xref: /openbmc/linux/arch/arm64/kvm/arm.c (revision a4767912)
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 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
50 
51 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
52 unsigned long kvm_arm_hyp_percpu_base[NR_CPUS];
53 
54 /* The VMID used in the VTTBR */
55 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
56 static u32 kvm_next_vmid;
57 static DEFINE_SPINLOCK(kvm_vmid_lock);
58 
59 static bool vgic_present;
60 
61 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
62 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
63 
64 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
65 {
66 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
67 }
68 
69 int kvm_arch_hardware_setup(void *opaque)
70 {
71 	return 0;
72 }
73 
74 int kvm_arch_check_processor_compat(void *opaque)
75 {
76 	return 0;
77 }
78 
79 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
80 			    struct kvm_enable_cap *cap)
81 {
82 	int r;
83 
84 	if (cap->flags)
85 		return -EINVAL;
86 
87 	switch (cap->cap) {
88 	case KVM_CAP_ARM_NISV_TO_USER:
89 		r = 0;
90 		kvm->arch.return_nisv_io_abort_to_user = true;
91 		break;
92 	default:
93 		r = -EINVAL;
94 		break;
95 	}
96 
97 	return r;
98 }
99 
100 static int kvm_arm_default_max_vcpus(void)
101 {
102 	return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
103 }
104 
105 static void set_default_csv2(struct kvm *kvm)
106 {
107 	/*
108 	 * The default is to expose CSV2 == 1 if the HW isn't affected.
109 	 * Although this is a per-CPU feature, we make it global because
110 	 * asymmetric systems are just a nuisance.
111 	 *
112 	 * Userspace can override this as long as it doesn't promise
113 	 * the impossible.
114 	 */
115 	if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
116 		kvm->arch.pfr0_csv2 = 1;
117 }
118 
119 /**
120  * kvm_arch_init_vm - initializes a VM data structure
121  * @kvm:	pointer to the KVM struct
122  */
123 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
124 {
125 	int ret;
126 
127 	ret = kvm_arm_setup_stage2(kvm, type);
128 	if (ret)
129 		return ret;
130 
131 	ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
132 	if (ret)
133 		return ret;
134 
135 	ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
136 	if (ret)
137 		goto out_free_stage2_pgd;
138 
139 	kvm_vgic_early_init(kvm);
140 
141 	/* The maximum number of VCPUs is limited by the host's GIC model */
142 	kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
143 
144 	set_default_csv2(kvm);
145 
146 	return ret;
147 out_free_stage2_pgd:
148 	kvm_free_stage2_pgd(&kvm->arch.mmu);
149 	return ret;
150 }
151 
152 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
153 {
154 	return VM_FAULT_SIGBUS;
155 }
156 
157 
158 /**
159  * kvm_arch_destroy_vm - destroy the VM data structure
160  * @kvm:	pointer to the KVM struct
161  */
162 void kvm_arch_destroy_vm(struct kvm *kvm)
163 {
164 	int i;
165 
166 	bitmap_free(kvm->arch.pmu_filter);
167 
168 	kvm_vgic_destroy(kvm);
169 
170 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
171 		if (kvm->vcpus[i]) {
172 			kvm_vcpu_destroy(kvm->vcpus[i]);
173 			kvm->vcpus[i] = NULL;
174 		}
175 	}
176 	atomic_set(&kvm->online_vcpus, 0);
177 }
178 
179 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
180 {
181 	int r;
182 	switch (ext) {
183 	case KVM_CAP_IRQCHIP:
184 		r = vgic_present;
185 		break;
186 	case KVM_CAP_IOEVENTFD:
187 	case KVM_CAP_DEVICE_CTRL:
188 	case KVM_CAP_USER_MEMORY:
189 	case KVM_CAP_SYNC_MMU:
190 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
191 	case KVM_CAP_ONE_REG:
192 	case KVM_CAP_ARM_PSCI:
193 	case KVM_CAP_ARM_PSCI_0_2:
194 	case KVM_CAP_READONLY_MEM:
195 	case KVM_CAP_MP_STATE:
196 	case KVM_CAP_IMMEDIATE_EXIT:
197 	case KVM_CAP_VCPU_EVENTS:
198 	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
199 	case KVM_CAP_ARM_NISV_TO_USER:
200 	case KVM_CAP_ARM_INJECT_EXT_DABT:
201 		r = 1;
202 		break;
203 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
204 		r = 1;
205 		break;
206 	case KVM_CAP_NR_VCPUS:
207 		r = num_online_cpus();
208 		break;
209 	case KVM_CAP_MAX_VCPUS:
210 	case KVM_CAP_MAX_VCPU_ID:
211 		if (kvm)
212 			r = kvm->arch.max_vcpus;
213 		else
214 			r = kvm_arm_default_max_vcpus();
215 		break;
216 	case KVM_CAP_MSI_DEVID:
217 		if (!kvm)
218 			r = -EINVAL;
219 		else
220 			r = kvm->arch.vgic.msis_require_devid;
221 		break;
222 	case KVM_CAP_ARM_USER_IRQ:
223 		/*
224 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
225 		 * (bump this number if adding more devices)
226 		 */
227 		r = 1;
228 		break;
229 	case KVM_CAP_STEAL_TIME:
230 		r = kvm_arm_pvtime_supported();
231 		break;
232 	default:
233 		r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
234 		break;
235 	}
236 	return r;
237 }
238 
239 long kvm_arch_dev_ioctl(struct file *filp,
240 			unsigned int ioctl, unsigned long arg)
241 {
242 	return -EINVAL;
243 }
244 
245 struct kvm *kvm_arch_alloc_vm(void)
246 {
247 	if (!has_vhe())
248 		return kzalloc(sizeof(struct kvm), GFP_KERNEL);
249 
250 	return vzalloc(sizeof(struct kvm));
251 }
252 
253 void kvm_arch_free_vm(struct kvm *kvm)
254 {
255 	if (!has_vhe())
256 		kfree(kvm);
257 	else
258 		vfree(kvm);
259 }
260 
261 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
262 {
263 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
264 		return -EBUSY;
265 
266 	if (id >= kvm->arch.max_vcpus)
267 		return -EINVAL;
268 
269 	return 0;
270 }
271 
272 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
273 {
274 	int err;
275 
276 	/* Force users to call KVM_ARM_VCPU_INIT */
277 	vcpu->arch.target = -1;
278 	bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
279 
280 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
281 
282 	/* Set up the timer */
283 	kvm_timer_vcpu_init(vcpu);
284 
285 	kvm_pmu_vcpu_init(vcpu);
286 
287 	kvm_arm_reset_debug_ptr(vcpu);
288 
289 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
290 
291 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
292 
293 	err = kvm_vgic_vcpu_init(vcpu);
294 	if (err)
295 		return err;
296 
297 	return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
298 }
299 
300 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
301 {
302 }
303 
304 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
305 {
306 	if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
307 		static_branch_dec(&userspace_irqchip_in_use);
308 
309 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
310 	kvm_timer_vcpu_terminate(vcpu);
311 	kvm_pmu_vcpu_destroy(vcpu);
312 
313 	kvm_arm_vcpu_destroy(vcpu);
314 }
315 
316 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
317 {
318 	return kvm_timer_is_pending(vcpu);
319 }
320 
321 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
322 {
323 	/*
324 	 * If we're about to block (most likely because we've just hit a
325 	 * WFI), we need to sync back the state of the GIC CPU interface
326 	 * so that we have the latest PMR and group enables. This ensures
327 	 * that kvm_arch_vcpu_runnable has up-to-date data to decide
328 	 * whether we have pending interrupts.
329 	 *
330 	 * For the same reason, we want to tell GICv4 that we need
331 	 * doorbells to be signalled, should an interrupt become pending.
332 	 */
333 	preempt_disable();
334 	kvm_vgic_vmcr_sync(vcpu);
335 	vgic_v4_put(vcpu, true);
336 	preempt_enable();
337 }
338 
339 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
340 {
341 	preempt_disable();
342 	vgic_v4_load(vcpu);
343 	preempt_enable();
344 }
345 
346 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
347 {
348 	struct kvm_s2_mmu *mmu;
349 	int *last_ran;
350 
351 	mmu = vcpu->arch.hw_mmu;
352 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
353 
354 	/*
355 	 * We might get preempted before the vCPU actually runs, but
356 	 * over-invalidation doesn't affect correctness.
357 	 */
358 	if (*last_ran != vcpu->vcpu_id) {
359 		kvm_call_hyp(__kvm_tlb_flush_local_vmid, mmu);
360 		*last_ran = vcpu->vcpu_id;
361 	}
362 
363 	vcpu->cpu = cpu;
364 
365 	kvm_vgic_load(vcpu);
366 	kvm_timer_vcpu_load(vcpu);
367 	if (has_vhe())
368 		kvm_vcpu_load_sysregs_vhe(vcpu);
369 	kvm_arch_vcpu_load_fp(vcpu);
370 	kvm_vcpu_pmu_restore_guest(vcpu);
371 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
372 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
373 
374 	if (single_task_running())
375 		vcpu_clear_wfx_traps(vcpu);
376 	else
377 		vcpu_set_wfx_traps(vcpu);
378 
379 	if (vcpu_has_ptrauth(vcpu))
380 		vcpu_ptrauth_disable(vcpu);
381 }
382 
383 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
384 {
385 	kvm_arch_vcpu_put_fp(vcpu);
386 	if (has_vhe())
387 		kvm_vcpu_put_sysregs_vhe(vcpu);
388 	kvm_timer_vcpu_put(vcpu);
389 	kvm_vgic_put(vcpu);
390 	kvm_vcpu_pmu_restore_host(vcpu);
391 
392 	vcpu->cpu = -1;
393 }
394 
395 static void vcpu_power_off(struct kvm_vcpu *vcpu)
396 {
397 	vcpu->arch.power_off = true;
398 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
399 	kvm_vcpu_kick(vcpu);
400 }
401 
402 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
403 				    struct kvm_mp_state *mp_state)
404 {
405 	if (vcpu->arch.power_off)
406 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
407 	else
408 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
409 
410 	return 0;
411 }
412 
413 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
414 				    struct kvm_mp_state *mp_state)
415 {
416 	int ret = 0;
417 
418 	switch (mp_state->mp_state) {
419 	case KVM_MP_STATE_RUNNABLE:
420 		vcpu->arch.power_off = false;
421 		break;
422 	case KVM_MP_STATE_STOPPED:
423 		vcpu_power_off(vcpu);
424 		break;
425 	default:
426 		ret = -EINVAL;
427 	}
428 
429 	return ret;
430 }
431 
432 /**
433  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
434  * @v:		The VCPU pointer
435  *
436  * If the guest CPU is not waiting for interrupts or an interrupt line is
437  * asserted, the CPU is by definition runnable.
438  */
439 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
440 {
441 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
442 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
443 		&& !v->arch.power_off && !v->arch.pause);
444 }
445 
446 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
447 {
448 	return vcpu_mode_priv(vcpu);
449 }
450 
451 /* Just ensure a guest exit from a particular CPU */
452 static void exit_vm_noop(void *info)
453 {
454 }
455 
456 void force_vm_exit(const cpumask_t *mask)
457 {
458 	preempt_disable();
459 	smp_call_function_many(mask, exit_vm_noop, NULL, true);
460 	preempt_enable();
461 }
462 
463 /**
464  * need_new_vmid_gen - check that the VMID is still valid
465  * @vmid: The VMID to check
466  *
467  * return true if there is a new generation of VMIDs being used
468  *
469  * The hardware supports a limited set of values with the value zero reserved
470  * for the host, so we check if an assigned value belongs to a previous
471  * generation, which requires us to assign a new value. If we're the first to
472  * use a VMID for the new generation, we must flush necessary caches and TLBs
473  * on all CPUs.
474  */
475 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
476 {
477 	u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
478 	smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
479 	return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
480 }
481 
482 /**
483  * update_vmid - Update the vmid with a valid VMID for the current generation
484  * @vmid: The stage-2 VMID information struct
485  */
486 static void update_vmid(struct kvm_vmid *vmid)
487 {
488 	if (!need_new_vmid_gen(vmid))
489 		return;
490 
491 	spin_lock(&kvm_vmid_lock);
492 
493 	/*
494 	 * We need to re-check the vmid_gen here to ensure that if another vcpu
495 	 * already allocated a valid vmid for this vm, then this vcpu should
496 	 * use the same vmid.
497 	 */
498 	if (!need_new_vmid_gen(vmid)) {
499 		spin_unlock(&kvm_vmid_lock);
500 		return;
501 	}
502 
503 	/* First user of a new VMID generation? */
504 	if (unlikely(kvm_next_vmid == 0)) {
505 		atomic64_inc(&kvm_vmid_gen);
506 		kvm_next_vmid = 1;
507 
508 		/*
509 		 * On SMP we know no other CPUs can use this CPU's or each
510 		 * other's VMID after force_vm_exit returns since the
511 		 * kvm_vmid_lock blocks them from reentry to the guest.
512 		 */
513 		force_vm_exit(cpu_all_mask);
514 		/*
515 		 * Now broadcast TLB + ICACHE invalidation over the inner
516 		 * shareable domain to make sure all data structures are
517 		 * clean.
518 		 */
519 		kvm_call_hyp(__kvm_flush_vm_context);
520 	}
521 
522 	vmid->vmid = kvm_next_vmid;
523 	kvm_next_vmid++;
524 	kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
525 
526 	smp_wmb();
527 	WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
528 
529 	spin_unlock(&kvm_vmid_lock);
530 }
531 
532 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
533 {
534 	struct kvm *kvm = vcpu->kvm;
535 	int ret = 0;
536 
537 	if (likely(vcpu->arch.has_run_once))
538 		return 0;
539 
540 	if (!kvm_arm_vcpu_is_finalized(vcpu))
541 		return -EPERM;
542 
543 	vcpu->arch.has_run_once = true;
544 
545 	if (likely(irqchip_in_kernel(kvm))) {
546 		/*
547 		 * Map the VGIC hardware resources before running a vcpu the
548 		 * first time on this VM.
549 		 */
550 		if (unlikely(!vgic_ready(kvm))) {
551 			ret = kvm_vgic_map_resources(kvm);
552 			if (ret)
553 				return ret;
554 		}
555 	} else {
556 		/*
557 		 * Tell the rest of the code that there are userspace irqchip
558 		 * VMs in the wild.
559 		 */
560 		static_branch_inc(&userspace_irqchip_in_use);
561 	}
562 
563 	ret = kvm_timer_enable(vcpu);
564 	if (ret)
565 		return ret;
566 
567 	ret = kvm_arm_pmu_v3_enable(vcpu);
568 
569 	return ret;
570 }
571 
572 bool kvm_arch_intc_initialized(struct kvm *kvm)
573 {
574 	return vgic_initialized(kvm);
575 }
576 
577 void kvm_arm_halt_guest(struct kvm *kvm)
578 {
579 	int i;
580 	struct kvm_vcpu *vcpu;
581 
582 	kvm_for_each_vcpu(i, vcpu, kvm)
583 		vcpu->arch.pause = true;
584 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
585 }
586 
587 void kvm_arm_resume_guest(struct kvm *kvm)
588 {
589 	int i;
590 	struct kvm_vcpu *vcpu;
591 
592 	kvm_for_each_vcpu(i, vcpu, kvm) {
593 		vcpu->arch.pause = false;
594 		rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
595 	}
596 }
597 
598 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
599 {
600 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
601 
602 	rcuwait_wait_event(wait,
603 			   (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
604 			   TASK_INTERRUPTIBLE);
605 
606 	if (vcpu->arch.power_off || vcpu->arch.pause) {
607 		/* Awaken to handle a signal, request we sleep again later. */
608 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
609 	}
610 
611 	/*
612 	 * Make sure we will observe a potential reset request if we've
613 	 * observed a change to the power state. Pairs with the smp_wmb() in
614 	 * kvm_psci_vcpu_on().
615 	 */
616 	smp_rmb();
617 }
618 
619 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
620 {
621 	return vcpu->arch.target >= 0;
622 }
623 
624 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
625 {
626 	if (kvm_request_pending(vcpu)) {
627 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
628 			vcpu_req_sleep(vcpu);
629 
630 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
631 			kvm_reset_vcpu(vcpu);
632 
633 		/*
634 		 * Clear IRQ_PENDING requests that were made to guarantee
635 		 * that a VCPU sees new virtual interrupts.
636 		 */
637 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
638 
639 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
640 			kvm_update_stolen_time(vcpu);
641 
642 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
643 			/* The distributor enable bits were changed */
644 			preempt_disable();
645 			vgic_v4_put(vcpu, false);
646 			vgic_v4_load(vcpu);
647 			preempt_enable();
648 		}
649 	}
650 }
651 
652 /**
653  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
654  * @vcpu:	The VCPU pointer
655  *
656  * This function is called through the VCPU_RUN ioctl called from user space. It
657  * will execute VM code in a loop until the time slice for the process is used
658  * or some emulation is needed from user space in which case the function will
659  * return with return value 0 and with the kvm_run structure filled in with the
660  * required data for the requested emulation.
661  */
662 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
663 {
664 	struct kvm_run *run = vcpu->run;
665 	int ret;
666 
667 	if (unlikely(!kvm_vcpu_initialized(vcpu)))
668 		return -ENOEXEC;
669 
670 	ret = kvm_vcpu_first_run_init(vcpu);
671 	if (ret)
672 		return ret;
673 
674 	if (run->exit_reason == KVM_EXIT_MMIO) {
675 		ret = kvm_handle_mmio_return(vcpu);
676 		if (ret)
677 			return ret;
678 	}
679 
680 	if (run->immediate_exit)
681 		return -EINTR;
682 
683 	vcpu_load(vcpu);
684 
685 	kvm_sigset_activate(vcpu);
686 
687 	ret = 1;
688 	run->exit_reason = KVM_EXIT_UNKNOWN;
689 	while (ret > 0) {
690 		/*
691 		 * Check conditions before entering the guest
692 		 */
693 		cond_resched();
694 
695 		update_vmid(&vcpu->arch.hw_mmu->vmid);
696 
697 		check_vcpu_requests(vcpu);
698 
699 		/*
700 		 * Preparing the interrupts to be injected also
701 		 * involves poking the GIC, which must be done in a
702 		 * non-preemptible context.
703 		 */
704 		preempt_disable();
705 
706 		kvm_pmu_flush_hwstate(vcpu);
707 
708 		local_irq_disable();
709 
710 		kvm_vgic_flush_hwstate(vcpu);
711 
712 		/*
713 		 * Exit if we have a signal pending so that we can deliver the
714 		 * signal to user space.
715 		 */
716 		if (signal_pending(current)) {
717 			ret = -EINTR;
718 			run->exit_reason = KVM_EXIT_INTR;
719 		}
720 
721 		/*
722 		 * If we're using a userspace irqchip, then check if we need
723 		 * to tell a userspace irqchip about timer or PMU level
724 		 * changes and if so, exit to userspace (the actual level
725 		 * state gets updated in kvm_timer_update_run and
726 		 * kvm_pmu_update_run below).
727 		 */
728 		if (static_branch_unlikely(&userspace_irqchip_in_use)) {
729 			if (kvm_timer_should_notify_user(vcpu) ||
730 			    kvm_pmu_should_notify_user(vcpu)) {
731 				ret = -EINTR;
732 				run->exit_reason = KVM_EXIT_INTR;
733 			}
734 		}
735 
736 		/*
737 		 * Ensure we set mode to IN_GUEST_MODE after we disable
738 		 * interrupts and before the final VCPU requests check.
739 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
740 		 * Documentation/virt/kvm/vcpu-requests.rst
741 		 */
742 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
743 
744 		if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
745 		    kvm_request_pending(vcpu)) {
746 			vcpu->mode = OUTSIDE_GUEST_MODE;
747 			isb(); /* Ensure work in x_flush_hwstate is committed */
748 			kvm_pmu_sync_hwstate(vcpu);
749 			if (static_branch_unlikely(&userspace_irqchip_in_use))
750 				kvm_timer_sync_user(vcpu);
751 			kvm_vgic_sync_hwstate(vcpu);
752 			local_irq_enable();
753 			preempt_enable();
754 			continue;
755 		}
756 
757 		kvm_arm_setup_debug(vcpu);
758 
759 		/**************************************************************
760 		 * Enter the guest
761 		 */
762 		trace_kvm_entry(*vcpu_pc(vcpu));
763 		guest_enter_irqoff();
764 
765 		ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
766 
767 		vcpu->mode = OUTSIDE_GUEST_MODE;
768 		vcpu->stat.exits++;
769 		/*
770 		 * Back from guest
771 		 *************************************************************/
772 
773 		kvm_arm_clear_debug(vcpu);
774 
775 		/*
776 		 * We must sync the PMU state before the vgic state so
777 		 * that the vgic can properly sample the updated state of the
778 		 * interrupt line.
779 		 */
780 		kvm_pmu_sync_hwstate(vcpu);
781 
782 		/*
783 		 * Sync the vgic state before syncing the timer state because
784 		 * the timer code needs to know if the virtual timer
785 		 * interrupts are active.
786 		 */
787 		kvm_vgic_sync_hwstate(vcpu);
788 
789 		/*
790 		 * Sync the timer hardware state before enabling interrupts as
791 		 * we don't want vtimer interrupts to race with syncing the
792 		 * timer virtual interrupt state.
793 		 */
794 		if (static_branch_unlikely(&userspace_irqchip_in_use))
795 			kvm_timer_sync_user(vcpu);
796 
797 		kvm_arch_vcpu_ctxsync_fp(vcpu);
798 
799 		/*
800 		 * We may have taken a host interrupt in HYP mode (ie
801 		 * while executing the guest). This interrupt is still
802 		 * pending, as we haven't serviced it yet!
803 		 *
804 		 * We're now back in SVC mode, with interrupts
805 		 * disabled.  Enabling the interrupts now will have
806 		 * the effect of taking the interrupt again, in SVC
807 		 * mode this time.
808 		 */
809 		local_irq_enable();
810 
811 		/*
812 		 * We do local_irq_enable() before calling guest_exit() so
813 		 * that if a timer interrupt hits while running the guest we
814 		 * account that tick as being spent in the guest.  We enable
815 		 * preemption after calling guest_exit() so that if we get
816 		 * preempted we make sure ticks after that is not counted as
817 		 * guest time.
818 		 */
819 		guest_exit();
820 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
821 
822 		/* Exit types that need handling before we can be preempted */
823 		handle_exit_early(vcpu, ret);
824 
825 		preempt_enable();
826 
827 		/*
828 		 * The ARMv8 architecture doesn't give the hypervisor
829 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
830 		 * if implemented by the CPU. If we spot the guest in such
831 		 * state and that we decided it wasn't supposed to do so (like
832 		 * with the asymmetric AArch32 case), return to userspace with
833 		 * a fatal error.
834 		 */
835 		if (!system_supports_32bit_el0() && vcpu_mode_is_32bit(vcpu)) {
836 			/*
837 			 * As we have caught the guest red-handed, decide that
838 			 * it isn't fit for purpose anymore by making the vcpu
839 			 * invalid. The VMM can try and fix it by issuing  a
840 			 * KVM_ARM_VCPU_INIT if it really wants to.
841 			 */
842 			vcpu->arch.target = -1;
843 			ret = ARM_EXCEPTION_IL;
844 		}
845 
846 		ret = handle_exit(vcpu, ret);
847 	}
848 
849 	/* Tell userspace about in-kernel device output levels */
850 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
851 		kvm_timer_update_run(vcpu);
852 		kvm_pmu_update_run(vcpu);
853 	}
854 
855 	kvm_sigset_deactivate(vcpu);
856 
857 	vcpu_put(vcpu);
858 	return ret;
859 }
860 
861 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
862 {
863 	int bit_index;
864 	bool set;
865 	unsigned long *hcr;
866 
867 	if (number == KVM_ARM_IRQ_CPU_IRQ)
868 		bit_index = __ffs(HCR_VI);
869 	else /* KVM_ARM_IRQ_CPU_FIQ */
870 		bit_index = __ffs(HCR_VF);
871 
872 	hcr = vcpu_hcr(vcpu);
873 	if (level)
874 		set = test_and_set_bit(bit_index, hcr);
875 	else
876 		set = test_and_clear_bit(bit_index, hcr);
877 
878 	/*
879 	 * If we didn't change anything, no need to wake up or kick other CPUs
880 	 */
881 	if (set == level)
882 		return 0;
883 
884 	/*
885 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
886 	 * trigger a world-switch round on the running physical CPU to set the
887 	 * virtual IRQ/FIQ fields in the HCR appropriately.
888 	 */
889 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
890 	kvm_vcpu_kick(vcpu);
891 
892 	return 0;
893 }
894 
895 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
896 			  bool line_status)
897 {
898 	u32 irq = irq_level->irq;
899 	unsigned int irq_type, vcpu_idx, irq_num;
900 	int nrcpus = atomic_read(&kvm->online_vcpus);
901 	struct kvm_vcpu *vcpu = NULL;
902 	bool level = irq_level->level;
903 
904 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
905 	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
906 	vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
907 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
908 
909 	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
910 
911 	switch (irq_type) {
912 	case KVM_ARM_IRQ_TYPE_CPU:
913 		if (irqchip_in_kernel(kvm))
914 			return -ENXIO;
915 
916 		if (vcpu_idx >= nrcpus)
917 			return -EINVAL;
918 
919 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
920 		if (!vcpu)
921 			return -EINVAL;
922 
923 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
924 			return -EINVAL;
925 
926 		return vcpu_interrupt_line(vcpu, irq_num, level);
927 	case KVM_ARM_IRQ_TYPE_PPI:
928 		if (!irqchip_in_kernel(kvm))
929 			return -ENXIO;
930 
931 		if (vcpu_idx >= nrcpus)
932 			return -EINVAL;
933 
934 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
935 		if (!vcpu)
936 			return -EINVAL;
937 
938 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
939 			return -EINVAL;
940 
941 		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
942 	case KVM_ARM_IRQ_TYPE_SPI:
943 		if (!irqchip_in_kernel(kvm))
944 			return -ENXIO;
945 
946 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
947 			return -EINVAL;
948 
949 		return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
950 	}
951 
952 	return -EINVAL;
953 }
954 
955 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
956 			       const struct kvm_vcpu_init *init)
957 {
958 	unsigned int i, ret;
959 	int phys_target = kvm_target_cpu();
960 
961 	if (init->target != phys_target)
962 		return -EINVAL;
963 
964 	/*
965 	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
966 	 * use the same target.
967 	 */
968 	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
969 		return -EINVAL;
970 
971 	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
972 	for (i = 0; i < sizeof(init->features) * 8; i++) {
973 		bool set = (init->features[i / 32] & (1 << (i % 32)));
974 
975 		if (set && i >= KVM_VCPU_MAX_FEATURES)
976 			return -ENOENT;
977 
978 		/*
979 		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
980 		 * use the same feature set.
981 		 */
982 		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
983 		    test_bit(i, vcpu->arch.features) != set)
984 			return -EINVAL;
985 
986 		if (set)
987 			set_bit(i, vcpu->arch.features);
988 	}
989 
990 	vcpu->arch.target = phys_target;
991 
992 	/* Now we know what it is, we can reset it. */
993 	ret = kvm_reset_vcpu(vcpu);
994 	if (ret) {
995 		vcpu->arch.target = -1;
996 		bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
997 	}
998 
999 	return ret;
1000 }
1001 
1002 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1003 					 struct kvm_vcpu_init *init)
1004 {
1005 	int ret;
1006 
1007 	ret = kvm_vcpu_set_target(vcpu, init);
1008 	if (ret)
1009 		return ret;
1010 
1011 	/*
1012 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1013 	 * guest MMU is turned off and flush the caches as needed.
1014 	 *
1015 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1016 	 * ensuring that the data side is always coherent. We still
1017 	 * need to invalidate the I-cache though, as FWB does *not*
1018 	 * imply CTR_EL0.DIC.
1019 	 */
1020 	if (vcpu->arch.has_run_once) {
1021 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1022 			stage2_unmap_vm(vcpu->kvm);
1023 		else
1024 			__flush_icache_all();
1025 	}
1026 
1027 	vcpu_reset_hcr(vcpu);
1028 
1029 	/*
1030 	 * Handle the "start in power-off" case.
1031 	 */
1032 	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1033 		vcpu_power_off(vcpu);
1034 	else
1035 		vcpu->arch.power_off = false;
1036 
1037 	return 0;
1038 }
1039 
1040 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1041 				 struct kvm_device_attr *attr)
1042 {
1043 	int ret = -ENXIO;
1044 
1045 	switch (attr->group) {
1046 	default:
1047 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1048 		break;
1049 	}
1050 
1051 	return ret;
1052 }
1053 
1054 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1055 				 struct kvm_device_attr *attr)
1056 {
1057 	int ret = -ENXIO;
1058 
1059 	switch (attr->group) {
1060 	default:
1061 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1062 		break;
1063 	}
1064 
1065 	return ret;
1066 }
1067 
1068 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1069 				 struct kvm_device_attr *attr)
1070 {
1071 	int ret = -ENXIO;
1072 
1073 	switch (attr->group) {
1074 	default:
1075 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1076 		break;
1077 	}
1078 
1079 	return ret;
1080 }
1081 
1082 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1083 				   struct kvm_vcpu_events *events)
1084 {
1085 	memset(events, 0, sizeof(*events));
1086 
1087 	return __kvm_arm_vcpu_get_events(vcpu, events);
1088 }
1089 
1090 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1091 				   struct kvm_vcpu_events *events)
1092 {
1093 	int i;
1094 
1095 	/* check whether the reserved field is zero */
1096 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1097 		if (events->reserved[i])
1098 			return -EINVAL;
1099 
1100 	/* check whether the pad field is zero */
1101 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1102 		if (events->exception.pad[i])
1103 			return -EINVAL;
1104 
1105 	return __kvm_arm_vcpu_set_events(vcpu, events);
1106 }
1107 
1108 long kvm_arch_vcpu_ioctl(struct file *filp,
1109 			 unsigned int ioctl, unsigned long arg)
1110 {
1111 	struct kvm_vcpu *vcpu = filp->private_data;
1112 	void __user *argp = (void __user *)arg;
1113 	struct kvm_device_attr attr;
1114 	long r;
1115 
1116 	switch (ioctl) {
1117 	case KVM_ARM_VCPU_INIT: {
1118 		struct kvm_vcpu_init init;
1119 
1120 		r = -EFAULT;
1121 		if (copy_from_user(&init, argp, sizeof(init)))
1122 			break;
1123 
1124 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1125 		break;
1126 	}
1127 	case KVM_SET_ONE_REG:
1128 	case KVM_GET_ONE_REG: {
1129 		struct kvm_one_reg reg;
1130 
1131 		r = -ENOEXEC;
1132 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1133 			break;
1134 
1135 		r = -EFAULT;
1136 		if (copy_from_user(&reg, argp, sizeof(reg)))
1137 			break;
1138 
1139 		if (ioctl == KVM_SET_ONE_REG)
1140 			r = kvm_arm_set_reg(vcpu, &reg);
1141 		else
1142 			r = kvm_arm_get_reg(vcpu, &reg);
1143 		break;
1144 	}
1145 	case KVM_GET_REG_LIST: {
1146 		struct kvm_reg_list __user *user_list = argp;
1147 		struct kvm_reg_list reg_list;
1148 		unsigned n;
1149 
1150 		r = -ENOEXEC;
1151 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1152 			break;
1153 
1154 		r = -EPERM;
1155 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1156 			break;
1157 
1158 		r = -EFAULT;
1159 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1160 			break;
1161 		n = reg_list.n;
1162 		reg_list.n = kvm_arm_num_regs(vcpu);
1163 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1164 			break;
1165 		r = -E2BIG;
1166 		if (n < reg_list.n)
1167 			break;
1168 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1169 		break;
1170 	}
1171 	case KVM_SET_DEVICE_ATTR: {
1172 		r = -EFAULT;
1173 		if (copy_from_user(&attr, argp, sizeof(attr)))
1174 			break;
1175 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1176 		break;
1177 	}
1178 	case KVM_GET_DEVICE_ATTR: {
1179 		r = -EFAULT;
1180 		if (copy_from_user(&attr, argp, sizeof(attr)))
1181 			break;
1182 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1183 		break;
1184 	}
1185 	case KVM_HAS_DEVICE_ATTR: {
1186 		r = -EFAULT;
1187 		if (copy_from_user(&attr, argp, sizeof(attr)))
1188 			break;
1189 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1190 		break;
1191 	}
1192 	case KVM_GET_VCPU_EVENTS: {
1193 		struct kvm_vcpu_events events;
1194 
1195 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1196 			return -EINVAL;
1197 
1198 		if (copy_to_user(argp, &events, sizeof(events)))
1199 			return -EFAULT;
1200 
1201 		return 0;
1202 	}
1203 	case KVM_SET_VCPU_EVENTS: {
1204 		struct kvm_vcpu_events events;
1205 
1206 		if (copy_from_user(&events, argp, sizeof(events)))
1207 			return -EFAULT;
1208 
1209 		return kvm_arm_vcpu_set_events(vcpu, &events);
1210 	}
1211 	case KVM_ARM_VCPU_FINALIZE: {
1212 		int what;
1213 
1214 		if (!kvm_vcpu_initialized(vcpu))
1215 			return -ENOEXEC;
1216 
1217 		if (get_user(what, (const int __user *)argp))
1218 			return -EFAULT;
1219 
1220 		return kvm_arm_vcpu_finalize(vcpu, what);
1221 	}
1222 	default:
1223 		r = -EINVAL;
1224 	}
1225 
1226 	return r;
1227 }
1228 
1229 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1230 {
1231 
1232 }
1233 
1234 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1235 					struct kvm_memory_slot *memslot)
1236 {
1237 	kvm_flush_remote_tlbs(kvm);
1238 }
1239 
1240 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1241 					struct kvm_arm_device_addr *dev_addr)
1242 {
1243 	unsigned long dev_id, type;
1244 
1245 	dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1246 		KVM_ARM_DEVICE_ID_SHIFT;
1247 	type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1248 		KVM_ARM_DEVICE_TYPE_SHIFT;
1249 
1250 	switch (dev_id) {
1251 	case KVM_ARM_DEVICE_VGIC_V2:
1252 		if (!vgic_present)
1253 			return -ENXIO;
1254 		return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1255 	default:
1256 		return -ENODEV;
1257 	}
1258 }
1259 
1260 long kvm_arch_vm_ioctl(struct file *filp,
1261 		       unsigned int ioctl, unsigned long arg)
1262 {
1263 	struct kvm *kvm = filp->private_data;
1264 	void __user *argp = (void __user *)arg;
1265 
1266 	switch (ioctl) {
1267 	case KVM_CREATE_IRQCHIP: {
1268 		int ret;
1269 		if (!vgic_present)
1270 			return -ENXIO;
1271 		mutex_lock(&kvm->lock);
1272 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1273 		mutex_unlock(&kvm->lock);
1274 		return ret;
1275 	}
1276 	case KVM_ARM_SET_DEVICE_ADDR: {
1277 		struct kvm_arm_device_addr dev_addr;
1278 
1279 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1280 			return -EFAULT;
1281 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1282 	}
1283 	case KVM_ARM_PREFERRED_TARGET: {
1284 		int err;
1285 		struct kvm_vcpu_init init;
1286 
1287 		err = kvm_vcpu_preferred_target(&init);
1288 		if (err)
1289 			return err;
1290 
1291 		if (copy_to_user(argp, &init, sizeof(init)))
1292 			return -EFAULT;
1293 
1294 		return 0;
1295 	}
1296 	default:
1297 		return -EINVAL;
1298 	}
1299 }
1300 
1301 static unsigned long nvhe_percpu_size(void)
1302 {
1303 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1304 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1305 }
1306 
1307 static unsigned long nvhe_percpu_order(void)
1308 {
1309 	unsigned long size = nvhe_percpu_size();
1310 
1311 	return size ? get_order(size) : 0;
1312 }
1313 
1314 static int kvm_map_vectors(void)
1315 {
1316 	/*
1317 	 * SV2  = ARM64_SPECTRE_V2
1318 	 * HEL2 = ARM64_HARDEN_EL2_VECTORS
1319 	 *
1320 	 * !SV2 + !HEL2 -> use direct vectors
1321 	 *  SV2 + !HEL2 -> use hardened vectors in place
1322 	 * !SV2 +  HEL2 -> allocate one vector slot and use exec mapping
1323 	 *  SV2 +  HEL2 -> use hardened vectors and use exec mapping
1324 	 */
1325 	if (cpus_have_const_cap(ARM64_SPECTRE_V2)) {
1326 		__kvm_bp_vect_base = kvm_ksym_ref(__bp_harden_hyp_vecs);
1327 		__kvm_bp_vect_base = kern_hyp_va(__kvm_bp_vect_base);
1328 	}
1329 
1330 	if (cpus_have_const_cap(ARM64_HARDEN_EL2_VECTORS)) {
1331 		phys_addr_t vect_pa = __pa_symbol(__bp_harden_hyp_vecs);
1332 		unsigned long size = __BP_HARDEN_HYP_VECS_SZ;
1333 
1334 		/*
1335 		 * Always allocate a spare vector slot, as we don't
1336 		 * know yet which CPUs have a BP hardening slot that
1337 		 * we can reuse.
1338 		 */
1339 		__kvm_harden_el2_vector_slot = atomic_inc_return(&arm64_el2_vector_last_slot);
1340 		BUG_ON(__kvm_harden_el2_vector_slot >= BP_HARDEN_EL2_SLOTS);
1341 		return create_hyp_exec_mappings(vect_pa, size,
1342 						&__kvm_bp_vect_base);
1343 	}
1344 
1345 	return 0;
1346 }
1347 
1348 static void cpu_init_hyp_mode(void)
1349 {
1350 	phys_addr_t pgd_ptr;
1351 	unsigned long hyp_stack_ptr;
1352 	unsigned long vector_ptr;
1353 	unsigned long tpidr_el2;
1354 	struct arm_smccc_res res;
1355 
1356 	/* Switch from the HYP stub to our own HYP init vector */
1357 	__hyp_set_vectors(kvm_get_idmap_vector());
1358 
1359 	/*
1360 	 * Calculate the raw per-cpu offset without a translation from the
1361 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1362 	 * so that we can use adr_l to access per-cpu variables in EL2.
1363 	 */
1364 	tpidr_el2 = (unsigned long)this_cpu_ptr_nvhe_sym(__per_cpu_start) -
1365 		    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1366 
1367 	pgd_ptr = kvm_mmu_get_httbr();
1368 	hyp_stack_ptr = __this_cpu_read(kvm_arm_hyp_stack_page) + PAGE_SIZE;
1369 	hyp_stack_ptr = kern_hyp_va(hyp_stack_ptr);
1370 	vector_ptr = (unsigned long)kern_hyp_va(kvm_ksym_ref(__kvm_hyp_host_vector));
1371 
1372 	/*
1373 	 * Call initialization code, and switch to the full blown HYP code.
1374 	 * If the cpucaps haven't been finalized yet, something has gone very
1375 	 * wrong, and hyp will crash and burn when it uses any
1376 	 * cpus_have_const_cap() wrapper.
1377 	 */
1378 	BUG_ON(!system_capabilities_finalized());
1379 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init),
1380 			  pgd_ptr, tpidr_el2, hyp_stack_ptr, vector_ptr, &res);
1381 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1382 
1383 	/*
1384 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1385 	 * at EL2.
1386 	 */
1387 	if (this_cpu_has_cap(ARM64_SSBS) &&
1388 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1389 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1390 	}
1391 }
1392 
1393 static void cpu_hyp_reset(void)
1394 {
1395 	if (!is_kernel_in_hyp_mode())
1396 		__hyp_reset_vectors();
1397 }
1398 
1399 static void cpu_hyp_reinit(void)
1400 {
1401 	kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1402 
1403 	cpu_hyp_reset();
1404 
1405 	*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)kvm_get_hyp_vector();
1406 
1407 	if (is_kernel_in_hyp_mode())
1408 		kvm_timer_init_vhe();
1409 	else
1410 		cpu_init_hyp_mode();
1411 
1412 	kvm_arm_init_debug();
1413 
1414 	if (vgic_present)
1415 		kvm_vgic_init_cpu_hardware();
1416 }
1417 
1418 static void _kvm_arch_hardware_enable(void *discard)
1419 {
1420 	if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1421 		cpu_hyp_reinit();
1422 		__this_cpu_write(kvm_arm_hardware_enabled, 1);
1423 	}
1424 }
1425 
1426 int kvm_arch_hardware_enable(void)
1427 {
1428 	_kvm_arch_hardware_enable(NULL);
1429 	return 0;
1430 }
1431 
1432 static void _kvm_arch_hardware_disable(void *discard)
1433 {
1434 	if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1435 		cpu_hyp_reset();
1436 		__this_cpu_write(kvm_arm_hardware_enabled, 0);
1437 	}
1438 }
1439 
1440 void kvm_arch_hardware_disable(void)
1441 {
1442 	_kvm_arch_hardware_disable(NULL);
1443 }
1444 
1445 #ifdef CONFIG_CPU_PM
1446 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1447 				    unsigned long cmd,
1448 				    void *v)
1449 {
1450 	/*
1451 	 * kvm_arm_hardware_enabled is left with its old value over
1452 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1453 	 * re-enable hyp.
1454 	 */
1455 	switch (cmd) {
1456 	case CPU_PM_ENTER:
1457 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1458 			/*
1459 			 * don't update kvm_arm_hardware_enabled here
1460 			 * so that the hardware will be re-enabled
1461 			 * when we resume. See below.
1462 			 */
1463 			cpu_hyp_reset();
1464 
1465 		return NOTIFY_OK;
1466 	case CPU_PM_ENTER_FAILED:
1467 	case CPU_PM_EXIT:
1468 		if (__this_cpu_read(kvm_arm_hardware_enabled))
1469 			/* The hardware was enabled before suspend. */
1470 			cpu_hyp_reinit();
1471 
1472 		return NOTIFY_OK;
1473 
1474 	default:
1475 		return NOTIFY_DONE;
1476 	}
1477 }
1478 
1479 static struct notifier_block hyp_init_cpu_pm_nb = {
1480 	.notifier_call = hyp_init_cpu_pm_notifier,
1481 };
1482 
1483 static void __init hyp_cpu_pm_init(void)
1484 {
1485 	cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1486 }
1487 static void __init hyp_cpu_pm_exit(void)
1488 {
1489 	cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1490 }
1491 #else
1492 static inline void hyp_cpu_pm_init(void)
1493 {
1494 }
1495 static inline void hyp_cpu_pm_exit(void)
1496 {
1497 }
1498 #endif
1499 
1500 static int init_common_resources(void)
1501 {
1502 	return kvm_set_ipa_limit();
1503 }
1504 
1505 static int init_subsystems(void)
1506 {
1507 	int err = 0;
1508 
1509 	/*
1510 	 * Enable hardware so that subsystem initialisation can access EL2.
1511 	 */
1512 	on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1513 
1514 	/*
1515 	 * Register CPU lower-power notifier
1516 	 */
1517 	hyp_cpu_pm_init();
1518 
1519 	/*
1520 	 * Init HYP view of VGIC
1521 	 */
1522 	err = kvm_vgic_hyp_init();
1523 	switch (err) {
1524 	case 0:
1525 		vgic_present = true;
1526 		break;
1527 	case -ENODEV:
1528 	case -ENXIO:
1529 		vgic_present = false;
1530 		err = 0;
1531 		break;
1532 	default:
1533 		goto out;
1534 	}
1535 
1536 	/*
1537 	 * Init HYP architected timer support
1538 	 */
1539 	err = kvm_timer_hyp_init(vgic_present);
1540 	if (err)
1541 		goto out;
1542 
1543 	kvm_perf_init();
1544 	kvm_coproc_table_init();
1545 
1546 out:
1547 	on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1548 
1549 	return err;
1550 }
1551 
1552 static void teardown_hyp_mode(void)
1553 {
1554 	int cpu;
1555 
1556 	free_hyp_pgds();
1557 	for_each_possible_cpu(cpu) {
1558 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1559 		free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1560 	}
1561 }
1562 
1563 /**
1564  * Inits Hyp-mode on all online CPUs
1565  */
1566 static int init_hyp_mode(void)
1567 {
1568 	int cpu;
1569 	int err = 0;
1570 
1571 	/*
1572 	 * Allocate Hyp PGD and setup Hyp identity mapping
1573 	 */
1574 	err = kvm_mmu_init();
1575 	if (err)
1576 		goto out_err;
1577 
1578 	/*
1579 	 * Allocate stack pages for Hypervisor-mode
1580 	 */
1581 	for_each_possible_cpu(cpu) {
1582 		unsigned long stack_page;
1583 
1584 		stack_page = __get_free_page(GFP_KERNEL);
1585 		if (!stack_page) {
1586 			err = -ENOMEM;
1587 			goto out_err;
1588 		}
1589 
1590 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1591 	}
1592 
1593 	/*
1594 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1595 	 */
1596 	for_each_possible_cpu(cpu) {
1597 		struct page *page;
1598 		void *page_addr;
1599 
1600 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1601 		if (!page) {
1602 			err = -ENOMEM;
1603 			goto out_err;
1604 		}
1605 
1606 		page_addr = page_address(page);
1607 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1608 		kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1609 	}
1610 
1611 	/*
1612 	 * Map the Hyp-code called directly from the host
1613 	 */
1614 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1615 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1616 	if (err) {
1617 		kvm_err("Cannot map world-switch code\n");
1618 		goto out_err;
1619 	}
1620 
1621 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1622 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1623 	if (err) {
1624 		kvm_err("Cannot map rodata section\n");
1625 		goto out_err;
1626 	}
1627 
1628 	err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1629 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1630 	if (err) {
1631 		kvm_err("Cannot map bss section\n");
1632 		goto out_err;
1633 	}
1634 
1635 	err = kvm_map_vectors();
1636 	if (err) {
1637 		kvm_err("Cannot map vectors\n");
1638 		goto out_err;
1639 	}
1640 
1641 	/*
1642 	 * Map the Hyp stack pages
1643 	 */
1644 	for_each_possible_cpu(cpu) {
1645 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1646 		err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1647 					  PAGE_HYP);
1648 
1649 		if (err) {
1650 			kvm_err("Cannot map hyp stack\n");
1651 			goto out_err;
1652 		}
1653 	}
1654 
1655 	/*
1656 	 * Map Hyp percpu pages
1657 	 */
1658 	for_each_possible_cpu(cpu) {
1659 		char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1660 		char *percpu_end = percpu_begin + nvhe_percpu_size();
1661 
1662 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1663 
1664 		if (err) {
1665 			kvm_err("Cannot map hyp percpu region\n");
1666 			goto out_err;
1667 		}
1668 	}
1669 
1670 	return 0;
1671 
1672 out_err:
1673 	teardown_hyp_mode();
1674 	kvm_err("error initializing Hyp mode: %d\n", err);
1675 	return err;
1676 }
1677 
1678 static void check_kvm_target_cpu(void *ret)
1679 {
1680 	*(int *)ret = kvm_target_cpu();
1681 }
1682 
1683 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1684 {
1685 	struct kvm_vcpu *vcpu;
1686 	int i;
1687 
1688 	mpidr &= MPIDR_HWID_BITMASK;
1689 	kvm_for_each_vcpu(i, vcpu, kvm) {
1690 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1691 			return vcpu;
1692 	}
1693 	return NULL;
1694 }
1695 
1696 bool kvm_arch_has_irq_bypass(void)
1697 {
1698 	return true;
1699 }
1700 
1701 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1702 				      struct irq_bypass_producer *prod)
1703 {
1704 	struct kvm_kernel_irqfd *irqfd =
1705 		container_of(cons, struct kvm_kernel_irqfd, consumer);
1706 
1707 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1708 					  &irqfd->irq_entry);
1709 }
1710 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1711 				      struct irq_bypass_producer *prod)
1712 {
1713 	struct kvm_kernel_irqfd *irqfd =
1714 		container_of(cons, struct kvm_kernel_irqfd, consumer);
1715 
1716 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1717 				     &irqfd->irq_entry);
1718 }
1719 
1720 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1721 {
1722 	struct kvm_kernel_irqfd *irqfd =
1723 		container_of(cons, struct kvm_kernel_irqfd, consumer);
1724 
1725 	kvm_arm_halt_guest(irqfd->kvm);
1726 }
1727 
1728 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1729 {
1730 	struct kvm_kernel_irqfd *irqfd =
1731 		container_of(cons, struct kvm_kernel_irqfd, consumer);
1732 
1733 	kvm_arm_resume_guest(irqfd->kvm);
1734 }
1735 
1736 /**
1737  * Initialize Hyp-mode and memory mappings on all CPUs.
1738  */
1739 int kvm_arch_init(void *opaque)
1740 {
1741 	int err;
1742 	int ret, cpu;
1743 	bool in_hyp_mode;
1744 
1745 	if (!is_hyp_mode_available()) {
1746 		kvm_info("HYP mode not available\n");
1747 		return -ENODEV;
1748 	}
1749 
1750 	in_hyp_mode = is_kernel_in_hyp_mode();
1751 
1752 	if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1753 		kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1754 		return -ENODEV;
1755 	}
1756 
1757 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
1758 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
1759 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
1760 			 "Only trusted guests should be used on this system.\n");
1761 
1762 	for_each_online_cpu(cpu) {
1763 		smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1764 		if (ret < 0) {
1765 			kvm_err("Error, CPU %d not supported!\n", cpu);
1766 			return -ENODEV;
1767 		}
1768 	}
1769 
1770 	err = init_common_resources();
1771 	if (err)
1772 		return err;
1773 
1774 	err = kvm_arm_init_sve();
1775 	if (err)
1776 		return err;
1777 
1778 	if (!in_hyp_mode) {
1779 		err = init_hyp_mode();
1780 		if (err)
1781 			goto out_err;
1782 	}
1783 
1784 	err = init_subsystems();
1785 	if (err)
1786 		goto out_hyp;
1787 
1788 	if (in_hyp_mode)
1789 		kvm_info("VHE mode initialized successfully\n");
1790 	else
1791 		kvm_info("Hyp mode initialized successfully\n");
1792 
1793 	return 0;
1794 
1795 out_hyp:
1796 	hyp_cpu_pm_exit();
1797 	if (!in_hyp_mode)
1798 		teardown_hyp_mode();
1799 out_err:
1800 	return err;
1801 }
1802 
1803 /* NOP: Compiling as a module not supported */
1804 void kvm_arch_exit(void)
1805 {
1806 	kvm_perf_teardown();
1807 }
1808 
1809 static int arm_init(void)
1810 {
1811 	int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1812 	return rc;
1813 }
1814 
1815 module_init(arm_init);
1816