Merge tag 'virtio-next-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[sfrench/cifs-2.6.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/errno.h>
21 #include <linux/err.h>
22 #include <linux/kvm_host.h>
23 #include <linux/module.h>
24 #include <linux/vmalloc.h>
25 #include <linux/fs.h>
26 #include <linux/mman.h>
27 #include <linux/sched.h>
28 #include <linux/kvm.h>
29 #include <trace/events/kvm.h>
30
31 #define CREATE_TRACE_POINTS
32 #include "trace.h"
33
34 #include <asm/uaccess.h>
35 #include <asm/ptrace.h>
36 #include <asm/mman.h>
37 #include <asm/tlbflush.h>
38 #include <asm/cacheflush.h>
39 #include <asm/virt.h>
40 #include <asm/kvm_arm.h>
41 #include <asm/kvm_asm.h>
42 #include <asm/kvm_mmu.h>
43 #include <asm/kvm_emulate.h>
44 #include <asm/kvm_coproc.h>
45 #include <asm/kvm_psci.h>
46
47 #ifdef REQUIRES_VIRT
48 __asm__(".arch_extension        virt");
49 #endif
50
51 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
52 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
53 static unsigned long hyp_default_vectors;
54
55 /* Per-CPU variable containing the currently running vcpu. */
56 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
57
58 /* The VMID used in the VTTBR */
59 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
60 static u8 kvm_next_vmid;
61 static DEFINE_SPINLOCK(kvm_vmid_lock);
62
63 static bool vgic_present;
64
65 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
66 {
67         BUG_ON(preemptible());
68         __get_cpu_var(kvm_arm_running_vcpu) = vcpu;
69 }
70
71 /**
72  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
73  * Must be called from non-preemptible context
74  */
75 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
76 {
77         BUG_ON(preemptible());
78         return __get_cpu_var(kvm_arm_running_vcpu);
79 }
80
81 /**
82  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
83  */
84 struct kvm_vcpu __percpu **kvm_get_running_vcpus(void)
85 {
86         return &kvm_arm_running_vcpu;
87 }
88
89 int kvm_arch_hardware_enable(void *garbage)
90 {
91         return 0;
92 }
93
94 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
95 {
96         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
97 }
98
99 void kvm_arch_hardware_disable(void *garbage)
100 {
101 }
102
103 int kvm_arch_hardware_setup(void)
104 {
105         return 0;
106 }
107
108 void kvm_arch_hardware_unsetup(void)
109 {
110 }
111
112 void kvm_arch_check_processor_compat(void *rtn)
113 {
114         *(int *)rtn = 0;
115 }
116
117 void kvm_arch_sync_events(struct kvm *kvm)
118 {
119 }
120
121 /**
122  * kvm_arch_init_vm - initializes a VM data structure
123  * @kvm:        pointer to the KVM struct
124  */
125 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
126 {
127         int ret = 0;
128
129         if (type)
130                 return -EINVAL;
131
132         ret = kvm_alloc_stage2_pgd(kvm);
133         if (ret)
134                 goto out_fail_alloc;
135
136         ret = create_hyp_mappings(kvm, kvm + 1);
137         if (ret)
138                 goto out_free_stage2_pgd;
139
140         /* Mark the initial VMID generation invalid */
141         kvm->arch.vmid_gen = 0;
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 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
156                            struct kvm_memory_slot *dont)
157 {
158 }
159
160 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
161 {
162         return 0;
163 }
164
165 /**
166  * kvm_arch_destroy_vm - destroy the VM data structure
167  * @kvm:        pointer to the KVM struct
168  */
169 void kvm_arch_destroy_vm(struct kvm *kvm)
170 {
171         int i;
172
173         kvm_free_stage2_pgd(kvm);
174
175         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
176                 if (kvm->vcpus[i]) {
177                         kvm_arch_vcpu_free(kvm->vcpus[i]);
178                         kvm->vcpus[i] = NULL;
179                 }
180         }
181 }
182
183 int kvm_dev_ioctl_check_extension(long ext)
184 {
185         int r;
186         switch (ext) {
187         case KVM_CAP_IRQCHIP:
188                 r = vgic_present;
189                 break;
190         case KVM_CAP_USER_MEMORY:
191         case KVM_CAP_SYNC_MMU:
192         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
193         case KVM_CAP_ONE_REG:
194         case KVM_CAP_ARM_PSCI:
195                 r = 1;
196                 break;
197         case KVM_CAP_COALESCED_MMIO:
198                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
199                 break;
200         case KVM_CAP_ARM_SET_DEVICE_ADDR:
201                 r = 1;
202                 break;
203         case KVM_CAP_NR_VCPUS:
204                 r = num_online_cpus();
205                 break;
206         case KVM_CAP_MAX_VCPUS:
207                 r = KVM_MAX_VCPUS;
208                 break;
209         default:
210                 r = kvm_arch_dev_ioctl_check_extension(ext);
211                 break;
212         }
213         return r;
214 }
215
216 long kvm_arch_dev_ioctl(struct file *filp,
217                         unsigned int ioctl, unsigned long arg)
218 {
219         return -EINVAL;
220 }
221
222 int kvm_arch_prepare_memory_region(struct kvm *kvm,
223                                    struct kvm_memory_slot *memslot,
224                                    struct kvm_userspace_memory_region *mem,
225                                    enum kvm_mr_change change)
226 {
227         return 0;
228 }
229
230 void kvm_arch_commit_memory_region(struct kvm *kvm,
231                                    struct kvm_userspace_memory_region *mem,
232                                    const struct kvm_memory_slot *old,
233                                    enum kvm_mr_change change)
234 {
235 }
236
237 void kvm_arch_flush_shadow_all(struct kvm *kvm)
238 {
239 }
240
241 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
242                                    struct kvm_memory_slot *slot)
243 {
244 }
245
246 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
247 {
248         int err;
249         struct kvm_vcpu *vcpu;
250
251         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
252         if (!vcpu) {
253                 err = -ENOMEM;
254                 goto out;
255         }
256
257         err = kvm_vcpu_init(vcpu, kvm, id);
258         if (err)
259                 goto free_vcpu;
260
261         err = create_hyp_mappings(vcpu, vcpu + 1);
262         if (err)
263                 goto vcpu_uninit;
264
265         return vcpu;
266 vcpu_uninit:
267         kvm_vcpu_uninit(vcpu);
268 free_vcpu:
269         kmem_cache_free(kvm_vcpu_cache, vcpu);
270 out:
271         return ERR_PTR(err);
272 }
273
274 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
275 {
276         return 0;
277 }
278
279 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
280 {
281         kvm_mmu_free_memory_caches(vcpu);
282         kvm_timer_vcpu_terminate(vcpu);
283         kmem_cache_free(kvm_vcpu_cache, vcpu);
284 }
285
286 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
287 {
288         kvm_arch_vcpu_free(vcpu);
289 }
290
291 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
292 {
293         return 0;
294 }
295
296 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
297 {
298         int ret;
299
300         /* Force users to call KVM_ARM_VCPU_INIT */
301         vcpu->arch.target = -1;
302
303         /* Set up VGIC */
304         ret = kvm_vgic_vcpu_init(vcpu);
305         if (ret)
306                 return ret;
307
308         /* Set up the timer */
309         kvm_timer_vcpu_init(vcpu);
310
311         return 0;
312 }
313
314 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
315 {
316 }
317
318 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
319 {
320         vcpu->cpu = cpu;
321         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
322
323         /*
324          * Check whether this vcpu requires the cache to be flushed on
325          * this physical CPU. This is a consequence of doing dcache
326          * operations by set/way on this vcpu. We do it here to be in
327          * a non-preemptible section.
328          */
329         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
330                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
331
332         kvm_arm_set_running_vcpu(vcpu);
333 }
334
335 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
336 {
337         kvm_arm_set_running_vcpu(NULL);
338 }
339
340 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
341                                         struct kvm_guest_debug *dbg)
342 {
343         return -EINVAL;
344 }
345
346
347 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
348                                     struct kvm_mp_state *mp_state)
349 {
350         return -EINVAL;
351 }
352
353 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
354                                     struct kvm_mp_state *mp_state)
355 {
356         return -EINVAL;
357 }
358
359 /**
360  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
361  * @v:          The VCPU pointer
362  *
363  * If the guest CPU is not waiting for interrupts or an interrupt line is
364  * asserted, the CPU is by definition runnable.
365  */
366 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
367 {
368         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
369 }
370
371 /* Just ensure a guest exit from a particular CPU */
372 static void exit_vm_noop(void *info)
373 {
374 }
375
376 void force_vm_exit(const cpumask_t *mask)
377 {
378         smp_call_function_many(mask, exit_vm_noop, NULL, true);
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
449         /* update vttbr to be used with the new vmid */
450         pgd_phys = virt_to_phys(kvm->arch.pgd);
451         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
452         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
453         kvm->arch.vttbr |= vmid;
454
455         spin_unlock(&kvm_vmid_lock);
456 }
457
458 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
459 {
460         if (likely(vcpu->arch.has_run_once))
461                 return 0;
462
463         vcpu->arch.has_run_once = true;
464
465         /*
466          * Initialize the VGIC before running a vcpu the first time on
467          * this VM.
468          */
469         if (irqchip_in_kernel(vcpu->kvm) &&
470             unlikely(!vgic_initialized(vcpu->kvm))) {
471                 int ret = kvm_vgic_init(vcpu->kvm);
472                 if (ret)
473                         return ret;
474         }
475
476         /*
477          * Handle the "start in power-off" case by calling into the
478          * PSCI code.
479          */
480         if (test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) {
481                 *vcpu_reg(vcpu, 0) = KVM_PSCI_FN_CPU_OFF;
482                 kvm_psci_call(vcpu);
483         }
484
485         return 0;
486 }
487
488 static void vcpu_pause(struct kvm_vcpu *vcpu)
489 {
490         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
491
492         wait_event_interruptible(*wq, !vcpu->arch.pause);
493 }
494
495 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
496 {
497         return vcpu->arch.target >= 0;
498 }
499
500 /**
501  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
502  * @vcpu:       The VCPU pointer
503  * @run:        The kvm_run structure pointer used for userspace state exchange
504  *
505  * This function is called through the VCPU_RUN ioctl called from user space. It
506  * will execute VM code in a loop until the time slice for the process is used
507  * or some emulation is needed from user space in which case the function will
508  * return with return value 0 and with the kvm_run structure filled in with the
509  * required data for the requested emulation.
510  */
511 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
512 {
513         int ret;
514         sigset_t sigsaved;
515
516         if (unlikely(!kvm_vcpu_initialized(vcpu)))
517                 return -ENOEXEC;
518
519         ret = kvm_vcpu_first_run_init(vcpu);
520         if (ret)
521                 return ret;
522
523         if (run->exit_reason == KVM_EXIT_MMIO) {
524                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
525                 if (ret)
526                         return ret;
527         }
528
529         if (vcpu->sigset_active)
530                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
531
532         ret = 1;
533         run->exit_reason = KVM_EXIT_UNKNOWN;
534         while (ret > 0) {
535                 /*
536                  * Check conditions before entering the guest
537                  */
538                 cond_resched();
539
540                 update_vttbr(vcpu->kvm);
541
542                 if (vcpu->arch.pause)
543                         vcpu_pause(vcpu);
544
545                 kvm_vgic_flush_hwstate(vcpu);
546                 kvm_timer_flush_hwstate(vcpu);
547
548                 local_irq_disable();
549
550                 /*
551                  * Re-check atomic conditions
552                  */
553                 if (signal_pending(current)) {
554                         ret = -EINTR;
555                         run->exit_reason = KVM_EXIT_INTR;
556                 }
557
558                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
559                         local_irq_enable();
560                         kvm_timer_sync_hwstate(vcpu);
561                         kvm_vgic_sync_hwstate(vcpu);
562                         continue;
563                 }
564
565                 /**************************************************************
566                  * Enter the guest
567                  */
568                 trace_kvm_entry(*vcpu_pc(vcpu));
569                 kvm_guest_enter();
570                 vcpu->mode = IN_GUEST_MODE;
571
572                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
573
574                 vcpu->mode = OUTSIDE_GUEST_MODE;
575                 vcpu->arch.last_pcpu = smp_processor_id();
576                 kvm_guest_exit();
577                 trace_kvm_exit(*vcpu_pc(vcpu));
578                 /*
579                  * We may have taken a host interrupt in HYP mode (ie
580                  * while executing the guest). This interrupt is still
581                  * pending, as we haven't serviced it yet!
582                  *
583                  * We're now back in SVC mode, with interrupts
584                  * disabled.  Enabling the interrupts now will have
585                  * the effect of taking the interrupt again, in SVC
586                  * mode this time.
587                  */
588                 local_irq_enable();
589
590                 /*
591                  * Back from guest
592                  *************************************************************/
593
594                 kvm_timer_sync_hwstate(vcpu);
595                 kvm_vgic_sync_hwstate(vcpu);
596
597                 ret = handle_exit(vcpu, run, ret);
598         }
599
600         if (vcpu->sigset_active)
601                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
602         return ret;
603 }
604
605 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
606 {
607         int bit_index;
608         bool set;
609         unsigned long *ptr;
610
611         if (number == KVM_ARM_IRQ_CPU_IRQ)
612                 bit_index = __ffs(HCR_VI);
613         else /* KVM_ARM_IRQ_CPU_FIQ */
614                 bit_index = __ffs(HCR_VF);
615
616         ptr = (unsigned long *)&vcpu->arch.irq_lines;
617         if (level)
618                 set = test_and_set_bit(bit_index, ptr);
619         else
620                 set = test_and_clear_bit(bit_index, ptr);
621
622         /*
623          * If we didn't change anything, no need to wake up or kick other CPUs
624          */
625         if (set == level)
626                 return 0;
627
628         /*
629          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
630          * trigger a world-switch round on the running physical CPU to set the
631          * virtual IRQ/FIQ fields in the HCR appropriately.
632          */
633         kvm_vcpu_kick(vcpu);
634
635         return 0;
636 }
637
638 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
639                           bool line_status)
640 {
641         u32 irq = irq_level->irq;
642         unsigned int irq_type, vcpu_idx, irq_num;
643         int nrcpus = atomic_read(&kvm->online_vcpus);
644         struct kvm_vcpu *vcpu = NULL;
645         bool level = irq_level->level;
646
647         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
648         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
649         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
650
651         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
652
653         switch (irq_type) {
654         case KVM_ARM_IRQ_TYPE_CPU:
655                 if (irqchip_in_kernel(kvm))
656                         return -ENXIO;
657
658                 if (vcpu_idx >= nrcpus)
659                         return -EINVAL;
660
661                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
662                 if (!vcpu)
663                         return -EINVAL;
664
665                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
666                         return -EINVAL;
667
668                 return vcpu_interrupt_line(vcpu, irq_num, level);
669         case KVM_ARM_IRQ_TYPE_PPI:
670                 if (!irqchip_in_kernel(kvm))
671                         return -ENXIO;
672
673                 if (vcpu_idx >= nrcpus)
674                         return -EINVAL;
675
676                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
677                 if (!vcpu)
678                         return -EINVAL;
679
680                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
681                         return -EINVAL;
682
683                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
684         case KVM_ARM_IRQ_TYPE_SPI:
685                 if (!irqchip_in_kernel(kvm))
686                         return -ENXIO;
687
688                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
689                     irq_num > KVM_ARM_IRQ_GIC_MAX)
690                         return -EINVAL;
691
692                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
693         }
694
695         return -EINVAL;
696 }
697
698 long kvm_arch_vcpu_ioctl(struct file *filp,
699                          unsigned int ioctl, unsigned long arg)
700 {
701         struct kvm_vcpu *vcpu = filp->private_data;
702         void __user *argp = (void __user *)arg;
703
704         switch (ioctl) {
705         case KVM_ARM_VCPU_INIT: {
706                 struct kvm_vcpu_init init;
707
708                 if (copy_from_user(&init, argp, sizeof(init)))
709                         return -EFAULT;
710
711                 return kvm_vcpu_set_target(vcpu, &init);
712
713         }
714         case KVM_SET_ONE_REG:
715         case KVM_GET_ONE_REG: {
716                 struct kvm_one_reg reg;
717
718                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
719                         return -ENOEXEC;
720
721                 if (copy_from_user(&reg, argp, sizeof(reg)))
722                         return -EFAULT;
723                 if (ioctl == KVM_SET_ONE_REG)
724                         return kvm_arm_set_reg(vcpu, &reg);
725                 else
726                         return kvm_arm_get_reg(vcpu, &reg);
727         }
728         case KVM_GET_REG_LIST: {
729                 struct kvm_reg_list __user *user_list = argp;
730                 struct kvm_reg_list reg_list;
731                 unsigned n;
732
733                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
734                         return -ENOEXEC;
735
736                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
737                         return -EFAULT;
738                 n = reg_list.n;
739                 reg_list.n = kvm_arm_num_regs(vcpu);
740                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
741                         return -EFAULT;
742                 if (n < reg_list.n)
743                         return -E2BIG;
744                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
745         }
746         default:
747                 return -EINVAL;
748         }
749 }
750
751 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
752 {
753         return -EINVAL;
754 }
755
756 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
757                                         struct kvm_arm_device_addr *dev_addr)
758 {
759         unsigned long dev_id, type;
760
761         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
762                 KVM_ARM_DEVICE_ID_SHIFT;
763         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
764                 KVM_ARM_DEVICE_TYPE_SHIFT;
765
766         switch (dev_id) {
767         case KVM_ARM_DEVICE_VGIC_V2:
768                 if (!vgic_present)
769                         return -ENXIO;
770                 return kvm_vgic_set_addr(kvm, type, dev_addr->addr);
771         default:
772                 return -ENODEV;
773         }
774 }
775
776 long kvm_arch_vm_ioctl(struct file *filp,
777                        unsigned int ioctl, unsigned long arg)
778 {
779         struct kvm *kvm = filp->private_data;
780         void __user *argp = (void __user *)arg;
781
782         switch (ioctl) {
783         case KVM_CREATE_IRQCHIP: {
784                 if (vgic_present)
785                         return kvm_vgic_create(kvm);
786                 else
787                         return -ENXIO;
788         }
789         case KVM_ARM_SET_DEVICE_ADDR: {
790                 struct kvm_arm_device_addr dev_addr;
791
792                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
793                         return -EFAULT;
794                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
795         }
796         default:
797                 return -EINVAL;
798         }
799 }
800
801 static void cpu_init_hyp_mode(void *dummy)
802 {
803         phys_addr_t boot_pgd_ptr;
804         phys_addr_t pgd_ptr;
805         unsigned long hyp_stack_ptr;
806         unsigned long stack_page;
807         unsigned long vector_ptr;
808
809         /* Switch from the HYP stub to our own HYP init vector */
810         __hyp_set_vectors(kvm_get_idmap_vector());
811
812         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
813         pgd_ptr = kvm_mmu_get_httbr();
814         stack_page = __get_cpu_var(kvm_arm_hyp_stack_page);
815         hyp_stack_ptr = stack_page + PAGE_SIZE;
816         vector_ptr = (unsigned long)__kvm_hyp_vector;
817
818         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
819 }
820
821 static int hyp_init_cpu_notify(struct notifier_block *self,
822                                unsigned long action, void *cpu)
823 {
824         switch (action) {
825         case CPU_STARTING:
826         case CPU_STARTING_FROZEN:
827                 cpu_init_hyp_mode(NULL);
828                 break;
829         }
830
831         return NOTIFY_OK;
832 }
833
834 static struct notifier_block hyp_init_cpu_nb = {
835         .notifier_call = hyp_init_cpu_notify,
836 };
837
838 /**
839  * Inits Hyp-mode on all online CPUs
840  */
841 static int init_hyp_mode(void)
842 {
843         int cpu;
844         int err = 0;
845
846         /*
847          * Allocate Hyp PGD and setup Hyp identity mapping
848          */
849         err = kvm_mmu_init();
850         if (err)
851                 goto out_err;
852
853         /*
854          * It is probably enough to obtain the default on one
855          * CPU. It's unlikely to be different on the others.
856          */
857         hyp_default_vectors = __hyp_get_vectors();
858
859         /*
860          * Allocate stack pages for Hypervisor-mode
861          */
862         for_each_possible_cpu(cpu) {
863                 unsigned long stack_page;
864
865                 stack_page = __get_free_page(GFP_KERNEL);
866                 if (!stack_page) {
867                         err = -ENOMEM;
868                         goto out_free_stack_pages;
869                 }
870
871                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
872         }
873
874         /*
875          * Map the Hyp-code called directly from the host
876          */
877         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
878         if (err) {
879                 kvm_err("Cannot map world-switch code\n");
880                 goto out_free_mappings;
881         }
882
883         /*
884          * Map the Hyp stack pages
885          */
886         for_each_possible_cpu(cpu) {
887                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
888                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
889
890                 if (err) {
891                         kvm_err("Cannot map hyp stack\n");
892                         goto out_free_mappings;
893                 }
894         }
895
896         /*
897          * Map the host CPU structures
898          */
899         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
900         if (!kvm_host_cpu_state) {
901                 err = -ENOMEM;
902                 kvm_err("Cannot allocate host CPU state\n");
903                 goto out_free_mappings;
904         }
905
906         for_each_possible_cpu(cpu) {
907                 kvm_cpu_context_t *cpu_ctxt;
908
909                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
910                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
911
912                 if (err) {
913                         kvm_err("Cannot map host CPU state: %d\n", err);
914                         goto out_free_context;
915                 }
916         }
917
918         /*
919          * Execute the init code on each CPU.
920          */
921         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
922
923         /*
924          * Init HYP view of VGIC
925          */
926         err = kvm_vgic_hyp_init();
927         if (err)
928                 goto out_free_context;
929
930 #ifdef CONFIG_KVM_ARM_VGIC
931                 vgic_present = true;
932 #endif
933
934         /*
935          * Init HYP architected timer support
936          */
937         err = kvm_timer_hyp_init();
938         if (err)
939                 goto out_free_mappings;
940
941 #ifndef CONFIG_HOTPLUG_CPU
942         free_boot_hyp_pgd();
943 #endif
944
945         kvm_perf_init();
946
947         kvm_info("Hyp mode initialized successfully\n");
948
949         return 0;
950 out_free_context:
951         free_percpu(kvm_host_cpu_state);
952 out_free_mappings:
953         free_hyp_pgds();
954 out_free_stack_pages:
955         for_each_possible_cpu(cpu)
956                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
957 out_err:
958         kvm_err("error initializing Hyp mode: %d\n", err);
959         return err;
960 }
961
962 static void check_kvm_target_cpu(void *ret)
963 {
964         *(int *)ret = kvm_target_cpu();
965 }
966
967 /**
968  * Initialize Hyp-mode and memory mappings on all CPUs.
969  */
970 int kvm_arch_init(void *opaque)
971 {
972         int err;
973         int ret, cpu;
974
975         if (!is_hyp_mode_available()) {
976                 kvm_err("HYP mode not available\n");
977                 return -ENODEV;
978         }
979
980         for_each_online_cpu(cpu) {
981                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
982                 if (ret < 0) {
983                         kvm_err("Error, CPU %d not supported!\n", cpu);
984                         return -ENODEV;
985                 }
986         }
987
988         err = init_hyp_mode();
989         if (err)
990                 goto out_err;
991
992         err = register_cpu_notifier(&hyp_init_cpu_nb);
993         if (err) {
994                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
995                 goto out_err;
996         }
997
998         kvm_coproc_table_init();
999         return 0;
1000 out_err:
1001         return err;
1002 }
1003
1004 /* NOP: Compiling as a module not supported */
1005 void kvm_arch_exit(void)
1006 {
1007         kvm_perf_teardown();
1008 }
1009
1010 static int arm_init(void)
1011 {
1012         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1013         return rc;
1014 }
1015
1016 module_init(arm_init);