Loading...
Note: File does not exist in v3.1.
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 <linux/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
54static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
55static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
56static unsigned long hyp_default_vectors;
57
58/* Per-CPU variable containing the currently running vcpu. */
59static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
60
61/* The VMID used in the VTTBR */
62static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
63static u32 kvm_next_vmid;
64static unsigned int kvm_vmid_bits __read_mostly;
65static DEFINE_SPINLOCK(kvm_vmid_lock);
66
67static bool vgic_present;
68
69static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
70
71static 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 */
81struct 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 */
90struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
91{
92 return &kvm_arm_running_vcpu;
93}
94
95int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
96{
97 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
98}
99
100int kvm_arch_hardware_setup(void)
101{
102 return 0;
103}
104
105void 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 */
115int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
116{
117 int ret, cpu;
118
119 if (type)
120 return -EINVAL;
121
122 kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
123 if (!kvm->arch.last_vcpu_ran)
124 return -ENOMEM;
125
126 for_each_possible_cpu(cpu)
127 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
128
129 ret = kvm_alloc_stage2_pgd(kvm);
130 if (ret)
131 goto out_fail_alloc;
132
133 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
134 if (ret)
135 goto out_free_stage2_pgd;
136
137 kvm_vgic_early_init(kvm);
138 kvm_timer_init(kvm);
139
140 /* Mark the initial VMID generation invalid */
141 kvm->arch.vmid_gen = 0;
142
143 /* The maximum number of VCPUs is limited by the host's GIC model */
144 kvm->arch.max_vcpus = vgic_present ?
145 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
146
147 return ret;
148out_free_stage2_pgd:
149 kvm_free_stage2_pgd(kvm);
150out_fail_alloc:
151 free_percpu(kvm->arch.last_vcpu_ran);
152 kvm->arch.last_vcpu_ran = NULL;
153 return ret;
154}
155
156bool kvm_arch_has_vcpu_debugfs(void)
157{
158 return false;
159}
160
161int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
162{
163 return 0;
164}
165
166int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
167{
168 return VM_FAULT_SIGBUS;
169}
170
171
172/**
173 * kvm_arch_destroy_vm - destroy the VM data structure
174 * @kvm: pointer to the KVM struct
175 */
176void kvm_arch_destroy_vm(struct kvm *kvm)
177{
178 int i;
179
180 free_percpu(kvm->arch.last_vcpu_ran);
181 kvm->arch.last_vcpu_ran = NULL;
182
183 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
184 if (kvm->vcpus[i]) {
185 kvm_arch_vcpu_free(kvm->vcpus[i]);
186 kvm->vcpus[i] = NULL;
187 }
188 }
189
190 kvm_vgic_destroy(kvm);
191}
192
193int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
194{
195 int r;
196 switch (ext) {
197 case KVM_CAP_IRQCHIP:
198 r = vgic_present;
199 break;
200 case KVM_CAP_IOEVENTFD:
201 case KVM_CAP_DEVICE_CTRL:
202 case KVM_CAP_USER_MEMORY:
203 case KVM_CAP_SYNC_MMU:
204 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
205 case KVM_CAP_ONE_REG:
206 case KVM_CAP_ARM_PSCI:
207 case KVM_CAP_ARM_PSCI_0_2:
208 case KVM_CAP_READONLY_MEM:
209 case KVM_CAP_MP_STATE:
210 r = 1;
211 break;
212 case KVM_CAP_COALESCED_MMIO:
213 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
214 break;
215 case KVM_CAP_ARM_SET_DEVICE_ADDR:
216 r = 1;
217 break;
218 case KVM_CAP_NR_VCPUS:
219 r = num_online_cpus();
220 break;
221 case KVM_CAP_MAX_VCPUS:
222 r = KVM_MAX_VCPUS;
223 break;
224 case KVM_CAP_MSI_DEVID:
225 if (!kvm)
226 r = -EINVAL;
227 else
228 r = kvm->arch.vgic.msis_require_devid;
229 break;
230 default:
231 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
232 break;
233 }
234 return r;
235}
236
237long kvm_arch_dev_ioctl(struct file *filp,
238 unsigned int ioctl, unsigned long arg)
239{
240 return -EINVAL;
241}
242
243
244struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
245{
246 int err;
247 struct kvm_vcpu *vcpu;
248
249 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
250 err = -EBUSY;
251 goto out;
252 }
253
254 if (id >= kvm->arch.max_vcpus) {
255 err = -EINVAL;
256 goto out;
257 }
258
259 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
260 if (!vcpu) {
261 err = -ENOMEM;
262 goto out;
263 }
264
265 err = kvm_vcpu_init(vcpu, kvm, id);
266 if (err)
267 goto free_vcpu;
268
269 err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
270 if (err)
271 goto vcpu_uninit;
272
273 return vcpu;
274vcpu_uninit:
275 kvm_vcpu_uninit(vcpu);
276free_vcpu:
277 kmem_cache_free(kvm_vcpu_cache, vcpu);
278out:
279 return ERR_PTR(err);
280}
281
282void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
283{
284 kvm_vgic_vcpu_early_init(vcpu);
285}
286
287void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
288{
289 kvm_mmu_free_memory_caches(vcpu);
290 kvm_timer_vcpu_terminate(vcpu);
291 kvm_vgic_vcpu_destroy(vcpu);
292 kvm_pmu_vcpu_destroy(vcpu);
293 kvm_vcpu_uninit(vcpu);
294 kmem_cache_free(kvm_vcpu_cache, vcpu);
295}
296
297void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
298{
299 kvm_arch_vcpu_free(vcpu);
300}
301
302int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
303{
304 return kvm_timer_should_fire(vcpu);
305}
306
307void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
308{
309 kvm_timer_schedule(vcpu);
310}
311
312void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
313{
314 kvm_timer_unschedule(vcpu);
315}
316
317int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
318{
319 /* Force users to call KVM_ARM_VCPU_INIT */
320 vcpu->arch.target = -1;
321 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
322
323 /* Set up the timer */
324 kvm_timer_vcpu_init(vcpu);
325
326 kvm_arm_reset_debug_ptr(vcpu);
327
328 return 0;
329}
330
331void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
332{
333 int *last_ran;
334
335 last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
336
337 /*
338 * We might get preempted before the vCPU actually runs, but
339 * over-invalidation doesn't affect correctness.
340 */
341 if (*last_ran != vcpu->vcpu_id) {
342 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
343 *last_ran = vcpu->vcpu_id;
344 }
345
346 vcpu->cpu = cpu;
347 vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
348
349 kvm_arm_set_running_vcpu(vcpu);
350}
351
352void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
353{
354 /*
355 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
356 * if the vcpu is no longer assigned to a cpu. This is used for the
357 * optimized make_all_cpus_request path.
358 */
359 vcpu->cpu = -1;
360
361 kvm_arm_set_running_vcpu(NULL);
362 kvm_timer_vcpu_put(vcpu);
363}
364
365int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
366 struct kvm_mp_state *mp_state)
367{
368 if (vcpu->arch.power_off)
369 mp_state->mp_state = KVM_MP_STATE_STOPPED;
370 else
371 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
372
373 return 0;
374}
375
376int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
377 struct kvm_mp_state *mp_state)
378{
379 switch (mp_state->mp_state) {
380 case KVM_MP_STATE_RUNNABLE:
381 vcpu->arch.power_off = false;
382 break;
383 case KVM_MP_STATE_STOPPED:
384 vcpu->arch.power_off = true;
385 break;
386 default:
387 return -EINVAL;
388 }
389
390 return 0;
391}
392
393/**
394 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
395 * @v: The VCPU pointer
396 *
397 * If the guest CPU is not waiting for interrupts or an interrupt line is
398 * asserted, the CPU is by definition runnable.
399 */
400int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
401{
402 return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
403 && !v->arch.power_off && !v->arch.pause);
404}
405
406/* Just ensure a guest exit from a particular CPU */
407static void exit_vm_noop(void *info)
408{
409}
410
411void force_vm_exit(const cpumask_t *mask)
412{
413 preempt_disable();
414 smp_call_function_many(mask, exit_vm_noop, NULL, true);
415 preempt_enable();
416}
417
418/**
419 * need_new_vmid_gen - check that the VMID is still valid
420 * @kvm: The VM's VMID to check
421 *
422 * return true if there is a new generation of VMIDs being used
423 *
424 * The hardware supports only 256 values with the value zero reserved for the
425 * host, so we check if an assigned value belongs to a previous generation,
426 * which which requires us to assign a new value. If we're the first to use a
427 * VMID for the new generation, we must flush necessary caches and TLBs on all
428 * CPUs.
429 */
430static bool need_new_vmid_gen(struct kvm *kvm)
431{
432 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
433}
434
435/**
436 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
437 * @kvm The guest that we are about to run
438 *
439 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
440 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
441 * caches and TLBs.
442 */
443static void update_vttbr(struct kvm *kvm)
444{
445 phys_addr_t pgd_phys;
446 u64 vmid;
447
448 if (!need_new_vmid_gen(kvm))
449 return;
450
451 spin_lock(&kvm_vmid_lock);
452
453 /*
454 * We need to re-check the vmid_gen here to ensure that if another vcpu
455 * already allocated a valid vmid for this vm, then this vcpu should
456 * use the same vmid.
457 */
458 if (!need_new_vmid_gen(kvm)) {
459 spin_unlock(&kvm_vmid_lock);
460 return;
461 }
462
463 /* First user of a new VMID generation? */
464 if (unlikely(kvm_next_vmid == 0)) {
465 atomic64_inc(&kvm_vmid_gen);
466 kvm_next_vmid = 1;
467
468 /*
469 * On SMP we know no other CPUs can use this CPU's or each
470 * other's VMID after force_vm_exit returns since the
471 * kvm_vmid_lock blocks them from reentry to the guest.
472 */
473 force_vm_exit(cpu_all_mask);
474 /*
475 * Now broadcast TLB + ICACHE invalidation over the inner
476 * shareable domain to make sure all data structures are
477 * clean.
478 */
479 kvm_call_hyp(__kvm_flush_vm_context);
480 }
481
482 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
483 kvm->arch.vmid = kvm_next_vmid;
484 kvm_next_vmid++;
485 kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
486
487 /* update vttbr to be used with the new vmid */
488 pgd_phys = virt_to_phys(kvm->arch.pgd);
489 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
490 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
491 kvm->arch.vttbr = pgd_phys | vmid;
492
493 spin_unlock(&kvm_vmid_lock);
494}
495
496static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
497{
498 struct kvm *kvm = vcpu->kvm;
499 int ret = 0;
500
501 if (likely(vcpu->arch.has_run_once))
502 return 0;
503
504 vcpu->arch.has_run_once = true;
505
506 /*
507 * Map the VGIC hardware resources before running a vcpu the first
508 * time on this VM.
509 */
510 if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
511 ret = kvm_vgic_map_resources(kvm);
512 if (ret)
513 return ret;
514 }
515
516 /*
517 * Enable the arch timers only if we have an in-kernel VGIC
518 * and it has been properly initialized, since we cannot handle
519 * interrupts from the virtual timer with a userspace gic.
520 */
521 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
522 ret = kvm_timer_enable(vcpu);
523
524 return ret;
525}
526
527bool kvm_arch_intc_initialized(struct kvm *kvm)
528{
529 return vgic_initialized(kvm);
530}
531
532void kvm_arm_halt_guest(struct kvm *kvm)
533{
534 int i;
535 struct kvm_vcpu *vcpu;
536
537 kvm_for_each_vcpu(i, vcpu, kvm)
538 vcpu->arch.pause = true;
539 kvm_make_all_cpus_request(kvm, KVM_REQ_VCPU_EXIT);
540}
541
542void kvm_arm_halt_vcpu(struct kvm_vcpu *vcpu)
543{
544 vcpu->arch.pause = true;
545 kvm_vcpu_kick(vcpu);
546}
547
548void kvm_arm_resume_vcpu(struct kvm_vcpu *vcpu)
549{
550 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
551
552 vcpu->arch.pause = false;
553 swake_up(wq);
554}
555
556void kvm_arm_resume_guest(struct kvm *kvm)
557{
558 int i;
559 struct kvm_vcpu *vcpu;
560
561 kvm_for_each_vcpu(i, vcpu, kvm)
562 kvm_arm_resume_vcpu(vcpu);
563}
564
565static void vcpu_sleep(struct kvm_vcpu *vcpu)
566{
567 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
568
569 swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
570 (!vcpu->arch.pause)));
571}
572
573static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
574{
575 return vcpu->arch.target >= 0;
576}
577
578/**
579 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
580 * @vcpu: The VCPU pointer
581 * @run: The kvm_run structure pointer used for userspace state exchange
582 *
583 * This function is called through the VCPU_RUN ioctl called from user space. It
584 * will execute VM code in a loop until the time slice for the process is used
585 * or some emulation is needed from user space in which case the function will
586 * return with return value 0 and with the kvm_run structure filled in with the
587 * required data for the requested emulation.
588 */
589int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
590{
591 int ret;
592 sigset_t sigsaved;
593
594 if (unlikely(!kvm_vcpu_initialized(vcpu)))
595 return -ENOEXEC;
596
597 ret = kvm_vcpu_first_run_init(vcpu);
598 if (ret)
599 return ret;
600
601 if (run->exit_reason == KVM_EXIT_MMIO) {
602 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
603 if (ret)
604 return ret;
605 }
606
607 if (vcpu->sigset_active)
608 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
609
610 ret = 1;
611 run->exit_reason = KVM_EXIT_UNKNOWN;
612 while (ret > 0) {
613 /*
614 * Check conditions before entering the guest
615 */
616 cond_resched();
617
618 update_vttbr(vcpu->kvm);
619
620 if (vcpu->arch.power_off || vcpu->arch.pause)
621 vcpu_sleep(vcpu);
622
623 /*
624 * Preparing the interrupts to be injected also
625 * involves poking the GIC, which must be done in a
626 * non-preemptible context.
627 */
628 preempt_disable();
629 kvm_pmu_flush_hwstate(vcpu);
630 kvm_timer_flush_hwstate(vcpu);
631 kvm_vgic_flush_hwstate(vcpu);
632
633 local_irq_disable();
634
635 /*
636 * Re-check atomic conditions
637 */
638 if (signal_pending(current)) {
639 ret = -EINTR;
640 run->exit_reason = KVM_EXIT_INTR;
641 }
642
643 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
644 vcpu->arch.power_off || vcpu->arch.pause) {
645 local_irq_enable();
646 kvm_pmu_sync_hwstate(vcpu);
647 kvm_timer_sync_hwstate(vcpu);
648 kvm_vgic_sync_hwstate(vcpu);
649 preempt_enable();
650 continue;
651 }
652
653 kvm_arm_setup_debug(vcpu);
654
655 /**************************************************************
656 * Enter the guest
657 */
658 trace_kvm_entry(*vcpu_pc(vcpu));
659 guest_enter_irqoff();
660 vcpu->mode = IN_GUEST_MODE;
661
662 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
663
664 vcpu->mode = OUTSIDE_GUEST_MODE;
665 vcpu->stat.exits++;
666 /*
667 * Back from guest
668 *************************************************************/
669
670 kvm_arm_clear_debug(vcpu);
671
672 /*
673 * We may have taken a host interrupt in HYP mode (ie
674 * while executing the guest). This interrupt is still
675 * pending, as we haven't serviced it yet!
676 *
677 * We're now back in SVC mode, with interrupts
678 * disabled. Enabling the interrupts now will have
679 * the effect of taking the interrupt again, in SVC
680 * mode this time.
681 */
682 local_irq_enable();
683
684 /*
685 * We do local_irq_enable() before calling guest_exit() so
686 * that if a timer interrupt hits while running the guest we
687 * account that tick as being spent in the guest. We enable
688 * preemption after calling guest_exit() so that if we get
689 * preempted we make sure ticks after that is not counted as
690 * guest time.
691 */
692 guest_exit();
693 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
694
695 /*
696 * We must sync the PMU and timer state before the vgic state so
697 * that the vgic can properly sample the updated state of the
698 * interrupt line.
699 */
700 kvm_pmu_sync_hwstate(vcpu);
701 kvm_timer_sync_hwstate(vcpu);
702
703 kvm_vgic_sync_hwstate(vcpu);
704
705 preempt_enable();
706
707 ret = handle_exit(vcpu, run, ret);
708 }
709
710 if (vcpu->sigset_active)
711 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
712 return ret;
713}
714
715static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
716{
717 int bit_index;
718 bool set;
719 unsigned long *ptr;
720
721 if (number == KVM_ARM_IRQ_CPU_IRQ)
722 bit_index = __ffs(HCR_VI);
723 else /* KVM_ARM_IRQ_CPU_FIQ */
724 bit_index = __ffs(HCR_VF);
725
726 ptr = (unsigned long *)&vcpu->arch.irq_lines;
727 if (level)
728 set = test_and_set_bit(bit_index, ptr);
729 else
730 set = test_and_clear_bit(bit_index, ptr);
731
732 /*
733 * If we didn't change anything, no need to wake up or kick other CPUs
734 */
735 if (set == level)
736 return 0;
737
738 /*
739 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
740 * trigger a world-switch round on the running physical CPU to set the
741 * virtual IRQ/FIQ fields in the HCR appropriately.
742 */
743 kvm_vcpu_kick(vcpu);
744
745 return 0;
746}
747
748int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
749 bool line_status)
750{
751 u32 irq = irq_level->irq;
752 unsigned int irq_type, vcpu_idx, irq_num;
753 int nrcpus = atomic_read(&kvm->online_vcpus);
754 struct kvm_vcpu *vcpu = NULL;
755 bool level = irq_level->level;
756
757 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
758 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
759 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
760
761 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
762
763 switch (irq_type) {
764 case KVM_ARM_IRQ_TYPE_CPU:
765 if (irqchip_in_kernel(kvm))
766 return -ENXIO;
767
768 if (vcpu_idx >= nrcpus)
769 return -EINVAL;
770
771 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
772 if (!vcpu)
773 return -EINVAL;
774
775 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
776 return -EINVAL;
777
778 return vcpu_interrupt_line(vcpu, irq_num, level);
779 case KVM_ARM_IRQ_TYPE_PPI:
780 if (!irqchip_in_kernel(kvm))
781 return -ENXIO;
782
783 if (vcpu_idx >= nrcpus)
784 return -EINVAL;
785
786 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
787 if (!vcpu)
788 return -EINVAL;
789
790 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
791 return -EINVAL;
792
793 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
794 case KVM_ARM_IRQ_TYPE_SPI:
795 if (!irqchip_in_kernel(kvm))
796 return -ENXIO;
797
798 if (irq_num < VGIC_NR_PRIVATE_IRQS)
799 return -EINVAL;
800
801 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
802 }
803
804 return -EINVAL;
805}
806
807static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
808 const struct kvm_vcpu_init *init)
809{
810 unsigned int i;
811 int phys_target = kvm_target_cpu();
812
813 if (init->target != phys_target)
814 return -EINVAL;
815
816 /*
817 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
818 * use the same target.
819 */
820 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
821 return -EINVAL;
822
823 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
824 for (i = 0; i < sizeof(init->features) * 8; i++) {
825 bool set = (init->features[i / 32] & (1 << (i % 32)));
826
827 if (set && i >= KVM_VCPU_MAX_FEATURES)
828 return -ENOENT;
829
830 /*
831 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
832 * use the same feature set.
833 */
834 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
835 test_bit(i, vcpu->arch.features) != set)
836 return -EINVAL;
837
838 if (set)
839 set_bit(i, vcpu->arch.features);
840 }
841
842 vcpu->arch.target = phys_target;
843
844 /* Now we know what it is, we can reset it. */
845 return kvm_reset_vcpu(vcpu);
846}
847
848
849static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
850 struct kvm_vcpu_init *init)
851{
852 int ret;
853
854 ret = kvm_vcpu_set_target(vcpu, init);
855 if (ret)
856 return ret;
857
858 /*
859 * Ensure a rebooted VM will fault in RAM pages and detect if the
860 * guest MMU is turned off and flush the caches as needed.
861 */
862 if (vcpu->arch.has_run_once)
863 stage2_unmap_vm(vcpu->kvm);
864
865 vcpu_reset_hcr(vcpu);
866
867 /*
868 * Handle the "start in power-off" case.
869 */
870 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
871 vcpu->arch.power_off = true;
872 else
873 vcpu->arch.power_off = false;
874
875 return 0;
876}
877
878static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
879 struct kvm_device_attr *attr)
880{
881 int ret = -ENXIO;
882
883 switch (attr->group) {
884 default:
885 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
886 break;
887 }
888
889 return ret;
890}
891
892static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
893 struct kvm_device_attr *attr)
894{
895 int ret = -ENXIO;
896
897 switch (attr->group) {
898 default:
899 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
900 break;
901 }
902
903 return ret;
904}
905
906static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
907 struct kvm_device_attr *attr)
908{
909 int ret = -ENXIO;
910
911 switch (attr->group) {
912 default:
913 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
914 break;
915 }
916
917 return ret;
918}
919
920long kvm_arch_vcpu_ioctl(struct file *filp,
921 unsigned int ioctl, unsigned long arg)
922{
923 struct kvm_vcpu *vcpu = filp->private_data;
924 void __user *argp = (void __user *)arg;
925 struct kvm_device_attr attr;
926
927 switch (ioctl) {
928 case KVM_ARM_VCPU_INIT: {
929 struct kvm_vcpu_init init;
930
931 if (copy_from_user(&init, argp, sizeof(init)))
932 return -EFAULT;
933
934 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
935 }
936 case KVM_SET_ONE_REG:
937 case KVM_GET_ONE_REG: {
938 struct kvm_one_reg reg;
939
940 if (unlikely(!kvm_vcpu_initialized(vcpu)))
941 return -ENOEXEC;
942
943 if (copy_from_user(®, argp, sizeof(reg)))
944 return -EFAULT;
945 if (ioctl == KVM_SET_ONE_REG)
946 return kvm_arm_set_reg(vcpu, ®);
947 else
948 return kvm_arm_get_reg(vcpu, ®);
949 }
950 case KVM_GET_REG_LIST: {
951 struct kvm_reg_list __user *user_list = argp;
952 struct kvm_reg_list reg_list;
953 unsigned n;
954
955 if (unlikely(!kvm_vcpu_initialized(vcpu)))
956 return -ENOEXEC;
957
958 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
959 return -EFAULT;
960 n = reg_list.n;
961 reg_list.n = kvm_arm_num_regs(vcpu);
962 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
963 return -EFAULT;
964 if (n < reg_list.n)
965 return -E2BIG;
966 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
967 }
968 case KVM_SET_DEVICE_ATTR: {
969 if (copy_from_user(&attr, argp, sizeof(attr)))
970 return -EFAULT;
971 return kvm_arm_vcpu_set_attr(vcpu, &attr);
972 }
973 case KVM_GET_DEVICE_ATTR: {
974 if (copy_from_user(&attr, argp, sizeof(attr)))
975 return -EFAULT;
976 return kvm_arm_vcpu_get_attr(vcpu, &attr);
977 }
978 case KVM_HAS_DEVICE_ATTR: {
979 if (copy_from_user(&attr, argp, sizeof(attr)))
980 return -EFAULT;
981 return kvm_arm_vcpu_has_attr(vcpu, &attr);
982 }
983 default:
984 return -EINVAL;
985 }
986}
987
988/**
989 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
990 * @kvm: kvm instance
991 * @log: slot id and address to which we copy the log
992 *
993 * Steps 1-4 below provide general overview of dirty page logging. See
994 * kvm_get_dirty_log_protect() function description for additional details.
995 *
996 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
997 * always flush the TLB (step 4) even if previous step failed and the dirty
998 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
999 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1000 * writes will be marked dirty for next log read.
1001 *
1002 * 1. Take a snapshot of the bit and clear it if needed.
1003 * 2. Write protect the corresponding page.
1004 * 3. Copy the snapshot to the userspace.
1005 * 4. Flush TLB's if needed.
1006 */
1007int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1008{
1009 bool is_dirty = false;
1010 int r;
1011
1012 mutex_lock(&kvm->slots_lock);
1013
1014 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1015
1016 if (is_dirty)
1017 kvm_flush_remote_tlbs(kvm);
1018
1019 mutex_unlock(&kvm->slots_lock);
1020 return r;
1021}
1022
1023static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1024 struct kvm_arm_device_addr *dev_addr)
1025{
1026 unsigned long dev_id, type;
1027
1028 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1029 KVM_ARM_DEVICE_ID_SHIFT;
1030 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1031 KVM_ARM_DEVICE_TYPE_SHIFT;
1032
1033 switch (dev_id) {
1034 case KVM_ARM_DEVICE_VGIC_V2:
1035 if (!vgic_present)
1036 return -ENXIO;
1037 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1038 default:
1039 return -ENODEV;
1040 }
1041}
1042
1043long kvm_arch_vm_ioctl(struct file *filp,
1044 unsigned int ioctl, unsigned long arg)
1045{
1046 struct kvm *kvm = filp->private_data;
1047 void __user *argp = (void __user *)arg;
1048
1049 switch (ioctl) {
1050 case KVM_CREATE_IRQCHIP: {
1051 int ret;
1052 if (!vgic_present)
1053 return -ENXIO;
1054 mutex_lock(&kvm->lock);
1055 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1056 mutex_unlock(&kvm->lock);
1057 return ret;
1058 }
1059 case KVM_ARM_SET_DEVICE_ADDR: {
1060 struct kvm_arm_device_addr dev_addr;
1061
1062 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1063 return -EFAULT;
1064 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1065 }
1066 case KVM_ARM_PREFERRED_TARGET: {
1067 int err;
1068 struct kvm_vcpu_init init;
1069
1070 err = kvm_vcpu_preferred_target(&init);
1071 if (err)
1072 return err;
1073
1074 if (copy_to_user(argp, &init, sizeof(init)))
1075 return -EFAULT;
1076
1077 return 0;
1078 }
1079 default:
1080 return -EINVAL;
1081 }
1082}
1083
1084static void cpu_init_hyp_mode(void *dummy)
1085{
1086 phys_addr_t pgd_ptr;
1087 unsigned long hyp_stack_ptr;
1088 unsigned long stack_page;
1089 unsigned long vector_ptr;
1090
1091 /* Switch from the HYP stub to our own HYP init vector */
1092 __hyp_set_vectors(kvm_get_idmap_vector());
1093
1094 pgd_ptr = kvm_mmu_get_httbr();
1095 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1096 hyp_stack_ptr = stack_page + PAGE_SIZE;
1097 vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1098
1099 __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1100 __cpu_init_stage2();
1101
1102 if (is_kernel_in_hyp_mode())
1103 kvm_timer_init_vhe();
1104
1105 kvm_arm_init_debug();
1106}
1107
1108static void cpu_hyp_reinit(void)
1109{
1110 if (is_kernel_in_hyp_mode()) {
1111 /*
1112 * __cpu_init_stage2() is safe to call even if the PM
1113 * event was cancelled before the CPU was reset.
1114 */
1115 __cpu_init_stage2();
1116 } else {
1117 if (__hyp_get_vectors() == hyp_default_vectors)
1118 cpu_init_hyp_mode(NULL);
1119 }
1120}
1121
1122static void cpu_hyp_reset(void)
1123{
1124 if (!is_kernel_in_hyp_mode())
1125 __cpu_reset_hyp_mode(hyp_default_vectors,
1126 kvm_get_idmap_start());
1127}
1128
1129static void _kvm_arch_hardware_enable(void *discard)
1130{
1131 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1132 cpu_hyp_reinit();
1133 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1134 }
1135}
1136
1137int kvm_arch_hardware_enable(void)
1138{
1139 _kvm_arch_hardware_enable(NULL);
1140 return 0;
1141}
1142
1143static void _kvm_arch_hardware_disable(void *discard)
1144{
1145 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1146 cpu_hyp_reset();
1147 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1148 }
1149}
1150
1151void kvm_arch_hardware_disable(void)
1152{
1153 _kvm_arch_hardware_disable(NULL);
1154}
1155
1156#ifdef CONFIG_CPU_PM
1157static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1158 unsigned long cmd,
1159 void *v)
1160{
1161 /*
1162 * kvm_arm_hardware_enabled is left with its old value over
1163 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1164 * re-enable hyp.
1165 */
1166 switch (cmd) {
1167 case CPU_PM_ENTER:
1168 if (__this_cpu_read(kvm_arm_hardware_enabled))
1169 /*
1170 * don't update kvm_arm_hardware_enabled here
1171 * so that the hardware will be re-enabled
1172 * when we resume. See below.
1173 */
1174 cpu_hyp_reset();
1175
1176 return NOTIFY_OK;
1177 case CPU_PM_EXIT:
1178 if (__this_cpu_read(kvm_arm_hardware_enabled))
1179 /* The hardware was enabled before suspend. */
1180 cpu_hyp_reinit();
1181
1182 return NOTIFY_OK;
1183
1184 default:
1185 return NOTIFY_DONE;
1186 }
1187}
1188
1189static struct notifier_block hyp_init_cpu_pm_nb = {
1190 .notifier_call = hyp_init_cpu_pm_notifier,
1191};
1192
1193static void __init hyp_cpu_pm_init(void)
1194{
1195 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1196}
1197static void __init hyp_cpu_pm_exit(void)
1198{
1199 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1200}
1201#else
1202static inline void hyp_cpu_pm_init(void)
1203{
1204}
1205static inline void hyp_cpu_pm_exit(void)
1206{
1207}
1208#endif
1209
1210static void teardown_common_resources(void)
1211{
1212 free_percpu(kvm_host_cpu_state);
1213}
1214
1215static int init_common_resources(void)
1216{
1217 kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1218 if (!kvm_host_cpu_state) {
1219 kvm_err("Cannot allocate host CPU state\n");
1220 return -ENOMEM;
1221 }
1222
1223 /* set size of VMID supported by CPU */
1224 kvm_vmid_bits = kvm_get_vmid_bits();
1225 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1226
1227 return 0;
1228}
1229
1230static int init_subsystems(void)
1231{
1232 int err = 0;
1233
1234 /*
1235 * Enable hardware so that subsystem initialisation can access EL2.
1236 */
1237 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1238
1239 /*
1240 * Register CPU lower-power notifier
1241 */
1242 hyp_cpu_pm_init();
1243
1244 /*
1245 * Init HYP view of VGIC
1246 */
1247 err = kvm_vgic_hyp_init();
1248 switch (err) {
1249 case 0:
1250 vgic_present = true;
1251 break;
1252 case -ENODEV:
1253 case -ENXIO:
1254 vgic_present = false;
1255 err = 0;
1256 break;
1257 default:
1258 goto out;
1259 }
1260
1261 /*
1262 * Init HYP architected timer support
1263 */
1264 err = kvm_timer_hyp_init();
1265 if (err)
1266 goto out;
1267
1268 kvm_perf_init();
1269 kvm_coproc_table_init();
1270
1271out:
1272 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1273
1274 return err;
1275}
1276
1277static void teardown_hyp_mode(void)
1278{
1279 int cpu;
1280
1281 if (is_kernel_in_hyp_mode())
1282 return;
1283
1284 free_hyp_pgds();
1285 for_each_possible_cpu(cpu)
1286 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1287 hyp_cpu_pm_exit();
1288}
1289
1290static int init_vhe_mode(void)
1291{
1292 kvm_info("VHE mode initialized successfully\n");
1293 return 0;
1294}
1295
1296/**
1297 * Inits Hyp-mode on all online CPUs
1298 */
1299static int init_hyp_mode(void)
1300{
1301 int cpu;
1302 int err = 0;
1303
1304 /*
1305 * Allocate Hyp PGD and setup Hyp identity mapping
1306 */
1307 err = kvm_mmu_init();
1308 if (err)
1309 goto out_err;
1310
1311 /*
1312 * It is probably enough to obtain the default on one
1313 * CPU. It's unlikely to be different on the others.
1314 */
1315 hyp_default_vectors = __hyp_get_vectors();
1316
1317 /*
1318 * Allocate stack pages for Hypervisor-mode
1319 */
1320 for_each_possible_cpu(cpu) {
1321 unsigned long stack_page;
1322
1323 stack_page = __get_free_page(GFP_KERNEL);
1324 if (!stack_page) {
1325 err = -ENOMEM;
1326 goto out_err;
1327 }
1328
1329 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1330 }
1331
1332 /*
1333 * Map the Hyp-code called directly from the host
1334 */
1335 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1336 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1337 if (err) {
1338 kvm_err("Cannot map world-switch code\n");
1339 goto out_err;
1340 }
1341
1342 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1343 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1344 if (err) {
1345 kvm_err("Cannot map rodata section\n");
1346 goto out_err;
1347 }
1348
1349 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1350 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1351 if (err) {
1352 kvm_err("Cannot map bss section\n");
1353 goto out_err;
1354 }
1355
1356 /*
1357 * Map the Hyp stack pages
1358 */
1359 for_each_possible_cpu(cpu) {
1360 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1361 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1362 PAGE_HYP);
1363
1364 if (err) {
1365 kvm_err("Cannot map hyp stack\n");
1366 goto out_err;
1367 }
1368 }
1369
1370 for_each_possible_cpu(cpu) {
1371 kvm_cpu_context_t *cpu_ctxt;
1372
1373 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1374 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1375
1376 if (err) {
1377 kvm_err("Cannot map host CPU state: %d\n", err);
1378 goto out_err;
1379 }
1380 }
1381
1382 kvm_info("Hyp mode initialized successfully\n");
1383
1384 return 0;
1385
1386out_err:
1387 teardown_hyp_mode();
1388 kvm_err("error initializing Hyp mode: %d\n", err);
1389 return err;
1390}
1391
1392static void check_kvm_target_cpu(void *ret)
1393{
1394 *(int *)ret = kvm_target_cpu();
1395}
1396
1397struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1398{
1399 struct kvm_vcpu *vcpu;
1400 int i;
1401
1402 mpidr &= MPIDR_HWID_BITMASK;
1403 kvm_for_each_vcpu(i, vcpu, kvm) {
1404 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1405 return vcpu;
1406 }
1407 return NULL;
1408}
1409
1410/**
1411 * Initialize Hyp-mode and memory mappings on all CPUs.
1412 */
1413int kvm_arch_init(void *opaque)
1414{
1415 int err;
1416 int ret, cpu;
1417
1418 if (!is_hyp_mode_available()) {
1419 kvm_err("HYP mode not available\n");
1420 return -ENODEV;
1421 }
1422
1423 for_each_online_cpu(cpu) {
1424 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1425 if (ret < 0) {
1426 kvm_err("Error, CPU %d not supported!\n", cpu);
1427 return -ENODEV;
1428 }
1429 }
1430
1431 err = init_common_resources();
1432 if (err)
1433 return err;
1434
1435 if (is_kernel_in_hyp_mode())
1436 err = init_vhe_mode();
1437 else
1438 err = init_hyp_mode();
1439 if (err)
1440 goto out_err;
1441
1442 err = init_subsystems();
1443 if (err)
1444 goto out_hyp;
1445
1446 return 0;
1447
1448out_hyp:
1449 teardown_hyp_mode();
1450out_err:
1451 teardown_common_resources();
1452 return err;
1453}
1454
1455/* NOP: Compiling as a module not supported */
1456void kvm_arch_exit(void)
1457{
1458 kvm_perf_teardown();
1459}
1460
1461static int arm_init(void)
1462{
1463 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1464 return rc;
1465}
1466
1467module_init(arm_init);