xref: /openbmc/linux/arch/arm64/kvm/arm.c (revision 2cf1c348)
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  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
802  * @vcpu:	The VCPU pointer
803  *
804  * This function is called through the VCPU_RUN ioctl called from user space. It
805  * will execute VM code in a loop until the time slice for the process is used
806  * or some emulation is needed from user space in which case the function will
807  * return with return value 0 and with the kvm_run structure filled in with the
808  * required data for the requested emulation.
809  */
810 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
811 {
812 	struct kvm_run *run = vcpu->run;
813 	int ret;
814 
815 	if (run->exit_reason == KVM_EXIT_MMIO) {
816 		ret = kvm_handle_mmio_return(vcpu);
817 		if (ret)
818 			return ret;
819 	}
820 
821 	vcpu_load(vcpu);
822 
823 	if (run->immediate_exit) {
824 		ret = -EINTR;
825 		goto out;
826 	}
827 
828 	kvm_sigset_activate(vcpu);
829 
830 	ret = 1;
831 	run->exit_reason = KVM_EXIT_UNKNOWN;
832 	while (ret > 0) {
833 		/*
834 		 * Check conditions before entering the guest
835 		 */
836 		ret = xfer_to_guest_mode_handle_work(vcpu);
837 		if (!ret)
838 			ret = 1;
839 
840 		update_vmid(&vcpu->arch.hw_mmu->vmid);
841 
842 		check_vcpu_requests(vcpu);
843 
844 		/*
845 		 * Preparing the interrupts to be injected also
846 		 * involves poking the GIC, which must be done in a
847 		 * non-preemptible context.
848 		 */
849 		preempt_disable();
850 
851 		kvm_pmu_flush_hwstate(vcpu);
852 
853 		local_irq_disable();
854 
855 		kvm_vgic_flush_hwstate(vcpu);
856 
857 		/*
858 		 * Ensure we set mode to IN_GUEST_MODE after we disable
859 		 * interrupts and before the final VCPU requests check.
860 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
861 		 * Documentation/virt/kvm/vcpu-requests.rst
862 		 */
863 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
864 
865 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
866 			vcpu->mode = OUTSIDE_GUEST_MODE;
867 			isb(); /* Ensure work in x_flush_hwstate is committed */
868 			kvm_pmu_sync_hwstate(vcpu);
869 			if (static_branch_unlikely(&userspace_irqchip_in_use))
870 				kvm_timer_sync_user(vcpu);
871 			kvm_vgic_sync_hwstate(vcpu);
872 			local_irq_enable();
873 			preempt_enable();
874 			continue;
875 		}
876 
877 		kvm_arm_setup_debug(vcpu);
878 		kvm_arch_vcpu_ctxflush_fp(vcpu);
879 
880 		/**************************************************************
881 		 * Enter the guest
882 		 */
883 		trace_kvm_entry(*vcpu_pc(vcpu));
884 		guest_enter_irqoff();
885 
886 		ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
887 
888 		vcpu->mode = OUTSIDE_GUEST_MODE;
889 		vcpu->stat.exits++;
890 		/*
891 		 * Back from guest
892 		 *************************************************************/
893 
894 		kvm_arm_clear_debug(vcpu);
895 
896 		/*
897 		 * We must sync the PMU state before the vgic state so
898 		 * that the vgic can properly sample the updated state of the
899 		 * interrupt line.
900 		 */
901 		kvm_pmu_sync_hwstate(vcpu);
902 
903 		/*
904 		 * Sync the vgic state before syncing the timer state because
905 		 * the timer code needs to know if the virtual timer
906 		 * interrupts are active.
907 		 */
908 		kvm_vgic_sync_hwstate(vcpu);
909 
910 		/*
911 		 * Sync the timer hardware state before enabling interrupts as
912 		 * we don't want vtimer interrupts to race with syncing the
913 		 * timer virtual interrupt state.
914 		 */
915 		if (static_branch_unlikely(&userspace_irqchip_in_use))
916 			kvm_timer_sync_user(vcpu);
917 
918 		kvm_arch_vcpu_ctxsync_fp(vcpu);
919 
920 		/*
921 		 * We may have taken a host interrupt in HYP mode (ie
922 		 * while executing the guest). This interrupt is still
923 		 * pending, as we haven't serviced it yet!
924 		 *
925 		 * We're now back in SVC mode, with interrupts
926 		 * disabled.  Enabling the interrupts now will have
927 		 * the effect of taking the interrupt again, in SVC
928 		 * mode this time.
929 		 */
930 		local_irq_enable();
931 
932 		/*
933 		 * We do local_irq_enable() before calling guest_exit() so
934 		 * that if a timer interrupt hits while running the guest we
935 		 * account that tick as being spent in the guest.  We enable
936 		 * preemption after calling guest_exit() so that if we get
937 		 * preempted we make sure ticks after that is not counted as
938 		 * guest time.
939 		 */
940 		guest_exit();
941 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
942 
943 		/* Exit types that need handling before we can be preempted */
944 		handle_exit_early(vcpu, ret);
945 
946 		preempt_enable();
947 
948 		/*
949 		 * The ARMv8 architecture doesn't give the hypervisor
950 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
951 		 * if implemented by the CPU. If we spot the guest in such
952 		 * state and that we decided it wasn't supposed to do so (like
953 		 * with the asymmetric AArch32 case), return to userspace with
954 		 * a fatal error.
955 		 */
956 		if (vcpu_mode_is_bad_32bit(vcpu)) {
957 			/*
958 			 * As we have caught the guest red-handed, decide that
959 			 * it isn't fit for purpose anymore by making the vcpu
960 			 * invalid. The VMM can try and fix it by issuing  a
961 			 * KVM_ARM_VCPU_INIT if it really wants to.
962 			 */
963 			vcpu->arch.target = -1;
964 			ret = ARM_EXCEPTION_IL;
965 		}
966 
967 		ret = handle_exit(vcpu, ret);
968 	}
969 
970 	/* Tell userspace about in-kernel device output levels */
971 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
972 		kvm_timer_update_run(vcpu);
973 		kvm_pmu_update_run(vcpu);
974 	}
975 
976 	kvm_sigset_deactivate(vcpu);
977 
978 out:
979 	/*
980 	 * In the unlikely event that we are returning to userspace
981 	 * with pending exceptions or PC adjustment, commit these
982 	 * adjustments in order to give userspace a consistent view of
983 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
984 	 * being preempt-safe on VHE.
985 	 */
986 	if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
987 					 KVM_ARM64_INCREMENT_PC)))
988 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
989 
990 	vcpu_put(vcpu);
991 	return ret;
992 }
993 
994 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
995 {
996 	int bit_index;
997 	bool set;
998 	unsigned long *hcr;
999 
1000 	if (number == KVM_ARM_IRQ_CPU_IRQ)
1001 		bit_index = __ffs(HCR_VI);
1002 	else /* KVM_ARM_IRQ_CPU_FIQ */
1003 		bit_index = __ffs(HCR_VF);
1004 
1005 	hcr = vcpu_hcr(vcpu);
1006 	if (level)
1007 		set = test_and_set_bit(bit_index, hcr);
1008 	else
1009 		set = test_and_clear_bit(bit_index, hcr);
1010 
1011 	/*
1012 	 * If we didn't change anything, no need to wake up or kick other CPUs
1013 	 */
1014 	if (set == level)
1015 		return 0;
1016 
1017 	/*
1018 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1019 	 * trigger a world-switch round on the running physical CPU to set the
1020 	 * virtual IRQ/FIQ fields in the HCR appropriately.
1021 	 */
1022 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1023 	kvm_vcpu_kick(vcpu);
1024 
1025 	return 0;
1026 }
1027 
1028 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1029 			  bool line_status)
1030 {
1031 	u32 irq = irq_level->irq;
1032 	unsigned int irq_type, vcpu_idx, irq_num;
1033 	int nrcpus = atomic_read(&kvm->online_vcpus);
1034 	struct kvm_vcpu *vcpu = NULL;
1035 	bool level = irq_level->level;
1036 
1037 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1038 	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1039 	vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1040 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1041 
1042 	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1043 
1044 	switch (irq_type) {
1045 	case KVM_ARM_IRQ_TYPE_CPU:
1046 		if (irqchip_in_kernel(kvm))
1047 			return -ENXIO;
1048 
1049 		if (vcpu_idx >= nrcpus)
1050 			return -EINVAL;
1051 
1052 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1053 		if (!vcpu)
1054 			return -EINVAL;
1055 
1056 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1057 			return -EINVAL;
1058 
1059 		return vcpu_interrupt_line(vcpu, irq_num, level);
1060 	case KVM_ARM_IRQ_TYPE_PPI:
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 < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1072 			return -EINVAL;
1073 
1074 		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1075 	case KVM_ARM_IRQ_TYPE_SPI:
1076 		if (!irqchip_in_kernel(kvm))
1077 			return -ENXIO;
1078 
1079 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1080 			return -EINVAL;
1081 
1082 		return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1083 	}
1084 
1085 	return -EINVAL;
1086 }
1087 
1088 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1089 			       const struct kvm_vcpu_init *init)
1090 {
1091 	unsigned int i, ret;
1092 	u32 phys_target = kvm_target_cpu();
1093 
1094 	if (init->target != phys_target)
1095 		return -EINVAL;
1096 
1097 	/*
1098 	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1099 	 * use the same target.
1100 	 */
1101 	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1102 		return -EINVAL;
1103 
1104 	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1105 	for (i = 0; i < sizeof(init->features) * 8; i++) {
1106 		bool set = (init->features[i / 32] & (1 << (i % 32)));
1107 
1108 		if (set && i >= KVM_VCPU_MAX_FEATURES)
1109 			return -ENOENT;
1110 
1111 		/*
1112 		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1113 		 * use the same feature set.
1114 		 */
1115 		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1116 		    test_bit(i, vcpu->arch.features) != set)
1117 			return -EINVAL;
1118 
1119 		if (set)
1120 			set_bit(i, vcpu->arch.features);
1121 	}
1122 
1123 	vcpu->arch.target = phys_target;
1124 
1125 	/* Now we know what it is, we can reset it. */
1126 	ret = kvm_reset_vcpu(vcpu);
1127 	if (ret) {
1128 		vcpu->arch.target = -1;
1129 		bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1130 	}
1131 
1132 	return ret;
1133 }
1134 
1135 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1136 					 struct kvm_vcpu_init *init)
1137 {
1138 	int ret;
1139 
1140 	ret = kvm_vcpu_set_target(vcpu, init);
1141 	if (ret)
1142 		return ret;
1143 
1144 	/*
1145 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1146 	 * guest MMU is turned off and flush the caches as needed.
1147 	 *
1148 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1149 	 * ensuring that the data side is always coherent. We still
1150 	 * need to invalidate the I-cache though, as FWB does *not*
1151 	 * imply CTR_EL0.DIC.
1152 	 */
1153 	if (vcpu_has_run_once(vcpu)) {
1154 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1155 			stage2_unmap_vm(vcpu->kvm);
1156 		else
1157 			icache_inval_all_pou();
1158 	}
1159 
1160 	vcpu_reset_hcr(vcpu);
1161 	vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1162 
1163 	/*
1164 	 * Handle the "start in power-off" case.
1165 	 */
1166 	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1167 		vcpu_power_off(vcpu);
1168 	else
1169 		vcpu->arch.power_off = false;
1170 
1171 	return 0;
1172 }
1173 
1174 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1175 				 struct kvm_device_attr *attr)
1176 {
1177 	int ret = -ENXIO;
1178 
1179 	switch (attr->group) {
1180 	default:
1181 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1182 		break;
1183 	}
1184 
1185 	return ret;
1186 }
1187 
1188 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1189 				 struct kvm_device_attr *attr)
1190 {
1191 	int ret = -ENXIO;
1192 
1193 	switch (attr->group) {
1194 	default:
1195 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1196 		break;
1197 	}
1198 
1199 	return ret;
1200 }
1201 
1202 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1203 				 struct kvm_device_attr *attr)
1204 {
1205 	int ret = -ENXIO;
1206 
1207 	switch (attr->group) {
1208 	default:
1209 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1210 		break;
1211 	}
1212 
1213 	return ret;
1214 }
1215 
1216 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1217 				   struct kvm_vcpu_events *events)
1218 {
1219 	memset(events, 0, sizeof(*events));
1220 
1221 	return __kvm_arm_vcpu_get_events(vcpu, events);
1222 }
1223 
1224 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1225 				   struct kvm_vcpu_events *events)
1226 {
1227 	int i;
1228 
1229 	/* check whether the reserved field is zero */
1230 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1231 		if (events->reserved[i])
1232 			return -EINVAL;
1233 
1234 	/* check whether the pad field is zero */
1235 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1236 		if (events->exception.pad[i])
1237 			return -EINVAL;
1238 
1239 	return __kvm_arm_vcpu_set_events(vcpu, events);
1240 }
1241 
1242 long kvm_arch_vcpu_ioctl(struct file *filp,
1243 			 unsigned int ioctl, unsigned long arg)
1244 {
1245 	struct kvm_vcpu *vcpu = filp->private_data;
1246 	void __user *argp = (void __user *)arg;
1247 	struct kvm_device_attr attr;
1248 	long r;
1249 
1250 	switch (ioctl) {
1251 	case KVM_ARM_VCPU_INIT: {
1252 		struct kvm_vcpu_init init;
1253 
1254 		r = -EFAULT;
1255 		if (copy_from_user(&init, argp, sizeof(init)))
1256 			break;
1257 
1258 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1259 		break;
1260 	}
1261 	case KVM_SET_ONE_REG:
1262 	case KVM_GET_ONE_REG: {
1263 		struct kvm_one_reg reg;
1264 
1265 		r = -ENOEXEC;
1266 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1267 			break;
1268 
1269 		r = -EFAULT;
1270 		if (copy_from_user(&reg, argp, sizeof(reg)))
1271 			break;
1272 
1273 		/*
1274 		 * We could owe a reset due to PSCI. Handle the pending reset
1275 		 * here to ensure userspace register accesses are ordered after
1276 		 * the reset.
1277 		 */
1278 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1279 			kvm_reset_vcpu(vcpu);
1280 
1281 		if (ioctl == KVM_SET_ONE_REG)
1282 			r = kvm_arm_set_reg(vcpu, &reg);
1283 		else
1284 			r = kvm_arm_get_reg(vcpu, &reg);
1285 		break;
1286 	}
1287 	case KVM_GET_REG_LIST: {
1288 		struct kvm_reg_list __user *user_list = argp;
1289 		struct kvm_reg_list reg_list;
1290 		unsigned n;
1291 
1292 		r = -ENOEXEC;
1293 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1294 			break;
1295 
1296 		r = -EPERM;
1297 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1298 			break;
1299 
1300 		r = -EFAULT;
1301 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1302 			break;
1303 		n = reg_list.n;
1304 		reg_list.n = kvm_arm_num_regs(vcpu);
1305 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1306 			break;
1307 		r = -E2BIG;
1308 		if (n < reg_list.n)
1309 			break;
1310 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1311 		break;
1312 	}
1313 	case KVM_SET_DEVICE_ATTR: {
1314 		r = -EFAULT;
1315 		if (copy_from_user(&attr, argp, sizeof(attr)))
1316 			break;
1317 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1318 		break;
1319 	}
1320 	case KVM_GET_DEVICE_ATTR: {
1321 		r = -EFAULT;
1322 		if (copy_from_user(&attr, argp, sizeof(attr)))
1323 			break;
1324 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1325 		break;
1326 	}
1327 	case KVM_HAS_DEVICE_ATTR: {
1328 		r = -EFAULT;
1329 		if (copy_from_user(&attr, argp, sizeof(attr)))
1330 			break;
1331 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1332 		break;
1333 	}
1334 	case KVM_GET_VCPU_EVENTS: {
1335 		struct kvm_vcpu_events events;
1336 
1337 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1338 			return -EINVAL;
1339 
1340 		if (copy_to_user(argp, &events, sizeof(events)))
1341 			return -EFAULT;
1342 
1343 		return 0;
1344 	}
1345 	case KVM_SET_VCPU_EVENTS: {
1346 		struct kvm_vcpu_events events;
1347 
1348 		if (copy_from_user(&events, argp, sizeof(events)))
1349 			return -EFAULT;
1350 
1351 		return kvm_arm_vcpu_set_events(vcpu, &events);
1352 	}
1353 	case KVM_ARM_VCPU_FINALIZE: {
1354 		int what;
1355 
1356 		if (!kvm_vcpu_initialized(vcpu))
1357 			return -ENOEXEC;
1358 
1359 		if (get_user(what, (const int __user *)argp))
1360 			return -EFAULT;
1361 
1362 		return kvm_arm_vcpu_finalize(vcpu, what);
1363 	}
1364 	default:
1365 		r = -EINVAL;
1366 	}
1367 
1368 	return r;
1369 }
1370 
1371 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1372 {
1373 
1374 }
1375 
1376 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1377 					const struct kvm_memory_slot *memslot)
1378 {
1379 	kvm_flush_remote_tlbs(kvm);
1380 }
1381 
1382 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1383 					struct kvm_arm_device_addr *dev_addr)
1384 {
1385 	unsigned long dev_id, type;
1386 
1387 	dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1388 		KVM_ARM_DEVICE_ID_SHIFT;
1389 	type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1390 		KVM_ARM_DEVICE_TYPE_SHIFT;
1391 
1392 	switch (dev_id) {
1393 	case KVM_ARM_DEVICE_VGIC_V2:
1394 		if (!vgic_present)
1395 			return -ENXIO;
1396 		return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1397 	default:
1398 		return -ENODEV;
1399 	}
1400 }
1401 
1402 long kvm_arch_vm_ioctl(struct file *filp,
1403 		       unsigned int ioctl, unsigned long arg)
1404 {
1405 	struct kvm *kvm = filp->private_data;
1406 	void __user *argp = (void __user *)arg;
1407 
1408 	switch (ioctl) {
1409 	case KVM_CREATE_IRQCHIP: {
1410 		int ret;
1411 		if (!vgic_present)
1412 			return -ENXIO;
1413 		mutex_lock(&kvm->lock);
1414 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1415 		mutex_unlock(&kvm->lock);
1416 		return ret;
1417 	}
1418 	case KVM_ARM_SET_DEVICE_ADDR: {
1419 		struct kvm_arm_device_addr dev_addr;
1420 
1421 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1422 			return -EFAULT;
1423 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1424 	}
1425 	case KVM_ARM_PREFERRED_TARGET: {
1426 		struct kvm_vcpu_init init;
1427 
1428 		kvm_vcpu_preferred_target(&init);
1429 
1430 		if (copy_to_user(argp, &init, sizeof(init)))
1431 			return -EFAULT;
1432 
1433 		return 0;
1434 	}
1435 	case KVM_ARM_MTE_COPY_TAGS: {
1436 		struct kvm_arm_copy_mte_tags copy_tags;
1437 
1438 		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1439 			return -EFAULT;
1440 		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1441 	}
1442 	default:
1443 		return -EINVAL;
1444 	}
1445 }
1446 
1447 static unsigned long nvhe_percpu_size(void)
1448 {
1449 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1450 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1451 }
1452 
1453 static unsigned long nvhe_percpu_order(void)
1454 {
1455 	unsigned long size = nvhe_percpu_size();
1456 
1457 	return size ? get_order(size) : 0;
1458 }
1459 
1460 /* A lookup table holding the hypervisor VA for each vector slot */
1461 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1462 
1463 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1464 {
1465 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1466 }
1467 
1468 static int kvm_init_vector_slots(void)
1469 {
1470 	int err;
1471 	void *base;
1472 
1473 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1474 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1475 
1476 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1477 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1478 
1479 	if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1480 		return 0;
1481 
1482 	if (!has_vhe()) {
1483 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1484 					       __BP_HARDEN_HYP_VECS_SZ, &base);
1485 		if (err)
1486 			return err;
1487 	}
1488 
1489 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1490 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1491 	return 0;
1492 }
1493 
1494 static void cpu_prepare_hyp_mode(int cpu)
1495 {
1496 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1497 	unsigned long tcr;
1498 
1499 	/*
1500 	 * Calculate the raw per-cpu offset without a translation from the
1501 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1502 	 * so that we can use adr_l to access per-cpu variables in EL2.
1503 	 * Also drop the KASAN tag which gets in the way...
1504 	 */
1505 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1506 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1507 
1508 	params->mair_el2 = read_sysreg(mair_el1);
1509 
1510 	/*
1511 	 * The ID map may be configured to use an extended virtual address
1512 	 * range. This is only the case if system RAM is out of range for the
1513 	 * currently configured page size and VA_BITS, in which case we will
1514 	 * also need the extended virtual range for the HYP ID map, or we won't
1515 	 * be able to enable the EL2 MMU.
1516 	 *
1517 	 * However, at EL2, there is only one TTBR register, and we can't switch
1518 	 * between translation tables *and* update TCR_EL2.T0SZ at the same
1519 	 * time. Bottom line: we need to use the extended range with *both* our
1520 	 * translation tables.
1521 	 *
1522 	 * So use the same T0SZ value we use for the ID map.
1523 	 */
1524 	tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1525 	tcr &= ~TCR_T0SZ_MASK;
1526 	tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1527 	params->tcr_el2 = tcr;
1528 
1529 	params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1530 	params->pgd_pa = kvm_mmu_get_httbr();
1531 	if (is_protected_kvm_enabled())
1532 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1533 	else
1534 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1535 	params->vttbr = params->vtcr = 0;
1536 
1537 	/*
1538 	 * Flush the init params from the data cache because the struct will
1539 	 * be read while the MMU is off.
1540 	 */
1541 	kvm_flush_dcache_to_poc(params, sizeof(*params));
1542 }
1543 
1544 static void hyp_install_host_vector(void)
1545 {
1546 	struct kvm_nvhe_init_params *params;
1547 	struct arm_smccc_res res;
1548 
1549 	/* Switch from the HYP stub to our own HYP init vector */
1550 	__hyp_set_vectors(kvm_get_idmap_vector());
1551 
1552 	/*
1553 	 * Call initialization code, and switch to the full blown HYP code.
1554 	 * If the cpucaps haven't been finalized yet, something has gone very
1555 	 * wrong, and hyp will crash and burn when it uses any
1556 	 * cpus_have_const_cap() wrapper.
1557 	 */
1558 	BUG_ON(!system_capabilities_finalized());
1559 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1560 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1561 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1562 }
1563 
1564 static void cpu_init_hyp_mode(void)
1565 {
1566 	hyp_install_host_vector();
1567 
1568 	/*
1569 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1570 	 * at EL2.
1571 	 */
1572 	if (this_cpu_has_cap(ARM64_SSBS) &&
1573 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1574 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1575 	}
1576 }
1577 
1578 static void cpu_hyp_reset(void)
1579 {
1580 	if (!is_kernel_in_hyp_mode())
1581 		__hyp_reset_vectors();
1582 }
1583 
1584 /*
1585  * EL2 vectors can be mapped and rerouted in a number of ways,
1586  * depending on the kernel configuration and CPU present:
1587  *
1588  * - If the CPU is affected by Spectre-v2, the hardening sequence is
1589  *   placed in one of the vector slots, which is executed before jumping
1590  *   to the real vectors.
1591  *
1592  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1593  *   containing the hardening sequence is mapped next to the idmap page,
1594  *   and executed before jumping to the real vectors.
1595  *
1596  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1597  *   empty slot is selected, mapped next to the idmap page, and
1598  *   executed before jumping to the real vectors.
1599  *
1600  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1601  * VHE, as we don't have hypervisor-specific mappings. If the system
1602  * is VHE and yet selects this capability, it will be ignored.
1603  */
1604 static void cpu_set_hyp_vector(void)
1605 {
1606 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1607 	void *vector = hyp_spectre_vector_selector[data->slot];
1608 
1609 	if (!is_protected_kvm_enabled())
1610 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1611 	else
1612 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1613 }
1614 
1615 static void cpu_hyp_init_context(void)
1616 {
1617 	kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1618 
1619 	if (!is_kernel_in_hyp_mode())
1620 		cpu_init_hyp_mode();
1621 }
1622 
1623 static void cpu_hyp_init_features(void)
1624 {
1625 	cpu_set_hyp_vector();
1626 	kvm_arm_init_debug();
1627 
1628 	if (is_kernel_in_hyp_mode())
1629 		kvm_timer_init_vhe();
1630 
1631 	if (vgic_present)
1632 		kvm_vgic_init_cpu_hardware();
1633 }
1634 
1635 static void cpu_hyp_reinit(void)
1636 {
1637 	cpu_hyp_reset();
1638 	cpu_hyp_init_context();
1639 	cpu_hyp_init_features();
1640 }
1641 
1642 static void _kvm_arch_hardware_enable(void *discard)
1643 {
1644 	if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1645 		cpu_hyp_reinit();
1646 		__this_cpu_write(kvm_arm_hardware_enabled, 1);
1647 	}
1648 }
1649 
1650 int kvm_arch_hardware_enable(void)
1651 {
1652 	_kvm_arch_hardware_enable(NULL);
1653 	return 0;
1654 }
1655 
1656 static void _kvm_arch_hardware_disable(void *discard)
1657 {
1658 	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1659 		cpu_hyp_reset();
1660 		__this_cpu_write(kvm_arm_hardware_enabled, 0);
1661 	}
1662 }
1663 
1664 void kvm_arch_hardware_disable(void)
1665 {
1666 	if (!is_protected_kvm_enabled())
1667 		_kvm_arch_hardware_disable(NULL);
1668 }
1669 
1670 #ifdef CONFIG_CPU_PM
1671 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1672 				    unsigned long cmd,
1673 				    void *v)
1674 {
1675 	/*
1676 	 * kvm_arm_hardware_enabled is left with its old value over
1677 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1678 	 * re-enable hyp.
1679 	 */
1680 	switch (cmd) {
1681 	case CPU_PM_ENTER:
1682 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1683 			/*
1684 			 * don't update kvm_arm_hardware_enabled here
1685 			 * so that the hardware will be re-enabled
1686 			 * when we resume. See below.
1687 			 */
1688 			cpu_hyp_reset();
1689 
1690 		return NOTIFY_OK;
1691 	case CPU_PM_ENTER_FAILED:
1692 	case CPU_PM_EXIT:
1693 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1694 			/* The hardware was enabled before suspend. */
1695 			cpu_hyp_reinit();
1696 
1697 		return NOTIFY_OK;
1698 
1699 	default:
1700 		return NOTIFY_DONE;
1701 	}
1702 }
1703 
1704 static struct notifier_block hyp_init_cpu_pm_nb = {
1705 	.notifier_call = hyp_init_cpu_pm_notifier,
1706 };
1707 
1708 static void hyp_cpu_pm_init(void)
1709 {
1710 	if (!is_protected_kvm_enabled())
1711 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1712 }
1713 static void hyp_cpu_pm_exit(void)
1714 {
1715 	if (!is_protected_kvm_enabled())
1716 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1717 }
1718 #else
1719 static inline void hyp_cpu_pm_init(void)
1720 {
1721 }
1722 static inline void hyp_cpu_pm_exit(void)
1723 {
1724 }
1725 #endif
1726 
1727 static void init_cpu_logical_map(void)
1728 {
1729 	unsigned int cpu;
1730 
1731 	/*
1732 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1733 	 * Only copy the set of online CPUs whose features have been chacked
1734 	 * against the finalized system capabilities. The hypervisor will not
1735 	 * allow any other CPUs from the `possible` set to boot.
1736 	 */
1737 	for_each_online_cpu(cpu)
1738 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1739 }
1740 
1741 #define init_psci_0_1_impl_state(config, what)	\
1742 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
1743 
1744 static bool init_psci_relay(void)
1745 {
1746 	/*
1747 	 * If PSCI has not been initialized, protected KVM cannot install
1748 	 * itself on newly booted CPUs.
1749 	 */
1750 	if (!psci_ops.get_version) {
1751 		kvm_err("Cannot initialize protected mode without PSCI\n");
1752 		return false;
1753 	}
1754 
1755 	kvm_host_psci_config.version = psci_ops.get_version();
1756 
1757 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1758 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1759 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1760 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1761 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1762 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1763 	}
1764 	return true;
1765 }
1766 
1767 static int init_subsystems(void)
1768 {
1769 	int err = 0;
1770 
1771 	/*
1772 	 * Enable hardware so that subsystem initialisation can access EL2.
1773 	 */
1774 	on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1775 
1776 	/*
1777 	 * Register CPU lower-power notifier
1778 	 */
1779 	hyp_cpu_pm_init();
1780 
1781 	/*
1782 	 * Init HYP view of VGIC
1783 	 */
1784 	err = kvm_vgic_hyp_init();
1785 	switch (err) {
1786 	case 0:
1787 		vgic_present = true;
1788 		break;
1789 	case -ENODEV:
1790 	case -ENXIO:
1791 		vgic_present = false;
1792 		err = 0;
1793 		break;
1794 	default:
1795 		goto out;
1796 	}
1797 
1798 	/*
1799 	 * Init HYP architected timer support
1800 	 */
1801 	err = kvm_timer_hyp_init(vgic_present);
1802 	if (err)
1803 		goto out;
1804 
1805 	kvm_register_perf_callbacks(NULL);
1806 
1807 	kvm_sys_reg_table_init();
1808 
1809 out:
1810 	if (err || !is_protected_kvm_enabled())
1811 		on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1812 
1813 	return err;
1814 }
1815 
1816 static void teardown_hyp_mode(void)
1817 {
1818 	int cpu;
1819 
1820 	free_hyp_pgds();
1821 	for_each_possible_cpu(cpu) {
1822 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1823 		free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1824 	}
1825 }
1826 
1827 static int do_pkvm_init(u32 hyp_va_bits)
1828 {
1829 	void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1830 	int ret;
1831 
1832 	preempt_disable();
1833 	cpu_hyp_init_context();
1834 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1835 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
1836 				hyp_va_bits);
1837 	cpu_hyp_init_features();
1838 
1839 	/*
1840 	 * The stub hypercalls are now disabled, so set our local flag to
1841 	 * prevent a later re-init attempt in kvm_arch_hardware_enable().
1842 	 */
1843 	__this_cpu_write(kvm_arm_hardware_enabled, 1);
1844 	preempt_enable();
1845 
1846 	return ret;
1847 }
1848 
1849 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1850 {
1851 	void *addr = phys_to_virt(hyp_mem_base);
1852 	int ret;
1853 
1854 	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1855 	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1856 	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
1857 	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
1858 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1859 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1860 	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
1861 
1862 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1863 	if (ret)
1864 		return ret;
1865 
1866 	ret = do_pkvm_init(hyp_va_bits);
1867 	if (ret)
1868 		return ret;
1869 
1870 	free_hyp_pgds();
1871 
1872 	return 0;
1873 }
1874 
1875 /**
1876  * Inits Hyp-mode on all online CPUs
1877  */
1878 static int init_hyp_mode(void)
1879 {
1880 	u32 hyp_va_bits;
1881 	int cpu;
1882 	int err = -ENOMEM;
1883 
1884 	/*
1885 	 * The protected Hyp-mode cannot be initialized if the memory pool
1886 	 * allocation has failed.
1887 	 */
1888 	if (is_protected_kvm_enabled() && !hyp_mem_base)
1889 		goto out_err;
1890 
1891 	/*
1892 	 * Allocate Hyp PGD and setup Hyp identity mapping
1893 	 */
1894 	err = kvm_mmu_init(&hyp_va_bits);
1895 	if (err)
1896 		goto out_err;
1897 
1898 	/*
1899 	 * Allocate stack pages for Hypervisor-mode
1900 	 */
1901 	for_each_possible_cpu(cpu) {
1902 		unsigned long stack_page;
1903 
1904 		stack_page = __get_free_page(GFP_KERNEL);
1905 		if (!stack_page) {
1906 			err = -ENOMEM;
1907 			goto out_err;
1908 		}
1909 
1910 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1911 	}
1912 
1913 	/*
1914 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1915 	 */
1916 	for_each_possible_cpu(cpu) {
1917 		struct page *page;
1918 		void *page_addr;
1919 
1920 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1921 		if (!page) {
1922 			err = -ENOMEM;
1923 			goto out_err;
1924 		}
1925 
1926 		page_addr = page_address(page);
1927 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1928 		kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1929 	}
1930 
1931 	/*
1932 	 * Map the Hyp-code called directly from the host
1933 	 */
1934 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1935 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1936 	if (err) {
1937 		kvm_err("Cannot map world-switch code\n");
1938 		goto out_err;
1939 	}
1940 
1941 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1942 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1943 	if (err) {
1944 		kvm_err("Cannot map .hyp.rodata section\n");
1945 		goto out_err;
1946 	}
1947 
1948 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1949 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1950 	if (err) {
1951 		kvm_err("Cannot map rodata section\n");
1952 		goto out_err;
1953 	}
1954 
1955 	/*
1956 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1957 	 * section thanks to an assertion in the linker script. Map it RW and
1958 	 * the rest of .bss RO.
1959 	 */
1960 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1961 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1962 	if (err) {
1963 		kvm_err("Cannot map hyp bss section: %d\n", err);
1964 		goto out_err;
1965 	}
1966 
1967 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1968 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1969 	if (err) {
1970 		kvm_err("Cannot map bss section\n");
1971 		goto out_err;
1972 	}
1973 
1974 	/*
1975 	 * Map the Hyp stack pages
1976 	 */
1977 	for_each_possible_cpu(cpu) {
1978 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1979 		err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1980 					  PAGE_HYP);
1981 
1982 		if (err) {
1983 			kvm_err("Cannot map hyp stack\n");
1984 			goto out_err;
1985 		}
1986 	}
1987 
1988 	for_each_possible_cpu(cpu) {
1989 		char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1990 		char *percpu_end = percpu_begin + nvhe_percpu_size();
1991 
1992 		/* Map Hyp percpu pages */
1993 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1994 		if (err) {
1995 			kvm_err("Cannot map hyp percpu region\n");
1996 			goto out_err;
1997 		}
1998 
1999 		/* Prepare the CPU initialization parameters */
2000 		cpu_prepare_hyp_mode(cpu);
2001 	}
2002 
2003 	if (is_protected_kvm_enabled()) {
2004 		init_cpu_logical_map();
2005 
2006 		if (!init_psci_relay()) {
2007 			err = -ENODEV;
2008 			goto out_err;
2009 		}
2010 	}
2011 
2012 	if (is_protected_kvm_enabled()) {
2013 		err = kvm_hyp_init_protection(hyp_va_bits);
2014 		if (err) {
2015 			kvm_err("Failed to init hyp memory protection\n");
2016 			goto out_err;
2017 		}
2018 	}
2019 
2020 	return 0;
2021 
2022 out_err:
2023 	teardown_hyp_mode();
2024 	kvm_err("error initializing Hyp mode: %d\n", err);
2025 	return err;
2026 }
2027 
2028 static void _kvm_host_prot_finalize(void *arg)
2029 {
2030 	int *err = arg;
2031 
2032 	if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)))
2033 		WRITE_ONCE(*err, -EINVAL);
2034 }
2035 
2036 static int pkvm_drop_host_privileges(void)
2037 {
2038 	int ret = 0;
2039 
2040 	/*
2041 	 * Flip the static key upfront as that may no longer be possible
2042 	 * once the host stage 2 is installed.
2043 	 */
2044 	static_branch_enable(&kvm_protected_mode_initialized);
2045 	on_each_cpu(_kvm_host_prot_finalize, &ret, 1);
2046 	return ret;
2047 }
2048 
2049 static int finalize_hyp_mode(void)
2050 {
2051 	if (!is_protected_kvm_enabled())
2052 		return 0;
2053 
2054 	/*
2055 	 * Exclude HYP BSS from kmemleak so that it doesn't get peeked
2056 	 * at, which would end badly once the section is inaccessible.
2057 	 * None of other sections should ever be introspected.
2058 	 */
2059 	kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
2060 	return pkvm_drop_host_privileges();
2061 }
2062 
2063 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2064 {
2065 	struct kvm_vcpu *vcpu;
2066 	unsigned long i;
2067 
2068 	mpidr &= MPIDR_HWID_BITMASK;
2069 	kvm_for_each_vcpu(i, vcpu, kvm) {
2070 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2071 			return vcpu;
2072 	}
2073 	return NULL;
2074 }
2075 
2076 bool kvm_arch_has_irq_bypass(void)
2077 {
2078 	return true;
2079 }
2080 
2081 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2082 				      struct irq_bypass_producer *prod)
2083 {
2084 	struct kvm_kernel_irqfd *irqfd =
2085 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2086 
2087 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2088 					  &irqfd->irq_entry);
2089 }
2090 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2091 				      struct irq_bypass_producer *prod)
2092 {
2093 	struct kvm_kernel_irqfd *irqfd =
2094 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2095 
2096 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2097 				     &irqfd->irq_entry);
2098 }
2099 
2100 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2101 {
2102 	struct kvm_kernel_irqfd *irqfd =
2103 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2104 
2105 	kvm_arm_halt_guest(irqfd->kvm);
2106 }
2107 
2108 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2109 {
2110 	struct kvm_kernel_irqfd *irqfd =
2111 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2112 
2113 	kvm_arm_resume_guest(irqfd->kvm);
2114 }
2115 
2116 /**
2117  * Initialize Hyp-mode and memory mappings on all CPUs.
2118  */
2119 int kvm_arch_init(void *opaque)
2120 {
2121 	int err;
2122 	bool in_hyp_mode;
2123 
2124 	if (!is_hyp_mode_available()) {
2125 		kvm_info("HYP mode not available\n");
2126 		return -ENODEV;
2127 	}
2128 
2129 	if (kvm_get_mode() == KVM_MODE_NONE) {
2130 		kvm_info("KVM disabled from command line\n");
2131 		return -ENODEV;
2132 	}
2133 
2134 	in_hyp_mode = is_kernel_in_hyp_mode();
2135 
2136 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2137 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2138 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2139 			 "Only trusted guests should be used on this system.\n");
2140 
2141 	err = kvm_set_ipa_limit();
2142 	if (err)
2143 		return err;
2144 
2145 	err = kvm_arm_init_sve();
2146 	if (err)
2147 		return err;
2148 
2149 	if (!in_hyp_mode) {
2150 		err = init_hyp_mode();
2151 		if (err)
2152 			goto out_err;
2153 	}
2154 
2155 	err = kvm_init_vector_slots();
2156 	if (err) {
2157 		kvm_err("Cannot initialise vector slots\n");
2158 		goto out_err;
2159 	}
2160 
2161 	err = init_subsystems();
2162 	if (err)
2163 		goto out_hyp;
2164 
2165 	if (!in_hyp_mode) {
2166 		err = finalize_hyp_mode();
2167 		if (err) {
2168 			kvm_err("Failed to finalize Hyp protection\n");
2169 			goto out_hyp;
2170 		}
2171 	}
2172 
2173 	if (is_protected_kvm_enabled()) {
2174 		kvm_info("Protected nVHE mode initialized successfully\n");
2175 	} else if (in_hyp_mode) {
2176 		kvm_info("VHE mode initialized successfully\n");
2177 	} else {
2178 		kvm_info("Hyp mode initialized successfully\n");
2179 	}
2180 
2181 	return 0;
2182 
2183 out_hyp:
2184 	hyp_cpu_pm_exit();
2185 	if (!in_hyp_mode)
2186 		teardown_hyp_mode();
2187 out_err:
2188 	return err;
2189 }
2190 
2191 /* NOP: Compiling as a module not supported */
2192 void kvm_arch_exit(void)
2193 {
2194 	kvm_unregister_perf_callbacks();
2195 }
2196 
2197 static int __init early_kvm_mode_cfg(char *arg)
2198 {
2199 	if (!arg)
2200 		return -EINVAL;
2201 
2202 	if (strcmp(arg, "protected") == 0) {
2203 		kvm_mode = KVM_MODE_PROTECTED;
2204 		return 0;
2205 	}
2206 
2207 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2208 		kvm_mode = KVM_MODE_DEFAULT;
2209 		return 0;
2210 	}
2211 
2212 	if (strcmp(arg, "none") == 0) {
2213 		kvm_mode = KVM_MODE_NONE;
2214 		return 0;
2215 	}
2216 
2217 	return -EINVAL;
2218 }
2219 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2220 
2221 enum kvm_mode kvm_get_mode(void)
2222 {
2223 	return kvm_mode;
2224 }
2225 
2226 static int arm_init(void)
2227 {
2228 	int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2229 	return rc;
2230 }
2231 
2232 module_init(arm_init);
2233