Merge tag 'ceph-for-4.9-rc1' of git://github.com/ceph/ceph-client
[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 bool kvm_arch_has_vcpu_debugfs(void)
148 {
149         return false;
150 }
151
152 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
153 {
154         return 0;
155 }
156
157 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
158 {
159         return VM_FAULT_SIGBUS;
160 }
161
162
163 /**
164  * kvm_arch_destroy_vm - destroy the VM data structure
165  * @kvm:        pointer to the KVM struct
166  */
167 void kvm_arch_destroy_vm(struct kvm *kvm)
168 {
169         int i;
170
171         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
172                 if (kvm->vcpus[i]) {
173                         kvm_arch_vcpu_free(kvm->vcpus[i]);
174                         kvm->vcpus[i] = NULL;
175                 }
176         }
177
178         kvm_vgic_destroy(kvm);
179 }
180
181 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
182 {
183         int r;
184         switch (ext) {
185         case KVM_CAP_IRQCHIP:
186                 r = vgic_present;
187                 break;
188         case KVM_CAP_IOEVENTFD:
189         case KVM_CAP_DEVICE_CTRL:
190         case KVM_CAP_USER_MEMORY:
191         case KVM_CAP_SYNC_MMU:
192         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
193         case KVM_CAP_ONE_REG:
194         case KVM_CAP_ARM_PSCI:
195         case KVM_CAP_ARM_PSCI_0_2:
196         case KVM_CAP_READONLY_MEM:
197         case KVM_CAP_MP_STATE:
198                 r = 1;
199                 break;
200         case KVM_CAP_COALESCED_MMIO:
201                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
202                 break;
203         case KVM_CAP_ARM_SET_DEVICE_ADDR:
204                 r = 1;
205                 break;
206         case KVM_CAP_NR_VCPUS:
207                 r = num_online_cpus();
208                 break;
209         case KVM_CAP_MAX_VCPUS:
210                 r = KVM_MAX_VCPUS;
211                 break;
212         default:
213                 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
214                 break;
215         }
216         return r;
217 }
218
219 long kvm_arch_dev_ioctl(struct file *filp,
220                         unsigned int ioctl, unsigned long arg)
221 {
222         return -EINVAL;
223 }
224
225
226 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
227 {
228         int err;
229         struct kvm_vcpu *vcpu;
230
231         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
232                 err = -EBUSY;
233                 goto out;
234         }
235
236         if (id >= kvm->arch.max_vcpus) {
237                 err = -EINVAL;
238                 goto out;
239         }
240
241         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
242         if (!vcpu) {
243                 err = -ENOMEM;
244                 goto out;
245         }
246
247         err = kvm_vcpu_init(vcpu, kvm, id);
248         if (err)
249                 goto free_vcpu;
250
251         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
252         if (err)
253                 goto vcpu_uninit;
254
255         return vcpu;
256 vcpu_uninit:
257         kvm_vcpu_uninit(vcpu);
258 free_vcpu:
259         kmem_cache_free(kvm_vcpu_cache, vcpu);
260 out:
261         return ERR_PTR(err);
262 }
263
264 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
265 {
266         kvm_vgic_vcpu_early_init(vcpu);
267 }
268
269 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
270 {
271         kvm_mmu_free_memory_caches(vcpu);
272         kvm_timer_vcpu_terminate(vcpu);
273         kvm_vgic_vcpu_destroy(vcpu);
274         kvm_pmu_vcpu_destroy(vcpu);
275         kvm_vcpu_uninit(vcpu);
276         kmem_cache_free(kvm_vcpu_cache, vcpu);
277 }
278
279 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
280 {
281         kvm_arch_vcpu_free(vcpu);
282 }
283
284 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
285 {
286         return kvm_timer_should_fire(vcpu);
287 }
288
289 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
290 {
291         kvm_timer_schedule(vcpu);
292 }
293
294 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
295 {
296         kvm_timer_unschedule(vcpu);
297 }
298
299 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
300 {
301         /* Force users to call KVM_ARM_VCPU_INIT */
302         vcpu->arch.target = -1;
303         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
304
305         /* Set up the timer */
306         kvm_timer_vcpu_init(vcpu);
307
308         kvm_arm_reset_debug_ptr(vcpu);
309
310         return 0;
311 }
312
313 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
314 {
315         vcpu->cpu = cpu;
316         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
317
318         kvm_arm_set_running_vcpu(vcpu);
319 }
320
321 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
322 {
323         /*
324          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
325          * if the vcpu is no longer assigned to a cpu.  This is used for the
326          * optimized make_all_cpus_request path.
327          */
328         vcpu->cpu = -1;
329
330         kvm_arm_set_running_vcpu(NULL);
331         kvm_timer_vcpu_put(vcpu);
332 }
333
334 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
335                                     struct kvm_mp_state *mp_state)
336 {
337         if (vcpu->arch.power_off)
338                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
339         else
340                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
341
342         return 0;
343 }
344
345 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
346                                     struct kvm_mp_state *mp_state)
347 {
348         switch (mp_state->mp_state) {
349         case KVM_MP_STATE_RUNNABLE:
350                 vcpu->arch.power_off = false;
351                 break;
352         case KVM_MP_STATE_STOPPED:
353                 vcpu->arch.power_off = true;
354                 break;
355         default:
356                 return -EINVAL;
357         }
358
359         return 0;
360 }
361
362 /**
363  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
364  * @v:          The VCPU pointer
365  *
366  * If the guest CPU is not waiting for interrupts or an interrupt line is
367  * asserted, the CPU is by definition runnable.
368  */
369 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
370 {
371         return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
372                 && !v->arch.power_off && !v->arch.pause);
373 }
374
375 /* Just ensure a guest exit from a particular CPU */
376 static void exit_vm_noop(void *info)
377 {
378 }
379
380 void force_vm_exit(const cpumask_t *mask)
381 {
382         preempt_disable();
383         smp_call_function_many(mask, exit_vm_noop, NULL, true);
384         preempt_enable();
385 }
386
387 /**
388  * need_new_vmid_gen - check that the VMID is still valid
389  * @kvm: The VM's VMID to check
390  *
391  * return true if there is a new generation of VMIDs being used
392  *
393  * The hardware supports only 256 values with the value zero reserved for the
394  * host, so we check if an assigned value belongs to a previous generation,
395  * which which requires us to assign a new value. If we're the first to use a
396  * VMID for the new generation, we must flush necessary caches and TLBs on all
397  * CPUs.
398  */
399 static bool need_new_vmid_gen(struct kvm *kvm)
400 {
401         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
402 }
403
404 /**
405  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
406  * @kvm The guest that we are about to run
407  *
408  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
409  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
410  * caches and TLBs.
411  */
412 static void update_vttbr(struct kvm *kvm)
413 {
414         phys_addr_t pgd_phys;
415         u64 vmid;
416
417         if (!need_new_vmid_gen(kvm))
418                 return;
419
420         spin_lock(&kvm_vmid_lock);
421
422         /*
423          * We need to re-check the vmid_gen here to ensure that if another vcpu
424          * already allocated a valid vmid for this vm, then this vcpu should
425          * use the same vmid.
426          */
427         if (!need_new_vmid_gen(kvm)) {
428                 spin_unlock(&kvm_vmid_lock);
429                 return;
430         }
431
432         /* First user of a new VMID generation? */
433         if (unlikely(kvm_next_vmid == 0)) {
434                 atomic64_inc(&kvm_vmid_gen);
435                 kvm_next_vmid = 1;
436
437                 /*
438                  * On SMP we know no other CPUs can use this CPU's or each
439                  * other's VMID after force_vm_exit returns since the
440                  * kvm_vmid_lock blocks them from reentry to the guest.
441                  */
442                 force_vm_exit(cpu_all_mask);
443                 /*
444                  * Now broadcast TLB + ICACHE invalidation over the inner
445                  * shareable domain to make sure all data structures are
446                  * clean.
447                  */
448                 kvm_call_hyp(__kvm_flush_vm_context);
449         }
450
451         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
452         kvm->arch.vmid = kvm_next_vmid;
453         kvm_next_vmid++;
454         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
455
456         /* update vttbr to be used with the new vmid */
457         pgd_phys = virt_to_phys(kvm->arch.pgd);
458         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
459         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
460         kvm->arch.vttbr = pgd_phys | vmid;
461
462         spin_unlock(&kvm_vmid_lock);
463 }
464
465 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
466 {
467         struct kvm *kvm = vcpu->kvm;
468         int ret = 0;
469
470         if (likely(vcpu->arch.has_run_once))
471                 return 0;
472
473         vcpu->arch.has_run_once = true;
474
475         /*
476          * Map the VGIC hardware resources before running a vcpu the first
477          * time on this VM.
478          */
479         if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
480                 ret = kvm_vgic_map_resources(kvm);
481                 if (ret)
482                         return ret;
483         }
484
485         /*
486          * Enable the arch timers only if we have an in-kernel VGIC
487          * and it has been properly initialized, since we cannot handle
488          * interrupts from the virtual timer with a userspace gic.
489          */
490         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
491                 ret = kvm_timer_enable(vcpu);
492
493         return ret;
494 }
495
496 bool kvm_arch_intc_initialized(struct kvm *kvm)
497 {
498         return vgic_initialized(kvm);
499 }
500
501 void kvm_arm_halt_guest(struct kvm *kvm)
502 {
503         int i;
504         struct kvm_vcpu *vcpu;
505
506         kvm_for_each_vcpu(i, vcpu, kvm)
507                 vcpu->arch.pause = true;
508         kvm_make_all_cpus_request(kvm, KVM_REQ_VCPU_EXIT);
509 }
510
511 void kvm_arm_halt_vcpu(struct kvm_vcpu *vcpu)
512 {
513         vcpu->arch.pause = true;
514         kvm_vcpu_kick(vcpu);
515 }
516
517 void kvm_arm_resume_vcpu(struct kvm_vcpu *vcpu)
518 {
519         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
520
521         vcpu->arch.pause = false;
522         swake_up(wq);
523 }
524
525 void kvm_arm_resume_guest(struct kvm *kvm)
526 {
527         int i;
528         struct kvm_vcpu *vcpu;
529
530         kvm_for_each_vcpu(i, vcpu, kvm)
531                 kvm_arm_resume_vcpu(vcpu);
532 }
533
534 static void vcpu_sleep(struct kvm_vcpu *vcpu)
535 {
536         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
537
538         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
539                                        (!vcpu->arch.pause)));
540 }
541
542 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
543 {
544         return vcpu->arch.target >= 0;
545 }
546
547 /**
548  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
549  * @vcpu:       The VCPU pointer
550  * @run:        The kvm_run structure pointer used for userspace state exchange
551  *
552  * This function is called through the VCPU_RUN ioctl called from user space. It
553  * will execute VM code in a loop until the time slice for the process is used
554  * or some emulation is needed from user space in which case the function will
555  * return with return value 0 and with the kvm_run structure filled in with the
556  * required data for the requested emulation.
557  */
558 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
559 {
560         int ret;
561         sigset_t sigsaved;
562
563         if (unlikely(!kvm_vcpu_initialized(vcpu)))
564                 return -ENOEXEC;
565
566         ret = kvm_vcpu_first_run_init(vcpu);
567         if (ret)
568                 return ret;
569
570         if (run->exit_reason == KVM_EXIT_MMIO) {
571                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
572                 if (ret)
573                         return ret;
574         }
575
576         if (vcpu->sigset_active)
577                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
578
579         ret = 1;
580         run->exit_reason = KVM_EXIT_UNKNOWN;
581         while (ret > 0) {
582                 /*
583                  * Check conditions before entering the guest
584                  */
585                 cond_resched();
586
587                 update_vttbr(vcpu->kvm);
588
589                 if (vcpu->arch.power_off || vcpu->arch.pause)
590                         vcpu_sleep(vcpu);
591
592                 /*
593                  * Preparing the interrupts to be injected also
594                  * involves poking the GIC, which must be done in a
595                  * non-preemptible context.
596                  */
597                 preempt_disable();
598                 kvm_pmu_flush_hwstate(vcpu);
599                 kvm_timer_flush_hwstate(vcpu);
600                 kvm_vgic_flush_hwstate(vcpu);
601
602                 local_irq_disable();
603
604                 /*
605                  * Re-check atomic conditions
606                  */
607                 if (signal_pending(current)) {
608                         ret = -EINTR;
609                         run->exit_reason = KVM_EXIT_INTR;
610                 }
611
612                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
613                         vcpu->arch.power_off || vcpu->arch.pause) {
614                         local_irq_enable();
615                         kvm_pmu_sync_hwstate(vcpu);
616                         kvm_timer_sync_hwstate(vcpu);
617                         kvm_vgic_sync_hwstate(vcpu);
618                         preempt_enable();
619                         continue;
620                 }
621
622                 kvm_arm_setup_debug(vcpu);
623
624                 /**************************************************************
625                  * Enter the guest
626                  */
627                 trace_kvm_entry(*vcpu_pc(vcpu));
628                 guest_enter_irqoff();
629                 vcpu->mode = IN_GUEST_MODE;
630
631                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
632
633                 vcpu->mode = OUTSIDE_GUEST_MODE;
634                 vcpu->stat.exits++;
635                 /*
636                  * Back from guest
637                  *************************************************************/
638
639                 kvm_arm_clear_debug(vcpu);
640
641                 /*
642                  * We may have taken a host interrupt in HYP mode (ie
643                  * while executing the guest). This interrupt is still
644                  * pending, as we haven't serviced it yet!
645                  *
646                  * We're now back in SVC mode, with interrupts
647                  * disabled.  Enabling the interrupts now will have
648                  * the effect of taking the interrupt again, in SVC
649                  * mode this time.
650                  */
651                 local_irq_enable();
652
653                 /*
654                  * We do local_irq_enable() before calling guest_exit() so
655                  * that if a timer interrupt hits while running the guest we
656                  * account that tick as being spent in the guest.  We enable
657                  * preemption after calling guest_exit() so that if we get
658                  * preempted we make sure ticks after that is not counted as
659                  * guest time.
660                  */
661                 guest_exit();
662                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
663
664                 /*
665                  * We must sync the PMU and timer state before the vgic state so
666                  * that the vgic can properly sample the updated state of the
667                  * interrupt line.
668                  */
669                 kvm_pmu_sync_hwstate(vcpu);
670                 kvm_timer_sync_hwstate(vcpu);
671
672                 kvm_vgic_sync_hwstate(vcpu);
673
674                 preempt_enable();
675
676                 ret = handle_exit(vcpu, run, ret);
677         }
678
679         if (vcpu->sigset_active)
680                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
681         return ret;
682 }
683
684 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
685 {
686         int bit_index;
687         bool set;
688         unsigned long *ptr;
689
690         if (number == KVM_ARM_IRQ_CPU_IRQ)
691                 bit_index = __ffs(HCR_VI);
692         else /* KVM_ARM_IRQ_CPU_FIQ */
693                 bit_index = __ffs(HCR_VF);
694
695         ptr = (unsigned long *)&vcpu->arch.irq_lines;
696         if (level)
697                 set = test_and_set_bit(bit_index, ptr);
698         else
699                 set = test_and_clear_bit(bit_index, ptr);
700
701         /*
702          * If we didn't change anything, no need to wake up or kick other CPUs
703          */
704         if (set == level)
705                 return 0;
706
707         /*
708          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
709          * trigger a world-switch round on the running physical CPU to set the
710          * virtual IRQ/FIQ fields in the HCR appropriately.
711          */
712         kvm_vcpu_kick(vcpu);
713
714         return 0;
715 }
716
717 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
718                           bool line_status)
719 {
720         u32 irq = irq_level->irq;
721         unsigned int irq_type, vcpu_idx, irq_num;
722         int nrcpus = atomic_read(&kvm->online_vcpus);
723         struct kvm_vcpu *vcpu = NULL;
724         bool level = irq_level->level;
725
726         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
727         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
728         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
729
730         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
731
732         switch (irq_type) {
733         case KVM_ARM_IRQ_TYPE_CPU:
734                 if (irqchip_in_kernel(kvm))
735                         return -ENXIO;
736
737                 if (vcpu_idx >= nrcpus)
738                         return -EINVAL;
739
740                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
741                 if (!vcpu)
742                         return -EINVAL;
743
744                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
745                         return -EINVAL;
746
747                 return vcpu_interrupt_line(vcpu, irq_num, level);
748         case KVM_ARM_IRQ_TYPE_PPI:
749                 if (!irqchip_in_kernel(kvm))
750                         return -ENXIO;
751
752                 if (vcpu_idx >= nrcpus)
753                         return -EINVAL;
754
755                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
756                 if (!vcpu)
757                         return -EINVAL;
758
759                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
760                         return -EINVAL;
761
762                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
763         case KVM_ARM_IRQ_TYPE_SPI:
764                 if (!irqchip_in_kernel(kvm))
765                         return -ENXIO;
766
767                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
768                         return -EINVAL;
769
770                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
771         }
772
773         return -EINVAL;
774 }
775
776 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
777                                const struct kvm_vcpu_init *init)
778 {
779         unsigned int i;
780         int phys_target = kvm_target_cpu();
781
782         if (init->target != phys_target)
783                 return -EINVAL;
784
785         /*
786          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
787          * use the same target.
788          */
789         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
790                 return -EINVAL;
791
792         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
793         for (i = 0; i < sizeof(init->features) * 8; i++) {
794                 bool set = (init->features[i / 32] & (1 << (i % 32)));
795
796                 if (set && i >= KVM_VCPU_MAX_FEATURES)
797                         return -ENOENT;
798
799                 /*
800                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
801                  * use the same feature set.
802                  */
803                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
804                     test_bit(i, vcpu->arch.features) != set)
805                         return -EINVAL;
806
807                 if (set)
808                         set_bit(i, vcpu->arch.features);
809         }
810
811         vcpu->arch.target = phys_target;
812
813         /* Now we know what it is, we can reset it. */
814         return kvm_reset_vcpu(vcpu);
815 }
816
817
818 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
819                                          struct kvm_vcpu_init *init)
820 {
821         int ret;
822
823         ret = kvm_vcpu_set_target(vcpu, init);
824         if (ret)
825                 return ret;
826
827         /*
828          * Ensure a rebooted VM will fault in RAM pages and detect if the
829          * guest MMU is turned off and flush the caches as needed.
830          */
831         if (vcpu->arch.has_run_once)
832                 stage2_unmap_vm(vcpu->kvm);
833
834         vcpu_reset_hcr(vcpu);
835
836         /*
837          * Handle the "start in power-off" case.
838          */
839         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
840                 vcpu->arch.power_off = true;
841         else
842                 vcpu->arch.power_off = false;
843
844         return 0;
845 }
846
847 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
848                                  struct kvm_device_attr *attr)
849 {
850         int ret = -ENXIO;
851
852         switch (attr->group) {
853         default:
854                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
855                 break;
856         }
857
858         return ret;
859 }
860
861 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
862                                  struct kvm_device_attr *attr)
863 {
864         int ret = -ENXIO;
865
866         switch (attr->group) {
867         default:
868                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
869                 break;
870         }
871
872         return ret;
873 }
874
875 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
876                                  struct kvm_device_attr *attr)
877 {
878         int ret = -ENXIO;
879
880         switch (attr->group) {
881         default:
882                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
883                 break;
884         }
885
886         return ret;
887 }
888
889 long kvm_arch_vcpu_ioctl(struct file *filp,
890                          unsigned int ioctl, unsigned long arg)
891 {
892         struct kvm_vcpu *vcpu = filp->private_data;
893         void __user *argp = (void __user *)arg;
894         struct kvm_device_attr attr;
895
896         switch (ioctl) {
897         case KVM_ARM_VCPU_INIT: {
898                 struct kvm_vcpu_init init;
899
900                 if (copy_from_user(&init, argp, sizeof(init)))
901                         return -EFAULT;
902
903                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
904         }
905         case KVM_SET_ONE_REG:
906         case KVM_GET_ONE_REG: {
907                 struct kvm_one_reg reg;
908
909                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
910                         return -ENOEXEC;
911
912                 if (copy_from_user(&reg, argp, sizeof(reg)))
913                         return -EFAULT;
914                 if (ioctl == KVM_SET_ONE_REG)
915                         return kvm_arm_set_reg(vcpu, &reg);
916                 else
917                         return kvm_arm_get_reg(vcpu, &reg);
918         }
919         case KVM_GET_REG_LIST: {
920                 struct kvm_reg_list __user *user_list = argp;
921                 struct kvm_reg_list reg_list;
922                 unsigned n;
923
924                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
925                         return -ENOEXEC;
926
927                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
928                         return -EFAULT;
929                 n = reg_list.n;
930                 reg_list.n = kvm_arm_num_regs(vcpu);
931                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
932                         return -EFAULT;
933                 if (n < reg_list.n)
934                         return -E2BIG;
935                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
936         }
937         case KVM_SET_DEVICE_ATTR: {
938                 if (copy_from_user(&attr, argp, sizeof(attr)))
939                         return -EFAULT;
940                 return kvm_arm_vcpu_set_attr(vcpu, &attr);
941         }
942         case KVM_GET_DEVICE_ATTR: {
943                 if (copy_from_user(&attr, argp, sizeof(attr)))
944                         return -EFAULT;
945                 return kvm_arm_vcpu_get_attr(vcpu, &attr);
946         }
947         case KVM_HAS_DEVICE_ATTR: {
948                 if (copy_from_user(&attr, argp, sizeof(attr)))
949                         return -EFAULT;
950                 return kvm_arm_vcpu_has_attr(vcpu, &attr);
951         }
952         default:
953                 return -EINVAL;
954         }
955 }
956
957 /**
958  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
959  * @kvm: kvm instance
960  * @log: slot id and address to which we copy the log
961  *
962  * Steps 1-4 below provide general overview of dirty page logging. See
963  * kvm_get_dirty_log_protect() function description for additional details.
964  *
965  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
966  * always flush the TLB (step 4) even if previous step failed  and the dirty
967  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
968  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
969  * writes will be marked dirty for next log read.
970  *
971  *   1. Take a snapshot of the bit and clear it if needed.
972  *   2. Write protect the corresponding page.
973  *   3. Copy the snapshot to the userspace.
974  *   4. Flush TLB's if needed.
975  */
976 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
977 {
978         bool is_dirty = false;
979         int r;
980
981         mutex_lock(&kvm->slots_lock);
982
983         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
984
985         if (is_dirty)
986                 kvm_flush_remote_tlbs(kvm);
987
988         mutex_unlock(&kvm->slots_lock);
989         return r;
990 }
991
992 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
993                                         struct kvm_arm_device_addr *dev_addr)
994 {
995         unsigned long dev_id, type;
996
997         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
998                 KVM_ARM_DEVICE_ID_SHIFT;
999         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1000                 KVM_ARM_DEVICE_TYPE_SHIFT;
1001
1002         switch (dev_id) {
1003         case KVM_ARM_DEVICE_VGIC_V2:
1004                 if (!vgic_present)
1005                         return -ENXIO;
1006                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1007         default:
1008                 return -ENODEV;
1009         }
1010 }
1011
1012 long kvm_arch_vm_ioctl(struct file *filp,
1013                        unsigned int ioctl, unsigned long arg)
1014 {
1015         struct kvm *kvm = filp->private_data;
1016         void __user *argp = (void __user *)arg;
1017
1018         switch (ioctl) {
1019         case KVM_CREATE_IRQCHIP: {
1020                 int ret;
1021                 if (!vgic_present)
1022                         return -ENXIO;
1023                 mutex_lock(&kvm->lock);
1024                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1025                 mutex_unlock(&kvm->lock);
1026                 return ret;
1027         }
1028         case KVM_ARM_SET_DEVICE_ADDR: {
1029                 struct kvm_arm_device_addr dev_addr;
1030
1031                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1032                         return -EFAULT;
1033                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1034         }
1035         case KVM_ARM_PREFERRED_TARGET: {
1036                 int err;
1037                 struct kvm_vcpu_init init;
1038
1039                 err = kvm_vcpu_preferred_target(&init);
1040                 if (err)
1041                         return err;
1042
1043                 if (copy_to_user(argp, &init, sizeof(init)))
1044                         return -EFAULT;
1045
1046                 return 0;
1047         }
1048         default:
1049                 return -EINVAL;
1050         }
1051 }
1052
1053 static void cpu_init_hyp_mode(void *dummy)
1054 {
1055         phys_addr_t pgd_ptr;
1056         unsigned long hyp_stack_ptr;
1057         unsigned long stack_page;
1058         unsigned long vector_ptr;
1059
1060         /* Switch from the HYP stub to our own HYP init vector */
1061         __hyp_set_vectors(kvm_get_idmap_vector());
1062
1063         pgd_ptr = kvm_mmu_get_httbr();
1064         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1065         hyp_stack_ptr = stack_page + PAGE_SIZE;
1066         vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1067
1068         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1069         __cpu_init_stage2();
1070
1071         kvm_arm_init_debug();
1072 }
1073
1074 static void cpu_hyp_reinit(void)
1075 {
1076         if (is_kernel_in_hyp_mode()) {
1077                 /*
1078                  * __cpu_init_stage2() is safe to call even if the PM
1079                  * event was cancelled before the CPU was reset.
1080                  */
1081                 __cpu_init_stage2();
1082         } else {
1083                 if (__hyp_get_vectors() == hyp_default_vectors)
1084                         cpu_init_hyp_mode(NULL);
1085         }
1086 }
1087
1088 static void cpu_hyp_reset(void)
1089 {
1090         if (!is_kernel_in_hyp_mode())
1091                 __cpu_reset_hyp_mode(hyp_default_vectors,
1092                                      kvm_get_idmap_start());
1093 }
1094
1095 static void _kvm_arch_hardware_enable(void *discard)
1096 {
1097         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1098                 cpu_hyp_reinit();
1099                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1100         }
1101 }
1102
1103 int kvm_arch_hardware_enable(void)
1104 {
1105         _kvm_arch_hardware_enable(NULL);
1106         return 0;
1107 }
1108
1109 static void _kvm_arch_hardware_disable(void *discard)
1110 {
1111         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1112                 cpu_hyp_reset();
1113                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1114         }
1115 }
1116
1117 void kvm_arch_hardware_disable(void)
1118 {
1119         _kvm_arch_hardware_disable(NULL);
1120 }
1121
1122 #ifdef CONFIG_CPU_PM
1123 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1124                                     unsigned long cmd,
1125                                     void *v)
1126 {
1127         /*
1128          * kvm_arm_hardware_enabled is left with its old value over
1129          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1130          * re-enable hyp.
1131          */
1132         switch (cmd) {
1133         case CPU_PM_ENTER:
1134                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1135                         /*
1136                          * don't update kvm_arm_hardware_enabled here
1137                          * so that the hardware will be re-enabled
1138                          * when we resume. See below.
1139                          */
1140                         cpu_hyp_reset();
1141
1142                 return NOTIFY_OK;
1143         case CPU_PM_EXIT:
1144                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1145                         /* The hardware was enabled before suspend. */
1146                         cpu_hyp_reinit();
1147
1148                 return NOTIFY_OK;
1149
1150         default:
1151                 return NOTIFY_DONE;
1152         }
1153 }
1154
1155 static struct notifier_block hyp_init_cpu_pm_nb = {
1156         .notifier_call = hyp_init_cpu_pm_notifier,
1157 };
1158
1159 static void __init hyp_cpu_pm_init(void)
1160 {
1161         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1162 }
1163 static void __init hyp_cpu_pm_exit(void)
1164 {
1165         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1166 }
1167 #else
1168 static inline void hyp_cpu_pm_init(void)
1169 {
1170 }
1171 static inline void hyp_cpu_pm_exit(void)
1172 {
1173 }
1174 #endif
1175
1176 static void teardown_common_resources(void)
1177 {
1178         free_percpu(kvm_host_cpu_state);
1179 }
1180
1181 static int init_common_resources(void)
1182 {
1183         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1184         if (!kvm_host_cpu_state) {
1185                 kvm_err("Cannot allocate host CPU state\n");
1186                 return -ENOMEM;
1187         }
1188
1189         /* set size of VMID supported by CPU */
1190         kvm_vmid_bits = kvm_get_vmid_bits();
1191         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1192
1193         return 0;
1194 }
1195
1196 static int init_subsystems(void)
1197 {
1198         int err = 0;
1199
1200         /*
1201          * Enable hardware so that subsystem initialisation can access EL2.
1202          */
1203         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1204
1205         /*
1206          * Register CPU lower-power notifier
1207          */
1208         hyp_cpu_pm_init();
1209
1210         /*
1211          * Init HYP view of VGIC
1212          */
1213         err = kvm_vgic_hyp_init();
1214         switch (err) {
1215         case 0:
1216                 vgic_present = true;
1217                 break;
1218         case -ENODEV:
1219         case -ENXIO:
1220                 vgic_present = false;
1221                 err = 0;
1222                 break;
1223         default:
1224                 goto out;
1225         }
1226
1227         /*
1228          * Init HYP architected timer support
1229          */
1230         err = kvm_timer_hyp_init();
1231         if (err)
1232                 goto out;
1233
1234         kvm_perf_init();
1235         kvm_coproc_table_init();
1236
1237 out:
1238         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1239
1240         return err;
1241 }
1242
1243 static void teardown_hyp_mode(void)
1244 {
1245         int cpu;
1246
1247         if (is_kernel_in_hyp_mode())
1248                 return;
1249
1250         free_hyp_pgds();
1251         for_each_possible_cpu(cpu)
1252                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1253         hyp_cpu_pm_exit();
1254 }
1255
1256 static int init_vhe_mode(void)
1257 {
1258         kvm_info("VHE mode initialized successfully\n");
1259         return 0;
1260 }
1261
1262 /**
1263  * Inits Hyp-mode on all online CPUs
1264  */
1265 static int init_hyp_mode(void)
1266 {
1267         int cpu;
1268         int err = 0;
1269
1270         /*
1271          * Allocate Hyp PGD and setup Hyp identity mapping
1272          */
1273         err = kvm_mmu_init();
1274         if (err)
1275                 goto out_err;
1276
1277         /*
1278          * It is probably enough to obtain the default on one
1279          * CPU. It's unlikely to be different on the others.
1280          */
1281         hyp_default_vectors = __hyp_get_vectors();
1282
1283         /*
1284          * Allocate stack pages for Hypervisor-mode
1285          */
1286         for_each_possible_cpu(cpu) {
1287                 unsigned long stack_page;
1288
1289                 stack_page = __get_free_page(GFP_KERNEL);
1290                 if (!stack_page) {
1291                         err = -ENOMEM;
1292                         goto out_err;
1293                 }
1294
1295                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1296         }
1297
1298         /*
1299          * Map the Hyp-code called directly from the host
1300          */
1301         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1302                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1303         if (err) {
1304                 kvm_err("Cannot map world-switch code\n");
1305                 goto out_err;
1306         }
1307
1308         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1309                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1310         if (err) {
1311                 kvm_err("Cannot map rodata section\n");
1312                 goto out_err;
1313         }
1314
1315         /*
1316          * Map the Hyp stack pages
1317          */
1318         for_each_possible_cpu(cpu) {
1319                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1320                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1321                                           PAGE_HYP);
1322
1323                 if (err) {
1324                         kvm_err("Cannot map hyp stack\n");
1325                         goto out_err;
1326                 }
1327         }
1328
1329         for_each_possible_cpu(cpu) {
1330                 kvm_cpu_context_t *cpu_ctxt;
1331
1332                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1333                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1334
1335                 if (err) {
1336                         kvm_err("Cannot map host CPU state: %d\n", err);
1337                         goto out_err;
1338                 }
1339         }
1340
1341         kvm_info("Hyp mode initialized successfully\n");
1342
1343         return 0;
1344
1345 out_err:
1346         teardown_hyp_mode();
1347         kvm_err("error initializing Hyp mode: %d\n", err);
1348         return err;
1349 }
1350
1351 static void check_kvm_target_cpu(void *ret)
1352 {
1353         *(int *)ret = kvm_target_cpu();
1354 }
1355
1356 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1357 {
1358         struct kvm_vcpu *vcpu;
1359         int i;
1360
1361         mpidr &= MPIDR_HWID_BITMASK;
1362         kvm_for_each_vcpu(i, vcpu, kvm) {
1363                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1364                         return vcpu;
1365         }
1366         return NULL;
1367 }
1368
1369 /**
1370  * Initialize Hyp-mode and memory mappings on all CPUs.
1371  */
1372 int kvm_arch_init(void *opaque)
1373 {
1374         int err;
1375         int ret, cpu;
1376
1377         if (!is_hyp_mode_available()) {
1378                 kvm_err("HYP mode not available\n");
1379                 return -ENODEV;
1380         }
1381
1382         for_each_online_cpu(cpu) {
1383                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1384                 if (ret < 0) {
1385                         kvm_err("Error, CPU %d not supported!\n", cpu);
1386                         return -ENODEV;
1387                 }
1388         }
1389
1390         err = init_common_resources();
1391         if (err)
1392                 return err;
1393
1394         if (is_kernel_in_hyp_mode())
1395                 err = init_vhe_mode();
1396         else
1397                 err = init_hyp_mode();
1398         if (err)
1399                 goto out_err;
1400
1401         err = init_subsystems();
1402         if (err)
1403                 goto out_hyp;
1404
1405         return 0;
1406
1407 out_hyp:
1408         teardown_hyp_mode();
1409 out_err:
1410         teardown_common_resources();
1411         return err;
1412 }
1413
1414 /* NOP: Compiling as a module not supported */
1415 void kvm_arch_exit(void)
1416 {
1417         kvm_perf_teardown();
1418 }
1419
1420 static int arm_init(void)
1421 {
1422         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1423         return rc;
1424 }
1425
1426 module_init(arm_init);