1/*
2 * Copyright (C) 2012 - Virtual Open Systems and Columbia University
3 * Author: Christoffer Dall <c.dall@virtualopensystems.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2, as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
17 */
18
19#include <linux/cpu.h>
20#include <linux/cpu_pm.h>
21#include <linux/errno.h>
22#include <linux/err.h>
23#include <linux/kvm_host.h>
24#include <linux/module.h>
25#include <linux/vmalloc.h>
26#include <linux/fs.h>
27#include <linux/mman.h>
28#include <linux/sched.h>
29#include <linux/kvm.h>
30#include <trace/events/kvm.h>
31
32#define CREATE_TRACE_POINTS
33#include "trace.h"
34
35#include <asm/uaccess.h>
36#include <asm/ptrace.h>
37#include <asm/mman.h>
38#include <asm/tlbflush.h>
39#include <asm/cacheflush.h>
40#include <asm/virt.h>
41#include <asm/kvm_arm.h>
42#include <asm/kvm_asm.h>
43#include <asm/kvm_mmu.h>
44#include <asm/kvm_emulate.h>
45#include <asm/kvm_coproc.h>
46#include <asm/kvm_psci.h>
47
48#ifdef REQUIRES_VIRT
49__asm__(".arch_extension	virt");
50#endif
51
52static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
53static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
54static unsigned long hyp_default_vectors;
55
56/* Per-CPU variable containing the currently running vcpu. */
57static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
58
59/* The VMID used in the VTTBR */
60static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
61static u8 kvm_next_vmid;
62static DEFINE_SPINLOCK(kvm_vmid_lock);
63
64static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
65{
66	BUG_ON(preemptible());
67	__this_cpu_write(kvm_arm_running_vcpu, vcpu);
68}
69
70/**
71 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
72 * Must be called from non-preemptible context
73 */
74struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
75{
76	BUG_ON(preemptible());
77	return __this_cpu_read(kvm_arm_running_vcpu);
78}
79
80/**
81 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
82 */
83struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
84{
85	return &kvm_arm_running_vcpu;
86}
87
88int kvm_arch_hardware_enable(void)
89{
90	return 0;
91}
92
93int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
94{
95	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
96}
97
98int kvm_arch_hardware_setup(void)
99{
100	return 0;
101}
102
103void kvm_arch_check_processor_compat(void *rtn)
104{
105	*(int *)rtn = 0;
106}
107
108
109/**
110 * kvm_arch_init_vm - initializes a VM data structure
111 * @kvm:	pointer to the KVM struct
112 */
113int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
114{
115	int ret = 0;
116
117	if (type)
118		return -EINVAL;
119
120	ret = kvm_alloc_stage2_pgd(kvm);
121	if (ret)
122		goto out_fail_alloc;
123
124	ret = create_hyp_mappings(kvm, kvm + 1);
125	if (ret)
126		goto out_free_stage2_pgd;
127
128	kvm_vgic_early_init(kvm);
129	kvm_timer_init(kvm);
130
131	/* Mark the initial VMID generation invalid */
132	kvm->arch.vmid_gen = 0;
133
134	/* The maximum number of VCPUs is limited by the host's GIC model */
135	kvm->arch.max_vcpus = kvm_vgic_get_max_vcpus();
136
137	return ret;
138out_free_stage2_pgd:
139	kvm_free_stage2_pgd(kvm);
140out_fail_alloc:
141	return ret;
142}
143
144int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
145{
146	return VM_FAULT_SIGBUS;
147}
148
149
150/**
151 * kvm_arch_destroy_vm - destroy the VM data structure
152 * @kvm:	pointer to the KVM struct
153 */
154void kvm_arch_destroy_vm(struct kvm *kvm)
155{
156	int i;
157
158	kvm_free_stage2_pgd(kvm);
159
160	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
161		if (kvm->vcpus[i]) {
162			kvm_arch_vcpu_free(kvm->vcpus[i]);
163			kvm->vcpus[i] = NULL;
164		}
165	}
166
167	kvm_vgic_destroy(kvm);
168}
169
170int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
171{
172	int r;
173	switch (ext) {
174	case KVM_CAP_IRQCHIP:
175	case KVM_CAP_IOEVENTFD:
176	case KVM_CAP_DEVICE_CTRL:
177	case KVM_CAP_USER_MEMORY:
178	case KVM_CAP_SYNC_MMU:
179	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
180	case KVM_CAP_ONE_REG:
181	case KVM_CAP_ARM_PSCI:
182	case KVM_CAP_ARM_PSCI_0_2:
183	case KVM_CAP_READONLY_MEM:
184	case KVM_CAP_MP_STATE:
185		r = 1;
186		break;
187	case KVM_CAP_COALESCED_MMIO:
188		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
189		break;
190	case KVM_CAP_ARM_SET_DEVICE_ADDR:
191		r = 1;
192		break;
193	case KVM_CAP_NR_VCPUS:
194		r = num_online_cpus();
195		break;
196	case KVM_CAP_MAX_VCPUS:
197		r = KVM_MAX_VCPUS;
198		break;
199	default:
200		r = kvm_arch_dev_ioctl_check_extension(ext);
201		break;
202	}
203	return r;
204}
205
206long kvm_arch_dev_ioctl(struct file *filp,
207			unsigned int ioctl, unsigned long arg)
208{
209	return -EINVAL;
210}
211
212
213struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
214{
215	int err;
216	struct kvm_vcpu *vcpu;
217
218	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
219		err = -EBUSY;
220		goto out;
221	}
222
223	if (id >= kvm->arch.max_vcpus) {
224		err = -EINVAL;
225		goto out;
226	}
227
228	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
229	if (!vcpu) {
230		err = -ENOMEM;
231		goto out;
232	}
233
234	err = kvm_vcpu_init(vcpu, kvm, id);
235	if (err)
236		goto free_vcpu;
237
238	err = create_hyp_mappings(vcpu, vcpu + 1);
239	if (err)
240		goto vcpu_uninit;
241
242	return vcpu;
243vcpu_uninit:
244	kvm_vcpu_uninit(vcpu);
245free_vcpu:
246	kmem_cache_free(kvm_vcpu_cache, vcpu);
247out:
248	return ERR_PTR(err);
249}
250
251void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
252{
253	kvm_vgic_vcpu_early_init(vcpu);
254}
255
256void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
257{
258	kvm_mmu_free_memory_caches(vcpu);
259	kvm_timer_vcpu_terminate(vcpu);
260	kvm_vgic_vcpu_destroy(vcpu);
261	kmem_cache_free(kvm_vcpu_cache, vcpu);
262}
263
264void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
265{
266	kvm_arch_vcpu_free(vcpu);
267}
268
269int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
270{
271	return kvm_timer_should_fire(vcpu);
272}
273
274void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
275{
276	kvm_timer_schedule(vcpu);
277}
278
279void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
280{
281	kvm_timer_unschedule(vcpu);
282}
283
284int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
285{
286	/* Force users to call KVM_ARM_VCPU_INIT */
287	vcpu->arch.target = -1;
288	bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
289
290	/* Set up the timer */
291	kvm_timer_vcpu_init(vcpu);
292
293	kvm_arm_reset_debug_ptr(vcpu);
294
295	return 0;
296}
297
298void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
299{
300	vcpu->cpu = cpu;
301	vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
302
303	kvm_arm_set_running_vcpu(vcpu);
304}
305
306void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
307{
308	/*
309	 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
310	 * if the vcpu is no longer assigned to a cpu.  This is used for the
311	 * optimized make_all_cpus_request path.
312	 */
313	vcpu->cpu = -1;
314
315	kvm_arm_set_running_vcpu(NULL);
316}
317
318int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
319				    struct kvm_mp_state *mp_state)
320{
321	if (vcpu->arch.power_off)
322		mp_state->mp_state = KVM_MP_STATE_STOPPED;
323	else
324		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
325
326	return 0;
327}
328
329int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
330				    struct kvm_mp_state *mp_state)
331{
332	switch (mp_state->mp_state) {
333	case KVM_MP_STATE_RUNNABLE:
334		vcpu->arch.power_off = false;
335		break;
336	case KVM_MP_STATE_STOPPED:
337		vcpu->arch.power_off = true;
338		break;
339	default:
340		return -EINVAL;
341	}
342
343	return 0;
344}
345
346/**
347 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
348 * @v:		The VCPU pointer
349 *
350 * If the guest CPU is not waiting for interrupts or an interrupt line is
351 * asserted, the CPU is by definition runnable.
352 */
353int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
354{
355	return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
356		&& !v->arch.power_off && !v->arch.pause);
357}
358
359/* Just ensure a guest exit from a particular CPU */
360static void exit_vm_noop(void *info)
361{
362}
363
364void force_vm_exit(const cpumask_t *mask)
365{
366	smp_call_function_many(mask, exit_vm_noop, NULL, true);
367}
368
369/**
370 * need_new_vmid_gen - check that the VMID is still valid
371 * @kvm: The VM's VMID to checkt
372 *
373 * return true if there is a new generation of VMIDs being used
374 *
375 * The hardware supports only 256 values with the value zero reserved for the
376 * host, so we check if an assigned value belongs to a previous generation,
377 * which which requires us to assign a new value. If we're the first to use a
378 * VMID for the new generation, we must flush necessary caches and TLBs on all
379 * CPUs.
380 */
381static bool need_new_vmid_gen(struct kvm *kvm)
382{
383	return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
384}
385
386/**
387 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
388 * @kvm	The guest that we are about to run
389 *
390 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
391 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
392 * caches and TLBs.
393 */
394static void update_vttbr(struct kvm *kvm)
395{
396	phys_addr_t pgd_phys;
397	u64 vmid;
398
399	if (!need_new_vmid_gen(kvm))
400		return;
401
402	spin_lock(&kvm_vmid_lock);
403
404	/*
405	 * We need to re-check the vmid_gen here to ensure that if another vcpu
406	 * already allocated a valid vmid for this vm, then this vcpu should
407	 * use the same vmid.
408	 */
409	if (!need_new_vmid_gen(kvm)) {
410		spin_unlock(&kvm_vmid_lock);
411		return;
412	}
413
414	/* First user of a new VMID generation? */
415	if (unlikely(kvm_next_vmid == 0)) {
416		atomic64_inc(&kvm_vmid_gen);
417		kvm_next_vmid = 1;
418
419		/*
420		 * On SMP we know no other CPUs can use this CPU's or each
421		 * other's VMID after force_vm_exit returns since the
422		 * kvm_vmid_lock blocks them from reentry to the guest.
423		 */
424		force_vm_exit(cpu_all_mask);
425		/*
426		 * Now broadcast TLB + ICACHE invalidation over the inner
427		 * shareable domain to make sure all data structures are
428		 * clean.
429		 */
430		kvm_call_hyp(__kvm_flush_vm_context);
431	}
432
433	kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
434	kvm->arch.vmid = kvm_next_vmid;
435	kvm_next_vmid++;
436
437	/* update vttbr to be used with the new vmid */
438	pgd_phys = virt_to_phys(kvm_get_hwpgd(kvm));
439	BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
440	vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
441	kvm->arch.vttbr = pgd_phys | vmid;
442
443	spin_unlock(&kvm_vmid_lock);
444}
445
446static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
447{
448	struct kvm *kvm = vcpu->kvm;
449	int ret;
450
451	if (likely(vcpu->arch.has_run_once))
452		return 0;
453
454	vcpu->arch.has_run_once = true;
455
456	/*
457	 * Map the VGIC hardware resources before running a vcpu the first
458	 * time on this VM.
459	 */
460	if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
461		ret = kvm_vgic_map_resources(kvm);
462		if (ret)
463			return ret;
464	}
465
466	/*
467	 * Enable the arch timers only if we have an in-kernel VGIC
468	 * and it has been properly initialized, since we cannot handle
469	 * interrupts from the virtual timer with a userspace gic.
470	 */
471	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
472		kvm_timer_enable(kvm);
473
474	return 0;
475}
476
477bool kvm_arch_intc_initialized(struct kvm *kvm)
478{
479	return vgic_initialized(kvm);
480}
481
482static void kvm_arm_halt_guest(struct kvm *kvm) __maybe_unused;
483static void kvm_arm_resume_guest(struct kvm *kvm) __maybe_unused;
484
485static void kvm_arm_halt_guest(struct kvm *kvm)
486{
487	int i;
488	struct kvm_vcpu *vcpu;
489
490	kvm_for_each_vcpu(i, vcpu, kvm)
491		vcpu->arch.pause = true;
492	force_vm_exit(cpu_all_mask);
493}
494
495static void kvm_arm_resume_guest(struct kvm *kvm)
496{
497	int i;
498	struct kvm_vcpu *vcpu;
499
500	kvm_for_each_vcpu(i, vcpu, kvm) {
501		wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
502
503		vcpu->arch.pause = false;
504		wake_up_interruptible(wq);
505	}
506}
507
508static void vcpu_sleep(struct kvm_vcpu *vcpu)
509{
510	wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
511
512	wait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
513				       (!vcpu->arch.pause)));
514}
515
516static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
517{
518	return vcpu->arch.target >= 0;
519}
520
521/**
522 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
523 * @vcpu:	The VCPU pointer
524 * @run:	The kvm_run structure pointer used for userspace state exchange
525 *
526 * This function is called through the VCPU_RUN ioctl called from user space. It
527 * will execute VM code in a loop until the time slice for the process is used
528 * or some emulation is needed from user space in which case the function will
529 * return with return value 0 and with the kvm_run structure filled in with the
530 * required data for the requested emulation.
531 */
532int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
533{
534	int ret;
535	sigset_t sigsaved;
536
537	if (unlikely(!kvm_vcpu_initialized(vcpu)))
538		return -ENOEXEC;
539
540	ret = kvm_vcpu_first_run_init(vcpu);
541	if (ret)
542		return ret;
543
544	if (run->exit_reason == KVM_EXIT_MMIO) {
545		ret = kvm_handle_mmio_return(vcpu, vcpu->run);
546		if (ret)
547			return ret;
548	}
549
550	if (vcpu->sigset_active)
551		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
552
553	ret = 1;
554	run->exit_reason = KVM_EXIT_UNKNOWN;
555	while (ret > 0) {
556		/*
557		 * Check conditions before entering the guest
558		 */
559		cond_resched();
560
561		update_vttbr(vcpu->kvm);
562
563		if (vcpu->arch.power_off || vcpu->arch.pause)
564			vcpu_sleep(vcpu);
565
566		/*
567		 * Preparing the interrupts to be injected also
568		 * involves poking the GIC, which must be done in a
569		 * non-preemptible context.
570		 */
571		preempt_disable();
572		kvm_timer_flush_hwstate(vcpu);
573		kvm_vgic_flush_hwstate(vcpu);
574
575		local_irq_disable();
576
577		/*
578		 * Re-check atomic conditions
579		 */
580		if (signal_pending(current)) {
581			ret = -EINTR;
582			run->exit_reason = KVM_EXIT_INTR;
583		}
584
585		if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
586			vcpu->arch.power_off || vcpu->arch.pause) {
587			local_irq_enable();
588			kvm_timer_sync_hwstate(vcpu);
589			kvm_vgic_sync_hwstate(vcpu);
590			preempt_enable();
591			continue;
592		}
593
594		kvm_arm_setup_debug(vcpu);
595
596		/**************************************************************
597		 * Enter the guest
598		 */
599		trace_kvm_entry(*vcpu_pc(vcpu));
600		__kvm_guest_enter();
601		vcpu->mode = IN_GUEST_MODE;
602
603		ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
604
605		vcpu->mode = OUTSIDE_GUEST_MODE;
606		/*
607		 * Back from guest
608		 *************************************************************/
609
610		kvm_arm_clear_debug(vcpu);
611
612		/*
613		 * We may have taken a host interrupt in HYP mode (ie
614		 * while executing the guest). This interrupt is still
615		 * pending, as we haven't serviced it yet!
616		 *
617		 * We're now back in SVC mode, with interrupts
618		 * disabled.  Enabling the interrupts now will have
619		 * the effect of taking the interrupt again, in SVC
620		 * mode this time.
621		 */
622		local_irq_enable();
623
624		/*
625		 * We do local_irq_enable() before calling kvm_guest_exit() so
626		 * that if a timer interrupt hits while running the guest we
627		 * account that tick as being spent in the guest.  We enable
628		 * preemption after calling kvm_guest_exit() so that if we get
629		 * preempted we make sure ticks after that is not counted as
630		 * guest time.
631		 */
632		kvm_guest_exit();
633		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
634
635		/*
636		 * We must sync the timer state before the vgic state so that
637		 * the vgic can properly sample the updated state of the
638		 * interrupt line.
639		 */
640		kvm_timer_sync_hwstate(vcpu);
641
642		kvm_vgic_sync_hwstate(vcpu);
643
644		preempt_enable();
645
646		ret = handle_exit(vcpu, run, ret);
647	}
648
649	if (vcpu->sigset_active)
650		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
651	return ret;
652}
653
654static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
655{
656	int bit_index;
657	bool set;
658	unsigned long *ptr;
659
660	if (number == KVM_ARM_IRQ_CPU_IRQ)
661		bit_index = __ffs(HCR_VI);
662	else /* KVM_ARM_IRQ_CPU_FIQ */
663		bit_index = __ffs(HCR_VF);
664
665	ptr = (unsigned long *)&vcpu->arch.irq_lines;
666	if (level)
667		set = test_and_set_bit(bit_index, ptr);
668	else
669		set = test_and_clear_bit(bit_index, ptr);
670
671	/*
672	 * If we didn't change anything, no need to wake up or kick other CPUs
673	 */
674	if (set == level)
675		return 0;
676
677	/*
678	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
679	 * trigger a world-switch round on the running physical CPU to set the
680	 * virtual IRQ/FIQ fields in the HCR appropriately.
681	 */
682	kvm_vcpu_kick(vcpu);
683
684	return 0;
685}
686
687int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
688			  bool line_status)
689{
690	u32 irq = irq_level->irq;
691	unsigned int irq_type, vcpu_idx, irq_num;
692	int nrcpus = atomic_read(&kvm->online_vcpus);
693	struct kvm_vcpu *vcpu = NULL;
694	bool level = irq_level->level;
695
696	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
697	vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
698	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
699
700	trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
701
702	switch (irq_type) {
703	case KVM_ARM_IRQ_TYPE_CPU:
704		if (irqchip_in_kernel(kvm))
705			return -ENXIO;
706
707		if (vcpu_idx >= nrcpus)
708			return -EINVAL;
709
710		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
711		if (!vcpu)
712			return -EINVAL;
713
714		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
715			return -EINVAL;
716
717		return vcpu_interrupt_line(vcpu, irq_num, level);
718	case KVM_ARM_IRQ_TYPE_PPI:
719		if (!irqchip_in_kernel(kvm))
720			return -ENXIO;
721
722		if (vcpu_idx >= nrcpus)
723			return -EINVAL;
724
725		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
726		if (!vcpu)
727			return -EINVAL;
728
729		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
730			return -EINVAL;
731
732		return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
733	case KVM_ARM_IRQ_TYPE_SPI:
734		if (!irqchip_in_kernel(kvm))
735			return -ENXIO;
736
737		if (irq_num < VGIC_NR_PRIVATE_IRQS)
738			return -EINVAL;
739
740		return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
741	}
742
743	return -EINVAL;
744}
745
746static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
747			       const struct kvm_vcpu_init *init)
748{
749	unsigned int i;
750	int phys_target = kvm_target_cpu();
751
752	if (init->target != phys_target)
753		return -EINVAL;
754
755	/*
756	 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
757	 * use the same target.
758	 */
759	if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
760		return -EINVAL;
761
762	/* -ENOENT for unknown features, -EINVAL for invalid combinations. */
763	for (i = 0; i < sizeof(init->features) * 8; i++) {
764		bool set = (init->features[i / 32] & (1 << (i % 32)));
765
766		if (set && i >= KVM_VCPU_MAX_FEATURES)
767			return -ENOENT;
768
769		/*
770		 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
771		 * use the same feature set.
772		 */
773		if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
774		    test_bit(i, vcpu->arch.features) != set)
775			return -EINVAL;
776
777		if (set)
778			set_bit(i, vcpu->arch.features);
779	}
780
781	vcpu->arch.target = phys_target;
782
783	/* Now we know what it is, we can reset it. */
784	return kvm_reset_vcpu(vcpu);
785}
786
787
788static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
789					 struct kvm_vcpu_init *init)
790{
791	int ret;
792
793	ret = kvm_vcpu_set_target(vcpu, init);
794	if (ret)
795		return ret;
796
797	/*
798	 * Ensure a rebooted VM will fault in RAM pages and detect if the
799	 * guest MMU is turned off and flush the caches as needed.
800	 */
801	if (vcpu->arch.has_run_once)
802		stage2_unmap_vm(vcpu->kvm);
803
804	vcpu_reset_hcr(vcpu);
805
806	/*
807	 * Handle the "start in power-off" case.
808	 */
809	if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
810		vcpu->arch.power_off = true;
811	else
812		vcpu->arch.power_off = false;
813
814	return 0;
815}
816
817long kvm_arch_vcpu_ioctl(struct file *filp,
818			 unsigned int ioctl, unsigned long arg)
819{
820	struct kvm_vcpu *vcpu = filp->private_data;
821	void __user *argp = (void __user *)arg;
822
823	switch (ioctl) {
824	case KVM_ARM_VCPU_INIT: {
825		struct kvm_vcpu_init init;
826
827		if (copy_from_user(&init, argp, sizeof(init)))
828			return -EFAULT;
829
830		return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
831	}
832	case KVM_SET_ONE_REG:
833	case KVM_GET_ONE_REG: {
834		struct kvm_one_reg reg;
835
836		if (unlikely(!kvm_vcpu_initialized(vcpu)))
837			return -ENOEXEC;
838
839		if (copy_from_user(&reg, argp, sizeof(reg)))
840			return -EFAULT;
841		if (ioctl == KVM_SET_ONE_REG)
842			return kvm_arm_set_reg(vcpu, &reg);
843		else
844			return kvm_arm_get_reg(vcpu, &reg);
845	}
846	case KVM_GET_REG_LIST: {
847		struct kvm_reg_list __user *user_list = argp;
848		struct kvm_reg_list reg_list;
849		unsigned n;
850
851		if (unlikely(!kvm_vcpu_initialized(vcpu)))
852			return -ENOEXEC;
853
854		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
855			return -EFAULT;
856		n = reg_list.n;
857		reg_list.n = kvm_arm_num_regs(vcpu);
858		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
859			return -EFAULT;
860		if (n < reg_list.n)
861			return -E2BIG;
862		return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
863	}
864	default:
865		return -EINVAL;
866	}
867}
868
869/**
870 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
871 * @kvm: kvm instance
872 * @log: slot id and address to which we copy the log
873 *
874 * Steps 1-4 below provide general overview of dirty page logging. See
875 * kvm_get_dirty_log_protect() function description for additional details.
876 *
877 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
878 * always flush the TLB (step 4) even if previous step failed  and the dirty
879 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
880 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
881 * writes will be marked dirty for next log read.
882 *
883 *   1. Take a snapshot of the bit and clear it if needed.
884 *   2. Write protect the corresponding page.
885 *   3. Copy the snapshot to the userspace.
886 *   4. Flush TLB's if needed.
887 */
888int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
889{
890	bool is_dirty = false;
891	int r;
892
893	mutex_lock(&kvm->slots_lock);
894
895	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
896
897	if (is_dirty)
898		kvm_flush_remote_tlbs(kvm);
899
900	mutex_unlock(&kvm->slots_lock);
901	return r;
902}
903
904static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
905					struct kvm_arm_device_addr *dev_addr)
906{
907	unsigned long dev_id, type;
908
909	dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
910		KVM_ARM_DEVICE_ID_SHIFT;
911	type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
912		KVM_ARM_DEVICE_TYPE_SHIFT;
913
914	switch (dev_id) {
915	case KVM_ARM_DEVICE_VGIC_V2:
916		return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
917	default:
918		return -ENODEV;
919	}
920}
921
922long kvm_arch_vm_ioctl(struct file *filp,
923		       unsigned int ioctl, unsigned long arg)
924{
925	struct kvm *kvm = filp->private_data;
926	void __user *argp = (void __user *)arg;
927
928	switch (ioctl) {
929	case KVM_CREATE_IRQCHIP: {
930		return kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
931	}
932	case KVM_ARM_SET_DEVICE_ADDR: {
933		struct kvm_arm_device_addr dev_addr;
934
935		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
936			return -EFAULT;
937		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
938	}
939	case KVM_ARM_PREFERRED_TARGET: {
940		int err;
941		struct kvm_vcpu_init init;
942
943		err = kvm_vcpu_preferred_target(&init);
944		if (err)
945			return err;
946
947		if (copy_to_user(argp, &init, sizeof(init)))
948			return -EFAULT;
949
950		return 0;
951	}
952	default:
953		return -EINVAL;
954	}
955}
956
957static void cpu_init_hyp_mode(void *dummy)
958{
959	phys_addr_t boot_pgd_ptr;
960	phys_addr_t pgd_ptr;
961	unsigned long hyp_stack_ptr;
962	unsigned long stack_page;
963	unsigned long vector_ptr;
964
965	/* Switch from the HYP stub to our own HYP init vector */
966	__hyp_set_vectors(kvm_get_idmap_vector());
967
968	boot_pgd_ptr = kvm_mmu_get_boot_httbr();
969	pgd_ptr = kvm_mmu_get_httbr();
970	stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
971	hyp_stack_ptr = stack_page + PAGE_SIZE;
972	vector_ptr = (unsigned long)__kvm_hyp_vector;
973
974	__cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
975
976	kvm_arm_init_debug();
977}
978
979static int hyp_init_cpu_notify(struct notifier_block *self,
980			       unsigned long action, void *cpu)
981{
982	switch (action) {
983	case CPU_STARTING:
984	case CPU_STARTING_FROZEN:
985		if (__hyp_get_vectors() == hyp_default_vectors)
986			cpu_init_hyp_mode(NULL);
987		break;
988	}
989
990	return NOTIFY_OK;
991}
992
993static struct notifier_block hyp_init_cpu_nb = {
994	.notifier_call = hyp_init_cpu_notify,
995};
996
997#ifdef CONFIG_CPU_PM
998static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
999				    unsigned long cmd,
1000				    void *v)
1001{
1002	if (cmd == CPU_PM_EXIT &&
1003	    __hyp_get_vectors() == hyp_default_vectors) {
1004		cpu_init_hyp_mode(NULL);
1005		return NOTIFY_OK;
1006	}
1007
1008	return NOTIFY_DONE;
1009}
1010
1011static struct notifier_block hyp_init_cpu_pm_nb = {
1012	.notifier_call = hyp_init_cpu_pm_notifier,
1013};
1014
1015static void __init hyp_cpu_pm_init(void)
1016{
1017	cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1018}
1019#else
1020static inline void hyp_cpu_pm_init(void)
1021{
1022}
1023#endif
1024
1025/**
1026 * Inits Hyp-mode on all online CPUs
1027 */
1028static int init_hyp_mode(void)
1029{
1030	int cpu;
1031	int err = 0;
1032
1033	/*
1034	 * Allocate Hyp PGD and setup Hyp identity mapping
1035	 */
1036	err = kvm_mmu_init();
1037	if (err)
1038		goto out_err;
1039
1040	/*
1041	 * It is probably enough to obtain the default on one
1042	 * CPU. It's unlikely to be different on the others.
1043	 */
1044	hyp_default_vectors = __hyp_get_vectors();
1045
1046	/*
1047	 * Allocate stack pages for Hypervisor-mode
1048	 */
1049	for_each_possible_cpu(cpu) {
1050		unsigned long stack_page;
1051
1052		stack_page = __get_free_page(GFP_KERNEL);
1053		if (!stack_page) {
1054			err = -ENOMEM;
1055			goto out_free_stack_pages;
1056		}
1057
1058		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1059	}
1060
1061	/*
1062	 * Map the Hyp-code called directly from the host
1063	 */
1064	err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
1065	if (err) {
1066		kvm_err("Cannot map world-switch code\n");
1067		goto out_free_mappings;
1068	}
1069
1070	/*
1071	 * Map the Hyp stack pages
1072	 */
1073	for_each_possible_cpu(cpu) {
1074		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1075		err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
1076
1077		if (err) {
1078			kvm_err("Cannot map hyp stack\n");
1079			goto out_free_mappings;
1080		}
1081	}
1082
1083	/*
1084	 * Map the host CPU structures
1085	 */
1086	kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1087	if (!kvm_host_cpu_state) {
1088		err = -ENOMEM;
1089		kvm_err("Cannot allocate host CPU state\n");
1090		goto out_free_mappings;
1091	}
1092
1093	for_each_possible_cpu(cpu) {
1094		kvm_cpu_context_t *cpu_ctxt;
1095
1096		cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1097		err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
1098
1099		if (err) {
1100			kvm_err("Cannot map host CPU state: %d\n", err);
1101			goto out_free_context;
1102		}
1103	}
1104
1105	/*
1106	 * Execute the init code on each CPU.
1107	 */
1108	on_each_cpu(cpu_init_hyp_mode, NULL, 1);
1109
1110	/*
1111	 * Init HYP view of VGIC
1112	 */
1113	err = kvm_vgic_hyp_init();
1114	if (err)
1115		goto out_free_context;
1116
1117	/*
1118	 * Init HYP architected timer support
1119	 */
1120	err = kvm_timer_hyp_init();
1121	if (err)
1122		goto out_free_context;
1123
1124#ifndef CONFIG_HOTPLUG_CPU
1125	free_boot_hyp_pgd();
1126#endif
1127
1128	kvm_perf_init();
1129
1130	kvm_info("Hyp mode initialized successfully\n");
1131
1132	return 0;
1133out_free_context:
1134	free_percpu(kvm_host_cpu_state);
1135out_free_mappings:
1136	free_hyp_pgds();
1137out_free_stack_pages:
1138	for_each_possible_cpu(cpu)
1139		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1140out_err:
1141	kvm_err("error initializing Hyp mode: %d\n", err);
1142	return err;
1143}
1144
1145static void check_kvm_target_cpu(void *ret)
1146{
1147	*(int *)ret = kvm_target_cpu();
1148}
1149
1150struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1151{
1152	struct kvm_vcpu *vcpu;
1153	int i;
1154
1155	mpidr &= MPIDR_HWID_BITMASK;
1156	kvm_for_each_vcpu(i, vcpu, kvm) {
1157		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1158			return vcpu;
1159	}
1160	return NULL;
1161}
1162
1163/**
1164 * Initialize Hyp-mode and memory mappings on all CPUs.
1165 */
1166int kvm_arch_init(void *opaque)
1167{
1168	int err;
1169	int ret, cpu;
1170
1171	if (!is_hyp_mode_available()) {
1172		kvm_err("HYP mode not available\n");
1173		return -ENODEV;
1174	}
1175
1176	for_each_online_cpu(cpu) {
1177		smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1178		if (ret < 0) {
1179			kvm_err("Error, CPU %d not supported!\n", cpu);
1180			return -ENODEV;
1181		}
1182	}
1183
1184	cpu_notifier_register_begin();
1185
1186	err = init_hyp_mode();
1187	if (err)
1188		goto out_err;
1189
1190	err = __register_cpu_notifier(&hyp_init_cpu_nb);
1191	if (err) {
1192		kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1193		goto out_err;
1194	}
1195
1196	cpu_notifier_register_done();
1197
1198	hyp_cpu_pm_init();
1199
1200	kvm_coproc_table_init();
1201	return 0;
1202out_err:
1203	cpu_notifier_register_done();
1204	return err;
1205}
1206
1207/* NOP: Compiling as a module not supported */
1208void kvm_arch_exit(void)
1209{
1210	kvm_perf_teardown();
1211}
1212
1213static int arm_init(void)
1214{
1215	int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1216	return rc;
1217}
1218
1219module_init(arm_init);
1220