devices)
237 		 */
238 		r = 1;
239 		break;
240 	case KVM_CAP_STEAL_TIME:
241 		r = kvm_arm_pvtime_supported();
242 		break;
243 	case KVM_CAP_ARM_EL1_32BIT:
244 		r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
245 		break;
246 	case KVM_CAP_GUEST_DEBUG_HW_BPS:
247 		r = get_num_brps();
248 		break;
249 	case KVM_CAP_GUEST_DEBUG_HW_WPS:
250 		r = get_num_wrps();
251 		break;
252 	case KVM_CAP_ARM_PMU_V3:
253 		r = kvm_arm_support_pmu_v3();
254 		break;
255 	case KVM_CAP_ARM_INJECT_SERROR_ESR:
256 		r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
257 		break;
258 	case KVM_CAP_ARM_VM_IPA_SIZE:
259 		r = get_kvm_ipa_limit();
260 		break;
261 	case KVM_CAP_ARM_SVE:
262 		r = system_supports_sve();
263 		break;
264 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
265 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
266 		r = system_has_full_ptr_auth();
267 		break;
268 	default:
269 		r = 0;
270 	}
271 
272 	return r;
273 }
274 
275 long kvm_arch_dev_ioctl(struct file *filp,
276 			unsigned int ioctl, unsigned long arg)
277 {
278 	return -EINVAL;
279 }
280 
281 struct kvm *kvm_arch_alloc_vm(void)
282 {
283 	if (!has_vhe())
284 		return kzalloc(sizeof(struct kvm), GFP_KERNEL);
285 
286 	return vzalloc(sizeof(struct kvm));
287 }
288 
289 void kvm_arch_free_vm(struct kvm *kvm)
290 {
291 	if (!has_vhe())
292 		kfree(kvm);
293 	else
294 		vfree(kvm);
295 }
296 
297 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
298 {
299 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
300 		return -EBUSY;
301 
302 	if (id >= kvm->arch.max_vcpus)
303 		return -EINVAL;
304 
305 	return 0;
306 }
307 
308 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
309 {
310 	int err;
311 
312 	/* Force users to call KVM_ARM_VCPU_INIT */
313 	vcpu->arch.target = -1;
314 	bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
315 
316 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
317 
318 	/* Set up the timer */
319 	kvm_timer_vcpu_init(vcpu);
320 
321 	kvm_pmu_vcpu_init(vcpu);
322 
323 	kvm_arm_reset_debug_ptr(vcpu);
324 
325 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
326 
327 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
328 
329 	err = kvm_vgic_vcpu_init(vcpu);
330 	if (err)
331 		return err;
332 
333 	return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
334 }
335 
336 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
337 {
338 }
339 
340 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
341 {
342 	if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
343 		static_branch_dec(&userspace_irqchip_in_use);
344 
345 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
346 	kvm_timer_vcpu_terminate(vcpu);
347 	kvm_pmu_vcpu_destroy(vcpu);
348 
349 	kvm_arm_vcpu_destroy(vcpu);
350 }
351 
352 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
353 {
354 	return kvm_timer_is_pending(vcpu);
355 }
356 
357 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
358 {
359 	/*
360 	 * If we're about to block (most likely because we've just hit a
361 	 * WFI), we need to sync back the state of the GIC CPU interface
362 	 * so that we have the latest PMR and group enables. This ensures
363 	 * that kvm_arch_vcpu_runnable has up-to-date data to decide
364 	 * whether we have pending interrupts.
365 	 *
366 	 * For the same reason, we want to tell GICv4 that we need
367 	 * doorbells to be signalled, should an interrupt become pending.
368 	 */
369 	preempt_disable();
370 	kvm_vgic_vmcr_sync(vcpu);
371 	vgic_v4_put(vcpu, true);
372 	preempt_enable();
373 }
374 
375 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
376 {
377 	preempt_disable();
378 	vgic_v4_load(vcpu);
379 	preempt_enable();
380 }
381 
382 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
383 {
384 	struct kvm_s2_mmu *mmu;
385 	int *last_ran;
386 
387 	mmu = vcpu->arch.hw_mmu;
388 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
389 
390 	/*
391 	 * We guarantee that both TLBs and I-cache are private to each
392 	 * vcpu. If detecting that a vcpu from the same VM has
393 	 * previously run on the same physical CPU, call into the
394 	 * hypervisor code to nuke the relevant contexts.
395 	 *
396 	 * We might get preempted before the vCPU actually runs, but
397 	 * over-invalidation doesn't affect correctness.
398 	 */
399 	if (*last_ran != vcpu->vcpu_id) {
400 		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
401 		*last_ran = vcpu->vcpu_id;
402 	}
403 
404 	vcpu->cpu = cpu;
405 
406 	kvm_vgic_load(vcpu);
407 	kvm_timer_vcpu_load(vcpu);
408 	if (has_vhe())
409 		kvm_vcpu_load_sysregs_vhe(vcpu);
410 	kvm_arch_vcpu_load_fp(vcpu);
411 	kvm_vcpu_pmu_restore_guest(vcpu);
412 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
413 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
414 
415 	if (single_task_running())
416 		vcpu_clear_wfx_traps(vcpu);
417 	else
418 		vcpu_set_wfx_traps(vcpu);
419 
420 	if (vcpu_has_ptrauth(vcpu))
421 		vcpu_ptrauth_disable(vcpu);
422 	kvm_arch_vcpu_load_debug_state_flags(vcpu);
423 }
424 
425 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
426 {
427 	kvm_arch_vcpu_put_debug_state_flags(vcpu);
428 	kvm_arch_vcpu_put_fp(vcpu);
429 	if (has_vhe())
430 		kvm_vcpu_put_sysregs_vhe(vcpu);
431 	kvm_timer_vcpu_put(vcpu);
432 	kvm_vgic_put(vcpu);
433 	kvm_vcpu_pmu_restore_host(vcpu);
434 
435 	vcpu->cpu = -1;
436 }
437 
438 static void vcpu_power_off(struct kvm_vcpu *vcpu)
439 {
440 	vcpu->arch.power_off = true;
441 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
442 	kvm_vcpu_kick(vcpu);
443 }
444 
445 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
446 				    struct kvm_mp_state *mp_state)
447 {
448 	if (vcpu->arch.power_off)
449 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
450 	else
451 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
452 
453 	return 0;
454 }
455 
456 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
457 				    struct kvm_mp_state *mp_state)
458 {
459 	int ret = 0;
460 
461 	switch (mp_state->mp_state) {
462 	case KVM_MP_STATE_RUNNABLE:
463 		vcpu->arch.power_off = false;
464 		break;
465 	case KVM_MP_STATE_STOPPED:
466 		vcpu_power_off(vcpu);
467 		break;
468 	default:
469 		ret = -EINVAL;
470 	}
471 
472 	return ret;
473 }
474 
475 /**
476  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
477  * @v:		The VCPU pointer
478  *
479  * If the guest CPU is not waiting for interrupts or an interrupt line is
480  * asserted, the CPU is by definition runnable.
481  */
482 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
483 {
484 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
485 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
486 		&& !v->arch.power_off && !v->arch.pause);
487 }
488 
489 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
490 {
491 	return vcpu_mode_priv(vcpu);
492 }
493 
494 /* Just ensure a guest exit from a particular CPU */
495 static void exit_vm_noop(void *info)
496 {
497 }
498 
499 void force_vm_exit(const cpumask_t *mask)
500 {
501 	preempt_disable();
502 	smp_call_function_many(mask, exit_vm_noop, NULL, true);
503 	preempt_enable();
504 }
505 
506 /**
507  * need_new_vmid_gen - check that the VMID is still valid
508  * @vmid: The VMID to check
509  *
510  * return true if there is a new generation of VMIDs being used
511  *
512  * The hardware supports a limited set of values with the value zero reserved
513  * for the host, so we check if an assigned value belongs to a previous
514  * generation, which requires us to assign a new value. If we're the first to
515  * use a VMID for the new generation, we must flush necessary caches and TLBs
516  * on all CPUs.
517  */
518 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
519 {
520 	u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
521 	smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
522 	return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
523 }
524 
525 /**
526  * update_vmid - Update the vmid with a valid VMID for the current generation
527  * @vmid: The stage-2 VMID information struct
528  */
529 static void update_vmid(struct kvm_vmid *vmid)
530 {
531 	if (!need_new_vmid_gen(vmid))
532 		return;
533 
534 	spin_lock(&kvm_vmid_lock);
535 
536 	/*
537 	 * We need to re-check the vmid_gen here to ensure that if another vcpu
538 	 * already allocated a valid vmid for this vm, then this vcpu should
539 	 * use the same vmid.
540 	 */
541 	if (!need_new_vmid_gen(vmid)) {
542 		spin_unlock(&kvm_vmid_lock);
543 		return;
544 	}
545 
546 	/* First user of a new VMID generation? */
547 	if (unlikely(kvm_next_vmid == 0)) {
548 		atomic64_inc(&kvm_vmid_gen);
549 		kvm_next_vmid = 1;
550 
551 		/*
552 		 * On SMP we know no other CPUs can use this CPU's or each
553 		 * other's VMID after force_vm_exit returns since the
554 		 * kvm_vmid_lock blocks them from reentry to the guest.
555 		 */
556 		force_vm_exit(cpu_all_mask);
557 		/*
558 		 * Now broadcast TLB + ICACHE invalidation over the inner
559 		 * shareable domain to make sure all data structures are
560 		 * clean.
561 		 */
562 		kvm_call_hyp(__kvm_flush_vm_context);
563 	}
564 
565 	vmid->vmid = kvm_next_vmid;
566 	kvm_next_vmid++;
567 	kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
568 
569 	smp_wmb();
570 	WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
571 
572 	spin_unlock(&kvm_vmid_lock);
573 }
574 
575 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
576 {
577 	struct kvm *kvm = vcpu->kvm;
578 	int ret = 0;
579 
580 	if (likely(vcpu->arch.has_run_once))
581 		return 0;
582 
583 	if (!kvm_arm_vcpu_is_finalized(vcpu))
584 		return -EPERM;
585 
586 	vcpu->arch.has_run_once = true;
587 
588 	kvm_arm_vcpu_init_debug(vcpu);
589 
590 	if (likely(irqchip_in_kernel(kvm))) {
591 		/*
592 		 * Map the VGIC hardware resources before running a vcpu the
593 		 * first time on this VM.
594 		 */
595 		ret = kvm_vgic_map_resources(kvm);
596 		if (ret)
597 			return ret;
598 	} else {
599 		/*
600 		 * Tell the rest of the code that there are userspace irqchip
601 		 * VMs in the wild.
602 		 */
603 		static_branch_inc(&userspace_irqchip_in_use);
604 	}
605 
606 	ret = kvm_timer_enable(vcpu);
607 	if (ret)
608 		return ret;
609 
610 	ret = kvm_arm_pmu_v3_enable(vcpu);
611 
612 	return ret;
613 }
614 
615 bool kvm_arch_intc_initialized(struct kvm *kvm)
616 {
617 	return vgic_initialized(kvm);
618 }
619 
620 void kvm_arm_halt_guest(struct kvm *kvm)
621 {
622 	int i;
623 	struct kvm_vcpu *vcpu;
624 
625 	kvm_for_each_vcpu(i, vcpu, kvm)
626 		vcpu->arch.pause = true;
627 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
628 }
629 
630 void kvm_arm_resume_guest(struct kvm *kvm)
631 {
632 	int i;
633 	struct kvm_vcpu *vcpu;
634 
635 	kvm_for_each_vcpu(i, vcpu, kvm) {
636 		vcpu->arch.pause = false;
637 		rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
638 	}
639 }
640 
641 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
642 {
643 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
644 
645 	rcuwait_wait_event(wait,
646 			   (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
647 			   TASK_INTERRUPTIBLE);
648 
649 	if (vcpu->arch.power_off || vcpu->arch.pause) {
650 		/* Awaken to handle a signal, request we sleep again later. */
651 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
652 	}
653 
654 	/*
655 	 * Make sure we will observe a potential reset request if we've
656 	 * observed a change to the power state. Pairs with the smp_wmb() in
657 	 * kvm_psci_vcpu_on().
658 	 */
659 	smp_rmb();
660 }
661 
662 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
663 {
664 	return vcpu->arch.target >= 0;
665 }
666 
667 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
668 {
669 	if (kvm_request_pending(vcpu)) {
670 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
671 			vcpu_req_sleep(vcpu);
672 
673 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
674 			kvm_reset_vcpu(vcpu);
675 
676 		/*
677 		 * Clear IRQ_PENDING requests that were made to guarantee
678 		 * that a VCPU sees new virtual interrupts.
679 		 */
680 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
681 
682 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
683 			kvm_update_stolen_time(vcpu);
684 
685 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
686 			/* The distributor enable bits were changed */
687 			preempt_disable();
688 			vgic_v4_put(vcpu, false);
689 			vgic_v4_load(vcpu);
690 			preempt_enable();
691 		}
692 	}
693 }
694 
695 /**
696  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
697  * @vcpu:	The VCPU pointer
698  *
699  * This function is called through the VCPU_RUN ioctl called from user space. It
700  * will execute VM code in a loop until the time slice for the process is used
701  * or some emulation is needed from user space in which case the function will
702  * return with return value 0 and with the kvm_run structure filled in with the
703  * required data for the requested emulation.
704  */
705 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
706 {
707 	struct kvm_run *run = vcpu->run;
708 	int ret;
709 
710 	if (unlikely(!kvm_vcpu_initialized(vcpu)))
711 		return -ENOEXEC;
712 
713 	ret = kvm_vcpu_first_run_init(vcpu);
714 	if (ret)
715 		return ret;
716 
717 	if (run->exit_reason == KVM_EXIT_MMIO) {
718 		ret = kvm_handle_mmio_return(vcpu);
719 		if (ret)
720 			return ret;
721 	}
722 
723 	vcpu_load(vcpu);
724 
725 	if (run->immediate_exit) {
726 		ret = -EINTR;
727 		goto out;
728 	}
729 
730 	kvm_sigset_activate(vcpu);
731 
732 	ret = 1;
733 	run->exit_reason = KVM_EXIT_UNKNOWN;
734 	while (ret > 0) {
735 		/*
736 		 * Check conditions before entering the guest
737 		 */
738 		cond_resched();
739 
740 		update_vmid(&vcpu->arch.hw_mmu->vmid);
741 
742 		check_vcpu_requests(vcpu);
743 
744 		/*
745 		 * Preparing the interrupts to be injected also
746 		 * involves poking the GIC, which must be done in a
747 		 * non-preemptible context.
748 		 */
749 		preempt_disable();
750 
751 		kvm_pmu_flush_hwstate(vcpu);
752 
753 		local_irq_disable();
754 
755 		kvm_vgic_flush_hwstate(vcpu);
756 
757 		/*
758 		 * Exit if we have a signal pending so that we can deliver the
759 		 * signal to user space.
760 		 */
761 		if (signal_pending(current)) {
762 			ret = -EINTR;
763 			run->exit_reason = KVM_EXIT_INTR;
764 		}
765 
766 		/*
767 		 * If we're using a userspace irqchip, then check if we need
768 		 * to tell a userspace irqchip about timer or PMU level
769 		 * changes and if so, exit to userspace (the actual level
770 		 * state gets updated in kvm_timer_update_run and
771 		 * kvm_pmu_update_run below).
772 		 */
773 		if (static_branch_unlikely(&userspace_irqchip_in_use)) {
774 			if (kvm_timer_should_notify_user(vcpu) ||
775 			    kvm_pmu_should_notify_user(vcpu)) {
776 				ret = -EINTR;
777 				run->exit_reason = KVM_EXIT_INTR;
778 			}
779 		}
780 
781 		/*
782 		 * Ensure we set mode to IN_GUEST_MODE after we disable
783 		 * interrupts and before the final VCPU requests check.
784 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
785 		 * Documentation/virt/kvm/vcpu-requests.rst
786 		 */
787 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
788 
789 		if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
790 		    kvm_request_pending(vcpu)) {
791 			vcpu->mode = OUTSIDE_GUEST_MODE;
792 			isb(); /* Ensure work in x_flush_hwstate is committed */
793 			kvm_pmu_sync_hwstate(vcpu);
794 			if (static_branch_unlikely(&userspace_irqchip_in_use))
795 				kvm_timer_sync_user(vcpu);
796 			kvm_vgic_sync_hwstate(vcpu);
797 			local_irq_enable();
798 			preempt_enable();
799 			continue;
800 		}
801 
802 		kvm_arm_setup_debug(vcpu);
803 
804 		/**************************************************************
805 		 * Enter the guest
806 		 */
807 		trace_kvm_entry(*vcpu_pc(vcpu));
808 		guest_enter_irqoff();
809 
810 		ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
811 
812 		vcpu->mode = OUTSIDE_GUEST_MODE;
813 		vcpu->stat.exits++;
814 		/*
815 		 * Back from guest
816 		 *************************************************************/
817 
818 		kvm_arm_clear_debug(vcpu);
819 
820 		/*
821 		 * We must sync the PMU state before the vgic state so
822 		 * that the vgic can properly sample the updated state of the
823 		 * interrupt line.
824 		 */
825 		kvm_pmu_sync_hwstate(vcpu);
826 
827 		/*
828 		 * Sync the vgic state before syncing the timer state because
829 		 * the timer code needs to know if the virtual timer
830 		 * interrupts are active.
831 		 */
832 		kvm_vgic_sync_hwstate(vcpu);
833 
834 		/*
835 		 * Sync the timer hardware state before enabling interrupts as
836 		 * we don't want vtimer interrupts to race with syncing the
837 		 * timer virtual interrupt state.
838 		 */
839 		if (static_branch_unlikely(&userspace_irqchip_in_use))
840 			kvm_timer_sync_user(vcpu);
841 
842 		kvm_arch_vcpu_ctxsync_fp(vcpu);
843 
844 		/*
845 		 * We may have taken a host interrupt in HYP mode (ie
846 		 * while executing the guest). This interrupt is still
847 		 * pending, as we haven't serviced it yet!
848 		 *
849 		 * We're now back in SVC mode, with interrupts
850 		 * disabled.  Enabling the interrupts now will have
851 		 * the effect of taking the interrupt again, in SVC
852 		 * mode this time.
853 		 */
854 		local_irq_enable();
855 
856 		/*
857 		 * We do local_irq_enable() before calling guest_exit() so
858 		 * that if a timer interrupt hits while running the guest we
859 		 * account that tick as being spent in the guest.  We enable
860 		 * preemption after calling guest_exit() so that if we get
861 		 * preempted we make sure ticks after that is not counted as
862 		 * guest time.
863 		 */
864 		guest_exit();
865 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
866 
867 		/* Exit types that need handling before we can be preempted */
868 		handle_exit_early(vcpu, ret);
869 
870 		preempt_enable();
871 
872 		/*
873 		 * The ARMv8 architecture doesn't give the hypervisor
874 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
875 		 * if implemented by the CPU. If we spot the guest in such
876 		 * state and that we decided it wasn't supposed to do so (like
877 		 * with the asymmetric AArch32 case), return to userspace with
878 		 * a fatal error.
879 		 */
880 		if (!system_supports_32bit_el0() && vcpu_mode_is_32bit(vcpu)) {
881 			/*
882 			 * As we have caught the guest red-handed, decide that
883 			 * it isn't fit for purpose anymore by making the vcpu
884 			 * invalid. The VMM can try and fix it by issuing  a
885 			 * KVM_ARM_VCPU_INIT if it really wants to.
886 			 */
887 			vcpu->arch.target = -1;
888 			ret = ARM_EXCEPTION_IL;
889 		}
890 
891 		ret = handle_exit(vcpu, ret);
892 	}
893 
894 	/* Tell userspace about in-kernel device output levels */
895 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
896 		kvm_timer_update_run(vcpu);
897 		kvm_pmu_update_run(vcpu);
898 	}
899 
900 	kvm_sigset_deactivate(vcpu);
901 
902 out:
903 	/*
904 	 * In the unlikely event that we are returning to userspace
905 	 * with pending exceptions or PC adjustment, commit these
906 	 * adjustments in order to give userspace a consistent view of
907 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
908 	 * being preempt-safe on VHE.
909 	 */
910 	if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
911 					 KVM_ARM64_INCREMENT_PC)))
912 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
913 
914 	vcpu_put(vcpu);
915 	return ret;
916 }
917 
918 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
919 {
920 	int bit_index;
921 	bool set;
922 	unsigned long *hcr;
923 
924 	if (number == KVM_ARM_IRQ_CPU_IRQ)
925 		bit_index = __ffs(HCR_VI);
926 	else /* KVM_ARM_IRQ_CPU_FIQ */
927 		bit_index = __ffs(HCR_VF);
928 
929 	hcr = vcpu_hcr(vcpu);
930 	if (level)
931 		set = test_and_set_bit(bit_index, hcr);
932 	else
933 		set = test_and_clear_bit(bit_index, hcr);
934 
935 	/*
936 	 * If we didn't change anything, no need to wake up or kick other CPUs
937 	 */
938 	if (set == level)
939 		return 0;
940 
941 	/*
942 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
943 	 * trigger a world-switch round on the running physical CPU to set the
944 	 * virtual IRQ/FIQ fields in the HCR appropriately.
945 	 */
946 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
947 	kvm_vcpu_kick(vcpu);
948 
949 	return 0;
950 }
951 
952 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
953 			  bool line_status)
954 {
955 	u32 irq = irq_level->irq;
956 	unsigned int irq_type, vcpu_idx, irq_num;
957 	int nrcpus = atomic_read(&kvm->online_vcpus);
958 	struct kvm_vcpu *vcpu = NULL;
959 	bool level = irq_level->level;
960 
961 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
962 	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
963 	vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
964 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
965 
966 	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
967 
968 	switch (irq_type) {
969 	case KVM_ARM_IRQ_TYPE_CPU:
970 		if (irqchip_in_kernel(kvm))
971 			return -ENXIO;
972 
973 		if (vcpu_idx >= nrcpus)
974 			return -EINVAL;
975 
976 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
977 		if (!vcpu)
978 			return -EINVAL;
979 
980 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
981 			return -EINVAL;
982 
983 		return vcpu_interrupt_line(vcpu, irq_num, level);
984 	case KVM_ARM_IRQ_TYPE_PPI:
985 		if (!irqchip_in_kernel(kvm))
986 			return -ENXIO;
987 
988 		if (vcpu_idx >= nrcpus)
989 			return -EINVAL;
990 
991 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
992 		if (!vcpu)
993 			return -EINVAL;
994 
995 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
996 			return -EINVAL;
997 
998 		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
999 	case KVM_ARM_IRQ_TYPE_SPI:
1000 		if (!irqchip_in_kernel(kvm))
1001 			return -ENXIO;
1002 
1003 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1004 			return -EINVAL;
1005 
1006 		return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1007 	}
1008 
1009 	return -EINVAL;
1010 }
1011 
1012 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1013 			       const struct kvm_vcpu_init *init)
1014 {
1015 	unsigned int i, ret;
1016 	int phys_target = kvm_target_cpu();
1017 
1018 	if (init->target != phys_target)
1019 		return -EINVAL;
1020 
1021 	/*
1022 	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1023 	 * use the same target.
1024 	 */
1025 	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1026 		return -EINVAL;
1027 
1028 	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1029 	for (i = 0; i < sizeof(init->features) * 8; i++) {
1030 		bool set = (init->features[i / 32] & (1 << (i % 32)));
1031 
1032 		if (set && i >= KVM_VCPU_MAX_FEATURES)
1033 			return -ENOENT;
1034 
1035 		/*
1036 		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1037 		 * use the same feature set.
1038 		 */
1039 		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1040 		    test_bit(i, vcpu->arch.features) != set)
1041 			return -EINVAL;
1042 
1043 		if (set)
1044 			set_bit(i, vcpu->arch.features);
1045 	}
1046 
1047 	vcpu->arch.target = phys_target;
1048 
1049 	/* Now we know what it is, we can reset it. */
1050 	ret = kvm_reset_vcpu(vcpu);
1051 	if (ret) {
1052 		vcpu->arch.target = -1;
1053 		bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1054 	}
1055 
1056 	return ret;
1057 }
1058 
1059 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1060 					 struct kvm_vcpu_init *init)
1061 {
1062 	int ret;
1063 
1064 	ret = kvm_vcpu_set_target(vcpu, init);
1065 	if (ret)
1066 		return ret;
1067 
1068 	/*
1069 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1070 	 * guest MMU is turned off and flush the caches as needed.
1071 	 *
1072 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1073 	 * ensuring that the data side is always coherent. We still
1074 	 * need to invalidate the I-cache though, as FWB does *not*
1075 	 * imply CTR_EL0.DIC.
1076 	 */
1077 	if (vcpu->arch.has_run_once) {
1078 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1079 			stage2_unmap_vm(vcpu->kvm);
1080 		else
1081 			__flush_icache_all();
1082 	}
1083 
1084 	vcpu_reset_hcr(vcpu);
1085 
1086 	/*
1087 	 * Handle the "start in power-off" case.
1088 	 */
1089 	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1090 		vcpu_power_off(vcpu);
1091 	else
1092 		vcpu->arch.power_off = false;
1093 
1094 	return 0;
1095 }
1096 
1097 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1098 				 struct kvm_device_attr *attr)
1099 {
1100 	int ret = -ENXIO;
1101 
1102 	switch (attr->group) {
1103 	default:
1104 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1105 		break;
1106 	}
1107 
1108 	return ret;
1109 }
1110 
1111 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1112 				 struct kvm_device_attr *attr)
1113 {
1114 	int ret = -ENXIO;
1115 
1116 	switch (attr->group) {
1117 	default:
1118 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1119 		break;
1120 	}
1121 
1122 	return ret;
1123 }
1124 
1125 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1126 				 struct kvm_device_attr *attr)
1127 {
1128 	int ret = -ENXIO;
1129 
1130 	switch (attr->group) {
1131 	default:
1132 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1133 		break;
1134 	}
1135 
1136 	return ret;
1137 }
1138 
1139 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1140 				   struct kvm_vcpu_events *events)
1141 {
1142 	memset(events, 0, sizeof(*events));
1143 
1144 	return __kvm_arm_vcpu_get_events(vcpu, events);
1145 }
1146 
1147 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1148 				   struct kvm_vcpu_events *events)
1149 {
1150 	int i;
1151 
1152 	/* check whether the reserved field is zero */
1153 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1154 		if (events->reserved[i])
1155 			return -EINVAL;
1156 
1157 	/* check whether the pad field is zero */
1158 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1159 		if (events->exception.pad[i])
1160 			return -EINVAL;
1161 
1162 	return __kvm_arm_vcpu_set_events(vcpu, events);
1163 }
1164 
1165 long kvm_arch_vcpu_ioctl(struct file *filp,
1166 			 unsigned int ioctl, unsigned long arg)
1167 {
1168 	struct kvm_vcpu *vcpu = filp->private_data;
1169 	void __user *argp = (void __user *)arg;
1170 	struct kvm_device_attr attr;
1171 	long r;
1172 
1173 	switch (ioctl) {
1174 	case KVM_ARM_VCPU_INIT: {
1175 		struct kvm_vcpu_init init;
1176 
1177 		r = -EFAULT;
1178 		if (copy_from_user(&init, argp, sizeof(init)))
1179 			break;
1180 
1181 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1182 		break;
1183 	}
1184 	case KVM_SET_ONE_REG:
1185 	case KVM_GET_ONE_REG: {
1186 		struct kvm_one_reg reg;
1187 
1188 		r = -ENOEXEC;
1189 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1190 			break;
1191 
1192 		r = -EFAULT;
1193 		if (copy_from_user(&reg, argp, sizeof(reg)))
1194 			break;
1195 
1196 		if (ioctl == KVM_SET_ONE_REG)
1197 			r = kvm_arm_set_reg(vcpu, &reg);
1198 		else
1199 			r = kvm_arm_get_reg(vcpu, &reg);
1200 		break;
1201 	}
1202 	case KVM_GET_REG_LIST: {
1203 		struct kvm_reg_list __user *user_list = argp;
1204 		struct kvm_reg_list reg_list;
1205 		unsigned n;
1206 
1207 		r = -ENOEXEC;
1208 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1209 			break;
1210 
1211 		r = -EPERM;
1212 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1213 			break;
1214 
1215 		r = -EFAULT;
1216 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1217 			break;
1218 		n = reg_list.n;
1219 		reg_list.n = kvm_arm_num_regs(vcpu);
1220 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1221 			break;
1222 		r = -E2BIG;
1223 		if (n < reg_list.n)
1224 			break;
1225 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1226 		break;
1227 	}
1228 	case KVM_SET_DEVICE_ATTR: {
1229 		r = -EFAULT;
1230 		if (copy_from_user(&attr, argp, sizeof(attr)))
1231 			break;
1232 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1233 		break;
1234 	}
1235 	case KVM_GET_DEVICE_ATTR: {
1236 		r = -EFAULT;
1237 		if (copy_from_user(&attr, argp, sizeof(attr)))
1238 			break;
1239 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1240 		break;
1241 	}
1242 	case KVM_HAS_DEVICE_ATTR: {
1243 		r = -EFAULT;
1244 		if (copy_from_user(&attr, argp, sizeof(attr)))
1245 			break;
1246 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1247 		break;
1248 	}
1249 	case KVM_GET_VCPU_EVENTS: {
1250 		struct kvm_vcpu_events events;
1251 
1252 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1253 			return -EINVAL;
1254 
1255 		if (copy_to_user(argp, &events, sizeof(events)))
1256 			return -EFAULT;
1257 
1258 		return 0;
1259 	}
1260 	case KVM_SET_VCPU_EVENTS: {
1261 		struct kvm_vcpu_events events;
1262 
1263 		if (copy_from_user(&events, argp, sizeof(events)))
1264 			return -EFAULT;
1265 
1266 		return kvm_arm_vcpu_set_events(vcpu, &events);
1267 	}
1268 	case KVM_ARM_VCPU_FINALIZE: {
1269 		int what;
1270 
1271 		if (!kvm_vcpu_initialized(vcpu))
1272 			return -ENOEXEC;
1273 
1274 		if (get_user(what, (const int __user *)argp))
1275 			return -EFAULT;
1276 
1277 		return kvm_arm_vcpu_finalize(vcpu, what);
1278 	}
1279 	default:
1280 		r = -EINVAL;
1281 	}
1282 
1283 	return r;
1284 }
1285 
1286 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1287 {
1288 
1289 }
1290 
1291 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1292 					const struct kvm_memory_slot *memslot)
1293 {
1294 	kvm_flush_remote_tlbs(kvm);
1295 }
1296 
1297 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1298 					struct kvm_arm_device_addr *dev_addr)
1299 {
1300 	unsigned long dev_id, type;
1301 
1302 	dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1303 		KVM_ARM_DEVICE_ID_SHIFT;
1304 	type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1305 		KVM_ARM_DEVICE_TYPE_SHIFT;
1306 
1307 	switch (dev_id) {
1308 	case KVM_ARM_DEVICE_VGIC_V2:
1309 		if (!vgic_present)
1310 			return -ENXIO;
1311 		return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1312 	default:
1313 		return -ENODEV;
1314 	}
1315 }
1316 
1317 long kvm_arch_vm_ioctl(struct file *filp,
1318 		       unsigned int ioctl, unsigned long arg)
1319 {
1320 	struct kvm *kvm = filp->private_data;
1321 	void __user *argp = (void __user *)arg;
1322 
1323 	switch (ioctl) {
1324 	case KVM_CREATE_IRQCHIP: {
1325 		int ret;
1326 		if (!vgic_present)
1327 			return -ENXIO;
1328 		mutex_lock(&kvm->lock);
1329 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1330 		mutex_unlock(&kvm->lock);
1331 		return ret;
1332 	}
1333 	case KVM_ARM_SET_DEVICE_ADDR: {
1334 		struct kvm_arm_device_addr dev_addr;
1335 
1336 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1337 			return -EFAULT;
1338 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1339 	}
1340 	case KVM_ARM_PREFERRED_TARGET: {
1341 		int err;
1342 		struct kvm_vcpu_init init;
1343 
1344 		err = kvm_vcpu_preferred_target(&init);
1345 		if (err)
1346 			return err;
1347 
1348 		if (copy_to_user(argp, &init, sizeof(init)))
1349 			return -EFAULT;
1350 
1351 		return 0;
1352 	}
1353 	default:
1354 		return -EINVAL;
1355 	}
1356 }
1357 
1358 static unsigned long nvhe_percpu_size(void)
1359 {
1360 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1361 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1362 }
1363 
1364 static unsigned long nvhe_percpu_order(void)
1365 {
1366 	unsigned long size = nvhe_percpu_size();
1367 
1368 	return size ? get_order(size) : 0;
1369 }
1370 
1371 /* A lookup table holding the hypervisor VA for each vector slot */
1372 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1373 
1374 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1375 {
1376 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1377 }
1378 
1379 static int kvm_init_vector_slots(void)
1380 {
1381 	int err;
1382 	void *base;
1383 
1384 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1385 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1386 
1387 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1388 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1389 
1390 	if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1391 		return 0;
1392 
1393 	if (!has_vhe()) {
1394 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1395 					       __BP_HARDEN_HYP_VECS_SZ, &base);
1396 		if (err)
1397 			return err;
1398 	}
1399 
1400 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1401 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1402 	return 0;
1403 }
1404 
1405 static void cpu_prepare_hyp_mode(int cpu)
1406 {
1407 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1408 	unsigned long tcr;
1409 
1410 	/*
1411 	 * Calculate the raw per-cpu offset without a translation from the
1412 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1413 	 * so that we can use adr_l to access per-cpu variables in EL2.
1414 	 * Also drop the KASAN tag which gets in the way...
1415 	 */
1416 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1417 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1418 
1419 	params->mair_el2 = read_sysreg(mair_el1);
1420 
1421 	/*
1422 	 * The ID map may be configured to use an extended virtual address
1423 	 * range. This is only the case if system RAM is out of range for the
1424 	 * currently configured page size and VA_BITS, in which case we will
1425 	 * also need the extended virtual range for the HYP ID map, or we won't
1426 	 * be able to enable the EL2 MMU.
1427 	 *
1428 	 * However, at EL2, there is only one TTBR register, and we can't switch
1429 	 * between translation tables *and* update TCR_EL2.T0SZ at the same
1430 	 * time. Bottom line: we need to use the extended range with *both* our
1431 	 * translation tables.
1432 	 *
1433 	 * So use the same T0SZ value we use for the ID map.
1434 	 */
1435 	tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1436 	tcr &= ~TCR_T0SZ_MASK;
1437 	tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1438 	params->tcr_el2 = tcr;
1439 
1440 	params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1441 	params->pgd_pa = kvm_mmu_get_httbr();
1442 	if (is_protected_kvm_enabled())
1443 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1444 	else
1445 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1446 	params->vttbr = params->vtcr = 0;
1447 
1448 	/*
1449 	 * Flush the init params from the data cache because the struct will
1450 	 * be read while the MMU is off.
1451 	 */
1452 	kvm_flush_dcache_to_poc(params, sizeof(*params));
1453 }
1454 
1455 static void hyp_install_host_vector(void)
1456 {
1457 	struct kvm_nvhe_init_params *params;
1458 	struct arm_smccc_res res;
1459 
1460 	/* Switch from the HYP stub to our own HYP init vector */
1461 	__hyp_set_vectors(kvm_get_idmap_vector());
1462 
1463 	/*
1464 	 * Call initialization code, and switch to the full blown HYP code.
1465 	 * If the cpucaps haven't been finalized yet, something has gone very
1466 	 * wrong, and hyp will crash and burn when it uses any
1467 	 * cpus_have_const_cap() wrapper.
1468 	 */
1469 	BUG_ON(!system_capabilities_finalized());
1470 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1471 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1472 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1473 }
1474 
1475 static void cpu_init_hyp_mode(void)
1476 {
1477 	hyp_install_host_vector();
1478 
1479 	/*
1480 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1481 	 * at EL2.
1482 	 */
1483 	if (this_cpu_has_cap(ARM64_SSBS) &&
1484 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1485 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1486 	}
1487 }
1488 
1489 static void cpu_hyp_reset(void)
1490 {
1491 	if (!is_kernel_in_hyp_mode())
1492 		__hyp_reset_vectors();
1493 }
1494 
1495 /*
1496  * EL2 vectors can be mapped and rerouted in a number of ways,
1497  * depending on the kernel configuration and CPU present:
1498  *
1499  * - If the CPU is affected by Spectre-v2, the hardening sequence is
1500  *   placed in one of the vector slots, which is executed before jumping
1501  *   to the real vectors.
1502  *
1503  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1504  *   containing the hardening sequence is mapped next to the idmap page,
1505  *   and executed before jumping to the real vectors.
1506  *
1507  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1508  *   empty slot is selected, mapped next to the idmap page, and
1509  *   executed before jumping to the real vectors.
1510  *
1511  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1512  * VHE, as we don't have hypervisor-specific mappings. If the system
1513  * is VHE and yet selects this capability, it will be ignored.
1514  */
1515 static void cpu_set_hyp_vector(void)
1516 {
1517 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1518 	void *vector = hyp_spectre_vector_selector[data->slot];
1519 
1520 	if (!is_protected_kvm_enabled())
1521 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1522 	else
1523 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1524 }
1525 
1526 static void cpu_hyp_reinit(void)
1527 {
1528 	kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1529 
1530 	cpu_hyp_reset();
1531 
1532 	if (is_kernel_in_hyp_mode())
1533 		kvm_timer_init_vhe();
1534 	else
1535 		cpu_init_hyp_mode();
1536 
1537 	cpu_set_hyp_vector();
1538 
1539 	kvm_arm_init_debug();
1540 
1541 	if (vgic_present)
1542 		kvm_vgic_init_cpu_hardware();
1543 }
1544 
1545 static void _kvm_arch_hardware_enable(void *discard)
1546 {
1547 	if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1548 		cpu_hyp_reinit();
1549 		__this_cpu_write(kvm_arm_hardware_enabled, 1);
1550 	}
1551 }
1552 
1553 int kvm_arch_hardware_enable(void)
1554 {
1555 	_kvm_arch_hardware_enable(NULL);
1556 	return 0;
1557 }
1558 
1559 static void _kvm_arch_hardware_disable(void *discard)
1560 {
1561 	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1562 		cpu_hyp_reset();
1563 		__this_cpu_write(kvm_arm_hardware_enabled, 0);
1564 	}
1565 }
1566 
1567 void kvm_arch_hardware_disable(void)
1568 {
1569 	if (!is_protected_kvm_enabled())
1570 		_kvm_arch_hardware_disable(NULL);
1571 }
1572 
1573 #ifdef CONFIG_CPU_PM
1574 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1575 				    unsigned long cmd,
1576 				    void *v)
1577 {
1578 	/*
1579 	 * kvm_arm_hardware_enabled is left with its old value over
1580 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1581 	 * re-enable hyp.
1582 	 */
1583 	switch (cmd) {
1584 	case CPU_PM_ENTER:
1585 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1586 			/*
1587 			 * don't update kvm_arm_hardware_enabled here
1588 			 * so that the hardware will be re-enabled
1589 			 * when we resume. See below.
1590 			 */
1591 			cpu_hyp_reset();
1592 
1593 		return NOTIFY_OK;
1594 	case CPU_PM_ENTER_FAILED:
1595 	case CPU_PM_EXIT:
1596 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1597 			/* The hardware was enabled before suspend. */
1598 			cpu_hyp_reinit();
1599 
1600 		return NOTIFY_OK;
1601 
1602 	default:
1603 		return NOTIFY_DONE;
1604 	}
1605 }
1606 
1607 static struct notifier_block hyp_init_cpu_pm_nb = {
1608 	.notifier_call = hyp_init_cpu_pm_notifier,
1609 };
1610 
1611 static void hyp_cpu_pm_init(void)
1612 {
1613 	if (!is_protected_kvm_enabled())
1614 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1615 }
1616 static void hyp_cpu_pm_exit(void)
1617 {
1618 	if (!is_protected_kvm_enabled())
1619 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1620 }
1621 #else
1622 static inline void hyp_cpu_pm_init(void)
1623 {
1624 }
1625 static inline void hyp_cpu_pm_exit(void)
1626 {
1627 }
1628 #endif
1629 
1630 static void init_cpu_logical_map(void)
1631 {
1632 	unsigned int cpu;
1633 
1634 	/*
1635 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1636 	 * Only copy the set of online CPUs whose features have been chacked
1637 	 * against the finalized system capabilities. The hypervisor will not
1638 	 * allow any other CPUs from the `possible` set to boot.
1639 	 */
1640 	for_each_online_cpu(cpu)
1641 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1642 }
1643 
1644 #define init_psci_0_1_impl_state(config, what)	\
1645 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
1646 
1647 static bool init_psci_relay(void)
1648 {
1649 	/*
1650 	 * If PSCI has not been initialized, protected KVM cannot install
1651 	 * itself on newly booted CPUs.
1652 	 */
1653 	if (!psci_ops.get_version) {
1654 		kvm_err("Cannot initialize protected mode without PSCI\n");
1655 		return false;
1656 	}
1657 
1658 	kvm_host_psci_config.version = psci_ops.get_version();
1659 
1660 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1661 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1662 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1663 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1664 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1665 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1666 	}
1667 	return true;
1668 }
1669 
1670 static int init_common_resources(void)
1671 {
1672 	return kvm_set_ipa_limit();
1673 }
1674 
1675 static int init_subsystems(void)
1676 {
1677 	int err = 0;
1678 
1679 	/*
1680 	 * Enable hardware so that subsystem initialisation can access EL2.
1681 	 */
1682 	on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1683 
1684 	/*
1685 	 * Register CPU lower-power notifier
1686 	 */
1687 	hyp_cpu_pm_init();
1688 
1689 	/*
1690 	 * Init HYP view of VGIC
1691 	 */
1692 	err = kvm_vgic_hyp_init();
1693 	switch (err) {
1694 	case 0:
1695 		vgic_present = true;
1696 		break;
1697 	case -ENODEV:
1698 	case -ENXIO:
1699 		vgic_present = false;
1700 		err = 0;
1701 		break;
1702 	default:
1703 		goto out;
1704 	}
1705 
1706 	/*
1707 	 * Init HYP architected timer support
1708 	 */
1709 	err = kvm_timer_hyp_init(vgic_present);
1710 	if (err)
1711 		goto out;
1712 
1713 	kvm_perf_init();
1714 	kvm_sys_reg_table_init();
1715 
1716 out:
1717 	if (err || !is_protected_kvm_enabled())
1718 		on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1719 
1720 	return err;
1721 }
1722 
1723 static void teardown_hyp_mode(void)
1724 {
1725 	int cpu;
1726 
1727 	free_hyp_pgds();
1728 	for_each_possible_cpu(cpu) {
1729 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1730 		free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1731 	}
1732 }
1733 
1734 static int do_pkvm_init(u32 hyp_va_bits)
1735 {
1736 	void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1737 	int ret;
1738 
1739 	preempt_disable();
1740 	hyp_install_host_vector();
1741 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1742 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
1743 				hyp_va_bits);
1744 	preempt_enable();
1745 
1746 	return ret;
1747 }
1748 
1749 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1750 {
1751 	void *addr = phys_to_virt(hyp_mem_base);
1752 	int ret;
1753 
1754 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1755 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1756 
1757 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1758 	if (ret)
1759 		return ret;
1760 
1761 	ret = do_pkvm_init(hyp_va_bits);
1762 	if (ret)
1763 		return ret;
1764 
1765 	free_hyp_pgds();
1766 
1767 	return 0;
1768 }
1769 
1770 /**
1771  * Inits Hyp-mode on all online CPUs
1772  */
1773 static int init_hyp_mode(void)
1774 {
1775 	u32 hyp_va_bits;
1776 	int cpu;
1777 	int err = -ENOMEM;
1778 
1779 	/*
1780 	 * The protected Hyp-mode cannot be initialized if the memory pool
1781 	 * allocation has failed.
1782 	 */
1783 	if (is_protected_kvm_enabled() && !hyp_mem_base)
1784 		goto out_err;
1785 
1786 	/*
1787 	 * Allocate Hyp PGD and setup Hyp identity mapping
1788 	 */
1789 	err = kvm_mmu_init(&hyp_va_bits);
1790 	if (err)
1791 		goto out_err;
1792 
1793 	/*
1794 	 * Allocate stack pages for Hypervisor-mode
1795 	 */
1796 	for_each_possible_cpu(cpu) {
1797 		unsigned long stack_page;
1798 
1799 		stack_page = __get_free_page(GFP_KERNEL);
1800 		if (!stack_page) {
1801 			err = -ENOMEM;
1802 			goto out_err;
1803 		}
1804 
1805 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1806 	}
1807 
1808 	/*
1809 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1810 	 */
1811 	for_each_possible_cpu(cpu) {
1812 		struct page *page;
1813 		void *page_addr;
1814 
1815 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1816 		if (!page) {
1817 			err = -ENOMEM;
1818 			goto out_err;
1819 		}
1820 
1821 		page_addr = page_address(page);
1822 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1823 		kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1824 	}
1825 
1826 	/*
1827 	 * Map the Hyp-code called directly from the host
1828 	 */
1829 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1830 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1831 	if (err) {
1832 		kvm_err("Cannot map world-switch code\n");
1833 		goto out_err;
1834 	}
1835 
1836 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1837 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1838 	if (err) {
1839 		kvm_err("Cannot map .hyp.rodata section\n");
1840 		goto out_err;
1841 	}
1842 
1843 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1844 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1845 	if (err) {
1846 		kvm_err("Cannot map rodata section\n");
1847 		goto out_err;
1848 	}
1849 
1850 	/*
1851 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1852 	 * section thanks to an assertion in the linker script. Map it RW and
1853 	 * the rest of .bss RO.
1854 	 */
1855 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1856 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1857 	if (err) {
1858 		kvm_err("Cannot map hyp bss section: %d\n", err);
1859 		goto out_err;
1860 	}
1861 
1862 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1863 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1864 	if (err) {
1865 		kvm_err("Cannot map bss section\n");
1866 		goto out_err;
1867 	}
1868 
1869 	/*
1870 	 * Map the Hyp stack pages
1871 	 */
1872 	for_each_possible_cpu(cpu) {
1873 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1874 		err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1875 					  PAGE_HYP);
1876 
1877 		if (err) {
1878 			kvm_err("Cannot map hyp stack\n");
1879 			goto out_err;
1880 		}
1881 	}
1882 
1883 	for_each_possible_cpu(cpu) {
1884 		char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1885 		char *percpu_end = percpu_begin + nvhe_percpu_size();
1886 
1887 		/* Map Hyp percpu pages */
1888 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1889 		if (err) {
1890 			kvm_err("Cannot map hyp percpu region\n");
1891 			goto out_err;
1892 		}
1893 
1894 		/* Prepare the CPU initialization parameters */
1895 		cpu_prepare_hyp_mode(cpu);
1896 	}
1897 
1898 	if (is_protected_kvm_enabled()) {
1899 		init_cpu_logical_map();
1900 
1901 		if (!init_psci_relay()) {
1902 			err = -ENODEV;
1903 			goto out_err;
1904 		}
1905 	}
1906 
1907 	if (is_protected_kvm_enabled()) {
1908 		err = kvm_hyp_init_protection(hyp_va_bits);
1909 		if (err) {
1910 			kvm_err("Failed to init hyp memory protection\n");
1911 			goto out_err;
1912 		}
1913 	}
1914 
1915 	return 0;
1916 
1917 out_err:
1918 	teardown_hyp_mode();
1919 	kvm_err("error initializing Hyp mode: %d\n", err);
1920 	return err;
1921 }
1922 
1923 static void _kvm_host_prot_finalize(void *discard)
1924 {
1925 	WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize));
1926 }
1927 
1928 static inline int pkvm_mark_hyp(phys_addr_t start, phys_addr_t end)
1929 {
1930 	return kvm_call_hyp_nvhe(__pkvm_mark_hyp, start, end);
1931 }
1932 
1933 #define pkvm_mark_hyp_section(__section)		\
1934 	pkvm_mark_hyp(__pa_symbol(__section##_start),	\
1935 			__pa_symbol(__section##_end))
1936 
1937 static int finalize_hyp_mode(void)
1938 {
1939 	int cpu, ret;
1940 
1941 	if (!is_protected_kvm_enabled())
1942 		return 0;
1943 
1944 	ret = pkvm_mark_hyp_section(__hyp_idmap_text);
1945 	if (ret)
1946 		return ret;
1947 
1948 	ret = pkvm_mark_hyp_section(__hyp_text);
1949 	if (ret)
1950 		return ret;
1951 
1952 	ret = pkvm_mark_hyp_section(__hyp_rodata);
1953 	if (ret)
1954 		return ret;
1955 
1956 	ret = pkvm_mark_hyp_section(__hyp_bss);
1957 	if (ret)
1958 		return ret;
1959 
1960 	ret = pkvm_mark_hyp(hyp_mem_base, hyp_mem_base + hyp_mem_size);
1961 	if (ret)
1962 		return ret;
1963 
1964 	for_each_possible_cpu(cpu) {
1965 		phys_addr_t start = virt_to_phys((void *)kvm_arm_hyp_percpu_base[cpu]);
1966 		phys_addr_t end = start + (PAGE_SIZE << nvhe_percpu_order());
1967 
1968 		ret = pkvm_mark_hyp(start, end);
1969 		if (ret)
1970 			return ret;
1971 
1972 		start = virt_to_phys((void *)per_cpu(kvm_arm_hyp_stack_page, cpu));
1973 		end = start + PAGE_SIZE;
1974 		ret = pkvm_mark_hyp(start, end);
1975 		if (ret)
1976 			return ret;
1977 	}
1978 
1979 	/*
1980 	 * Flip the static key upfront as that may no longer be possible
1981 	 * once the host stage 2 is installed.
1982 	 */
1983 	static_branch_enable(&kvm_protected_mode_initialized);
1984 	on_each_cpu(_kvm_host_prot_finalize, NULL, 1);
1985 
1986 	return 0;
1987 }
1988 
1989 static void check_kvm_target_cpu(void *ret)
1990 {
1991 	*(int *)ret = kvm_target_cpu();
1992 }
1993 
1994 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1995 {
1996 	struct kvm_vcpu *vcpu;
1997 	int i;
1998 
1999 	mpidr &= MPIDR_HWID_BITMASK;
2000 	kvm_for_each_vcpu(i, vcpu, kvm) {
2001 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2002 			return vcpu;
2003 	}
2004 	return NULL;
2005 }
2006 
2007 bool kvm_arch_has_irq_bypass(void)
2008 {
2009 	return true;
2010 }
2011 
2012 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2013 				      struct irq_bypass_producer *prod)
2014 {
2015 	struct kvm_kernel_irqfd *irqfd =
2016 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2017 
2018 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2019 					  &irqfd->irq_entry);
2020 }
2021 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2022 				      struct irq_bypass_producer *prod)
2023 {
2024 	struct kvm_kernel_irqfd *irqfd =
2025 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2026 
2027 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2028 				     &irqfd->irq_entry);
2029 }
2030 
2031 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2032 {
2033 	struct kvm_kernel_irqfd *irqfd =
2034 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2035 
2036 	kvm_arm_halt_guest(irqfd->kvm);
2037 }
2038 
2039 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2040 {
2041 	struct kvm_kernel_irqfd *irqfd =
2042 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2043 
2044 	kvm_arm_resume_guest(irqfd->kvm);
2045 }
2046 
2047 /**
2048  * Initialize Hyp-mode and memory mappings on all CPUs.
2049  */
2050 int kvm_arch_init(void *opaque)
2051 {
2052 	int err;
2053 	int ret, cpu;
2054 	bool in_hyp_mode;
2055 
2056 	if (!is_hyp_mode_available()) {
2057 		kvm_info("HYP mode not available\n");
2058 		return -ENODEV;
2059 	}
2060 
2061 	in_hyp_mode = is_kernel_in_hyp_mode();
2062 
2063 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2064 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2065 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2066 			 "Only trusted guests should be used on this system.\n");
2067 
2068 	for_each_online_cpu(cpu) {
2069 		smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
2070 		if (ret < 0) {
2071 			kvm_err("Error, CPU %d not supported!\n", cpu);
2072 			return -ENODEV;
2073 		}
2074 	}
2075 
2076 	err = init_common_resources();
2077 	if (err)
2078 		return err;
2079 
2080 	err = kvm_arm_init_sve();
2081 	if (err)
2082 		return err;
2083 
2084 	if (!in_hyp_mode) {
2085 		err = init_hyp_mode();
2086 		if (err)
2087 			goto out_err;
2088 	}
2089 
2090 	err = kvm_init_vector_slots();
2091 	if (err) {
2092 		kvm_err("Cannot initialise vector slots\n");
2093 		goto out_err;
2094 	}
2095 
2096 	err = init_subsystems();
2097 	if (err)
2098 		goto out_hyp;
2099 
2100 	if (!in_hyp_mode) {
2101 		err = finalize_hyp_mode();
2102 		if (err) {
2103 			kvm_err("Failed to finalize Hyp protection\n");
2104 			goto out_hyp;
2105 		}
2106 	}
2107 
2108 	if (is_protected_kvm_enabled()) {
2109 		kvm_info("Protected nVHE mode initialized successfully\n");
2110 	} else if (in_hyp_mode) {
2111 		kvm_info("VHE mode initialized successfully\n");
2112 	} else {
2113 		kvm_info("Hyp mode initialized successfully\n");
2114 	}
2115 
2116 	return 0;
2117 
2118 out_hyp:
2119 	hyp_cpu_pm_exit();
2120 	if (!in_hyp_mode)
2121 		teardown_hyp_mode();
2122 out_err:
2123 	return err;
2124 }
2125 
2126 /* NOP: Compiling as a module not supported */
2127 void kvm_arch_exit(void)
2128 {
2129 	kvm_perf_teardown();
2130 }
2131 
2132 static int __init early_kvm_mode_cfg(char *arg)
2133 {
2134 	if (!arg)
2135 		return -EINVAL;
2136 
2137 	if (strcmp(arg, "protected") == 0) {
2138 		kvm_mode = KVM_MODE_PROTECTED;
2139 		return 0;
2140 	}
2141 
2142 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode()))
2143 		return 0;
2144 
2145 	return -EINVAL;
2146 }
2147 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2148 
2149 enum kvm_mode kvm_get_mode(void)
2150 {
2151 	return kvm_mode;
2152 }
2153 
2154 static int arm_init(void)
2155 {
2156 	int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2157 	return rc;
2158 }
2159 
2160 module_init(arm_init);
2161