Merge tag 'pci-v4.6-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci
[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         smp_call_function_many(mask, exit_vm_noop, NULL, true);
377 }
378
379 /**
380  * need_new_vmid_gen - check that the VMID is still valid
381  * @kvm: The VM's VMID to checkt
382  *
383  * return true if there is a new generation of VMIDs being used
384  *
385  * The hardware supports only 256 values with the value zero reserved for the
386  * host, so we check if an assigned value belongs to a previous generation,
387  * which which requires us to assign a new value. If we're the first to use a
388  * VMID for the new generation, we must flush necessary caches and TLBs on all
389  * CPUs.
390  */
391 static bool need_new_vmid_gen(struct kvm *kvm)
392 {
393         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
394 }
395
396 /**
397  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
398  * @kvm The guest that we are about to run
399  *
400  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
401  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
402  * caches and TLBs.
403  */
404 static void update_vttbr(struct kvm *kvm)
405 {
406         phys_addr_t pgd_phys;
407         u64 vmid;
408
409         if (!need_new_vmid_gen(kvm))
410                 return;
411
412         spin_lock(&kvm_vmid_lock);
413
414         /*
415          * We need to re-check the vmid_gen here to ensure that if another vcpu
416          * already allocated a valid vmid for this vm, then this vcpu should
417          * use the same vmid.
418          */
419         if (!need_new_vmid_gen(kvm)) {
420                 spin_unlock(&kvm_vmid_lock);
421                 return;
422         }
423
424         /* First user of a new VMID generation? */
425         if (unlikely(kvm_next_vmid == 0)) {
426                 atomic64_inc(&kvm_vmid_gen);
427                 kvm_next_vmid = 1;
428
429                 /*
430                  * On SMP we know no other CPUs can use this CPU's or each
431                  * other's VMID after force_vm_exit returns since the
432                  * kvm_vmid_lock blocks them from reentry to the guest.
433                  */
434                 force_vm_exit(cpu_all_mask);
435                 /*
436                  * Now broadcast TLB + ICACHE invalidation over the inner
437                  * shareable domain to make sure all data structures are
438                  * clean.
439                  */
440                 kvm_call_hyp(__kvm_flush_vm_context);
441         }
442
443         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
444         kvm->arch.vmid = kvm_next_vmid;
445         kvm_next_vmid++;
446         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
447
448         /* update vttbr to be used with the new vmid */
449         pgd_phys = virt_to_phys(kvm_get_hwpgd(kvm));
450         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
451         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
452         kvm->arch.vttbr = pgd_phys | vmid;
453
454         spin_unlock(&kvm_vmid_lock);
455 }
456
457 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
458 {
459         struct kvm *kvm = vcpu->kvm;
460         int ret;
461
462         if (likely(vcpu->arch.has_run_once))
463                 return 0;
464
465         vcpu->arch.has_run_once = true;
466
467         /*
468          * Map the VGIC hardware resources before running a vcpu the first
469          * time on this VM.
470          */
471         if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
472                 ret = kvm_vgic_map_resources(kvm);
473                 if (ret)
474                         return ret;
475         }
476
477         /*
478          * Enable the arch timers only if we have an in-kernel VGIC
479          * and it has been properly initialized, since we cannot handle
480          * interrupts from the virtual timer with a userspace gic.
481          */
482         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
483                 kvm_timer_enable(kvm);
484
485         return 0;
486 }
487
488 bool kvm_arch_intc_initialized(struct kvm *kvm)
489 {
490         return vgic_initialized(kvm);
491 }
492
493 static void kvm_arm_halt_guest(struct kvm *kvm) __maybe_unused;
494 static void kvm_arm_resume_guest(struct kvm *kvm) __maybe_unused;
495
496 static void kvm_arm_halt_guest(struct kvm *kvm)
497 {
498         int i;
499         struct kvm_vcpu *vcpu;
500
501         kvm_for_each_vcpu(i, vcpu, kvm)
502                 vcpu->arch.pause = true;
503         force_vm_exit(cpu_all_mask);
504 }
505
506 static void kvm_arm_resume_guest(struct kvm *kvm)
507 {
508         int i;
509         struct kvm_vcpu *vcpu;
510
511         kvm_for_each_vcpu(i, vcpu, kvm) {
512                 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
513
514                 vcpu->arch.pause = false;
515                 swake_up(wq);
516         }
517 }
518
519 static void vcpu_sleep(struct kvm_vcpu *vcpu)
520 {
521         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
522
523         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
524                                        (!vcpu->arch.pause)));
525 }
526
527 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
528 {
529         return vcpu->arch.target >= 0;
530 }
531
532 /**
533  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
534  * @vcpu:       The VCPU pointer
535  * @run:        The kvm_run structure pointer used for userspace state exchange
536  *
537  * This function is called through the VCPU_RUN ioctl called from user space. It
538  * will execute VM code in a loop until the time slice for the process is used
539  * or some emulation is needed from user space in which case the function will
540  * return with return value 0 and with the kvm_run structure filled in with the
541  * required data for the requested emulation.
542  */
543 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
544 {
545         int ret;
546         sigset_t sigsaved;
547
548         if (unlikely(!kvm_vcpu_initialized(vcpu)))
549                 return -ENOEXEC;
550
551         ret = kvm_vcpu_first_run_init(vcpu);
552         if (ret)
553                 return ret;
554
555         if (run->exit_reason == KVM_EXIT_MMIO) {
556                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
557                 if (ret)
558                         return ret;
559         }
560
561         if (vcpu->sigset_active)
562                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
563
564         ret = 1;
565         run->exit_reason = KVM_EXIT_UNKNOWN;
566         while (ret > 0) {
567                 /*
568                  * Check conditions before entering the guest
569                  */
570                 cond_resched();
571
572                 update_vttbr(vcpu->kvm);
573
574                 if (vcpu->arch.power_off || vcpu->arch.pause)
575                         vcpu_sleep(vcpu);
576
577                 /*
578                  * Preparing the interrupts to be injected also
579                  * involves poking the GIC, which must be done in a
580                  * non-preemptible context.
581                  */
582                 preempt_disable();
583                 kvm_pmu_flush_hwstate(vcpu);
584                 kvm_timer_flush_hwstate(vcpu);
585                 kvm_vgic_flush_hwstate(vcpu);
586
587                 local_irq_disable();
588
589                 /*
590                  * Re-check atomic conditions
591                  */
592                 if (signal_pending(current)) {
593                         ret = -EINTR;
594                         run->exit_reason = KVM_EXIT_INTR;
595                 }
596
597                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
598                         vcpu->arch.power_off || vcpu->arch.pause) {
599                         local_irq_enable();
600                         kvm_pmu_sync_hwstate(vcpu);
601                         kvm_timer_sync_hwstate(vcpu);
602                         kvm_vgic_sync_hwstate(vcpu);
603                         preempt_enable();
604                         continue;
605                 }
606
607                 kvm_arm_setup_debug(vcpu);
608
609                 /**************************************************************
610                  * Enter the guest
611                  */
612                 trace_kvm_entry(*vcpu_pc(vcpu));
613                 __kvm_guest_enter();
614                 vcpu->mode = IN_GUEST_MODE;
615
616                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
617
618                 vcpu->mode = OUTSIDE_GUEST_MODE;
619                 vcpu->stat.exits++;
620                 /*
621                  * Back from guest
622                  *************************************************************/
623
624                 kvm_arm_clear_debug(vcpu);
625
626                 /*
627                  * We may have taken a host interrupt in HYP mode (ie
628                  * while executing the guest). This interrupt is still
629                  * pending, as we haven't serviced it yet!
630                  *
631                  * We're now back in SVC mode, with interrupts
632                  * disabled.  Enabling the interrupts now will have
633                  * the effect of taking the interrupt again, in SVC
634                  * mode this time.
635                  */
636                 local_irq_enable();
637
638                 /*
639                  * We do local_irq_enable() before calling kvm_guest_exit() so
640                  * that if a timer interrupt hits while running the guest we
641                  * account that tick as being spent in the guest.  We enable
642                  * preemption after calling kvm_guest_exit() so that if we get
643                  * preempted we make sure ticks after that is not counted as
644                  * guest time.
645                  */
646                 kvm_guest_exit();
647                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
648
649                 /*
650                  * We must sync the PMU and timer state before the vgic state so
651                  * that the vgic can properly sample the updated state of the
652                  * interrupt line.
653                  */
654                 kvm_pmu_sync_hwstate(vcpu);
655                 kvm_timer_sync_hwstate(vcpu);
656
657                 kvm_vgic_sync_hwstate(vcpu);
658
659                 preempt_enable();
660
661                 ret = handle_exit(vcpu, run, ret);
662         }
663
664         if (vcpu->sigset_active)
665                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
666         return ret;
667 }
668
669 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
670 {
671         int bit_index;
672         bool set;
673         unsigned long *ptr;
674
675         if (number == KVM_ARM_IRQ_CPU_IRQ)
676                 bit_index = __ffs(HCR_VI);
677         else /* KVM_ARM_IRQ_CPU_FIQ */
678                 bit_index = __ffs(HCR_VF);
679
680         ptr = (unsigned long *)&vcpu->arch.irq_lines;
681         if (level)
682                 set = test_and_set_bit(bit_index, ptr);
683         else
684                 set = test_and_clear_bit(bit_index, ptr);
685
686         /*
687          * If we didn't change anything, no need to wake up or kick other CPUs
688          */
689         if (set == level)
690                 return 0;
691
692         /*
693          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
694          * trigger a world-switch round on the running physical CPU to set the
695          * virtual IRQ/FIQ fields in the HCR appropriately.
696          */
697         kvm_vcpu_kick(vcpu);
698
699         return 0;
700 }
701
702 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
703                           bool line_status)
704 {
705         u32 irq = irq_level->irq;
706         unsigned int irq_type, vcpu_idx, irq_num;
707         int nrcpus = atomic_read(&kvm->online_vcpus);
708         struct kvm_vcpu *vcpu = NULL;
709         bool level = irq_level->level;
710
711         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
712         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
713         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
714
715         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
716
717         switch (irq_type) {
718         case KVM_ARM_IRQ_TYPE_CPU:
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 > KVM_ARM_IRQ_CPU_FIQ)
730                         return -EINVAL;
731
732                 return vcpu_interrupt_line(vcpu, irq_num, level);
733         case KVM_ARM_IRQ_TYPE_PPI:
734                 if (!irqchip_in_kernel(kvm))
735                         return -ENXIO;
736
737                 if (vcpu_idx >= nrcpus)
738                         return -EINVAL;
739
740                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
741                 if (!vcpu)
742                         return -EINVAL;
743
744                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
745                         return -EINVAL;
746
747                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
748         case KVM_ARM_IRQ_TYPE_SPI:
749                 if (!irqchip_in_kernel(kvm))
750                         return -ENXIO;
751
752                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
753                         return -EINVAL;
754
755                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
756         }
757
758         return -EINVAL;
759 }
760
761 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
762                                const struct kvm_vcpu_init *init)
763 {
764         unsigned int i;
765         int phys_target = kvm_target_cpu();
766
767         if (init->target != phys_target)
768                 return -EINVAL;
769
770         /*
771          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
772          * use the same target.
773          */
774         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
775                 return -EINVAL;
776
777         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
778         for (i = 0; i < sizeof(init->features) * 8; i++) {
779                 bool set = (init->features[i / 32] & (1 << (i % 32)));
780
781                 if (set && i >= KVM_VCPU_MAX_FEATURES)
782                         return -ENOENT;
783
784                 /*
785                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
786                  * use the same feature set.
787                  */
788                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
789                     test_bit(i, vcpu->arch.features) != set)
790                         return -EINVAL;
791
792                 if (set)
793                         set_bit(i, vcpu->arch.features);
794         }
795
796         vcpu->arch.target = phys_target;
797
798         /* Now we know what it is, we can reset it. */
799         return kvm_reset_vcpu(vcpu);
800 }
801
802
803 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
804                                          struct kvm_vcpu_init *init)
805 {
806         int ret;
807
808         ret = kvm_vcpu_set_target(vcpu, init);
809         if (ret)
810                 return ret;
811
812         /*
813          * Ensure a rebooted VM will fault in RAM pages and detect if the
814          * guest MMU is turned off and flush the caches as needed.
815          */
816         if (vcpu->arch.has_run_once)
817                 stage2_unmap_vm(vcpu->kvm);
818
819         vcpu_reset_hcr(vcpu);
820
821         /*
822          * Handle the "start in power-off" case.
823          */
824         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
825                 vcpu->arch.power_off = true;
826         else
827                 vcpu->arch.power_off = false;
828
829         return 0;
830 }
831
832 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
833                                  struct kvm_device_attr *attr)
834 {
835         int ret = -ENXIO;
836
837         switch (attr->group) {
838         default:
839                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
840                 break;
841         }
842
843         return ret;
844 }
845
846 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
847                                  struct kvm_device_attr *attr)
848 {
849         int ret = -ENXIO;
850
851         switch (attr->group) {
852         default:
853                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
854                 break;
855         }
856
857         return ret;
858 }
859
860 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
861                                  struct kvm_device_attr *attr)
862 {
863         int ret = -ENXIO;
864
865         switch (attr->group) {
866         default:
867                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
868                 break;
869         }
870
871         return ret;
872 }
873
874 long kvm_arch_vcpu_ioctl(struct file *filp,
875                          unsigned int ioctl, unsigned long arg)
876 {
877         struct kvm_vcpu *vcpu = filp->private_data;
878         void __user *argp = (void __user *)arg;
879         struct kvm_device_attr attr;
880
881         switch (ioctl) {
882         case KVM_ARM_VCPU_INIT: {
883                 struct kvm_vcpu_init init;
884
885                 if (copy_from_user(&init, argp, sizeof(init)))
886                         return -EFAULT;
887
888                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
889         }
890         case KVM_SET_ONE_REG:
891         case KVM_GET_ONE_REG: {
892                 struct kvm_one_reg reg;
893
894                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
895                         return -ENOEXEC;
896
897                 if (copy_from_user(&reg, argp, sizeof(reg)))
898                         return -EFAULT;
899                 if (ioctl == KVM_SET_ONE_REG)
900                         return kvm_arm_set_reg(vcpu, &reg);
901                 else
902                         return kvm_arm_get_reg(vcpu, &reg);
903         }
904         case KVM_GET_REG_LIST: {
905                 struct kvm_reg_list __user *user_list = argp;
906                 struct kvm_reg_list reg_list;
907                 unsigned n;
908
909                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
910                         return -ENOEXEC;
911
912                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
913                         return -EFAULT;
914                 n = reg_list.n;
915                 reg_list.n = kvm_arm_num_regs(vcpu);
916                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
917                         return -EFAULT;
918                 if (n < reg_list.n)
919                         return -E2BIG;
920                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
921         }
922         case KVM_SET_DEVICE_ATTR: {
923                 if (copy_from_user(&attr, argp, sizeof(attr)))
924                         return -EFAULT;
925                 return kvm_arm_vcpu_set_attr(vcpu, &attr);
926         }
927         case KVM_GET_DEVICE_ATTR: {
928                 if (copy_from_user(&attr, argp, sizeof(attr)))
929                         return -EFAULT;
930                 return kvm_arm_vcpu_get_attr(vcpu, &attr);
931         }
932         case KVM_HAS_DEVICE_ATTR: {
933                 if (copy_from_user(&attr, argp, sizeof(attr)))
934                         return -EFAULT;
935                 return kvm_arm_vcpu_has_attr(vcpu, &attr);
936         }
937         default:
938                 return -EINVAL;
939         }
940 }
941
942 /**
943  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
944  * @kvm: kvm instance
945  * @log: slot id and address to which we copy the log
946  *
947  * Steps 1-4 below provide general overview of dirty page logging. See
948  * kvm_get_dirty_log_protect() function description for additional details.
949  *
950  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
951  * always flush the TLB (step 4) even if previous step failed  and the dirty
952  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
953  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
954  * writes will be marked dirty for next log read.
955  *
956  *   1. Take a snapshot of the bit and clear it if needed.
957  *   2. Write protect the corresponding page.
958  *   3. Copy the snapshot to the userspace.
959  *   4. Flush TLB's if needed.
960  */
961 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
962 {
963         bool is_dirty = false;
964         int r;
965
966         mutex_lock(&kvm->slots_lock);
967
968         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
969
970         if (is_dirty)
971                 kvm_flush_remote_tlbs(kvm);
972
973         mutex_unlock(&kvm->slots_lock);
974         return r;
975 }
976
977 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
978                                         struct kvm_arm_device_addr *dev_addr)
979 {
980         unsigned long dev_id, type;
981
982         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
983                 KVM_ARM_DEVICE_ID_SHIFT;
984         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
985                 KVM_ARM_DEVICE_TYPE_SHIFT;
986
987         switch (dev_id) {
988         case KVM_ARM_DEVICE_VGIC_V2:
989                 if (!vgic_present)
990                         return -ENXIO;
991                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
992         default:
993                 return -ENODEV;
994         }
995 }
996
997 long kvm_arch_vm_ioctl(struct file *filp,
998                        unsigned int ioctl, unsigned long arg)
999 {
1000         struct kvm *kvm = filp->private_data;
1001         void __user *argp = (void __user *)arg;
1002
1003         switch (ioctl) {
1004         case KVM_CREATE_IRQCHIP: {
1005                 if (!vgic_present)
1006                         return -ENXIO;
1007                 return kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1008         }
1009         case KVM_ARM_SET_DEVICE_ADDR: {
1010                 struct kvm_arm_device_addr dev_addr;
1011
1012                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1013                         return -EFAULT;
1014                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1015         }
1016         case KVM_ARM_PREFERRED_TARGET: {
1017                 int err;
1018                 struct kvm_vcpu_init init;
1019
1020                 err = kvm_vcpu_preferred_target(&init);
1021                 if (err)
1022                         return err;
1023
1024                 if (copy_to_user(argp, &init, sizeof(init)))
1025                         return -EFAULT;
1026
1027                 return 0;
1028         }
1029         default:
1030                 return -EINVAL;
1031         }
1032 }
1033
1034 static void cpu_init_stage2(void *dummy)
1035 {
1036         __cpu_init_stage2();
1037 }
1038
1039 static void cpu_init_hyp_mode(void *dummy)
1040 {
1041         phys_addr_t boot_pgd_ptr;
1042         phys_addr_t pgd_ptr;
1043         unsigned long hyp_stack_ptr;
1044         unsigned long stack_page;
1045         unsigned long vector_ptr;
1046
1047         /* Switch from the HYP stub to our own HYP init vector */
1048         __hyp_set_vectors(kvm_get_idmap_vector());
1049
1050         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
1051         pgd_ptr = kvm_mmu_get_httbr();
1052         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1053         hyp_stack_ptr = stack_page + PAGE_SIZE;
1054         vector_ptr = (unsigned long)__kvm_hyp_vector;
1055
1056         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
1057         __cpu_init_stage2();
1058
1059         kvm_arm_init_debug();
1060 }
1061
1062 static int hyp_init_cpu_notify(struct notifier_block *self,
1063                                unsigned long action, void *cpu)
1064 {
1065         switch (action) {
1066         case CPU_STARTING:
1067         case CPU_STARTING_FROZEN:
1068                 if (__hyp_get_vectors() == hyp_default_vectors)
1069                         cpu_init_hyp_mode(NULL);
1070                 break;
1071         }
1072
1073         return NOTIFY_OK;
1074 }
1075
1076 static struct notifier_block hyp_init_cpu_nb = {
1077         .notifier_call = hyp_init_cpu_notify,
1078 };
1079
1080 #ifdef CONFIG_CPU_PM
1081 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1082                                     unsigned long cmd,
1083                                     void *v)
1084 {
1085         if (cmd == CPU_PM_EXIT &&
1086             __hyp_get_vectors() == hyp_default_vectors) {
1087                 cpu_init_hyp_mode(NULL);
1088                 return NOTIFY_OK;
1089         }
1090
1091         return NOTIFY_DONE;
1092 }
1093
1094 static struct notifier_block hyp_init_cpu_pm_nb = {
1095         .notifier_call = hyp_init_cpu_pm_notifier,
1096 };
1097
1098 static void __init hyp_cpu_pm_init(void)
1099 {
1100         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1101 }
1102 #else
1103 static inline void hyp_cpu_pm_init(void)
1104 {
1105 }
1106 #endif
1107
1108 static void teardown_common_resources(void)
1109 {
1110         free_percpu(kvm_host_cpu_state);
1111 }
1112
1113 static int init_common_resources(void)
1114 {
1115         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1116         if (!kvm_host_cpu_state) {
1117                 kvm_err("Cannot allocate host CPU state\n");
1118                 return -ENOMEM;
1119         }
1120
1121         return 0;
1122 }
1123
1124 static int init_subsystems(void)
1125 {
1126         int err;
1127
1128         /*
1129          * Init HYP view of VGIC
1130          */
1131         err = kvm_vgic_hyp_init();
1132         switch (err) {
1133         case 0:
1134                 vgic_present = true;
1135                 break;
1136         case -ENODEV:
1137         case -ENXIO:
1138                 vgic_present = false;
1139                 break;
1140         default:
1141                 return err;
1142         }
1143
1144         /*
1145          * Init HYP architected timer support
1146          */
1147         err = kvm_timer_hyp_init();
1148         if (err)
1149                 return err;
1150
1151         kvm_perf_init();
1152         kvm_coproc_table_init();
1153
1154         return 0;
1155 }
1156
1157 static void teardown_hyp_mode(void)
1158 {
1159         int cpu;
1160
1161         if (is_kernel_in_hyp_mode())
1162                 return;
1163
1164         free_hyp_pgds();
1165         for_each_possible_cpu(cpu)
1166                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1167 }
1168
1169 static int init_vhe_mode(void)
1170 {
1171         /*
1172          * Execute the init code on each CPU.
1173          */
1174         on_each_cpu(cpu_init_stage2, NULL, 1);
1175
1176         /* set size of VMID supported by CPU */
1177         kvm_vmid_bits = kvm_get_vmid_bits();
1178         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1179
1180         kvm_info("VHE mode initialized successfully\n");
1181         return 0;
1182 }
1183
1184 /**
1185  * Inits Hyp-mode on all online CPUs
1186  */
1187 static int init_hyp_mode(void)
1188 {
1189         int cpu;
1190         int err = 0;
1191
1192         /*
1193          * Allocate Hyp PGD and setup Hyp identity mapping
1194          */
1195         err = kvm_mmu_init();
1196         if (err)
1197                 goto out_err;
1198
1199         /*
1200          * It is probably enough to obtain the default on one
1201          * CPU. It's unlikely to be different on the others.
1202          */
1203         hyp_default_vectors = __hyp_get_vectors();
1204
1205         /*
1206          * Allocate stack pages for Hypervisor-mode
1207          */
1208         for_each_possible_cpu(cpu) {
1209                 unsigned long stack_page;
1210
1211                 stack_page = __get_free_page(GFP_KERNEL);
1212                 if (!stack_page) {
1213                         err = -ENOMEM;
1214                         goto out_err;
1215                 }
1216
1217                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1218         }
1219
1220         /*
1221          * Map the Hyp-code called directly from the host
1222          */
1223         err = create_hyp_mappings(__hyp_text_start, __hyp_text_end);
1224         if (err) {
1225                 kvm_err("Cannot map world-switch code\n");
1226                 goto out_err;
1227         }
1228
1229         err = create_hyp_mappings(__start_rodata, __end_rodata);
1230         if (err) {
1231                 kvm_err("Cannot map rodata section\n");
1232                 goto out_err;
1233         }
1234
1235         /*
1236          * Map the Hyp stack pages
1237          */
1238         for_each_possible_cpu(cpu) {
1239                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1240                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
1241
1242                 if (err) {
1243                         kvm_err("Cannot map hyp stack\n");
1244                         goto out_err;
1245                 }
1246         }
1247
1248         for_each_possible_cpu(cpu) {
1249                 kvm_cpu_context_t *cpu_ctxt;
1250
1251                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1252                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
1253
1254                 if (err) {
1255                         kvm_err("Cannot map host CPU state: %d\n", err);
1256                         goto out_err;
1257                 }
1258         }
1259
1260         /*
1261          * Execute the init code on each CPU.
1262          */
1263         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
1264
1265 #ifndef CONFIG_HOTPLUG_CPU
1266         free_boot_hyp_pgd();
1267 #endif
1268
1269         cpu_notifier_register_begin();
1270
1271         err = __register_cpu_notifier(&hyp_init_cpu_nb);
1272
1273         cpu_notifier_register_done();
1274
1275         if (err) {
1276                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1277                 goto out_err;
1278         }
1279
1280         hyp_cpu_pm_init();
1281
1282         /* set size of VMID supported by CPU */
1283         kvm_vmid_bits = kvm_get_vmid_bits();
1284         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1285
1286         kvm_info("Hyp mode initialized successfully\n");
1287
1288         return 0;
1289
1290 out_err:
1291         teardown_hyp_mode();
1292         kvm_err("error initializing Hyp mode: %d\n", err);
1293         return err;
1294 }
1295
1296 static void check_kvm_target_cpu(void *ret)
1297 {
1298         *(int *)ret = kvm_target_cpu();
1299 }
1300
1301 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1302 {
1303         struct kvm_vcpu *vcpu;
1304         int i;
1305
1306         mpidr &= MPIDR_HWID_BITMASK;
1307         kvm_for_each_vcpu(i, vcpu, kvm) {
1308                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1309                         return vcpu;
1310         }
1311         return NULL;
1312 }
1313
1314 /**
1315  * Initialize Hyp-mode and memory mappings on all CPUs.
1316  */
1317 int kvm_arch_init(void *opaque)
1318 {
1319         int err;
1320         int ret, cpu;
1321
1322         if (!is_hyp_mode_available()) {
1323                 kvm_err("HYP mode not available\n");
1324                 return -ENODEV;
1325         }
1326
1327         for_each_online_cpu(cpu) {
1328                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1329                 if (ret < 0) {
1330                         kvm_err("Error, CPU %d not supported!\n", cpu);
1331                         return -ENODEV;
1332                 }
1333         }
1334
1335         err = init_common_resources();
1336         if (err)
1337                 return err;
1338
1339         if (is_kernel_in_hyp_mode())
1340                 err = init_vhe_mode();
1341         else
1342                 err = init_hyp_mode();
1343         if (err)
1344                 goto out_err;
1345
1346         err = init_subsystems();
1347         if (err)
1348                 goto out_hyp;
1349
1350         return 0;
1351
1352 out_hyp:
1353         teardown_hyp_mode();
1354 out_err:
1355         teardown_common_resources();
1356         return err;
1357 }
1358
1359 /* NOP: Compiling as a module not supported */
1360 void kvm_arch_exit(void)
1361 {
1362         kvm_perf_teardown();
1363 }
1364
1365 static int arm_init(void)
1366 {
1367         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1368         return rc;
1369 }
1370
1371 module_init(arm_init);