xref: /openbmc/linux/arch/arm64/kvm/arm.c (revision 09cf57eba304246141367b95c89801fd2047ac96)
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 <trace/events/kvm.h>
23 
24 #define CREATE_TRACE_POINTS
25 #include "trace_arm.h"
26 
27 #include <linux/uaccess.h>
28 #include <asm/ptrace.h>
29 #include <asm/mman.h>
30 #include <asm/tlbflush.h>
31 #include <asm/cacheflush.h>
32 #include <asm/cpufeature.h>
33 #include <asm/virt.h>
34 #include <asm/kvm_arm.h>
35 #include <asm/kvm_asm.h>
36 #include <asm/kvm_mmu.h>
37 #include <asm/kvm_emulate.h>
38 #include <asm/kvm_coproc.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 DEFINE_PER_CPU(kvm_host_data_t, kvm_host_data);
50 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
51 
52 /* The VMID used in the VTTBR */
53 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
54 static u32 kvm_next_vmid;
55 static DEFINE_SPINLOCK(kvm_vmid_lock);
56 
57 static bool vgic_present;
58 
59 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
60 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
61 
62 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
63 {
64 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
65 }
66 
67 int kvm_arch_hardware_setup(void *opaque)
68 {
69 	return 0;
70 }
71 
72 int kvm_arch_check_processor_compat(void *opaque)
73 {
74 	return 0;
75 }
76 
77 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
78 			    struct kvm_enable_cap *cap)
79 {
80 	int r;
81 
82 	if (cap->flags)
83 		return -EINVAL;
84 
85 	switch (cap->cap) {
86 	case KVM_CAP_ARM_NISV_TO_USER:
87 		r = 0;
88 		kvm->arch.return_nisv_io_abort_to_user = true;
89 		break;
90 	default:
91 		r = -EINVAL;
92 		break;
93 	}
94 
95 	return r;
96 }
97 
98 static int kvm_arm_default_max_vcpus(void)
99 {
100 	return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
101 }
102 
103 /**
104  * kvm_arch_init_vm - initializes a VM data structure
105  * @kvm:	pointer to the KVM struct
106  */
107 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
108 {
109 	int ret, cpu;
110 
111 	ret = kvm_arm_setup_stage2(kvm, type);
112 	if (ret)
113 		return ret;
114 
115 	kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
116 	if (!kvm->arch.last_vcpu_ran)
117 		return -ENOMEM;
118 
119 	for_each_possible_cpu(cpu)
120 		*per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
121 
122 	ret = kvm_alloc_stage2_pgd(kvm);
123 	if (ret)
124 		goto out_fail_alloc;
125 
126 	ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
127 	if (ret)
128 		goto out_free_stage2_pgd;
129 
130 	kvm_vgic_early_init(kvm);
131 
132 	/* Mark the initial VMID generation invalid */
133 	kvm->arch.vmid.vmid_gen = 0;
134 
135 	/* The maximum number of VCPUs is limited by the host's GIC model */
136 	kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
137 
138 	return ret;
139 out_free_stage2_pgd:
140 	kvm_free_stage2_pgd(kvm);
141 out_fail_alloc:
142 	free_percpu(kvm->arch.last_vcpu_ran);
143 	kvm->arch.last_vcpu_ran = NULL;
144 	return ret;
145 }
146 
147 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
148 {
149 	return VM_FAULT_SIGBUS;
150 }
151 
152 
153 /**
154  * kvm_arch_destroy_vm - destroy the VM data structure
155  * @kvm:	pointer to the KVM struct
156  */
157 void kvm_arch_destroy_vm(struct kvm *kvm)
158 {
159 	int i;
160 
161 	kvm_vgic_destroy(kvm);
162 
163 	free_percpu(kvm->arch.last_vcpu_ran);
164 	kvm->arch.last_vcpu_ran = NULL;
165 
166 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
167 		if (kvm->vcpus[i]) {
168 			kvm_vcpu_destroy(kvm->vcpus[i]);
169 			kvm->vcpus[i] = NULL;
170 		}
171 	}
172 	atomic_set(&kvm->online_vcpus, 0);
173 }
174 
175 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
176 {
177 	int r;
178 	switch (ext) {
179 	case KVM_CAP_IRQCHIP:
180 		r = vgic_present;
181 		break;
182 	case KVM_CAP_IOEVENTFD:
183 	case KVM_CAP_DEVICE_CTRL:
184 	case KVM_CAP_USER_MEMORY:
185 	case KVM_CAP_SYNC_MMU:
186 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
187 	case KVM_CAP_ONE_REG:
188 	case KVM_CAP_ARM_PSCI:
189 	case KVM_CAP_ARM_PSCI_0_2:
190 	case KVM_CAP_READONLY_MEM:
191 	case KVM_CAP_MP_STATE:
192 	case KVM_CAP_IMMEDIATE_EXIT:
193 	case KVM_CAP_VCPU_EVENTS:
194 	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
195 	case KVM_CAP_ARM_NISV_TO_USER:
196 	case KVM_CAP_ARM_INJECT_EXT_DABT:
197 		r = 1;
198 		break;
199 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
200 		r = 1;
201 		break;
202 	case KVM_CAP_NR_VCPUS:
203 		r = num_online_cpus();
204 		break;
205 	case KVM_CAP_MAX_VCPUS:
206 	case KVM_CAP_MAX_VCPU_ID:
207 		if (kvm)
208 			r = kvm->arch.max_vcpus;
209 		else
210 			r = kvm_arm_default_max_vcpus();
211 		break;
212 	case KVM_CAP_MSI_DEVID:
213 		if (!kvm)
214 			r = -EINVAL;
215 		else
216 			r = kvm->arch.vgic.msis_require_devid;
217 		break;
218 	case KVM_CAP_ARM_USER_IRQ:
219 		/*
220 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
221 		 * (bump this number if adding more devices)
222 		 */
223 		r = 1;
224 		break;
225 	default:
226 		r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
227 		break;
228 	}
229 	return r;
230 }
231 
232 long kvm_arch_dev_ioctl(struct file *filp,
233 			unsigned int ioctl, unsigned long arg)
234 {
235 	return -EINVAL;
236 }
237 
238 struct kvm *kvm_arch_alloc_vm(void)
239 {
240 	if (!has_vhe())
241 		return kzalloc(sizeof(struct kvm), GFP_KERNEL);
242 
243 	return vzalloc(sizeof(struct kvm));
244 }
245 
246 void kvm_arch_free_vm(struct kvm *kvm)
247 {
248 	if (!has_vhe())
249 		kfree(kvm);
250 	else
251 		vfree(kvm);
252 }
253 
254 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
255 {
256 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
257 		return -EBUSY;
258 
259 	if (id >= kvm->arch.max_vcpus)
260 		return -EINVAL;
261 
262 	return 0;
263 }
264 
265 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
266 {
267 	int err;
268 
269 	/* Force users to call KVM_ARM_VCPU_INIT */
270 	vcpu->arch.target = -1;
271 	bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
272 
273 	/* Set up the timer */
274 	kvm_timer_vcpu_init(vcpu);
275 
276 	kvm_pmu_vcpu_init(vcpu);
277 
278 	kvm_arm_reset_debug_ptr(vcpu);
279 
280 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
281 
282 	err = kvm_vgic_vcpu_init(vcpu);
283 	if (err)
284 		return err;
285 
286 	return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
287 }
288 
289 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
290 {
291 }
292 
293 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
294 {
295 	if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
296 		static_branch_dec(&userspace_irqchip_in_use);
297 
298 	kvm_mmu_free_memory_caches(vcpu);
299 	kvm_timer_vcpu_terminate(vcpu);
300 	kvm_pmu_vcpu_destroy(vcpu);
301 
302 	kvm_arm_vcpu_destroy(vcpu);
303 }
304 
305 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
306 {
307 	return kvm_timer_is_pending(vcpu);
308 }
309 
310 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
311 {
312 	/*
313 	 * If we're about to block (most likely because we've just hit a
314 	 * WFI), we need to sync back the state of the GIC CPU interface
315 	 * so that we have the latest PMR and group enables. This ensures
316 	 * that kvm_arch_vcpu_runnable has up-to-date data to decide
317 	 * whether we have pending interrupts.
318 	 *
319 	 * For the same reason, we want to tell GICv4 that we need
320 	 * doorbells to be signalled, should an interrupt become pending.
321 	 */
322 	preempt_disable();
323 	kvm_vgic_vmcr_sync(vcpu);
324 	vgic_v4_put(vcpu, true);
325 	preempt_enable();
326 }
327 
328 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
329 {
330 	preempt_disable();
331 	vgic_v4_load(vcpu);
332 	preempt_enable();
333 }
334 
335 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
336 {
337 	int *last_ran;
338 
339 	last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
340 
341 	/*
342 	 * We might get preempted before the vCPU actually runs, but
343 	 * over-invalidation doesn't affect correctness.
344 	 */
345 	if (*last_ran != vcpu->vcpu_id) {
346 		kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
347 		*last_ran = vcpu->vcpu_id;
348 	}
349 
350 	vcpu->cpu = cpu;
351 
352 	kvm_vgic_load(vcpu);
353 	kvm_timer_vcpu_load(vcpu);
354 	kvm_vcpu_load_sysregs(vcpu);
355 	kvm_arch_vcpu_load_fp(vcpu);
356 	kvm_vcpu_pmu_restore_guest(vcpu);
357 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
358 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
359 
360 	if (single_task_running())
361 		vcpu_clear_wfx_traps(vcpu);
362 	else
363 		vcpu_set_wfx_traps(vcpu);
364 
365 	if (vcpu_has_ptrauth(vcpu))
366 		vcpu_ptrauth_disable(vcpu);
367 }
368 
369 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
370 {
371 	kvm_arch_vcpu_put_fp(vcpu);
372 	kvm_vcpu_put_sysregs(vcpu);
373 	kvm_timer_vcpu_put(vcpu);
374 	kvm_vgic_put(vcpu);
375 	kvm_vcpu_pmu_restore_host(vcpu);
376 
377 	vcpu->cpu = -1;
378 }
379 
380 static void vcpu_power_off(struct kvm_vcpu *vcpu)
381 {
382 	vcpu->arch.power_off = true;
383 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
384 	kvm_vcpu_kick(vcpu);
385 }
386 
387 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
388 				    struct kvm_mp_state *mp_state)
389 {
390 	if (vcpu->arch.power_off)
391 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
392 	else
393 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
394 
395 	return 0;
396 }
397 
398 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
399 				    struct kvm_mp_state *mp_state)
400 {
401 	int ret = 0;
402 
403 	switch (mp_state->mp_state) {
404 	case KVM_MP_STATE_RUNNABLE:
405 		vcpu->arch.power_off = false;
406 		break;
407 	case KVM_MP_STATE_STOPPED:
408 		vcpu_power_off(vcpu);
409 		break;
410 	default:
411 		ret = -EINVAL;
412 	}
413 
414 	return ret;
415 }
416 
417 /**
418  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
419  * @v:		The VCPU pointer
420  *
421  * If the guest CPU is not waiting for interrupts or an interrupt line is
422  * asserted, the CPU is by definition runnable.
423  */
424 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
425 {
426 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
427 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
428 		&& !v->arch.power_off && !v->arch.pause);
429 }
430 
431 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
432 {
433 	return vcpu_mode_priv(vcpu);
434 }
435 
436 /* Just ensure a guest exit from a particular CPU */
437 static void exit_vm_noop(void *info)
438 {
439 }
440 
441 void force_vm_exit(const cpumask_t *mask)
442 {
443 	preempt_disable();
444 	smp_call_function_many(mask, exit_vm_noop, NULL, true);
445 	preempt_enable();
446 }
447 
448 /**
449  * need_new_vmid_gen - check that the VMID is still valid
450  * @vmid: The VMID to check
451  *
452  * return true if there is a new generation of VMIDs being used
453  *
454  * The hardware supports a limited set of values with the value zero reserved
455  * for the host, so we check if an assigned value belongs to a previous
456  * generation, which requires us to assign a new value. If we're the first to
457  * use a VMID for the new generation, we must flush necessary caches and TLBs
458  * on all CPUs.
459  */
460 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
461 {
462 	u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
463 	smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
464 	return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
465 }
466 
467 /**
468  * update_vmid - Update the vmid with a valid VMID for the current generation
469  * @kvm: The guest that struct vmid belongs to
470  * @vmid: The stage-2 VMID information struct
471  */
472 static void update_vmid(struct kvm_vmid *vmid)
473 {
474 	if (!need_new_vmid_gen(vmid))
475 		return;
476 
477 	spin_lock(&kvm_vmid_lock);
478 
479 	/*
480 	 * We need to re-check the vmid_gen here to ensure that if another vcpu
481 	 * already allocated a valid vmid for this vm, then this vcpu should
482 	 * use the same vmid.
483 	 */
484 	if (!need_new_vmid_gen(vmid)) {
485 		spin_unlock(&kvm_vmid_lock);
486 		return;
487 	}
488 
489 	/* First user of a new VMID generation? */
490 	if (unlikely(kvm_next_vmid == 0)) {
491 		atomic64_inc(&kvm_vmid_gen);
492 		kvm_next_vmid = 1;
493 
494 		/*
495 		 * On SMP we know no other CPUs can use this CPU's or each
496 		 * other's VMID after force_vm_exit returns since the
497 		 * kvm_vmid_lock blocks them from reentry to the guest.
498 		 */
499 		force_vm_exit(cpu_all_mask);
500 		/*
501 		 * Now broadcast TLB + ICACHE invalidation over the inner
502 		 * shareable domain to make sure all data structures are
503 		 * clean.
504 		 */
505 		kvm_call_hyp(__kvm_flush_vm_context);
506 	}
507 
508 	vmid->vmid = kvm_next_vmid;
509 	kvm_next_vmid++;
510 	kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
511 
512 	smp_wmb();
513 	WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
514 
515 	spin_unlock(&kvm_vmid_lock);
516 }
517 
518 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
519 {
520 	struct kvm *kvm = vcpu->kvm;
521 	int ret = 0;
522 
523 	if (likely(vcpu->arch.has_run_once))
524 		return 0;
525 
526 	if (!kvm_arm_vcpu_is_finalized(vcpu))
527 		return -EPERM;
528 
529 	vcpu->arch.has_run_once = true;
530 
531 	if (likely(irqchip_in_kernel(kvm))) {
532 		/*
533 		 * Map the VGIC hardware resources before running a vcpu the
534 		 * first time on this VM.
535 		 */
536 		if (unlikely(!vgic_ready(kvm))) {
537 			ret = kvm_vgic_map_resources(kvm);
538 			if (ret)
539 				return ret;
540 		}
541 	} else {
542 		/*
543 		 * Tell the rest of the code that there are userspace irqchip
544 		 * VMs in the wild.
545 		 */
546 		static_branch_inc(&userspace_irqchip_in_use);
547 	}
548 
549 	ret = kvm_timer_enable(vcpu);
550 	if (ret)
551 		return ret;
552 
553 	ret = kvm_arm_pmu_v3_enable(vcpu);
554 
555 	return ret;
556 }
557 
558 bool kvm_arch_intc_initialized(struct kvm *kvm)
559 {
560 	return vgic_initialized(kvm);
561 }
562 
563 void kvm_arm_halt_guest(struct kvm *kvm)
564 {
565 	int i;
566 	struct kvm_vcpu *vcpu;
567 
568 	kvm_for_each_vcpu(i, vcpu, kvm)
569 		vcpu->arch.pause = true;
570 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
571 }
572 
573 void kvm_arm_resume_guest(struct kvm *kvm)
574 {
575 	int i;
576 	struct kvm_vcpu *vcpu;
577 
578 	kvm_for_each_vcpu(i, vcpu, kvm) {
579 		vcpu->arch.pause = false;
580 		rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
581 	}
582 }
583 
584 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
585 {
586 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
587 
588 	rcuwait_wait_event(wait,
589 			   (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
590 			   TASK_INTERRUPTIBLE);
591 
592 	if (vcpu->arch.power_off || vcpu->arch.pause) {
593 		/* Awaken to handle a signal, request we sleep again later. */
594 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
595 	}
596 
597 	/*
598 	 * Make sure we will observe a potential reset request if we've
599 	 * observed a change to the power state. Pairs with the smp_wmb() in
600 	 * kvm_psci_vcpu_on().
601 	 */
602 	smp_rmb();
603 }
604 
605 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
606 {
607 	return vcpu->arch.target >= 0;
608 }
609 
610 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
611 {
612 	if (kvm_request_pending(vcpu)) {
613 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
614 			vcpu_req_sleep(vcpu);
615 
616 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
617 			kvm_reset_vcpu(vcpu);
618 
619 		/*
620 		 * Clear IRQ_PENDING requests that were made to guarantee
621 		 * that a VCPU sees new virtual interrupts.
622 		 */
623 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
624 
625 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
626 			kvm_update_stolen_time(vcpu);
627 
628 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
629 			/* The distributor enable bits were changed */
630 			preempt_disable();
631 			vgic_v4_put(vcpu, false);
632 			vgic_v4_load(vcpu);
633 			preempt_enable();
634 		}
635 	}
636 }
637 
638 /**
639  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
640  * @vcpu:	The VCPU pointer
641  *
642  * This function is called through the VCPU_RUN ioctl called from user space. It
643  * will execute VM code in a loop until the time slice for the process is used
644  * or some emulation is needed from user space in which case the function will
645  * return with return value 0 and with the kvm_run structure filled in with the
646  * required data for the requested emulation.
647  */
648 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
649 {
650 	struct kvm_run *run = vcpu->run;
651 	int ret;
652 
653 	if (unlikely(!kvm_vcpu_initialized(vcpu)))
654 		return -ENOEXEC;
655 
656 	ret = kvm_vcpu_first_run_init(vcpu);
657 	if (ret)
658 		return ret;
659 
660 	if (run->exit_reason == KVM_EXIT_MMIO) {
661 		ret = kvm_handle_mmio_return(vcpu, run);
662 		if (ret)
663 			return ret;
664 	}
665 
666 	if (run->immediate_exit)
667 		return -EINTR;
668 
669 	vcpu_load(vcpu);
670 
671 	kvm_sigset_activate(vcpu);
672 
673 	ret = 1;
674 	run->exit_reason = KVM_EXIT_UNKNOWN;
675 	while (ret > 0) {
676 		/*
677 		 * Check conditions before entering the guest
678 		 */
679 		cond_resched();
680 
681 		update_vmid(&vcpu->kvm->arch.vmid);
682 
683 		check_vcpu_requests(vcpu);
684 
685 		/*
686 		 * Preparing the interrupts to be injected also
687 		 * involves poking the GIC, which must be done in a
688 		 * non-preemptible context.
689 		 */
690 		preempt_disable();
691 
692 		kvm_pmu_flush_hwstate(vcpu);
693 
694 		local_irq_disable();
695 
696 		kvm_vgic_flush_hwstate(vcpu);
697 
698 		/*
699 		 * Exit if we have a signal pending so that we can deliver the
700 		 * signal to user space.
701 		 */
702 		if (signal_pending(current)) {
703 			ret = -EINTR;
704 			run->exit_reason = KVM_EXIT_INTR;
705 		}
706 
707 		/*
708 		 * If we're using a userspace irqchip, then check if we need
709 		 * to tell a userspace irqchip about timer or PMU level
710 		 * changes and if so, exit to userspace (the actual level
711 		 * state gets updated in kvm_timer_update_run and
712 		 * kvm_pmu_update_run below).
713 		 */
714 		if (static_branch_unlikely(&userspace_irqchip_in_use)) {
715 			if (kvm_timer_should_notify_user(vcpu) ||
716 			    kvm_pmu_should_notify_user(vcpu)) {
717 				ret = -EINTR;
718 				run->exit_reason = KVM_EXIT_INTR;
719 			}
720 		}
721 
722 		/*
723 		 * Ensure we set mode to IN_GUEST_MODE after we disable
724 		 * interrupts and before the final VCPU requests check.
725 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
726 		 * Documentation/virt/kvm/vcpu-requests.rst
727 		 */
728 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
729 
730 		if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
731 		    kvm_request_pending(vcpu)) {
732 			vcpu->mode = OUTSIDE_GUEST_MODE;
733 			isb(); /* Ensure work in x_flush_hwstate is committed */
734 			kvm_pmu_sync_hwstate(vcpu);
735 			if (static_branch_unlikely(&userspace_irqchip_in_use))
736 				kvm_timer_sync_hwstate(vcpu);
737 			kvm_vgic_sync_hwstate(vcpu);
738 			local_irq_enable();
739 			preempt_enable();
740 			continue;
741 		}
742 
743 		kvm_arm_setup_debug(vcpu);
744 
745 		/**************************************************************
746 		 * Enter the guest
747 		 */
748 		trace_kvm_entry(*vcpu_pc(vcpu));
749 		guest_enter_irqoff();
750 
751 		ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
752 
753 		vcpu->mode = OUTSIDE_GUEST_MODE;
754 		vcpu->stat.exits++;
755 		/*
756 		 * Back from guest
757 		 *************************************************************/
758 
759 		kvm_arm_clear_debug(vcpu);
760 
761 		/*
762 		 * We must sync the PMU state before the vgic state so
763 		 * that the vgic can properly sample the updated state of the
764 		 * interrupt line.
765 		 */
766 		kvm_pmu_sync_hwstate(vcpu);
767 
768 		/*
769 		 * Sync the vgic state before syncing the timer state because
770 		 * the timer code needs to know if the virtual timer
771 		 * interrupts are active.
772 		 */
773 		kvm_vgic_sync_hwstate(vcpu);
774 
775 		/*
776 		 * Sync the timer hardware state before enabling interrupts as
777 		 * we don't want vtimer interrupts to race with syncing the
778 		 * timer virtual interrupt state.
779 		 */
780 		if (static_branch_unlikely(&userspace_irqchip_in_use))
781 			kvm_timer_sync_hwstate(vcpu);
782 
783 		kvm_arch_vcpu_ctxsync_fp(vcpu);
784 
785 		/*
786 		 * We may have taken a host interrupt in HYP mode (ie
787 		 * while executing the guest). This interrupt is still
788 		 * pending, as we haven't serviced it yet!
789 		 *
790 		 * We're now back in SVC mode, with interrupts
791 		 * disabled.  Enabling the interrupts now will have
792 		 * the effect of taking the interrupt again, in SVC
793 		 * mode this time.
794 		 */
795 		local_irq_enable();
796 
797 		/*
798 		 * We do local_irq_enable() before calling guest_exit() so
799 		 * that if a timer interrupt hits while running the guest we
800 		 * account that tick as being spent in the guest.  We enable
801 		 * preemption after calling guest_exit() so that if we get
802 		 * preempted we make sure ticks after that is not counted as
803 		 * guest time.
804 		 */
805 		guest_exit();
806 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
807 
808 		/* Exit types that need handling before we can be preempted */
809 		handle_exit_early(vcpu, run, ret);
810 
811 		preempt_enable();
812 
813 		ret = handle_exit(vcpu, run, ret);
814 	}
815 
816 	/* Tell userspace about in-kernel device output levels */
817 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
818 		kvm_timer_update_run(vcpu);
819 		kvm_pmu_update_run(vcpu);
820 	}
821 
822 	kvm_sigset_deactivate(vcpu);
823 
824 	vcpu_put(vcpu);
825 	return ret;
826 }
827 
828 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
829 {
830 	int bit_index;
831 	bool set;
832 	unsigned long *hcr;
833 
834 	if (number == KVM_ARM_IRQ_CPU_IRQ)
835 		bit_index = __ffs(HCR_VI);
836 	else /* KVM_ARM_IRQ_CPU_FIQ */
837 		bit_index = __ffs(HCR_VF);
838 
839 	hcr = vcpu_hcr(vcpu);
840 	if (level)
841 		set = test_and_set_bit(bit_index, hcr);
842 	else
843 		set = test_and_clear_bit(bit_index, hcr);
844 
845 	/*
846 	 * If we didn't change anything, no need to wake up or kick other CPUs
847 	 */
848 	if (set == level)
849 		return 0;
850 
851 	/*
852 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
853 	 * trigger a world-switch round on the running physical CPU to set the
854 	 * virtual IRQ/FIQ fields in the HCR appropriately.
855 	 */
856 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
857 	kvm_vcpu_kick(vcpu);
858 
859 	return 0;
860 }
861 
862 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
863 			  bool line_status)
864 {
865 	u32 irq = irq_level->irq;
866 	unsigned int irq_type, vcpu_idx, irq_num;
867 	int nrcpus = atomic_read(&kvm->online_vcpus);
868 	struct kvm_vcpu *vcpu = NULL;
869 	bool level = irq_level->level;
870 
871 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
872 	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
873 	vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
874 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
875 
876 	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
877 
878 	switch (irq_type) {
879 	case KVM_ARM_IRQ_TYPE_CPU:
880 		if (irqchip_in_kernel(kvm))
881 			return -ENXIO;
882 
883 		if (vcpu_idx >= nrcpus)
884 			return -EINVAL;
885 
886 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
887 		if (!vcpu)
888 			return -EINVAL;
889 
890 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
891 			return -EINVAL;
892 
893 		return vcpu_interrupt_line(vcpu, irq_num, level);
894 	case KVM_ARM_IRQ_TYPE_PPI:
895 		if (!irqchip_in_kernel(kvm))
896 			return -ENXIO;
897 
898 		if (vcpu_idx >= nrcpus)
899 			return -EINVAL;
900 
901 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
902 		if (!vcpu)
903 			return -EINVAL;
904 
905 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
906 			return -EINVAL;
907 
908 		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
909 	case KVM_ARM_IRQ_TYPE_SPI:
910 		if (!irqchip_in_kernel(kvm))
911 			return -ENXIO;
912 
913 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
914 			return -EINVAL;
915 
916 		return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
917 	}
918 
919 	return -EINVAL;
920 }
921 
922 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
923 			       const struct kvm_vcpu_init *init)
924 {
925 	unsigned int i, ret;
926 	int phys_target = kvm_target_cpu();
927 
928 	if (init->target != phys_target)
929 		return -EINVAL;
930 
931 	/*
932 	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
933 	 * use the same target.
934 	 */
935 	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
936 		return -EINVAL;
937 
938 	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
939 	for (i = 0; i < sizeof(init->features) * 8; i++) {
940 		bool set = (init->features[i / 32] & (1 << (i % 32)));
941 
942 		if (set && i >= KVM_VCPU_MAX_FEATURES)
943 			return -ENOENT;
944 
945 		/*
946 		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
947 		 * use the same feature set.
948 		 */
949 		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
950 		    test_bit(i, vcpu->arch.features) != set)
951 			return -EINVAL;
952 
953 		if (set)
954 			set_bit(i, vcpu->arch.features);
955 	}
956 
957 	vcpu->arch.target = phys_target;
958 
959 	/* Now we know what it is, we can reset it. */
960 	ret = kvm_reset_vcpu(vcpu);
961 	if (ret) {
962 		vcpu->arch.target = -1;
963 		bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
964 	}
965 
966 	return ret;
967 }
968 
969 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
970 					 struct kvm_vcpu_init *init)
971 {
972 	int ret;
973 
974 	ret = kvm_vcpu_set_target(vcpu, init);
975 	if (ret)
976 		return ret;
977 
978 	/*
979 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
980 	 * guest MMU is turned off and flush the caches as needed.
981 	 *
982 	 * S2FWB enforces all memory accesses to RAM being cacheable,
983 	 * ensuring that the data side is always coherent. We still
984 	 * need to invalidate the I-cache though, as FWB does *not*
985 	 * imply CTR_EL0.DIC.
986 	 */
987 	if (vcpu->arch.has_run_once) {
988 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
989 			stage2_unmap_vm(vcpu->kvm);
990 		else
991 			__flush_icache_all();
992 	}
993 
994 	vcpu_reset_hcr(vcpu);
995 
996 	/*
997 	 * Handle the "start in power-off" case.
998 	 */
999 	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1000 		vcpu_power_off(vcpu);
1001 	else
1002 		vcpu->arch.power_off = false;
1003 
1004 	return 0;
1005 }
1006 
1007 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1008 				 struct kvm_device_attr *attr)
1009 {
1010 	int ret = -ENXIO;
1011 
1012 	switch (attr->group) {
1013 	default:
1014 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1015 		break;
1016 	}
1017 
1018 	return ret;
1019 }
1020 
1021 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1022 				 struct kvm_device_attr *attr)
1023 {
1024 	int ret = -ENXIO;
1025 
1026 	switch (attr->group) {
1027 	default:
1028 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1029 		break;
1030 	}
1031 
1032 	return ret;
1033 }
1034 
1035 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1036 				 struct kvm_device_attr *attr)
1037 {
1038 	int ret = -ENXIO;
1039 
1040 	switch (attr->group) {
1041 	default:
1042 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1043 		break;
1044 	}
1045 
1046 	return ret;
1047 }
1048 
1049 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1050 				   struct kvm_vcpu_events *events)
1051 {
1052 	memset(events, 0, sizeof(*events));
1053 
1054 	return __kvm_arm_vcpu_get_events(vcpu, events);
1055 }
1056 
1057 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1058 				   struct kvm_vcpu_events *events)
1059 {
1060 	int i;
1061 
1062 	/* check whether the reserved field is zero */
1063 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1064 		if (events->reserved[i])
1065 			return -EINVAL;
1066 
1067 	/* check whether the pad field is zero */
1068 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1069 		if (events->exception.pad[i])
1070 			return -EINVAL;
1071 
1072 	return __kvm_arm_vcpu_set_events(vcpu, events);
1073 }
1074 
1075 long kvm_arch_vcpu_ioctl(struct file *filp,
1076 			 unsigned int ioctl, unsigned long arg)
1077 {
1078 	struct kvm_vcpu *vcpu = filp->private_data;
1079 	void __user *argp = (void __user *)arg;
1080 	struct kvm_device_attr attr;
1081 	long r;
1082 
1083 	switch (ioctl) {
1084 	case KVM_ARM_VCPU_INIT: {
1085 		struct kvm_vcpu_init init;
1086 
1087 		r = -EFAULT;
1088 		if (copy_from_user(&init, argp, sizeof(init)))
1089 			break;
1090 
1091 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1092 		break;
1093 	}
1094 	case KVM_SET_ONE_REG:
1095 	case KVM_GET_ONE_REG: {
1096 		struct kvm_one_reg reg;
1097 
1098 		r = -ENOEXEC;
1099 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1100 			break;
1101 
1102 		r = -EFAULT;
1103 		if (copy_from_user(&reg, argp, sizeof(reg)))
1104 			break;
1105 
1106 		if (ioctl == KVM_SET_ONE_REG)
1107 			r = kvm_arm_set_reg(vcpu, &reg);
1108 		else
1109 			r = kvm_arm_get_reg(vcpu, &reg);
1110 		break;
1111 	}
1112 	case KVM_GET_REG_LIST: {
1113 		struct kvm_reg_list __user *user_list = argp;
1114 		struct kvm_reg_list reg_list;
1115 		unsigned n;
1116 
1117 		r = -ENOEXEC;
1118 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1119 			break;
1120 
1121 		r = -EPERM;
1122 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1123 			break;
1124 
1125 		r = -EFAULT;
1126 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1127 			break;
1128 		n = reg_list.n;
1129 		reg_list.n = kvm_arm_num_regs(vcpu);
1130 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1131 			break;
1132 		r = -E2BIG;
1133 		if (n < reg_list.n)
1134 			break;
1135 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1136 		break;
1137 	}
1138 	case KVM_SET_DEVICE_ATTR: {
1139 		r = -EFAULT;
1140 		if (copy_from_user(&attr, argp, sizeof(attr)))
1141 			break;
1142 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1143 		break;
1144 	}
1145 	case KVM_GET_DEVICE_ATTR: {
1146 		r = -EFAULT;
1147 		if (copy_from_user(&attr, argp, sizeof(attr)))
1148 			break;
1149 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1150 		break;
1151 	}
1152 	case KVM_HAS_DEVICE_ATTR: {
1153 		r = -EFAULT;
1154 		if (copy_from_user(&attr, argp, sizeof(attr)))
1155 			break;
1156 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1157 		break;
1158 	}
1159 	case KVM_GET_VCPU_EVENTS: {
1160 		struct kvm_vcpu_events events;
1161 
1162 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1163 			return -EINVAL;
1164 
1165 		if (copy_to_user(argp, &events, sizeof(events)))
1166 			return -EFAULT;
1167 
1168 		return 0;
1169 	}
1170 	case KVM_SET_VCPU_EVENTS: {
1171 		struct kvm_vcpu_events events;
1172 
1173 		if (copy_from_user(&events, argp, sizeof(events)))
1174 			return -EFAULT;
1175 
1176 		return kvm_arm_vcpu_set_events(vcpu, &events);
1177 	}
1178 	case KVM_ARM_VCPU_FINALIZE: {
1179 		int what;
1180 
1181 		if (!kvm_vcpu_initialized(vcpu))
1182 			return -ENOEXEC;
1183 
1184 		if (get_user(what, (const int __user *)argp))
1185 			return -EFAULT;
1186 
1187 		return kvm_arm_vcpu_finalize(vcpu, what);
1188 	}
1189 	default:
1190 		r = -EINVAL;
1191 	}
1192 
1193 	return r;
1194 }
1195 
1196 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1197 {
1198 
1199 }
1200 
1201 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1202 					struct kvm_memory_slot *memslot)
1203 {
1204 	kvm_flush_remote_tlbs(kvm);
1205 }
1206 
1207 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1208 					struct kvm_arm_device_addr *dev_addr)
1209 {
1210 	unsigned long dev_id, type;
1211 
1212 	dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1213 		KVM_ARM_DEVICE_ID_SHIFT;
1214 	type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1215 		KVM_ARM_DEVICE_TYPE_SHIFT;
1216 
1217 	switch (dev_id) {
1218 	case KVM_ARM_DEVICE_VGIC_V2:
1219 		if (!vgic_present)
1220 			return -ENXIO;
1221 		return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1222 	default:
1223 		return -ENODEV;
1224 	}
1225 }
1226 
1227 long kvm_arch_vm_ioctl(struct file *filp,
1228 		       unsigned int ioctl, unsigned long arg)
1229 {
1230 	struct kvm *kvm = filp->private_data;
1231 	void __user *argp = (void __user *)arg;
1232 
1233 	switch (ioctl) {
1234 	case KVM_CREATE_IRQCHIP: {
1235 		int ret;
1236 		if (!vgic_present)
1237 			return -ENXIO;
1238 		mutex_lock(&kvm->lock);
1239 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1240 		mutex_unlock(&kvm->lock);
1241 		return ret;
1242 	}
1243 	case KVM_ARM_SET_DEVICE_ADDR: {
1244 		struct kvm_arm_device_addr dev_addr;
1245 
1246 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1247 			return -EFAULT;
1248 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1249 	}
1250 	case KVM_ARM_PREFERRED_TARGET: {
1251 		int err;
1252 		struct kvm_vcpu_init init;
1253 
1254 		err = kvm_vcpu_preferred_target(&init);
1255 		if (err)
1256 			return err;
1257 
1258 		if (copy_to_user(argp, &init, sizeof(init)))
1259 			return -EFAULT;
1260 
1261 		return 0;
1262 	}
1263 	default:
1264 		return -EINVAL;
1265 	}
1266 }
1267 
1268 static void cpu_init_hyp_mode(void)
1269 {
1270 	phys_addr_t pgd_ptr;
1271 	unsigned long hyp_stack_ptr;
1272 	unsigned long vector_ptr;
1273 	unsigned long tpidr_el2;
1274 
1275 	/* Switch from the HYP stub to our own HYP init vector */
1276 	__hyp_set_vectors(kvm_get_idmap_vector());
1277 
1278 	/*
1279 	 * Calculate the raw per-cpu offset without a translation from the
1280 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1281 	 * so that we can use adr_l to access per-cpu variables in EL2.
1282 	 */
1283 	tpidr_el2 = ((unsigned long)this_cpu_ptr(&kvm_host_data) -
1284 		     (unsigned long)kvm_ksym_ref(&kvm_host_data));
1285 
1286 	pgd_ptr = kvm_mmu_get_httbr();
1287 	hyp_stack_ptr = __this_cpu_read(kvm_arm_hyp_stack_page) + PAGE_SIZE;
1288 	vector_ptr = (unsigned long)kvm_get_hyp_vector();
1289 
1290 	/*
1291 	 * Call initialization code, and switch to the full blown HYP code.
1292 	 * If the cpucaps haven't been finalized yet, something has gone very
1293 	 * wrong, and hyp will crash and burn when it uses any
1294 	 * cpus_have_const_cap() wrapper.
1295 	 */
1296 	BUG_ON(!system_capabilities_finalized());
1297 	__kvm_call_hyp((void *)pgd_ptr, hyp_stack_ptr, vector_ptr, tpidr_el2);
1298 
1299 	/*
1300 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1301 	 * at EL2.
1302 	 */
1303 	if (this_cpu_has_cap(ARM64_SSBS) &&
1304 	    arm64_get_ssbd_state() == ARM64_SSBD_FORCE_DISABLE) {
1305 		kvm_call_hyp(__kvm_enable_ssbs);
1306 	}
1307 }
1308 
1309 static void cpu_hyp_reset(void)
1310 {
1311 	if (!is_kernel_in_hyp_mode())
1312 		__hyp_reset_vectors();
1313 }
1314 
1315 static void cpu_hyp_reinit(void)
1316 {
1317 	kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1318 
1319 	cpu_hyp_reset();
1320 
1321 	if (is_kernel_in_hyp_mode())
1322 		kvm_timer_init_vhe();
1323 	else
1324 		cpu_init_hyp_mode();
1325 
1326 	kvm_arm_init_debug();
1327 
1328 	if (vgic_present)
1329 		kvm_vgic_init_cpu_hardware();
1330 }
1331 
1332 static void _kvm_arch_hardware_enable(void *discard)
1333 {
1334 	if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1335 		cpu_hyp_reinit();
1336 		__this_cpu_write(kvm_arm_hardware_enabled, 1);
1337 	}
1338 }
1339 
1340 int kvm_arch_hardware_enable(void)
1341 {
1342 	_kvm_arch_hardware_enable(NULL);
1343 	return 0;
1344 }
1345 
1346 static void _kvm_arch_hardware_disable(void *discard)
1347 {
1348 	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1349 		cpu_hyp_reset();
1350 		__this_cpu_write(kvm_arm_hardware_enabled, 0);
1351 	}
1352 }
1353 
1354 void kvm_arch_hardware_disable(void)
1355 {
1356 	_kvm_arch_hardware_disable(NULL);
1357 }
1358 
1359 #ifdef CONFIG_CPU_PM
1360 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1361 				    unsigned long cmd,
1362 				    void *v)
1363 {
1364 	/*
1365 	 * kvm_arm_hardware_enabled is left with its old value over
1366 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1367 	 * re-enable hyp.
1368 	 */
1369 	switch (cmd) {
1370 	case CPU_PM_ENTER:
1371 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1372 			/*
1373 			 * don't update kvm_arm_hardware_enabled here
1374 			 * so that the hardware will be re-enabled
1375 			 * when we resume. See below.
1376 			 */
1377 			cpu_hyp_reset();
1378 
1379 		return NOTIFY_OK;
1380 	case CPU_PM_ENTER_FAILED:
1381 	case CPU_PM_EXIT:
1382 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1383 			/* The hardware was enabled before suspend. */
1384 			cpu_hyp_reinit();
1385 
1386 		return NOTIFY_OK;
1387 
1388 	default:
1389 		return NOTIFY_DONE;
1390 	}
1391 }
1392 
1393 static struct notifier_block hyp_init_cpu_pm_nb = {
1394 	.notifier_call = hyp_init_cpu_pm_notifier,
1395 };
1396 
1397 static void __init hyp_cpu_pm_init(void)
1398 {
1399 	cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1400 }
1401 static void __init hyp_cpu_pm_exit(void)
1402 {
1403 	cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1404 }
1405 #else
1406 static inline void hyp_cpu_pm_init(void)
1407 {
1408 }
1409 static inline void hyp_cpu_pm_exit(void)
1410 {
1411 }
1412 #endif
1413 
1414 static int init_common_resources(void)
1415 {
1416 	return kvm_set_ipa_limit();
1417 }
1418 
1419 static int init_subsystems(void)
1420 {
1421 	int err = 0;
1422 
1423 	/*
1424 	 * Enable hardware so that subsystem initialisation can access EL2.
1425 	 */
1426 	on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1427 
1428 	/*
1429 	 * Register CPU lower-power notifier
1430 	 */
1431 	hyp_cpu_pm_init();
1432 
1433 	/*
1434 	 * Init HYP view of VGIC
1435 	 */
1436 	err = kvm_vgic_hyp_init();
1437 	switch (err) {
1438 	case 0:
1439 		vgic_present = true;
1440 		break;
1441 	case -ENODEV:
1442 	case -ENXIO:
1443 		vgic_present = false;
1444 		err = 0;
1445 		break;
1446 	default:
1447 		goto out;
1448 	}
1449 
1450 	/*
1451 	 * Init HYP architected timer support
1452 	 */
1453 	err = kvm_timer_hyp_init(vgic_present);
1454 	if (err)
1455 		goto out;
1456 
1457 	kvm_perf_init();
1458 	kvm_coproc_table_init();
1459 
1460 out:
1461 	on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1462 
1463 	return err;
1464 }
1465 
1466 static void teardown_hyp_mode(void)
1467 {
1468 	int cpu;
1469 
1470 	free_hyp_pgds();
1471 	for_each_possible_cpu(cpu)
1472 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1473 }
1474 
1475 /**
1476  * Inits Hyp-mode on all online CPUs
1477  */
1478 static int init_hyp_mode(void)
1479 {
1480 	int cpu;
1481 	int err = 0;
1482 
1483 	/*
1484 	 * Allocate Hyp PGD and setup Hyp identity mapping
1485 	 */
1486 	err = kvm_mmu_init();
1487 	if (err)
1488 		goto out_err;
1489 
1490 	/*
1491 	 * Allocate stack pages for Hypervisor-mode
1492 	 */
1493 	for_each_possible_cpu(cpu) {
1494 		unsigned long stack_page;
1495 
1496 		stack_page = __get_free_page(GFP_KERNEL);
1497 		if (!stack_page) {
1498 			err = -ENOMEM;
1499 			goto out_err;
1500 		}
1501 
1502 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1503 	}
1504 
1505 	/*
1506 	 * Map the Hyp-code called directly from the host
1507 	 */
1508 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1509 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1510 	if (err) {
1511 		kvm_err("Cannot map world-switch code\n");
1512 		goto out_err;
1513 	}
1514 
1515 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1516 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1517 	if (err) {
1518 		kvm_err("Cannot map rodata section\n");
1519 		goto out_err;
1520 	}
1521 
1522 	err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1523 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1524 	if (err) {
1525 		kvm_err("Cannot map bss section\n");
1526 		goto out_err;
1527 	}
1528 
1529 	err = kvm_map_vectors();
1530 	if (err) {
1531 		kvm_err("Cannot map vectors\n");
1532 		goto out_err;
1533 	}
1534 
1535 	/*
1536 	 * Map the Hyp stack pages
1537 	 */
1538 	for_each_possible_cpu(cpu) {
1539 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1540 		err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1541 					  PAGE_HYP);
1542 
1543 		if (err) {
1544 			kvm_err("Cannot map hyp stack\n");
1545 			goto out_err;
1546 		}
1547 	}
1548 
1549 	for_each_possible_cpu(cpu) {
1550 		kvm_host_data_t *cpu_data;
1551 
1552 		cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1553 		err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1554 
1555 		if (err) {
1556 			kvm_err("Cannot map host CPU state: %d\n", err);
1557 			goto out_err;
1558 		}
1559 	}
1560 
1561 	err = hyp_map_aux_data();
1562 	if (err)
1563 		kvm_err("Cannot map host auxiliary data: %d\n", err);
1564 
1565 	return 0;
1566 
1567 out_err:
1568 	teardown_hyp_mode();
1569 	kvm_err("error initializing Hyp mode: %d\n", err);
1570 	return err;
1571 }
1572 
1573 static void check_kvm_target_cpu(void *ret)
1574 {
1575 	*(int *)ret = kvm_target_cpu();
1576 }
1577 
1578 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1579 {
1580 	struct kvm_vcpu *vcpu;
1581 	int i;
1582 
1583 	mpidr &= MPIDR_HWID_BITMASK;
1584 	kvm_for_each_vcpu(i, vcpu, kvm) {
1585 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1586 			return vcpu;
1587 	}
1588 	return NULL;
1589 }
1590 
1591 bool kvm_arch_has_irq_bypass(void)
1592 {
1593 	return true;
1594 }
1595 
1596 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1597 				      struct irq_bypass_producer *prod)
1598 {
1599 	struct kvm_kernel_irqfd *irqfd =
1600 		container_of(cons, struct kvm_kernel_irqfd, consumer);
1601 
1602 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1603 					  &irqfd->irq_entry);
1604 }
1605 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1606 				      struct irq_bypass_producer *prod)
1607 {
1608 	struct kvm_kernel_irqfd *irqfd =
1609 		container_of(cons, struct kvm_kernel_irqfd, consumer);
1610 
1611 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1612 				     &irqfd->irq_entry);
1613 }
1614 
1615 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1616 {
1617 	struct kvm_kernel_irqfd *irqfd =
1618 		container_of(cons, struct kvm_kernel_irqfd, consumer);
1619 
1620 	kvm_arm_halt_guest(irqfd->kvm);
1621 }
1622 
1623 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1624 {
1625 	struct kvm_kernel_irqfd *irqfd =
1626 		container_of(cons, struct kvm_kernel_irqfd, consumer);
1627 
1628 	kvm_arm_resume_guest(irqfd->kvm);
1629 }
1630 
1631 /**
1632  * Initialize Hyp-mode and memory mappings on all CPUs.
1633  */
1634 int kvm_arch_init(void *opaque)
1635 {
1636 	int err;
1637 	int ret, cpu;
1638 	bool in_hyp_mode;
1639 
1640 	if (!is_hyp_mode_available()) {
1641 		kvm_info("HYP mode not available\n");
1642 		return -ENODEV;
1643 	}
1644 
1645 	in_hyp_mode = is_kernel_in_hyp_mode();
1646 
1647 	if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1648 		kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1649 		return -ENODEV;
1650 	}
1651 
1652 	for_each_online_cpu(cpu) {
1653 		smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1654 		if (ret < 0) {
1655 			kvm_err("Error, CPU %d not supported!\n", cpu);
1656 			return -ENODEV;
1657 		}
1658 	}
1659 
1660 	err = init_common_resources();
1661 	if (err)
1662 		return err;
1663 
1664 	err = kvm_arm_init_sve();
1665 	if (err)
1666 		return err;
1667 
1668 	if (!in_hyp_mode) {
1669 		err = init_hyp_mode();
1670 		if (err)
1671 			goto out_err;
1672 	}
1673 
1674 	err = init_subsystems();
1675 	if (err)
1676 		goto out_hyp;
1677 
1678 	if (in_hyp_mode)
1679 		kvm_info("VHE mode initialized successfully\n");
1680 	else
1681 		kvm_info("Hyp mode initialized successfully\n");
1682 
1683 	return 0;
1684 
1685 out_hyp:
1686 	hyp_cpu_pm_exit();
1687 	if (!in_hyp_mode)
1688 		teardown_hyp_mode();
1689 out_err:
1690 	return err;
1691 }
1692 
1693 /* NOP: Compiling as a module not supported */
1694 void kvm_arch_exit(void)
1695 {
1696 	kvm_perf_teardown();
1697 }
1698 
1699 static int arm_init(void)
1700 {
1701 	int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1702 	return rc;
1703 }
1704 
1705 module_init(arm_init);
1706