ASoC: max98371 Remove duplicate entry in max98371_reg
[cascardo/linux.git] / arch / arm / kvm / arm.c
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 #include <kvm/arm_pmu.h>
32
33 #define CREATE_TRACE_POINTS
34 #include "trace.h"
35
36 #include <asm/uaccess.h>
37 #include <asm/ptrace.h>
38 #include <asm/mman.h>
39 #include <asm/tlbflush.h>
40 #include <asm/cacheflush.h>
41 #include <asm/virt.h>
42 #include <asm/kvm_arm.h>
43 #include <asm/kvm_asm.h>
44 #include <asm/kvm_mmu.h>
45 #include <asm/kvm_emulate.h>
46 #include <asm/kvm_coproc.h>
47 #include <asm/kvm_psci.h>
48 #include <asm/sections.h>
49
50 #ifdef REQUIRES_VIRT
51 __asm__(".arch_extension        virt");
52 #endif
53
54 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
55 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
56 static unsigned long hyp_default_vectors;
57
58 /* Per-CPU variable containing the currently running vcpu. */
59 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
60
61 /* The VMID used in the VTTBR */
62 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
63 static u32 kvm_next_vmid;
64 static unsigned int kvm_vmid_bits __read_mostly;
65 static DEFINE_SPINLOCK(kvm_vmid_lock);
66
67 static bool vgic_present;
68
69 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
70 {
71         BUG_ON(preemptible());
72         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
73 }
74
75 /**
76  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
77  * Must be called from non-preemptible context
78  */
79 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
80 {
81         BUG_ON(preemptible());
82         return __this_cpu_read(kvm_arm_running_vcpu);
83 }
84
85 /**
86  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
87  */
88 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
89 {
90         return &kvm_arm_running_vcpu;
91 }
92
93 int kvm_arch_hardware_enable(void)
94 {
95         return 0;
96 }
97
98 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
99 {
100         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
101 }
102
103 int kvm_arch_hardware_setup(void)
104 {
105         return 0;
106 }
107
108 void kvm_arch_check_processor_compat(void *rtn)
109 {
110         *(int *)rtn = 0;
111 }
112
113
114 /**
115  * kvm_arch_init_vm - initializes a VM data structure
116  * @kvm:        pointer to the KVM struct
117  */
118 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
119 {
120         int ret = 0;
121
122         if (type)
123                 return -EINVAL;
124
125         ret = kvm_alloc_stage2_pgd(kvm);
126         if (ret)
127                 goto out_fail_alloc;
128
129         ret = create_hyp_mappings(kvm, kvm + 1);
130         if (ret)
131                 goto out_free_stage2_pgd;
132
133         kvm_vgic_early_init(kvm);
134         kvm_timer_init(kvm);
135
136         /* Mark the initial VMID generation invalid */
137         kvm->arch.vmid_gen = 0;
138
139         /* The maximum number of VCPUs is limited by the host's GIC model */
140         kvm->arch.max_vcpus = vgic_present ?
141                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
142
143         return ret;
144 out_free_stage2_pgd:
145         kvm_free_stage2_pgd(kvm);
146 out_fail_alloc:
147         return ret;
148 }
149
150 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
151 {
152         return VM_FAULT_SIGBUS;
153 }
154
155
156 /**
157  * kvm_arch_destroy_vm - destroy the VM data structure
158  * @kvm:        pointer to the KVM struct
159  */
160 void kvm_arch_destroy_vm(struct kvm *kvm)
161 {
162         int i;
163
164         kvm_free_stage2_pgd(kvm);
165
166         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
167                 if (kvm->vcpus[i]) {
168                         kvm_arch_vcpu_free(kvm->vcpus[i]);
169                         kvm->vcpus[i] = NULL;
170                 }
171         }
172
173         kvm_vgic_destroy(kvm);
174 }
175
176 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
177 {
178         int r;
179         switch (ext) {
180         case KVM_CAP_IRQCHIP:
181                 r = vgic_present;
182                 break;
183         case KVM_CAP_IOEVENTFD:
184         case KVM_CAP_DEVICE_CTRL:
185         case KVM_CAP_USER_MEMORY:
186         case KVM_CAP_SYNC_MMU:
187         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
188         case KVM_CAP_ONE_REG:
189         case KVM_CAP_ARM_PSCI:
190         case KVM_CAP_ARM_PSCI_0_2:
191         case KVM_CAP_READONLY_MEM:
192         case KVM_CAP_MP_STATE:
193                 r = 1;
194                 break;
195         case KVM_CAP_COALESCED_MMIO:
196                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
197                 break;
198         case KVM_CAP_ARM_SET_DEVICE_ADDR:
199                 r = 1;
200                 break;
201         case KVM_CAP_NR_VCPUS:
202                 r = num_online_cpus();
203                 break;
204         case KVM_CAP_MAX_VCPUS:
205                 r = KVM_MAX_VCPUS;
206                 break;
207         default:
208                 r = kvm_arch_dev_ioctl_check_extension(ext);
209                 break;
210         }
211         return r;
212 }
213
214 long kvm_arch_dev_ioctl(struct file *filp,
215                         unsigned int ioctl, unsigned long arg)
216 {
217         return -EINVAL;
218 }
219
220
221 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
222 {
223         int err;
224         struct kvm_vcpu *vcpu;
225
226         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
227                 err = -EBUSY;
228                 goto out;
229         }
230
231         if (id >= kvm->arch.max_vcpus) {
232                 err = -EINVAL;
233                 goto out;
234         }
235
236         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
237         if (!vcpu) {
238                 err = -ENOMEM;
239                 goto out;
240         }
241
242         err = kvm_vcpu_init(vcpu, kvm, id);
243         if (err)
244                 goto free_vcpu;
245
246         err = create_hyp_mappings(vcpu, vcpu + 1);
247         if (err)
248                 goto vcpu_uninit;
249
250         return vcpu;
251 vcpu_uninit:
252         kvm_vcpu_uninit(vcpu);
253 free_vcpu:
254         kmem_cache_free(kvm_vcpu_cache, vcpu);
255 out:
256         return ERR_PTR(err);
257 }
258
259 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
260 {
261         kvm_vgic_vcpu_early_init(vcpu);
262 }
263
264 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
265 {
266         kvm_mmu_free_memory_caches(vcpu);
267         kvm_timer_vcpu_terminate(vcpu);
268         kvm_vgic_vcpu_destroy(vcpu);
269         kvm_pmu_vcpu_destroy(vcpu);
270         kmem_cache_free(kvm_vcpu_cache, vcpu);
271 }
272
273 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
274 {
275         kvm_arch_vcpu_free(vcpu);
276 }
277
278 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
279 {
280         return kvm_timer_should_fire(vcpu);
281 }
282
283 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
284 {
285         kvm_timer_schedule(vcpu);
286 }
287
288 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
289 {
290         kvm_timer_unschedule(vcpu);
291 }
292
293 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
294 {
295         /* Force users to call KVM_ARM_VCPU_INIT */
296         vcpu->arch.target = -1;
297         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
298
299         /* Set up the timer */
300         kvm_timer_vcpu_init(vcpu);
301
302         kvm_arm_reset_debug_ptr(vcpu);
303
304         return 0;
305 }
306
307 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
308 {
309         vcpu->cpu = cpu;
310         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
311
312         kvm_arm_set_running_vcpu(vcpu);
313 }
314
315 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
316 {
317         /*
318          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
319          * if the vcpu is no longer assigned to a cpu.  This is used for the
320          * optimized make_all_cpus_request path.
321          */
322         vcpu->cpu = -1;
323
324         kvm_arm_set_running_vcpu(NULL);
325         kvm_timer_vcpu_put(vcpu);
326 }
327
328 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
329                                     struct kvm_mp_state *mp_state)
330 {
331         if (vcpu->arch.power_off)
332                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
333         else
334                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
335
336         return 0;
337 }
338
339 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
340                                     struct kvm_mp_state *mp_state)
341 {
342         switch (mp_state->mp_state) {
343         case KVM_MP_STATE_RUNNABLE:
344                 vcpu->arch.power_off = false;
345                 break;
346         case KVM_MP_STATE_STOPPED:
347                 vcpu->arch.power_off = true;
348                 break;
349         default:
350                 return -EINVAL;
351         }
352
353         return 0;
354 }
355
356 /**
357  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
358  * @v:          The VCPU pointer
359  *
360  * If the guest CPU is not waiting for interrupts or an interrupt line is
361  * asserted, the CPU is by definition runnable.
362  */
363 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
364 {
365         return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
366                 && !v->arch.power_off && !v->arch.pause);
367 }
368
369 /* Just ensure a guest exit from a particular CPU */
370 static void exit_vm_noop(void *info)
371 {
372 }
373
374 void force_vm_exit(const cpumask_t *mask)
375 {
376         preempt_disable();
377         smp_call_function_many(mask, exit_vm_noop, NULL, true);
378         preempt_enable();
379 }
380
381 /**
382  * need_new_vmid_gen - check that the VMID is still valid
383  * @kvm: The VM's VMID to checkt
384  *
385  * return true if there is a new generation of VMIDs being used
386  *
387  * The hardware supports only 256 values with the value zero reserved for the
388  * host, so we check if an assigned value belongs to a previous generation,
389  * which which requires us to assign a new value. If we're the first to use a
390  * VMID for the new generation, we must flush necessary caches and TLBs on all
391  * CPUs.
392  */
393 static bool need_new_vmid_gen(struct kvm *kvm)
394 {
395         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
396 }
397
398 /**
399  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
400  * @kvm The guest that we are about to run
401  *
402  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
403  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
404  * caches and TLBs.
405  */
406 static void update_vttbr(struct kvm *kvm)
407 {
408         phys_addr_t pgd_phys;
409         u64 vmid;
410
411         if (!need_new_vmid_gen(kvm))
412                 return;
413
414         spin_lock(&kvm_vmid_lock);
415
416         /*
417          * We need to re-check the vmid_gen here to ensure that if another vcpu
418          * already allocated a valid vmid for this vm, then this vcpu should
419          * use the same vmid.
420          */
421         if (!need_new_vmid_gen(kvm)) {
422                 spin_unlock(&kvm_vmid_lock);
423                 return;
424         }
425
426         /* First user of a new VMID generation? */
427         if (unlikely(kvm_next_vmid == 0)) {
428                 atomic64_inc(&kvm_vmid_gen);
429                 kvm_next_vmid = 1;
430
431                 /*
432                  * On SMP we know no other CPUs can use this CPU's or each
433                  * other's VMID after force_vm_exit returns since the
434                  * kvm_vmid_lock blocks them from reentry to the guest.
435                  */
436                 force_vm_exit(cpu_all_mask);
437                 /*
438                  * Now broadcast TLB + ICACHE invalidation over the inner
439                  * shareable domain to make sure all data structures are
440                  * clean.
441                  */
442                 kvm_call_hyp(__kvm_flush_vm_context);
443         }
444
445         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
446         kvm->arch.vmid = kvm_next_vmid;
447         kvm_next_vmid++;
448         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
449
450         /* update vttbr to be used with the new vmid */
451         pgd_phys = virt_to_phys(kvm_get_hwpgd(kvm));
452         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
453         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
454         kvm->arch.vttbr = pgd_phys | vmid;
455
456         spin_unlock(&kvm_vmid_lock);
457 }
458
459 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
460 {
461         struct kvm *kvm = vcpu->kvm;
462         int ret;
463
464         if (likely(vcpu->arch.has_run_once))
465                 return 0;
466
467         vcpu->arch.has_run_once = true;
468
469         /*
470          * Map the VGIC hardware resources before running a vcpu the first
471          * time on this VM.
472          */
473         if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
474                 ret = kvm_vgic_map_resources(kvm);
475                 if (ret)
476                         return ret;
477         }
478
479         /*
480          * Enable the arch timers only if we have an in-kernel VGIC
481          * and it has been properly initialized, since we cannot handle
482          * interrupts from the virtual timer with a userspace gic.
483          */
484         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
485                 kvm_timer_enable(kvm);
486
487         return 0;
488 }
489
490 bool kvm_arch_intc_initialized(struct kvm *kvm)
491 {
492         return vgic_initialized(kvm);
493 }
494
495 static void kvm_arm_halt_guest(struct kvm *kvm) __maybe_unused;
496 static void kvm_arm_resume_guest(struct kvm *kvm) __maybe_unused;
497
498 static void kvm_arm_halt_guest(struct kvm *kvm)
499 {
500         int i;
501         struct kvm_vcpu *vcpu;
502
503         kvm_for_each_vcpu(i, vcpu, kvm)
504                 vcpu->arch.pause = true;
505         force_vm_exit(cpu_all_mask);
506 }
507
508 static void kvm_arm_resume_guest(struct kvm *kvm)
509 {
510         int i;
511         struct kvm_vcpu *vcpu;
512
513         kvm_for_each_vcpu(i, vcpu, kvm) {
514                 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
515
516                 vcpu->arch.pause = false;
517                 swake_up(wq);
518         }
519 }
520
521 static void vcpu_sleep(struct kvm_vcpu *vcpu)
522 {
523         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
524
525         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
526                                        (!vcpu->arch.pause)));
527 }
528
529 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
530 {
531         return vcpu->arch.target >= 0;
532 }
533
534 /**
535  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
536  * @vcpu:       The VCPU pointer
537  * @run:        The kvm_run structure pointer used for userspace state exchange
538  *
539  * This function is called through the VCPU_RUN ioctl called from user space. It
540  * will execute VM code in a loop until the time slice for the process is used
541  * or some emulation is needed from user space in which case the function will
542  * return with return value 0 and with the kvm_run structure filled in with the
543  * required data for the requested emulation.
544  */
545 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
546 {
547         int ret;
548         sigset_t sigsaved;
549
550         if (unlikely(!kvm_vcpu_initialized(vcpu)))
551                 return -ENOEXEC;
552
553         ret = kvm_vcpu_first_run_init(vcpu);
554         if (ret)
555                 return ret;
556
557         if (run->exit_reason == KVM_EXIT_MMIO) {
558                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
559                 if (ret)
560                         return ret;
561         }
562
563         if (vcpu->sigset_active)
564                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
565
566         ret = 1;
567         run->exit_reason = KVM_EXIT_UNKNOWN;
568         while (ret > 0) {
569                 /*
570                  * Check conditions before entering the guest
571                  */
572                 cond_resched();
573
574                 update_vttbr(vcpu->kvm);
575
576                 if (vcpu->arch.power_off || vcpu->arch.pause)
577                         vcpu_sleep(vcpu);
578
579                 /*
580                  * Preparing the interrupts to be injected also
581                  * involves poking the GIC, which must be done in a
582                  * non-preemptible context.
583                  */
584                 preempt_disable();
585                 kvm_pmu_flush_hwstate(vcpu);
586                 kvm_timer_flush_hwstate(vcpu);
587                 kvm_vgic_flush_hwstate(vcpu);
588
589                 local_irq_disable();
590
591                 /*
592                  * Re-check atomic conditions
593                  */
594                 if (signal_pending(current)) {
595                         ret = -EINTR;
596                         run->exit_reason = KVM_EXIT_INTR;
597                 }
598
599                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
600                         vcpu->arch.power_off || vcpu->arch.pause) {
601                         local_irq_enable();
602                         kvm_pmu_sync_hwstate(vcpu);
603                         kvm_timer_sync_hwstate(vcpu);
604                         kvm_vgic_sync_hwstate(vcpu);
605                         preempt_enable();
606                         continue;
607                 }
608
609                 kvm_arm_setup_debug(vcpu);
610
611                 /**************************************************************
612                  * Enter the guest
613                  */
614                 trace_kvm_entry(*vcpu_pc(vcpu));
615                 __kvm_guest_enter();
616                 vcpu->mode = IN_GUEST_MODE;
617
618                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
619
620                 vcpu->mode = OUTSIDE_GUEST_MODE;
621                 vcpu->stat.exits++;
622                 /*
623                  * Back from guest
624                  *************************************************************/
625
626                 kvm_arm_clear_debug(vcpu);
627
628                 /*
629                  * We may have taken a host interrupt in HYP mode (ie
630                  * while executing the guest). This interrupt is still
631                  * pending, as we haven't serviced it yet!
632                  *
633                  * We're now back in SVC mode, with interrupts
634                  * disabled.  Enabling the interrupts now will have
635                  * the effect of taking the interrupt again, in SVC
636                  * mode this time.
637                  */
638                 local_irq_enable();
639
640                 /*
641                  * We do local_irq_enable() before calling kvm_guest_exit() so
642                  * that if a timer interrupt hits while running the guest we
643                  * account that tick as being spent in the guest.  We enable
644                  * preemption after calling kvm_guest_exit() so that if we get
645                  * preempted we make sure ticks after that is not counted as
646                  * guest time.
647                  */
648                 kvm_guest_exit();
649                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
650
651                 /*
652                  * We must sync the PMU and timer state before the vgic state so
653                  * that the vgic can properly sample the updated state of the
654                  * interrupt line.
655                  */
656                 kvm_pmu_sync_hwstate(vcpu);
657                 kvm_timer_sync_hwstate(vcpu);
658
659                 kvm_vgic_sync_hwstate(vcpu);
660
661                 preempt_enable();
662
663                 ret = handle_exit(vcpu, run, ret);
664         }
665
666         if (vcpu->sigset_active)
667                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
668         return ret;
669 }
670
671 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
672 {
673         int bit_index;
674         bool set;
675         unsigned long *ptr;
676
677         if (number == KVM_ARM_IRQ_CPU_IRQ)
678                 bit_index = __ffs(HCR_VI);
679         else /* KVM_ARM_IRQ_CPU_FIQ */
680                 bit_index = __ffs(HCR_VF);
681
682         ptr = (unsigned long *)&vcpu->arch.irq_lines;
683         if (level)
684                 set = test_and_set_bit(bit_index, ptr);
685         else
686                 set = test_and_clear_bit(bit_index, ptr);
687
688         /*
689          * If we didn't change anything, no need to wake up or kick other CPUs
690          */
691         if (set == level)
692                 return 0;
693
694         /*
695          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
696          * trigger a world-switch round on the running physical CPU to set the
697          * virtual IRQ/FIQ fields in the HCR appropriately.
698          */
699         kvm_vcpu_kick(vcpu);
700
701         return 0;
702 }
703
704 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
705                           bool line_status)
706 {
707         u32 irq = irq_level->irq;
708         unsigned int irq_type, vcpu_idx, irq_num;
709         int nrcpus = atomic_read(&kvm->online_vcpus);
710         struct kvm_vcpu *vcpu = NULL;
711         bool level = irq_level->level;
712
713         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
714         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
715         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
716
717         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
718
719         switch (irq_type) {
720         case KVM_ARM_IRQ_TYPE_CPU:
721                 if (irqchip_in_kernel(kvm))
722                         return -ENXIO;
723
724                 if (vcpu_idx >= nrcpus)
725                         return -EINVAL;
726
727                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
728                 if (!vcpu)
729                         return -EINVAL;
730
731                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
732                         return -EINVAL;
733
734                 return vcpu_interrupt_line(vcpu, irq_num, level);
735         case KVM_ARM_IRQ_TYPE_PPI:
736                 if (!irqchip_in_kernel(kvm))
737                         return -ENXIO;
738
739                 if (vcpu_idx >= nrcpus)
740                         return -EINVAL;
741
742                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
743                 if (!vcpu)
744                         return -EINVAL;
745
746                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
747                         return -EINVAL;
748
749                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
750         case KVM_ARM_IRQ_TYPE_SPI:
751                 if (!irqchip_in_kernel(kvm))
752                         return -ENXIO;
753
754                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
755                         return -EINVAL;
756
757                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
758         }
759
760         return -EINVAL;
761 }
762
763 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
764                                const struct kvm_vcpu_init *init)
765 {
766         unsigned int i;
767         int phys_target = kvm_target_cpu();
768
769         if (init->target != phys_target)
770                 return -EINVAL;
771
772         /*
773          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
774          * use the same target.
775          */
776         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
777                 return -EINVAL;
778
779         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
780         for (i = 0; i < sizeof(init->features) * 8; i++) {
781                 bool set = (init->features[i / 32] & (1 << (i % 32)));
782
783                 if (set && i >= KVM_VCPU_MAX_FEATURES)
784                         return -ENOENT;
785
786                 /*
787                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
788                  * use the same feature set.
789                  */
790                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
791                     test_bit(i, vcpu->arch.features) != set)
792                         return -EINVAL;
793
794                 if (set)
795                         set_bit(i, vcpu->arch.features);
796         }
797
798         vcpu->arch.target = phys_target;
799
800         /* Now we know what it is, we can reset it. */
801         return kvm_reset_vcpu(vcpu);
802 }
803
804
805 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
806                                          struct kvm_vcpu_init *init)
807 {
808         int ret;
809
810         ret = kvm_vcpu_set_target(vcpu, init);
811         if (ret)
812                 return ret;
813
814         /*
815          * Ensure a rebooted VM will fault in RAM pages and detect if the
816          * guest MMU is turned off and flush the caches as needed.
817          */
818         if (vcpu->arch.has_run_once)
819                 stage2_unmap_vm(vcpu->kvm);
820
821         vcpu_reset_hcr(vcpu);
822
823         /*
824          * Handle the "start in power-off" case.
825          */
826         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
827                 vcpu->arch.power_off = true;
828         else
829                 vcpu->arch.power_off = false;
830
831         return 0;
832 }
833
834 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
835                                  struct kvm_device_attr *attr)
836 {
837         int ret = -ENXIO;
838
839         switch (attr->group) {
840         default:
841                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
842                 break;
843         }
844
845         return ret;
846 }
847
848 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
849                                  struct kvm_device_attr *attr)
850 {
851         int ret = -ENXIO;
852
853         switch (attr->group) {
854         default:
855                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
856                 break;
857         }
858
859         return ret;
860 }
861
862 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
863                                  struct kvm_device_attr *attr)
864 {
865         int ret = -ENXIO;
866
867         switch (attr->group) {
868         default:
869                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
870                 break;
871         }
872
873         return ret;
874 }
875
876 long kvm_arch_vcpu_ioctl(struct file *filp,
877                          unsigned int ioctl, unsigned long arg)
878 {
879         struct kvm_vcpu *vcpu = filp->private_data;
880         void __user *argp = (void __user *)arg;
881         struct kvm_device_attr attr;
882
883         switch (ioctl) {
884         case KVM_ARM_VCPU_INIT: {
885                 struct kvm_vcpu_init init;
886
887                 if (copy_from_user(&init, argp, sizeof(init)))
888                         return -EFAULT;
889
890                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
891         }
892         case KVM_SET_ONE_REG:
893         case KVM_GET_ONE_REG: {
894                 struct kvm_one_reg reg;
895
896                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
897                         return -ENOEXEC;
898
899                 if (copy_from_user(&reg, argp, sizeof(reg)))
900                         return -EFAULT;
901                 if (ioctl == KVM_SET_ONE_REG)
902                         return kvm_arm_set_reg(vcpu, &reg);
903                 else
904                         return kvm_arm_get_reg(vcpu, &reg);
905         }
906         case KVM_GET_REG_LIST: {
907                 struct kvm_reg_list __user *user_list = argp;
908                 struct kvm_reg_list reg_list;
909                 unsigned n;
910
911                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
912                         return -ENOEXEC;
913
914                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
915                         return -EFAULT;
916                 n = reg_list.n;
917                 reg_list.n = kvm_arm_num_regs(vcpu);
918                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
919                         return -EFAULT;
920                 if (n < reg_list.n)
921                         return -E2BIG;
922                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
923         }
924         case KVM_SET_DEVICE_ATTR: {
925                 if (copy_from_user(&attr, argp, sizeof(attr)))
926                         return -EFAULT;
927                 return kvm_arm_vcpu_set_attr(vcpu, &attr);
928         }
929         case KVM_GET_DEVICE_ATTR: {
930                 if (copy_from_user(&attr, argp, sizeof(attr)))
931                         return -EFAULT;
932                 return kvm_arm_vcpu_get_attr(vcpu, &attr);
933         }
934         case KVM_HAS_DEVICE_ATTR: {
935                 if (copy_from_user(&attr, argp, sizeof(attr)))
936                         return -EFAULT;
937                 return kvm_arm_vcpu_has_attr(vcpu, &attr);
938         }
939         default:
940                 return -EINVAL;
941         }
942 }
943
944 /**
945  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
946  * @kvm: kvm instance
947  * @log: slot id and address to which we copy the log
948  *
949  * Steps 1-4 below provide general overview of dirty page logging. See
950  * kvm_get_dirty_log_protect() function description for additional details.
951  *
952  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
953  * always flush the TLB (step 4) even if previous step failed  and the dirty
954  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
955  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
956  * writes will be marked dirty for next log read.
957  *
958  *   1. Take a snapshot of the bit and clear it if needed.
959  *   2. Write protect the corresponding page.
960  *   3. Copy the snapshot to the userspace.
961  *   4. Flush TLB's if needed.
962  */
963 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
964 {
965         bool is_dirty = false;
966         int r;
967
968         mutex_lock(&kvm->slots_lock);
969
970         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
971
972         if (is_dirty)
973                 kvm_flush_remote_tlbs(kvm);
974
975         mutex_unlock(&kvm->slots_lock);
976         return r;
977 }
978
979 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
980                                         struct kvm_arm_device_addr *dev_addr)
981 {
982         unsigned long dev_id, type;
983
984         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
985                 KVM_ARM_DEVICE_ID_SHIFT;
986         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
987                 KVM_ARM_DEVICE_TYPE_SHIFT;
988
989         switch (dev_id) {
990         case KVM_ARM_DEVICE_VGIC_V2:
991                 if (!vgic_present)
992                         return -ENXIO;
993                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
994         default:
995                 return -ENODEV;
996         }
997 }
998
999 long kvm_arch_vm_ioctl(struct file *filp,
1000                        unsigned int ioctl, unsigned long arg)
1001 {
1002         struct kvm *kvm = filp->private_data;
1003         void __user *argp = (void __user *)arg;
1004
1005         switch (ioctl) {
1006         case KVM_CREATE_IRQCHIP: {
1007                 if (!vgic_present)
1008                         return -ENXIO;
1009                 return kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1010         }
1011         case KVM_ARM_SET_DEVICE_ADDR: {
1012                 struct kvm_arm_device_addr dev_addr;
1013
1014                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1015                         return -EFAULT;
1016                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1017         }
1018         case KVM_ARM_PREFERRED_TARGET: {
1019                 int err;
1020                 struct kvm_vcpu_init init;
1021
1022                 err = kvm_vcpu_preferred_target(&init);
1023                 if (err)
1024                         return err;
1025
1026                 if (copy_to_user(argp, &init, sizeof(init)))
1027                         return -EFAULT;
1028
1029                 return 0;
1030         }
1031         default:
1032                 return -EINVAL;
1033         }
1034 }
1035
1036 static void cpu_init_stage2(void *dummy)
1037 {
1038         __cpu_init_stage2();
1039 }
1040
1041 static void cpu_init_hyp_mode(void *dummy)
1042 {
1043         phys_addr_t boot_pgd_ptr;
1044         phys_addr_t pgd_ptr;
1045         unsigned long hyp_stack_ptr;
1046         unsigned long stack_page;
1047         unsigned long vector_ptr;
1048
1049         /* Switch from the HYP stub to our own HYP init vector */
1050         __hyp_set_vectors(kvm_get_idmap_vector());
1051
1052         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
1053         pgd_ptr = kvm_mmu_get_httbr();
1054         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1055         hyp_stack_ptr = stack_page + PAGE_SIZE;
1056         vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1057
1058         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
1059         __cpu_init_stage2();
1060
1061         kvm_arm_init_debug();
1062 }
1063
1064 static int hyp_init_cpu_notify(struct notifier_block *self,
1065                                unsigned long action, void *cpu)
1066 {
1067         switch (action) {
1068         case CPU_STARTING:
1069         case CPU_STARTING_FROZEN:
1070                 if (__hyp_get_vectors() == hyp_default_vectors)
1071                         cpu_init_hyp_mode(NULL);
1072                 break;
1073         }
1074
1075         return NOTIFY_OK;
1076 }
1077
1078 static struct notifier_block hyp_init_cpu_nb = {
1079         .notifier_call = hyp_init_cpu_notify,
1080 };
1081
1082 #ifdef CONFIG_CPU_PM
1083 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1084                                     unsigned long cmd,
1085                                     void *v)
1086 {
1087         if (cmd == CPU_PM_EXIT &&
1088             __hyp_get_vectors() == hyp_default_vectors) {
1089                 cpu_init_hyp_mode(NULL);
1090                 return NOTIFY_OK;
1091         }
1092
1093         return NOTIFY_DONE;
1094 }
1095
1096 static struct notifier_block hyp_init_cpu_pm_nb = {
1097         .notifier_call = hyp_init_cpu_pm_notifier,
1098 };
1099
1100 static void __init hyp_cpu_pm_init(void)
1101 {
1102         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1103 }
1104 #else
1105 static inline void hyp_cpu_pm_init(void)
1106 {
1107 }
1108 #endif
1109
1110 static void teardown_common_resources(void)
1111 {
1112         free_percpu(kvm_host_cpu_state);
1113 }
1114
1115 static int init_common_resources(void)
1116 {
1117         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1118         if (!kvm_host_cpu_state) {
1119                 kvm_err("Cannot allocate host CPU state\n");
1120                 return -ENOMEM;
1121         }
1122
1123         return 0;
1124 }
1125
1126 static int init_subsystems(void)
1127 {
1128         int err;
1129
1130         /*
1131          * Init HYP view of VGIC
1132          */
1133         err = kvm_vgic_hyp_init();
1134         switch (err) {
1135         case 0:
1136                 vgic_present = true;
1137                 break;
1138         case -ENODEV:
1139         case -ENXIO:
1140                 vgic_present = false;
1141                 break;
1142         default:
1143                 return err;
1144         }
1145
1146         /*
1147          * Init HYP architected timer support
1148          */
1149         err = kvm_timer_hyp_init();
1150         if (err)
1151                 return err;
1152
1153         kvm_perf_init();
1154         kvm_coproc_table_init();
1155
1156         return 0;
1157 }
1158
1159 static void teardown_hyp_mode(void)
1160 {
1161         int cpu;
1162
1163         if (is_kernel_in_hyp_mode())
1164                 return;
1165
1166         free_hyp_pgds();
1167         for_each_possible_cpu(cpu)
1168                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1169 }
1170
1171 static int init_vhe_mode(void)
1172 {
1173         /*
1174          * Execute the init code on each CPU.
1175          */
1176         on_each_cpu(cpu_init_stage2, NULL, 1);
1177
1178         /* set size of VMID supported by CPU */
1179         kvm_vmid_bits = kvm_get_vmid_bits();
1180         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1181
1182         kvm_info("VHE mode initialized successfully\n");
1183         return 0;
1184 }
1185
1186 /**
1187  * Inits Hyp-mode on all online CPUs
1188  */
1189 static int init_hyp_mode(void)
1190 {
1191         int cpu;
1192         int err = 0;
1193
1194         /*
1195          * Allocate Hyp PGD and setup Hyp identity mapping
1196          */
1197         err = kvm_mmu_init();
1198         if (err)
1199                 goto out_err;
1200
1201         /*
1202          * It is probably enough to obtain the default on one
1203          * CPU. It's unlikely to be different on the others.
1204          */
1205         hyp_default_vectors = __hyp_get_vectors();
1206
1207         /*
1208          * Allocate stack pages for Hypervisor-mode
1209          */
1210         for_each_possible_cpu(cpu) {
1211                 unsigned long stack_page;
1212
1213                 stack_page = __get_free_page(GFP_KERNEL);
1214                 if (!stack_page) {
1215                         err = -ENOMEM;
1216                         goto out_err;
1217                 }
1218
1219                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1220         }
1221
1222         /*
1223          * Map the Hyp-code called directly from the host
1224          */
1225         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1226                                   kvm_ksym_ref(__hyp_text_end));
1227         if (err) {
1228                 kvm_err("Cannot map world-switch code\n");
1229                 goto out_err;
1230         }
1231
1232         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1233                                   kvm_ksym_ref(__end_rodata));
1234         if (err) {
1235                 kvm_err("Cannot map rodata section\n");
1236                 goto out_err;
1237         }
1238
1239         /*
1240          * Map the Hyp stack pages
1241          */
1242         for_each_possible_cpu(cpu) {
1243                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1244                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
1245
1246                 if (err) {
1247                         kvm_err("Cannot map hyp stack\n");
1248                         goto out_err;
1249                 }
1250         }
1251
1252         for_each_possible_cpu(cpu) {
1253                 kvm_cpu_context_t *cpu_ctxt;
1254
1255                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1256                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
1257
1258                 if (err) {
1259                         kvm_err("Cannot map host CPU state: %d\n", err);
1260                         goto out_err;
1261                 }
1262         }
1263
1264         /*
1265          * Execute the init code on each CPU.
1266          */
1267         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
1268
1269 #ifndef CONFIG_HOTPLUG_CPU
1270         free_boot_hyp_pgd();
1271 #endif
1272
1273         cpu_notifier_register_begin();
1274
1275         err = __register_cpu_notifier(&hyp_init_cpu_nb);
1276
1277         cpu_notifier_register_done();
1278
1279         if (err) {
1280                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1281                 goto out_err;
1282         }
1283
1284         hyp_cpu_pm_init();
1285
1286         /* set size of VMID supported by CPU */
1287         kvm_vmid_bits = kvm_get_vmid_bits();
1288         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1289
1290         kvm_info("Hyp mode initialized successfully\n");
1291
1292         return 0;
1293
1294 out_err:
1295         teardown_hyp_mode();
1296         kvm_err("error initializing Hyp mode: %d\n", err);
1297         return err;
1298 }
1299
1300 static void check_kvm_target_cpu(void *ret)
1301 {
1302         *(int *)ret = kvm_target_cpu();
1303 }
1304
1305 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1306 {
1307         struct kvm_vcpu *vcpu;
1308         int i;
1309
1310         mpidr &= MPIDR_HWID_BITMASK;
1311         kvm_for_each_vcpu(i, vcpu, kvm) {
1312                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1313                         return vcpu;
1314         }
1315         return NULL;
1316 }
1317
1318 /**
1319  * Initialize Hyp-mode and memory mappings on all CPUs.
1320  */
1321 int kvm_arch_init(void *opaque)
1322 {
1323         int err;
1324         int ret, cpu;
1325
1326         if (!is_hyp_mode_available()) {
1327                 kvm_err("HYP mode not available\n");
1328                 return -ENODEV;
1329         }
1330
1331         for_each_online_cpu(cpu) {
1332                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1333                 if (ret < 0) {
1334                         kvm_err("Error, CPU %d not supported!\n", cpu);
1335                         return -ENODEV;
1336                 }
1337         }
1338
1339         err = init_common_resources();
1340         if (err)
1341                 return err;
1342
1343         if (is_kernel_in_hyp_mode())
1344                 err = init_vhe_mode();
1345         else
1346                 err = init_hyp_mode();
1347         if (err)
1348                 goto out_err;
1349
1350         err = init_subsystems();
1351         if (err)
1352                 goto out_hyp;
1353
1354         return 0;
1355
1356 out_hyp:
1357         teardown_hyp_mode();
1358 out_err:
1359         teardown_common_resources();
1360         return err;
1361 }
1362
1363 /* NOP: Compiling as a module not supported */
1364 void kvm_arch_exit(void)
1365 {
1366         kvm_perf_teardown();
1367 }
1368
1369 static int arm_init(void)
1370 {
1371         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1372         return rc;
1373 }
1374
1375 module_init(arm_init);