Merge branch 'stable-4.9' of git://git.infradead.org/users/pcmoore/audit
[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         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
162                 if (kvm->vcpus[i]) {
163                         kvm_arch_vcpu_free(kvm->vcpus[i]);
164                         kvm->vcpus[i] = NULL;
165                 }
166         }
167
168         kvm_vgic_destroy(kvm);
169 }
170
171 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
172 {
173         int r;
174         switch (ext) {
175         case KVM_CAP_IRQCHIP:
176                 r = vgic_present;
177                 break;
178         case KVM_CAP_IOEVENTFD:
179         case KVM_CAP_DEVICE_CTRL:
180         case KVM_CAP_USER_MEMORY:
181         case KVM_CAP_SYNC_MMU:
182         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
183         case KVM_CAP_ONE_REG:
184         case KVM_CAP_ARM_PSCI:
185         case KVM_CAP_ARM_PSCI_0_2:
186         case KVM_CAP_READONLY_MEM:
187         case KVM_CAP_MP_STATE:
188                 r = 1;
189                 break;
190         case KVM_CAP_COALESCED_MMIO:
191                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
192                 break;
193         case KVM_CAP_ARM_SET_DEVICE_ADDR:
194                 r = 1;
195                 break;
196         case KVM_CAP_NR_VCPUS:
197                 r = num_online_cpus();
198                 break;
199         case KVM_CAP_MAX_VCPUS:
200                 r = KVM_MAX_VCPUS;
201                 break;
202         default:
203                 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
204                 break;
205         }
206         return r;
207 }
208
209 long kvm_arch_dev_ioctl(struct file *filp,
210                         unsigned int ioctl, unsigned long arg)
211 {
212         return -EINVAL;
213 }
214
215
216 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
217 {
218         int err;
219         struct kvm_vcpu *vcpu;
220
221         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
222                 err = -EBUSY;
223                 goto out;
224         }
225
226         if (id >= kvm->arch.max_vcpus) {
227                 err = -EINVAL;
228                 goto out;
229         }
230
231         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
232         if (!vcpu) {
233                 err = -ENOMEM;
234                 goto out;
235         }
236
237         err = kvm_vcpu_init(vcpu, kvm, id);
238         if (err)
239                 goto free_vcpu;
240
241         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
242         if (err)
243                 goto vcpu_uninit;
244
245         return vcpu;
246 vcpu_uninit:
247         kvm_vcpu_uninit(vcpu);
248 free_vcpu:
249         kmem_cache_free(kvm_vcpu_cache, vcpu);
250 out:
251         return ERR_PTR(err);
252 }
253
254 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
255 {
256         kvm_vgic_vcpu_early_init(vcpu);
257 }
258
259 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
260 {
261         kvm_mmu_free_memory_caches(vcpu);
262         kvm_timer_vcpu_terminate(vcpu);
263         kvm_vgic_vcpu_destroy(vcpu);
264         kvm_pmu_vcpu_destroy(vcpu);
265         kvm_vcpu_uninit(vcpu);
266         kmem_cache_free(kvm_vcpu_cache, vcpu);
267 }
268
269 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
270 {
271         kvm_arch_vcpu_free(vcpu);
272 }
273
274 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
275 {
276         return kvm_timer_should_fire(vcpu);
277 }
278
279 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
280 {
281         kvm_timer_schedule(vcpu);
282 }
283
284 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
285 {
286         kvm_timer_unschedule(vcpu);
287 }
288
289 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
290 {
291         /* Force users to call KVM_ARM_VCPU_INIT */
292         vcpu->arch.target = -1;
293         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
294
295         /* Set up the timer */
296         kvm_timer_vcpu_init(vcpu);
297
298         kvm_arm_reset_debug_ptr(vcpu);
299
300         return 0;
301 }
302
303 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
304 {
305         vcpu->cpu = cpu;
306         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
307
308         kvm_arm_set_running_vcpu(vcpu);
309 }
310
311 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
312 {
313         /*
314          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
315          * if the vcpu is no longer assigned to a cpu.  This is used for the
316          * optimized make_all_cpus_request path.
317          */
318         vcpu->cpu = -1;
319
320         kvm_arm_set_running_vcpu(NULL);
321         kvm_timer_vcpu_put(vcpu);
322 }
323
324 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
325                                     struct kvm_mp_state *mp_state)
326 {
327         if (vcpu->arch.power_off)
328                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
329         else
330                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
331
332         return 0;
333 }
334
335 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
336                                     struct kvm_mp_state *mp_state)
337 {
338         switch (mp_state->mp_state) {
339         case KVM_MP_STATE_RUNNABLE:
340                 vcpu->arch.power_off = false;
341                 break;
342         case KVM_MP_STATE_STOPPED:
343                 vcpu->arch.power_off = true;
344                 break;
345         default:
346                 return -EINVAL;
347         }
348
349         return 0;
350 }
351
352 /**
353  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
354  * @v:          The VCPU pointer
355  *
356  * If the guest CPU is not waiting for interrupts or an interrupt line is
357  * asserted, the CPU is by definition runnable.
358  */
359 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
360 {
361         return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
362                 && !v->arch.power_off && !v->arch.pause);
363 }
364
365 /* Just ensure a guest exit from a particular CPU */
366 static void exit_vm_noop(void *info)
367 {
368 }
369
370 void force_vm_exit(const cpumask_t *mask)
371 {
372         preempt_disable();
373         smp_call_function_many(mask, exit_vm_noop, NULL, true);
374         preempt_enable();
375 }
376
377 /**
378  * need_new_vmid_gen - check that the VMID is still valid
379  * @kvm: The VM's VMID to check
380  *
381  * return true if there is a new generation of VMIDs being used
382  *
383  * The hardware supports only 256 values with the value zero reserved for the
384  * host, so we check if an assigned value belongs to a previous generation,
385  * which which requires us to assign a new value. If we're the first to use a
386  * VMID for the new generation, we must flush necessary caches and TLBs on all
387  * CPUs.
388  */
389 static bool need_new_vmid_gen(struct kvm *kvm)
390 {
391         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
392 }
393
394 /**
395  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
396  * @kvm The guest that we are about to run
397  *
398  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
399  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
400  * caches and TLBs.
401  */
402 static void update_vttbr(struct kvm *kvm)
403 {
404         phys_addr_t pgd_phys;
405         u64 vmid;
406
407         if (!need_new_vmid_gen(kvm))
408                 return;
409
410         spin_lock(&kvm_vmid_lock);
411
412         /*
413          * We need to re-check the vmid_gen here to ensure that if another vcpu
414          * already allocated a valid vmid for this vm, then this vcpu should
415          * use the same vmid.
416          */
417         if (!need_new_vmid_gen(kvm)) {
418                 spin_unlock(&kvm_vmid_lock);
419                 return;
420         }
421
422         /* First user of a new VMID generation? */
423         if (unlikely(kvm_next_vmid == 0)) {
424                 atomic64_inc(&kvm_vmid_gen);
425                 kvm_next_vmid = 1;
426
427                 /*
428                  * On SMP we know no other CPUs can use this CPU's or each
429                  * other's VMID after force_vm_exit returns since the
430                  * kvm_vmid_lock blocks them from reentry to the guest.
431                  */
432                 force_vm_exit(cpu_all_mask);
433                 /*
434                  * Now broadcast TLB + ICACHE invalidation over the inner
435                  * shareable domain to make sure all data structures are
436                  * clean.
437                  */
438                 kvm_call_hyp(__kvm_flush_vm_context);
439         }
440
441         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
442         kvm->arch.vmid = kvm_next_vmid;
443         kvm_next_vmid++;
444         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
445
446         /* update vttbr to be used with the new vmid */
447         pgd_phys = virt_to_phys(kvm->arch.pgd);
448         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
449         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
450         kvm->arch.vttbr = pgd_phys | vmid;
451
452         spin_unlock(&kvm_vmid_lock);
453 }
454
455 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
456 {
457         struct kvm *kvm = vcpu->kvm;
458         int ret = 0;
459
460         if (likely(vcpu->arch.has_run_once))
461                 return 0;
462
463         vcpu->arch.has_run_once = true;
464
465         /*
466          * Map the VGIC hardware resources before running a vcpu the first
467          * time on this VM.
468          */
469         if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
470                 ret = kvm_vgic_map_resources(kvm);
471                 if (ret)
472                         return ret;
473         }
474
475         /*
476          * Enable the arch timers only if we have an in-kernel VGIC
477          * and it has been properly initialized, since we cannot handle
478          * interrupts from the virtual timer with a userspace gic.
479          */
480         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
481                 ret = kvm_timer_enable(vcpu);
482
483         return ret;
484 }
485
486 bool kvm_arch_intc_initialized(struct kvm *kvm)
487 {
488         return vgic_initialized(kvm);
489 }
490
491 void kvm_arm_halt_guest(struct kvm *kvm)
492 {
493         int i;
494         struct kvm_vcpu *vcpu;
495
496         kvm_for_each_vcpu(i, vcpu, kvm)
497                 vcpu->arch.pause = true;
498         kvm_make_all_cpus_request(kvm, KVM_REQ_VCPU_EXIT);
499 }
500
501 void kvm_arm_halt_vcpu(struct kvm_vcpu *vcpu)
502 {
503         vcpu->arch.pause = true;
504         kvm_vcpu_kick(vcpu);
505 }
506
507 void kvm_arm_resume_vcpu(struct kvm_vcpu *vcpu)
508 {
509         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
510
511         vcpu->arch.pause = false;
512         swake_up(wq);
513 }
514
515 void kvm_arm_resume_guest(struct kvm *kvm)
516 {
517         int i;
518         struct kvm_vcpu *vcpu;
519
520         kvm_for_each_vcpu(i, vcpu, kvm)
521                 kvm_arm_resume_vcpu(vcpu);
522 }
523
524 static void vcpu_sleep(struct kvm_vcpu *vcpu)
525 {
526         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
527
528         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
529                                        (!vcpu->arch.pause)));
530 }
531
532 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
533 {
534         return vcpu->arch.target >= 0;
535 }
536
537 /**
538  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
539  * @vcpu:       The VCPU pointer
540  * @run:        The kvm_run structure pointer used for userspace state exchange
541  *
542  * This function is called through the VCPU_RUN ioctl called from user space. It
543  * will execute VM code in a loop until the time slice for the process is used
544  * or some emulation is needed from user space in which case the function will
545  * return with return value 0 and with the kvm_run structure filled in with the
546  * required data for the requested emulation.
547  */
548 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
549 {
550         int ret;
551         sigset_t sigsaved;
552
553         if (unlikely(!kvm_vcpu_initialized(vcpu)))
554                 return -ENOEXEC;
555
556         ret = kvm_vcpu_first_run_init(vcpu);
557         if (ret)
558                 return ret;
559
560         if (run->exit_reason == KVM_EXIT_MMIO) {
561                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
562                 if (ret)
563                         return ret;
564         }
565
566         if (vcpu->sigset_active)
567                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
568
569         ret = 1;
570         run->exit_reason = KVM_EXIT_UNKNOWN;
571         while (ret > 0) {
572                 /*
573                  * Check conditions before entering the guest
574                  */
575                 cond_resched();
576
577                 update_vttbr(vcpu->kvm);
578
579                 if (vcpu->arch.power_off || vcpu->arch.pause)
580                         vcpu_sleep(vcpu);
581
582                 /*
583                  * Preparing the interrupts to be injected also
584                  * involves poking the GIC, which must be done in a
585                  * non-preemptible context.
586                  */
587                 preempt_disable();
588                 kvm_pmu_flush_hwstate(vcpu);
589                 kvm_timer_flush_hwstate(vcpu);
590                 kvm_vgic_flush_hwstate(vcpu);
591
592                 local_irq_disable();
593
594                 /*
595                  * Re-check atomic conditions
596                  */
597                 if (signal_pending(current)) {
598                         ret = -EINTR;
599                         run->exit_reason = KVM_EXIT_INTR;
600                 }
601
602                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
603                         vcpu->arch.power_off || vcpu->arch.pause) {
604                         local_irq_enable();
605                         kvm_pmu_sync_hwstate(vcpu);
606                         kvm_timer_sync_hwstate(vcpu);
607                         kvm_vgic_sync_hwstate(vcpu);
608                         preempt_enable();
609                         continue;
610                 }
611
612                 kvm_arm_setup_debug(vcpu);
613
614                 /**************************************************************
615                  * Enter the guest
616                  */
617                 trace_kvm_entry(*vcpu_pc(vcpu));
618                 guest_enter_irqoff();
619                 vcpu->mode = IN_GUEST_MODE;
620
621                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
622
623                 vcpu->mode = OUTSIDE_GUEST_MODE;
624                 vcpu->stat.exits++;
625                 /*
626                  * Back from guest
627                  *************************************************************/
628
629                 kvm_arm_clear_debug(vcpu);
630
631                 /*
632                  * We may have taken a host interrupt in HYP mode (ie
633                  * while executing the guest). This interrupt is still
634                  * pending, as we haven't serviced it yet!
635                  *
636                  * We're now back in SVC mode, with interrupts
637                  * disabled.  Enabling the interrupts now will have
638                  * the effect of taking the interrupt again, in SVC
639                  * mode this time.
640                  */
641                 local_irq_enable();
642
643                 /*
644                  * We do local_irq_enable() before calling guest_exit() so
645                  * that if a timer interrupt hits while running the guest we
646                  * account that tick as being spent in the guest.  We enable
647                  * preemption after calling guest_exit() so that if we get
648                  * preempted we make sure ticks after that is not counted as
649                  * guest time.
650                  */
651                 guest_exit();
652                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
653
654                 /*
655                  * We must sync the PMU and timer state before the vgic state so
656                  * that the vgic can properly sample the updated state of the
657                  * interrupt line.
658                  */
659                 kvm_pmu_sync_hwstate(vcpu);
660                 kvm_timer_sync_hwstate(vcpu);
661
662                 kvm_vgic_sync_hwstate(vcpu);
663
664                 preempt_enable();
665
666                 ret = handle_exit(vcpu, run, ret);
667         }
668
669         if (vcpu->sigset_active)
670                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
671         return ret;
672 }
673
674 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
675 {
676         int bit_index;
677         bool set;
678         unsigned long *ptr;
679
680         if (number == KVM_ARM_IRQ_CPU_IRQ)
681                 bit_index = __ffs(HCR_VI);
682         else /* KVM_ARM_IRQ_CPU_FIQ */
683                 bit_index = __ffs(HCR_VF);
684
685         ptr = (unsigned long *)&vcpu->arch.irq_lines;
686         if (level)
687                 set = test_and_set_bit(bit_index, ptr);
688         else
689                 set = test_and_clear_bit(bit_index, ptr);
690
691         /*
692          * If we didn't change anything, no need to wake up or kick other CPUs
693          */
694         if (set == level)
695                 return 0;
696
697         /*
698          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
699          * trigger a world-switch round on the running physical CPU to set the
700          * virtual IRQ/FIQ fields in the HCR appropriately.
701          */
702         kvm_vcpu_kick(vcpu);
703
704         return 0;
705 }
706
707 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
708                           bool line_status)
709 {
710         u32 irq = irq_level->irq;
711         unsigned int irq_type, vcpu_idx, irq_num;
712         int nrcpus = atomic_read(&kvm->online_vcpus);
713         struct kvm_vcpu *vcpu = NULL;
714         bool level = irq_level->level;
715
716         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
717         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
718         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
719
720         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
721
722         switch (irq_type) {
723         case KVM_ARM_IRQ_TYPE_CPU:
724                 if (irqchip_in_kernel(kvm))
725                         return -ENXIO;
726
727                 if (vcpu_idx >= nrcpus)
728                         return -EINVAL;
729
730                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
731                 if (!vcpu)
732                         return -EINVAL;
733
734                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
735                         return -EINVAL;
736
737                 return vcpu_interrupt_line(vcpu, irq_num, level);
738         case KVM_ARM_IRQ_TYPE_PPI:
739                 if (!irqchip_in_kernel(kvm))
740                         return -ENXIO;
741
742                 if (vcpu_idx >= nrcpus)
743                         return -EINVAL;
744
745                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
746                 if (!vcpu)
747                         return -EINVAL;
748
749                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
750                         return -EINVAL;
751
752                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
753         case KVM_ARM_IRQ_TYPE_SPI:
754                 if (!irqchip_in_kernel(kvm))
755                         return -ENXIO;
756
757                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
758                         return -EINVAL;
759
760                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
761         }
762
763         return -EINVAL;
764 }
765
766 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
767                                const struct kvm_vcpu_init *init)
768 {
769         unsigned int i;
770         int phys_target = kvm_target_cpu();
771
772         if (init->target != phys_target)
773                 return -EINVAL;
774
775         /*
776          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
777          * use the same target.
778          */
779         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
780                 return -EINVAL;
781
782         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
783         for (i = 0; i < sizeof(init->features) * 8; i++) {
784                 bool set = (init->features[i / 32] & (1 << (i % 32)));
785
786                 if (set && i >= KVM_VCPU_MAX_FEATURES)
787                         return -ENOENT;
788
789                 /*
790                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
791                  * use the same feature set.
792                  */
793                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
794                     test_bit(i, vcpu->arch.features) != set)
795                         return -EINVAL;
796
797                 if (set)
798                         set_bit(i, vcpu->arch.features);
799         }
800
801         vcpu->arch.target = phys_target;
802
803         /* Now we know what it is, we can reset it. */
804         return kvm_reset_vcpu(vcpu);
805 }
806
807
808 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
809                                          struct kvm_vcpu_init *init)
810 {
811         int ret;
812
813         ret = kvm_vcpu_set_target(vcpu, init);
814         if (ret)
815                 return ret;
816
817         /*
818          * Ensure a rebooted VM will fault in RAM pages and detect if the
819          * guest MMU is turned off and flush the caches as needed.
820          */
821         if (vcpu->arch.has_run_once)
822                 stage2_unmap_vm(vcpu->kvm);
823
824         vcpu_reset_hcr(vcpu);
825
826         /*
827          * Handle the "start in power-off" case.
828          */
829         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
830                 vcpu->arch.power_off = true;
831         else
832                 vcpu->arch.power_off = false;
833
834         return 0;
835 }
836
837 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
838                                  struct kvm_device_attr *attr)
839 {
840         int ret = -ENXIO;
841
842         switch (attr->group) {
843         default:
844                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
845                 break;
846         }
847
848         return ret;
849 }
850
851 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
852                                  struct kvm_device_attr *attr)
853 {
854         int ret = -ENXIO;
855
856         switch (attr->group) {
857         default:
858                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
859                 break;
860         }
861
862         return ret;
863 }
864
865 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
866                                  struct kvm_device_attr *attr)
867 {
868         int ret = -ENXIO;
869
870         switch (attr->group) {
871         default:
872                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
873                 break;
874         }
875
876         return ret;
877 }
878
879 long kvm_arch_vcpu_ioctl(struct file *filp,
880                          unsigned int ioctl, unsigned long arg)
881 {
882         struct kvm_vcpu *vcpu = filp->private_data;
883         void __user *argp = (void __user *)arg;
884         struct kvm_device_attr attr;
885
886         switch (ioctl) {
887         case KVM_ARM_VCPU_INIT: {
888                 struct kvm_vcpu_init init;
889
890                 if (copy_from_user(&init, argp, sizeof(init)))
891                         return -EFAULT;
892
893                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
894         }
895         case KVM_SET_ONE_REG:
896         case KVM_GET_ONE_REG: {
897                 struct kvm_one_reg reg;
898
899                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
900                         return -ENOEXEC;
901
902                 if (copy_from_user(&reg, argp, sizeof(reg)))
903                         return -EFAULT;
904                 if (ioctl == KVM_SET_ONE_REG)
905                         return kvm_arm_set_reg(vcpu, &reg);
906                 else
907                         return kvm_arm_get_reg(vcpu, &reg);
908         }
909         case KVM_GET_REG_LIST: {
910                 struct kvm_reg_list __user *user_list = argp;
911                 struct kvm_reg_list reg_list;
912                 unsigned n;
913
914                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
915                         return -ENOEXEC;
916
917                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
918                         return -EFAULT;
919                 n = reg_list.n;
920                 reg_list.n = kvm_arm_num_regs(vcpu);
921                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
922                         return -EFAULT;
923                 if (n < reg_list.n)
924                         return -E2BIG;
925                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
926         }
927         case KVM_SET_DEVICE_ATTR: {
928                 if (copy_from_user(&attr, argp, sizeof(attr)))
929                         return -EFAULT;
930                 return kvm_arm_vcpu_set_attr(vcpu, &attr);
931         }
932         case KVM_GET_DEVICE_ATTR: {
933                 if (copy_from_user(&attr, argp, sizeof(attr)))
934                         return -EFAULT;
935                 return kvm_arm_vcpu_get_attr(vcpu, &attr);
936         }
937         case KVM_HAS_DEVICE_ATTR: {
938                 if (copy_from_user(&attr, argp, sizeof(attr)))
939                         return -EFAULT;
940                 return kvm_arm_vcpu_has_attr(vcpu, &attr);
941         }
942         default:
943                 return -EINVAL;
944         }
945 }
946
947 /**
948  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
949  * @kvm: kvm instance
950  * @log: slot id and address to which we copy the log
951  *
952  * Steps 1-4 below provide general overview of dirty page logging. See
953  * kvm_get_dirty_log_protect() function description for additional details.
954  *
955  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
956  * always flush the TLB (step 4) even if previous step failed  and the dirty
957  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
958  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
959  * writes will be marked dirty for next log read.
960  *
961  *   1. Take a snapshot of the bit and clear it if needed.
962  *   2. Write protect the corresponding page.
963  *   3. Copy the snapshot to the userspace.
964  *   4. Flush TLB's if needed.
965  */
966 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
967 {
968         bool is_dirty = false;
969         int r;
970
971         mutex_lock(&kvm->slots_lock);
972
973         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
974
975         if (is_dirty)
976                 kvm_flush_remote_tlbs(kvm);
977
978         mutex_unlock(&kvm->slots_lock);
979         return r;
980 }
981
982 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
983                                         struct kvm_arm_device_addr *dev_addr)
984 {
985         unsigned long dev_id, type;
986
987         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
988                 KVM_ARM_DEVICE_ID_SHIFT;
989         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
990                 KVM_ARM_DEVICE_TYPE_SHIFT;
991
992         switch (dev_id) {
993         case KVM_ARM_DEVICE_VGIC_V2:
994                 if (!vgic_present)
995                         return -ENXIO;
996                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
997         default:
998                 return -ENODEV;
999         }
1000 }
1001
1002 long kvm_arch_vm_ioctl(struct file *filp,
1003                        unsigned int ioctl, unsigned long arg)
1004 {
1005         struct kvm *kvm = filp->private_data;
1006         void __user *argp = (void __user *)arg;
1007
1008         switch (ioctl) {
1009         case KVM_CREATE_IRQCHIP: {
1010                 int ret;
1011                 if (!vgic_present)
1012                         return -ENXIO;
1013                 mutex_lock(&kvm->lock);
1014                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1015                 mutex_unlock(&kvm->lock);
1016                 return ret;
1017         }
1018         case KVM_ARM_SET_DEVICE_ADDR: {
1019                 struct kvm_arm_device_addr dev_addr;
1020
1021                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1022                         return -EFAULT;
1023                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1024         }
1025         case KVM_ARM_PREFERRED_TARGET: {
1026                 int err;
1027                 struct kvm_vcpu_init init;
1028
1029                 err = kvm_vcpu_preferred_target(&init);
1030                 if (err)
1031                         return err;
1032
1033                 if (copy_to_user(argp, &init, sizeof(init)))
1034                         return -EFAULT;
1035
1036                 return 0;
1037         }
1038         default:
1039                 return -EINVAL;
1040         }
1041 }
1042
1043 static void cpu_init_hyp_mode(void *dummy)
1044 {
1045         phys_addr_t pgd_ptr;
1046         unsigned long hyp_stack_ptr;
1047         unsigned long stack_page;
1048         unsigned long vector_ptr;
1049
1050         /* Switch from the HYP stub to our own HYP init vector */
1051         __hyp_set_vectors(kvm_get_idmap_vector());
1052
1053         pgd_ptr = kvm_mmu_get_httbr();
1054         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1055         hyp_stack_ptr = stack_page + PAGE_SIZE;
1056         vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1057
1058         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1059         __cpu_init_stage2();
1060
1061         kvm_arm_init_debug();
1062 }
1063
1064 static void cpu_hyp_reinit(void)
1065 {
1066         if (is_kernel_in_hyp_mode()) {
1067                 /*
1068                  * __cpu_init_stage2() is safe to call even if the PM
1069                  * event was cancelled before the CPU was reset.
1070                  */
1071                 __cpu_init_stage2();
1072         } else {
1073                 if (__hyp_get_vectors() == hyp_default_vectors)
1074                         cpu_init_hyp_mode(NULL);
1075         }
1076 }
1077
1078 static void cpu_hyp_reset(void)
1079 {
1080         if (!is_kernel_in_hyp_mode())
1081                 __cpu_reset_hyp_mode(hyp_default_vectors,
1082                                      kvm_get_idmap_start());
1083 }
1084
1085 static void _kvm_arch_hardware_enable(void *discard)
1086 {
1087         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1088                 cpu_hyp_reinit();
1089                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1090         }
1091 }
1092
1093 int kvm_arch_hardware_enable(void)
1094 {
1095         _kvm_arch_hardware_enable(NULL);
1096         return 0;
1097 }
1098
1099 static void _kvm_arch_hardware_disable(void *discard)
1100 {
1101         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1102                 cpu_hyp_reset();
1103                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1104         }
1105 }
1106
1107 void kvm_arch_hardware_disable(void)
1108 {
1109         _kvm_arch_hardware_disable(NULL);
1110 }
1111
1112 #ifdef CONFIG_CPU_PM
1113 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1114                                     unsigned long cmd,
1115                                     void *v)
1116 {
1117         /*
1118          * kvm_arm_hardware_enabled is left with its old value over
1119          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1120          * re-enable hyp.
1121          */
1122         switch (cmd) {
1123         case CPU_PM_ENTER:
1124                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1125                         /*
1126                          * don't update kvm_arm_hardware_enabled here
1127                          * so that the hardware will be re-enabled
1128                          * when we resume. See below.
1129                          */
1130                         cpu_hyp_reset();
1131
1132                 return NOTIFY_OK;
1133         case CPU_PM_EXIT:
1134                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1135                         /* The hardware was enabled before suspend. */
1136                         cpu_hyp_reinit();
1137
1138                 return NOTIFY_OK;
1139
1140         default:
1141                 return NOTIFY_DONE;
1142         }
1143 }
1144
1145 static struct notifier_block hyp_init_cpu_pm_nb = {
1146         .notifier_call = hyp_init_cpu_pm_notifier,
1147 };
1148
1149 static void __init hyp_cpu_pm_init(void)
1150 {
1151         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1152 }
1153 static void __init hyp_cpu_pm_exit(void)
1154 {
1155         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1156 }
1157 #else
1158 static inline void hyp_cpu_pm_init(void)
1159 {
1160 }
1161 static inline void hyp_cpu_pm_exit(void)
1162 {
1163 }
1164 #endif
1165
1166 static void teardown_common_resources(void)
1167 {
1168         free_percpu(kvm_host_cpu_state);
1169 }
1170
1171 static int init_common_resources(void)
1172 {
1173         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1174         if (!kvm_host_cpu_state) {
1175                 kvm_err("Cannot allocate host CPU state\n");
1176                 return -ENOMEM;
1177         }
1178
1179         return 0;
1180 }
1181
1182 static int init_subsystems(void)
1183 {
1184         int err = 0;
1185
1186         /*
1187          * Enable hardware so that subsystem initialisation can access EL2.
1188          */
1189         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1190
1191         /*
1192          * Register CPU lower-power notifier
1193          */
1194         hyp_cpu_pm_init();
1195
1196         /*
1197          * Init HYP view of VGIC
1198          */
1199         err = kvm_vgic_hyp_init();
1200         switch (err) {
1201         case 0:
1202                 vgic_present = true;
1203                 break;
1204         case -ENODEV:
1205         case -ENXIO:
1206                 vgic_present = false;
1207                 err = 0;
1208                 break;
1209         default:
1210                 goto out;
1211         }
1212
1213         /*
1214          * Init HYP architected timer support
1215          */
1216         err = kvm_timer_hyp_init();
1217         if (err)
1218                 goto out;
1219
1220         kvm_perf_init();
1221         kvm_coproc_table_init();
1222
1223 out:
1224         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1225
1226         return err;
1227 }
1228
1229 static void teardown_hyp_mode(void)
1230 {
1231         int cpu;
1232
1233         if (is_kernel_in_hyp_mode())
1234                 return;
1235
1236         free_hyp_pgds();
1237         for_each_possible_cpu(cpu)
1238                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1239         hyp_cpu_pm_exit();
1240 }
1241
1242 static int init_vhe_mode(void)
1243 {
1244         /* set size of VMID supported by CPU */
1245         kvm_vmid_bits = kvm_get_vmid_bits();
1246         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1247
1248         kvm_info("VHE mode initialized successfully\n");
1249         return 0;
1250 }
1251
1252 /**
1253  * Inits Hyp-mode on all online CPUs
1254  */
1255 static int init_hyp_mode(void)
1256 {
1257         int cpu;
1258         int err = 0;
1259
1260         /*
1261          * Allocate Hyp PGD and setup Hyp identity mapping
1262          */
1263         err = kvm_mmu_init();
1264         if (err)
1265                 goto out_err;
1266
1267         /*
1268          * It is probably enough to obtain the default on one
1269          * CPU. It's unlikely to be different on the others.
1270          */
1271         hyp_default_vectors = __hyp_get_vectors();
1272
1273         /*
1274          * Allocate stack pages for Hypervisor-mode
1275          */
1276         for_each_possible_cpu(cpu) {
1277                 unsigned long stack_page;
1278
1279                 stack_page = __get_free_page(GFP_KERNEL);
1280                 if (!stack_page) {
1281                         err = -ENOMEM;
1282                         goto out_err;
1283                 }
1284
1285                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1286         }
1287
1288         /*
1289          * Map the Hyp-code called directly from the host
1290          */
1291         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1292                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1293         if (err) {
1294                 kvm_err("Cannot map world-switch code\n");
1295                 goto out_err;
1296         }
1297
1298         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1299                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1300         if (err) {
1301                 kvm_err("Cannot map rodata section\n");
1302                 goto out_err;
1303         }
1304
1305         /*
1306          * Map the Hyp stack pages
1307          */
1308         for_each_possible_cpu(cpu) {
1309                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1310                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1311                                           PAGE_HYP);
1312
1313                 if (err) {
1314                         kvm_err("Cannot map hyp stack\n");
1315                         goto out_err;
1316                 }
1317         }
1318
1319         for_each_possible_cpu(cpu) {
1320                 kvm_cpu_context_t *cpu_ctxt;
1321
1322                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1323                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1324
1325                 if (err) {
1326                         kvm_err("Cannot map host CPU state: %d\n", err);
1327                         goto out_err;
1328                 }
1329         }
1330
1331         /* set size of VMID supported by CPU */
1332         kvm_vmid_bits = kvm_get_vmid_bits();
1333         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1334
1335         kvm_info("Hyp mode initialized successfully\n");
1336
1337         return 0;
1338
1339 out_err:
1340         teardown_hyp_mode();
1341         kvm_err("error initializing Hyp mode: %d\n", err);
1342         return err;
1343 }
1344
1345 static void check_kvm_target_cpu(void *ret)
1346 {
1347         *(int *)ret = kvm_target_cpu();
1348 }
1349
1350 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1351 {
1352         struct kvm_vcpu *vcpu;
1353         int i;
1354
1355         mpidr &= MPIDR_HWID_BITMASK;
1356         kvm_for_each_vcpu(i, vcpu, kvm) {
1357                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1358                         return vcpu;
1359         }
1360         return NULL;
1361 }
1362
1363 /**
1364  * Initialize Hyp-mode and memory mappings on all CPUs.
1365  */
1366 int kvm_arch_init(void *opaque)
1367 {
1368         int err;
1369         int ret, cpu;
1370
1371         if (!is_hyp_mode_available()) {
1372                 kvm_err("HYP mode not available\n");
1373                 return -ENODEV;
1374         }
1375
1376         for_each_online_cpu(cpu) {
1377                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1378                 if (ret < 0) {
1379                         kvm_err("Error, CPU %d not supported!\n", cpu);
1380                         return -ENODEV;
1381                 }
1382         }
1383
1384         err = init_common_resources();
1385         if (err)
1386                 return err;
1387
1388         if (is_kernel_in_hyp_mode())
1389                 err = init_vhe_mode();
1390         else
1391                 err = init_hyp_mode();
1392         if (err)
1393                 goto out_err;
1394
1395         err = init_subsystems();
1396         if (err)
1397                 goto out_hyp;
1398
1399         return 0;
1400
1401 out_hyp:
1402         teardown_hyp_mode();
1403 out_err:
1404         teardown_common_resources();
1405         return err;
1406 }
1407
1408 /* NOP: Compiling as a module not supported */
1409 void kvm_arch_exit(void)
1410 {
1411         kvm_perf_teardown();
1412 }
1413
1414 static int arm_init(void)
1415 {
1416         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1417         return rc;
1418 }
1419
1420 module_init(arm_init);