Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[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_pm.h>
20 #include <linux/errno.h>
21 #include <linux/err.h>
22 #include <linux/kvm_host.h>
23 #include <linux/list.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 DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
70
71 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
72 {
73         BUG_ON(preemptible());
74         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
75 }
76
77 /**
78  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
79  * Must be called from non-preemptible context
80  */
81 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
82 {
83         BUG_ON(preemptible());
84         return __this_cpu_read(kvm_arm_running_vcpu);
85 }
86
87 /**
88  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
89  */
90 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
91 {
92         return &kvm_arm_running_vcpu;
93 }
94
95 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
96 {
97         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
98 }
99
100 int kvm_arch_hardware_setup(void)
101 {
102         return 0;
103 }
104
105 void kvm_arch_check_processor_compat(void *rtn)
106 {
107         *(int *)rtn = 0;
108 }
109
110
111 /**
112  * kvm_arch_init_vm - initializes a VM data structure
113  * @kvm:        pointer to the KVM struct
114  */
115 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
116 {
117         int ret = 0;
118
119         if (type)
120                 return -EINVAL;
121
122         ret = kvm_alloc_stage2_pgd(kvm);
123         if (ret)
124                 goto out_fail_alloc;
125
126         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
127         if (ret)
128                 goto out_free_stage2_pgd;
129
130         kvm_vgic_early_init(kvm);
131         kvm_timer_init(kvm);
132
133         /* Mark the initial VMID generation invalid */
134         kvm->arch.vmid_gen = 0;
135
136         /* The maximum number of VCPUs is limited by the host's GIC model */
137         kvm->arch.max_vcpus = vgic_present ?
138                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
139
140         return ret;
141 out_free_stage2_pgd:
142         kvm_free_stage2_pgd(kvm);
143 out_fail_alloc:
144         return ret;
145 }
146
147 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
148 {
149         return VM_FAULT_SIGBUS;
150 }
151
152
153 /**
154  * kvm_arch_destroy_vm - destroy the VM data structure
155  * @kvm:        pointer to the KVM struct
156  */
157 void kvm_arch_destroy_vm(struct kvm *kvm)
158 {
159         int i;
160
161         kvm_free_stage2_pgd(kvm);
162
163         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
164                 if (kvm->vcpus[i]) {
165                         kvm_arch_vcpu_free(kvm->vcpus[i]);
166                         kvm->vcpus[i] = NULL;
167                 }
168         }
169
170         kvm_vgic_destroy(kvm);
171 }
172
173 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
174 {
175         int r;
176         switch (ext) {
177         case KVM_CAP_IRQCHIP:
178                 r = vgic_present;
179                 break;
180         case KVM_CAP_IOEVENTFD:
181         case KVM_CAP_DEVICE_CTRL:
182         case KVM_CAP_USER_MEMORY:
183         case KVM_CAP_SYNC_MMU:
184         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
185         case KVM_CAP_ONE_REG:
186         case KVM_CAP_ARM_PSCI:
187         case KVM_CAP_ARM_PSCI_0_2:
188         case KVM_CAP_READONLY_MEM:
189         case KVM_CAP_MP_STATE:
190                 r = 1;
191                 break;
192         case KVM_CAP_COALESCED_MMIO:
193                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
194                 break;
195         case KVM_CAP_ARM_SET_DEVICE_ADDR:
196                 r = 1;
197                 break;
198         case KVM_CAP_NR_VCPUS:
199                 r = num_online_cpus();
200                 break;
201         case KVM_CAP_MAX_VCPUS:
202                 r = KVM_MAX_VCPUS;
203                 break;
204         default:
205                 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
206                 break;
207         }
208         return r;
209 }
210
211 long kvm_arch_dev_ioctl(struct file *filp,
212                         unsigned int ioctl, unsigned long arg)
213 {
214         return -EINVAL;
215 }
216
217
218 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
219 {
220         int err;
221         struct kvm_vcpu *vcpu;
222
223         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
224                 err = -EBUSY;
225                 goto out;
226         }
227
228         if (id >= kvm->arch.max_vcpus) {
229                 err = -EINVAL;
230                 goto out;
231         }
232
233         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
234         if (!vcpu) {
235                 err = -ENOMEM;
236                 goto out;
237         }
238
239         err = kvm_vcpu_init(vcpu, kvm, id);
240         if (err)
241                 goto free_vcpu;
242
243         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
244         if (err)
245                 goto vcpu_uninit;
246
247         return vcpu;
248 vcpu_uninit:
249         kvm_vcpu_uninit(vcpu);
250 free_vcpu:
251         kmem_cache_free(kvm_vcpu_cache, vcpu);
252 out:
253         return ERR_PTR(err);
254 }
255
256 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
257 {
258         kvm_vgic_vcpu_early_init(vcpu);
259 }
260
261 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
262 {
263         kvm_mmu_free_memory_caches(vcpu);
264         kvm_timer_vcpu_terminate(vcpu);
265         kvm_vgic_vcpu_destroy(vcpu);
266         kvm_pmu_vcpu_destroy(vcpu);
267         kvm_vcpu_uninit(vcpu);
268         kmem_cache_free(kvm_vcpu_cache, vcpu);
269 }
270
271 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
272 {
273         kvm_arch_vcpu_free(vcpu);
274 }
275
276 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
277 {
278         return kvm_timer_should_fire(vcpu);
279 }
280
281 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
282 {
283         kvm_timer_schedule(vcpu);
284 }
285
286 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
287 {
288         kvm_timer_unschedule(vcpu);
289 }
290
291 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
292 {
293         /* Force users to call KVM_ARM_VCPU_INIT */
294         vcpu->arch.target = -1;
295         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
296
297         /* Set up the timer */
298         kvm_timer_vcpu_init(vcpu);
299
300         kvm_arm_reset_debug_ptr(vcpu);
301
302         return 0;
303 }
304
305 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
306 {
307         vcpu->cpu = cpu;
308         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
309
310         kvm_arm_set_running_vcpu(vcpu);
311 }
312
313 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
314 {
315         /*
316          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
317          * if the vcpu is no longer assigned to a cpu.  This is used for the
318          * optimized make_all_cpus_request path.
319          */
320         vcpu->cpu = -1;
321
322         kvm_arm_set_running_vcpu(NULL);
323         kvm_timer_vcpu_put(vcpu);
324 }
325
326 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
327                                     struct kvm_mp_state *mp_state)
328 {
329         if (vcpu->arch.power_off)
330                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
331         else
332                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
333
334         return 0;
335 }
336
337 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
338                                     struct kvm_mp_state *mp_state)
339 {
340         switch (mp_state->mp_state) {
341         case KVM_MP_STATE_RUNNABLE:
342                 vcpu->arch.power_off = false;
343                 break;
344         case KVM_MP_STATE_STOPPED:
345                 vcpu->arch.power_off = true;
346                 break;
347         default:
348                 return -EINVAL;
349         }
350
351         return 0;
352 }
353
354 /**
355  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
356  * @v:          The VCPU pointer
357  *
358  * If the guest CPU is not waiting for interrupts or an interrupt line is
359  * asserted, the CPU is by definition runnable.
360  */
361 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
362 {
363         return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
364                 && !v->arch.power_off && !v->arch.pause);
365 }
366
367 /* Just ensure a guest exit from a particular CPU */
368 static void exit_vm_noop(void *info)
369 {
370 }
371
372 void force_vm_exit(const cpumask_t *mask)
373 {
374         preempt_disable();
375         smp_call_function_many(mask, exit_vm_noop, NULL, true);
376         preempt_enable();
377 }
378
379 /**
380  * need_new_vmid_gen - check that the VMID is still valid
381  * @kvm: The VM's VMID to check
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->arch.pgd);
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 = 0;
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                 ret = kvm_timer_enable(vcpu);
484
485         return ret;
486 }
487
488 bool kvm_arch_intc_initialized(struct kvm *kvm)
489 {
490         return vgic_initialized(kvm);
491 }
492
493 void kvm_arm_halt_guest(struct kvm *kvm)
494 {
495         int i;
496         struct kvm_vcpu *vcpu;
497
498         kvm_for_each_vcpu(i, vcpu, kvm)
499                 vcpu->arch.pause = true;
500         kvm_make_all_cpus_request(kvm, KVM_REQ_VCPU_EXIT);
501 }
502
503 void kvm_arm_halt_vcpu(struct kvm_vcpu *vcpu)
504 {
505         vcpu->arch.pause = true;
506         kvm_vcpu_kick(vcpu);
507 }
508
509 void kvm_arm_resume_vcpu(struct kvm_vcpu *vcpu)
510 {
511         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
512
513         vcpu->arch.pause = false;
514         swake_up(wq);
515 }
516
517 void kvm_arm_resume_guest(struct kvm *kvm)
518 {
519         int i;
520         struct kvm_vcpu *vcpu;
521
522         kvm_for_each_vcpu(i, vcpu, kvm)
523                 kvm_arm_resume_vcpu(vcpu);
524 }
525
526 static void vcpu_sleep(struct kvm_vcpu *vcpu)
527 {
528         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
529
530         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
531                                        (!vcpu->arch.pause)));
532 }
533
534 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
535 {
536         return vcpu->arch.target >= 0;
537 }
538
539 /**
540  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
541  * @vcpu:       The VCPU pointer
542  * @run:        The kvm_run structure pointer used for userspace state exchange
543  *
544  * This function is called through the VCPU_RUN ioctl called from user space. It
545  * will execute VM code in a loop until the time slice for the process is used
546  * or some emulation is needed from user space in which case the function will
547  * return with return value 0 and with the kvm_run structure filled in with the
548  * required data for the requested emulation.
549  */
550 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
551 {
552         int ret;
553         sigset_t sigsaved;
554
555         if (unlikely(!kvm_vcpu_initialized(vcpu)))
556                 return -ENOEXEC;
557
558         ret = kvm_vcpu_first_run_init(vcpu);
559         if (ret)
560                 return ret;
561
562         if (run->exit_reason == KVM_EXIT_MMIO) {
563                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
564                 if (ret)
565                         return ret;
566         }
567
568         if (vcpu->sigset_active)
569                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
570
571         ret = 1;
572         run->exit_reason = KVM_EXIT_UNKNOWN;
573         while (ret > 0) {
574                 /*
575                  * Check conditions before entering the guest
576                  */
577                 cond_resched();
578
579                 update_vttbr(vcpu->kvm);
580
581                 if (vcpu->arch.power_off || vcpu->arch.pause)
582                         vcpu_sleep(vcpu);
583
584                 /*
585                  * Preparing the interrupts to be injected also
586                  * involves poking the GIC, which must be done in a
587                  * non-preemptible context.
588                  */
589                 preempt_disable();
590                 kvm_pmu_flush_hwstate(vcpu);
591                 kvm_timer_flush_hwstate(vcpu);
592                 kvm_vgic_flush_hwstate(vcpu);
593
594                 local_irq_disable();
595
596                 /*
597                  * Re-check atomic conditions
598                  */
599                 if (signal_pending(current)) {
600                         ret = -EINTR;
601                         run->exit_reason = KVM_EXIT_INTR;
602                 }
603
604                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
605                         vcpu->arch.power_off || vcpu->arch.pause) {
606                         local_irq_enable();
607                         kvm_pmu_sync_hwstate(vcpu);
608                         kvm_timer_sync_hwstate(vcpu);
609                         kvm_vgic_sync_hwstate(vcpu);
610                         preempt_enable();
611                         continue;
612                 }
613
614                 kvm_arm_setup_debug(vcpu);
615
616                 /**************************************************************
617                  * Enter the guest
618                  */
619                 trace_kvm_entry(*vcpu_pc(vcpu));
620                 guest_enter_irqoff();
621                 vcpu->mode = IN_GUEST_MODE;
622
623                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
624
625                 vcpu->mode = OUTSIDE_GUEST_MODE;
626                 vcpu->stat.exits++;
627                 /*
628                  * Back from guest
629                  *************************************************************/
630
631                 kvm_arm_clear_debug(vcpu);
632
633                 /*
634                  * We may have taken a host interrupt in HYP mode (ie
635                  * while executing the guest). This interrupt is still
636                  * pending, as we haven't serviced it yet!
637                  *
638                  * We're now back in SVC mode, with interrupts
639                  * disabled.  Enabling the interrupts now will have
640                  * the effect of taking the interrupt again, in SVC
641                  * mode this time.
642                  */
643                 local_irq_enable();
644
645                 /*
646                  * We do local_irq_enable() before calling guest_exit() so
647                  * that if a timer interrupt hits while running the guest we
648                  * account that tick as being spent in the guest.  We enable
649                  * preemption after calling guest_exit() so that if we get
650                  * preempted we make sure ticks after that is not counted as
651                  * guest time.
652                  */
653                 guest_exit();
654                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
655
656                 /*
657                  * We must sync the PMU and timer state before the vgic state so
658                  * that the vgic can properly sample the updated state of the
659                  * interrupt line.
660                  */
661                 kvm_pmu_sync_hwstate(vcpu);
662                 kvm_timer_sync_hwstate(vcpu);
663
664                 kvm_vgic_sync_hwstate(vcpu);
665
666                 preempt_enable();
667
668                 ret = handle_exit(vcpu, run, ret);
669         }
670
671         if (vcpu->sigset_active)
672                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
673         return ret;
674 }
675
676 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
677 {
678         int bit_index;
679         bool set;
680         unsigned long *ptr;
681
682         if (number == KVM_ARM_IRQ_CPU_IRQ)
683                 bit_index = __ffs(HCR_VI);
684         else /* KVM_ARM_IRQ_CPU_FIQ */
685                 bit_index = __ffs(HCR_VF);
686
687         ptr = (unsigned long *)&vcpu->arch.irq_lines;
688         if (level)
689                 set = test_and_set_bit(bit_index, ptr);
690         else
691                 set = test_and_clear_bit(bit_index, ptr);
692
693         /*
694          * If we didn't change anything, no need to wake up or kick other CPUs
695          */
696         if (set == level)
697                 return 0;
698
699         /*
700          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
701          * trigger a world-switch round on the running physical CPU to set the
702          * virtual IRQ/FIQ fields in the HCR appropriately.
703          */
704         kvm_vcpu_kick(vcpu);
705
706         return 0;
707 }
708
709 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
710                           bool line_status)
711 {
712         u32 irq = irq_level->irq;
713         unsigned int irq_type, vcpu_idx, irq_num;
714         int nrcpus = atomic_read(&kvm->online_vcpus);
715         struct kvm_vcpu *vcpu = NULL;
716         bool level = irq_level->level;
717
718         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
719         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
720         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
721
722         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
723
724         switch (irq_type) {
725         case KVM_ARM_IRQ_TYPE_CPU:
726                 if (irqchip_in_kernel(kvm))
727                         return -ENXIO;
728
729                 if (vcpu_idx >= nrcpus)
730                         return -EINVAL;
731
732                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
733                 if (!vcpu)
734                         return -EINVAL;
735
736                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
737                         return -EINVAL;
738
739                 return vcpu_interrupt_line(vcpu, irq_num, level);
740         case KVM_ARM_IRQ_TYPE_PPI:
741                 if (!irqchip_in_kernel(kvm))
742                         return -ENXIO;
743
744                 if (vcpu_idx >= nrcpus)
745                         return -EINVAL;
746
747                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
748                 if (!vcpu)
749                         return -EINVAL;
750
751                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
752                         return -EINVAL;
753
754                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
755         case KVM_ARM_IRQ_TYPE_SPI:
756                 if (!irqchip_in_kernel(kvm))
757                         return -ENXIO;
758
759                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
760                         return -EINVAL;
761
762                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
763         }
764
765         return -EINVAL;
766 }
767
768 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
769                                const struct kvm_vcpu_init *init)
770 {
771         unsigned int i;
772         int phys_target = kvm_target_cpu();
773
774         if (init->target != phys_target)
775                 return -EINVAL;
776
777         /*
778          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
779          * use the same target.
780          */
781         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
782                 return -EINVAL;
783
784         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
785         for (i = 0; i < sizeof(init->features) * 8; i++) {
786                 bool set = (init->features[i / 32] & (1 << (i % 32)));
787
788                 if (set && i >= KVM_VCPU_MAX_FEATURES)
789                         return -ENOENT;
790
791                 /*
792                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
793                  * use the same feature set.
794                  */
795                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
796                     test_bit(i, vcpu->arch.features) != set)
797                         return -EINVAL;
798
799                 if (set)
800                         set_bit(i, vcpu->arch.features);
801         }
802
803         vcpu->arch.target = phys_target;
804
805         /* Now we know what it is, we can reset it. */
806         return kvm_reset_vcpu(vcpu);
807 }
808
809
810 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
811                                          struct kvm_vcpu_init *init)
812 {
813         int ret;
814
815         ret = kvm_vcpu_set_target(vcpu, init);
816         if (ret)
817                 return ret;
818
819         /*
820          * Ensure a rebooted VM will fault in RAM pages and detect if the
821          * guest MMU is turned off and flush the caches as needed.
822          */
823         if (vcpu->arch.has_run_once)
824                 stage2_unmap_vm(vcpu->kvm);
825
826         vcpu_reset_hcr(vcpu);
827
828         /*
829          * Handle the "start in power-off" case.
830          */
831         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
832                 vcpu->arch.power_off = true;
833         else
834                 vcpu->arch.power_off = false;
835
836         return 0;
837 }
838
839 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
840                                  struct kvm_device_attr *attr)
841 {
842         int ret = -ENXIO;
843
844         switch (attr->group) {
845         default:
846                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
847                 break;
848         }
849
850         return ret;
851 }
852
853 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
854                                  struct kvm_device_attr *attr)
855 {
856         int ret = -ENXIO;
857
858         switch (attr->group) {
859         default:
860                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
861                 break;
862         }
863
864         return ret;
865 }
866
867 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
868                                  struct kvm_device_attr *attr)
869 {
870         int ret = -ENXIO;
871
872         switch (attr->group) {
873         default:
874                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
875                 break;
876         }
877
878         return ret;
879 }
880
881 long kvm_arch_vcpu_ioctl(struct file *filp,
882                          unsigned int ioctl, unsigned long arg)
883 {
884         struct kvm_vcpu *vcpu = filp->private_data;
885         void __user *argp = (void __user *)arg;
886         struct kvm_device_attr attr;
887
888         switch (ioctl) {
889         case KVM_ARM_VCPU_INIT: {
890                 struct kvm_vcpu_init init;
891
892                 if (copy_from_user(&init, argp, sizeof(init)))
893                         return -EFAULT;
894
895                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
896         }
897         case KVM_SET_ONE_REG:
898         case KVM_GET_ONE_REG: {
899                 struct kvm_one_reg reg;
900
901                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
902                         return -ENOEXEC;
903
904                 if (copy_from_user(&reg, argp, sizeof(reg)))
905                         return -EFAULT;
906                 if (ioctl == KVM_SET_ONE_REG)
907                         return kvm_arm_set_reg(vcpu, &reg);
908                 else
909                         return kvm_arm_get_reg(vcpu, &reg);
910         }
911         case KVM_GET_REG_LIST: {
912                 struct kvm_reg_list __user *user_list = argp;
913                 struct kvm_reg_list reg_list;
914                 unsigned n;
915
916                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
917                         return -ENOEXEC;
918
919                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
920                         return -EFAULT;
921                 n = reg_list.n;
922                 reg_list.n = kvm_arm_num_regs(vcpu);
923                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
924                         return -EFAULT;
925                 if (n < reg_list.n)
926                         return -E2BIG;
927                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
928         }
929         case KVM_SET_DEVICE_ATTR: {
930                 if (copy_from_user(&attr, argp, sizeof(attr)))
931                         return -EFAULT;
932                 return kvm_arm_vcpu_set_attr(vcpu, &attr);
933         }
934         case KVM_GET_DEVICE_ATTR: {
935                 if (copy_from_user(&attr, argp, sizeof(attr)))
936                         return -EFAULT;
937                 return kvm_arm_vcpu_get_attr(vcpu, &attr);
938         }
939         case KVM_HAS_DEVICE_ATTR: {
940                 if (copy_from_user(&attr, argp, sizeof(attr)))
941                         return -EFAULT;
942                 return kvm_arm_vcpu_has_attr(vcpu, &attr);
943         }
944         default:
945                 return -EINVAL;
946         }
947 }
948
949 /**
950  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
951  * @kvm: kvm instance
952  * @log: slot id and address to which we copy the log
953  *
954  * Steps 1-4 below provide general overview of dirty page logging. See
955  * kvm_get_dirty_log_protect() function description for additional details.
956  *
957  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
958  * always flush the TLB (step 4) even if previous step failed  and the dirty
959  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
960  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
961  * writes will be marked dirty for next log read.
962  *
963  *   1. Take a snapshot of the bit and clear it if needed.
964  *   2. Write protect the corresponding page.
965  *   3. Copy the snapshot to the userspace.
966  *   4. Flush TLB's if needed.
967  */
968 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
969 {
970         bool is_dirty = false;
971         int r;
972
973         mutex_lock(&kvm->slots_lock);
974
975         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
976
977         if (is_dirty)
978                 kvm_flush_remote_tlbs(kvm);
979
980         mutex_unlock(&kvm->slots_lock);
981         return r;
982 }
983
984 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
985                                         struct kvm_arm_device_addr *dev_addr)
986 {
987         unsigned long dev_id, type;
988
989         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
990                 KVM_ARM_DEVICE_ID_SHIFT;
991         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
992                 KVM_ARM_DEVICE_TYPE_SHIFT;
993
994         switch (dev_id) {
995         case KVM_ARM_DEVICE_VGIC_V2:
996                 if (!vgic_present)
997                         return -ENXIO;
998                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
999         default:
1000                 return -ENODEV;
1001         }
1002 }
1003
1004 long kvm_arch_vm_ioctl(struct file *filp,
1005                        unsigned int ioctl, unsigned long arg)
1006 {
1007         struct kvm *kvm = filp->private_data;
1008         void __user *argp = (void __user *)arg;
1009
1010         switch (ioctl) {
1011         case KVM_CREATE_IRQCHIP: {
1012                 if (!vgic_present)
1013                         return -ENXIO;
1014                 return kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1015         }
1016         case KVM_ARM_SET_DEVICE_ADDR: {
1017                 struct kvm_arm_device_addr dev_addr;
1018
1019                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1020                         return -EFAULT;
1021                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1022         }
1023         case KVM_ARM_PREFERRED_TARGET: {
1024                 int err;
1025                 struct kvm_vcpu_init init;
1026
1027                 err = kvm_vcpu_preferred_target(&init);
1028                 if (err)
1029                         return err;
1030
1031                 if (copy_to_user(argp, &init, sizeof(init)))
1032                         return -EFAULT;
1033
1034                 return 0;
1035         }
1036         default:
1037                 return -EINVAL;
1038         }
1039 }
1040
1041 static void cpu_init_hyp_mode(void *dummy)
1042 {
1043         phys_addr_t pgd_ptr;
1044         unsigned long hyp_stack_ptr;
1045         unsigned long stack_page;
1046         unsigned long vector_ptr;
1047
1048         /* Switch from the HYP stub to our own HYP init vector */
1049         __hyp_set_vectors(kvm_get_idmap_vector());
1050
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_ksym_ref(__kvm_hyp_vector);
1055
1056         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1057         __cpu_init_stage2();
1058
1059         kvm_arm_init_debug();
1060 }
1061
1062 static void cpu_hyp_reinit(void)
1063 {
1064         if (is_kernel_in_hyp_mode()) {
1065                 /*
1066                  * __cpu_init_stage2() is safe to call even if the PM
1067                  * event was cancelled before the CPU was reset.
1068                  */
1069                 __cpu_init_stage2();
1070         } else {
1071                 if (__hyp_get_vectors() == hyp_default_vectors)
1072                         cpu_init_hyp_mode(NULL);
1073         }
1074 }
1075
1076 static void cpu_hyp_reset(void)
1077 {
1078         if (!is_kernel_in_hyp_mode())
1079                 __cpu_reset_hyp_mode(hyp_default_vectors,
1080                                      kvm_get_idmap_start());
1081 }
1082
1083 static void _kvm_arch_hardware_enable(void *discard)
1084 {
1085         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1086                 cpu_hyp_reinit();
1087                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1088         }
1089 }
1090
1091 int kvm_arch_hardware_enable(void)
1092 {
1093         _kvm_arch_hardware_enable(NULL);
1094         return 0;
1095 }
1096
1097 static void _kvm_arch_hardware_disable(void *discard)
1098 {
1099         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1100                 cpu_hyp_reset();
1101                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1102         }
1103 }
1104
1105 void kvm_arch_hardware_disable(void)
1106 {
1107         _kvm_arch_hardware_disable(NULL);
1108 }
1109
1110 #ifdef CONFIG_CPU_PM
1111 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1112                                     unsigned long cmd,
1113                                     void *v)
1114 {
1115         /*
1116          * kvm_arm_hardware_enabled is left with its old value over
1117          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1118          * re-enable hyp.
1119          */
1120         switch (cmd) {
1121         case CPU_PM_ENTER:
1122                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1123                         /*
1124                          * don't update kvm_arm_hardware_enabled here
1125                          * so that the hardware will be re-enabled
1126                          * when we resume. See below.
1127                          */
1128                         cpu_hyp_reset();
1129
1130                 return NOTIFY_OK;
1131         case CPU_PM_EXIT:
1132                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1133                         /* The hardware was enabled before suspend. */
1134                         cpu_hyp_reinit();
1135
1136                 return NOTIFY_OK;
1137
1138         default:
1139                 return NOTIFY_DONE;
1140         }
1141 }
1142
1143 static struct notifier_block hyp_init_cpu_pm_nb = {
1144         .notifier_call = hyp_init_cpu_pm_notifier,
1145 };
1146
1147 static void __init hyp_cpu_pm_init(void)
1148 {
1149         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1150 }
1151 static void __init hyp_cpu_pm_exit(void)
1152 {
1153         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1154 }
1155 #else
1156 static inline void hyp_cpu_pm_init(void)
1157 {
1158 }
1159 static inline void hyp_cpu_pm_exit(void)
1160 {
1161 }
1162 #endif
1163
1164 static void teardown_common_resources(void)
1165 {
1166         free_percpu(kvm_host_cpu_state);
1167 }
1168
1169 static int init_common_resources(void)
1170 {
1171         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1172         if (!kvm_host_cpu_state) {
1173                 kvm_err("Cannot allocate host CPU state\n");
1174                 return -ENOMEM;
1175         }
1176
1177         return 0;
1178 }
1179
1180 static int init_subsystems(void)
1181 {
1182         int err = 0;
1183
1184         /*
1185          * Enable hardware so that subsystem initialisation can access EL2.
1186          */
1187         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1188
1189         /*
1190          * Register CPU lower-power notifier
1191          */
1192         hyp_cpu_pm_init();
1193
1194         /*
1195          * Init HYP view of VGIC
1196          */
1197         err = kvm_vgic_hyp_init();
1198         switch (err) {
1199         case 0:
1200                 vgic_present = true;
1201                 break;
1202         case -ENODEV:
1203         case -ENXIO:
1204                 vgic_present = false;
1205                 err = 0;
1206                 break;
1207         default:
1208                 goto out;
1209         }
1210
1211         /*
1212          * Init HYP architected timer support
1213          */
1214         err = kvm_timer_hyp_init();
1215         if (err)
1216                 goto out;
1217
1218         kvm_perf_init();
1219         kvm_coproc_table_init();
1220
1221 out:
1222         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1223
1224         return err;
1225 }
1226
1227 static void teardown_hyp_mode(void)
1228 {
1229         int cpu;
1230
1231         if (is_kernel_in_hyp_mode())
1232                 return;
1233
1234         free_hyp_pgds();
1235         for_each_possible_cpu(cpu)
1236                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1237         hyp_cpu_pm_exit();
1238 }
1239
1240 static int init_vhe_mode(void)
1241 {
1242         /* set size of VMID supported by CPU */
1243         kvm_vmid_bits = kvm_get_vmid_bits();
1244         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1245
1246         kvm_info("VHE mode initialized successfully\n");
1247         return 0;
1248 }
1249
1250 /**
1251  * Inits Hyp-mode on all online CPUs
1252  */
1253 static int init_hyp_mode(void)
1254 {
1255         int cpu;
1256         int err = 0;
1257
1258         /*
1259          * Allocate Hyp PGD and setup Hyp identity mapping
1260          */
1261         err = kvm_mmu_init();
1262         if (err)
1263                 goto out_err;
1264
1265         /*
1266          * It is probably enough to obtain the default on one
1267          * CPU. It's unlikely to be different on the others.
1268          */
1269         hyp_default_vectors = __hyp_get_vectors();
1270
1271         /*
1272          * Allocate stack pages for Hypervisor-mode
1273          */
1274         for_each_possible_cpu(cpu) {
1275                 unsigned long stack_page;
1276
1277                 stack_page = __get_free_page(GFP_KERNEL);
1278                 if (!stack_page) {
1279                         err = -ENOMEM;
1280                         goto out_err;
1281                 }
1282
1283                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1284         }
1285
1286         /*
1287          * Map the Hyp-code called directly from the host
1288          */
1289         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1290                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1291         if (err) {
1292                 kvm_err("Cannot map world-switch code\n");
1293                 goto out_err;
1294         }
1295
1296         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1297                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1298         if (err) {
1299                 kvm_err("Cannot map rodata section\n");
1300                 goto out_err;
1301         }
1302
1303         /*
1304          * Map the Hyp stack pages
1305          */
1306         for_each_possible_cpu(cpu) {
1307                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1308                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1309                                           PAGE_HYP);
1310
1311                 if (err) {
1312                         kvm_err("Cannot map hyp stack\n");
1313                         goto out_err;
1314                 }
1315         }
1316
1317         for_each_possible_cpu(cpu) {
1318                 kvm_cpu_context_t *cpu_ctxt;
1319
1320                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1321                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1322
1323                 if (err) {
1324                         kvm_err("Cannot map host CPU state: %d\n", err);
1325                         goto out_err;
1326                 }
1327         }
1328
1329         /* set size of VMID supported by CPU */
1330         kvm_vmid_bits = kvm_get_vmid_bits();
1331         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1332
1333         kvm_info("Hyp mode initialized successfully\n");
1334
1335         return 0;
1336
1337 out_err:
1338         teardown_hyp_mode();
1339         kvm_err("error initializing Hyp mode: %d\n", err);
1340         return err;
1341 }
1342
1343 static void check_kvm_target_cpu(void *ret)
1344 {
1345         *(int *)ret = kvm_target_cpu();
1346 }
1347
1348 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1349 {
1350         struct kvm_vcpu *vcpu;
1351         int i;
1352
1353         mpidr &= MPIDR_HWID_BITMASK;
1354         kvm_for_each_vcpu(i, vcpu, kvm) {
1355                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1356                         return vcpu;
1357         }
1358         return NULL;
1359 }
1360
1361 /**
1362  * Initialize Hyp-mode and memory mappings on all CPUs.
1363  */
1364 int kvm_arch_init(void *opaque)
1365 {
1366         int err;
1367         int ret, cpu;
1368
1369         if (!is_hyp_mode_available()) {
1370                 kvm_err("HYP mode not available\n");
1371                 return -ENODEV;
1372         }
1373
1374         for_each_online_cpu(cpu) {
1375                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1376                 if (ret < 0) {
1377                         kvm_err("Error, CPU %d not supported!\n", cpu);
1378                         return -ENODEV;
1379                 }
1380         }
1381
1382         err = init_common_resources();
1383         if (err)
1384                 return err;
1385
1386         if (is_kernel_in_hyp_mode())
1387                 err = init_vhe_mode();
1388         else
1389                 err = init_hyp_mode();
1390         if (err)
1391                 goto out_err;
1392
1393         err = init_subsystems();
1394         if (err)
1395                 goto out_hyp;
1396
1397         return 0;
1398
1399 out_hyp:
1400         teardown_hyp_mode();
1401 out_err:
1402         teardown_common_resources();
1403         return err;
1404 }
1405
1406 /* NOP: Compiling as a module not supported */
1407 void kvm_arch_exit(void)
1408 {
1409         kvm_perf_teardown();
1410 }
1411
1412 static int arm_init(void)
1413 {
1414         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1415         return rc;
1416 }
1417
1418 module_init(arm_init);