1 /*
2 * Copyright (C) 2012 - Virtual Open Systems and Columbia University
3 * Author: Christoffer Dall <c.dall@virtualopensystems.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2, as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
17 */
18
19 #include <linux/cpu.h>
20 #include <linux/cpu_pm.h>
21 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/kvm_host.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
32 #define CREATE_TRACE_POINTS
33 #include "trace.h"
34
35 #include <asm/uaccess.h>
36 #include <asm/ptrace.h>
37 #include <asm/mman.h>
38 #include <asm/tlbflush.h>
39 #include <asm/cacheflush.h>
40 #include <asm/virt.h>
41 #include <asm/kvm_arm.h>
42 #include <asm/kvm_asm.h>
43 #include <asm/kvm_mmu.h>
44 #include <asm/kvm_emulate.h>
45 #include <asm/kvm_coproc.h>
46 #include <asm/kvm_psci.h>
47
48 #ifdef REQUIRES_VIRT
49 __asm__(".arch_extension virt");
50 #endif
51
52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
53 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
54 static unsigned long hyp_default_vectors;
55
56 /* Per-CPU variable containing the currently running vcpu. */
57 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
58
59 /* The VMID used in the VTTBR */
60 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
61 static u8 kvm_next_vmid;
62 static DEFINE_SPINLOCK(kvm_vmid_lock);
63
64 static bool vgic_present;
65
kvm_arm_set_running_vcpu(struct kvm_vcpu * vcpu)66 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
67 {
68 BUG_ON(preemptible());
69 __this_cpu_write(kvm_arm_running_vcpu, vcpu);
70 }
71
72 /**
73 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
74 * Must be called from non-preemptible context
75 */
kvm_arm_get_running_vcpu(void)76 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
77 {
78 BUG_ON(preemptible());
79 return __this_cpu_read(kvm_arm_running_vcpu);
80 }
81
82 /**
83 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
84 */
kvm_get_running_vcpus(void)85 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
86 {
87 return &kvm_arm_running_vcpu;
88 }
89
kvm_arch_hardware_enable(void)90 int kvm_arch_hardware_enable(void)
91 {
92 return 0;
93 }
94
kvm_arch_vcpu_should_kick(struct kvm_vcpu * vcpu)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
kvm_arch_hardware_setup(void)100 int kvm_arch_hardware_setup(void)
101 {
102 return 0;
103 }
104
kvm_arch_check_processor_compat(void * rtn)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 */
kvm_arch_init_vm(struct kvm * kvm,unsigned long type)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);
127 if (ret)
128 goto out_free_stage2_pgd;
129
130 kvm_timer_init(kvm);
131
132 /* Mark the initial VMID generation invalid */
133 kvm->arch.vmid_gen = 0;
134
135 return ret;
136 out_free_stage2_pgd:
137 kvm_free_stage2_pgd(kvm);
138 out_fail_alloc:
139 return ret;
140 }
141
kvm_arch_vcpu_fault(struct kvm_vcpu * vcpu,struct vm_fault * vmf)142 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
143 {
144 return VM_FAULT_SIGBUS;
145 }
146
147
148 /**
149 * kvm_arch_destroy_vm - destroy the VM data structure
150 * @kvm: pointer to the KVM struct
151 */
kvm_arch_destroy_vm(struct kvm * kvm)152 void kvm_arch_destroy_vm(struct kvm *kvm)
153 {
154 int i;
155
156 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
157 if (kvm->vcpus[i]) {
158 kvm_arch_vcpu_free(kvm->vcpus[i]);
159 kvm->vcpus[i] = NULL;
160 }
161 }
162
163 kvm_vgic_destroy(kvm);
164 }
165
kvm_vm_ioctl_check_extension(struct kvm * kvm,long ext)166 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
167 {
168 int r;
169 switch (ext) {
170 case KVM_CAP_IRQCHIP:
171 r = vgic_present;
172 break;
173 case KVM_CAP_DEVICE_CTRL:
174 case KVM_CAP_USER_MEMORY:
175 case KVM_CAP_SYNC_MMU:
176 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
177 case KVM_CAP_ONE_REG:
178 case KVM_CAP_ARM_PSCI:
179 case KVM_CAP_ARM_PSCI_0_2:
180 case KVM_CAP_READONLY_MEM:
181 r = 1;
182 break;
183 case KVM_CAP_COALESCED_MMIO:
184 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
185 break;
186 case KVM_CAP_ARM_SET_DEVICE_ADDR:
187 r = 1;
188 break;
189 case KVM_CAP_NR_VCPUS:
190 r = num_online_cpus();
191 break;
192 case KVM_CAP_MAX_VCPUS:
193 r = KVM_MAX_VCPUS;
194 break;
195 default:
196 r = kvm_arch_dev_ioctl_check_extension(ext);
197 break;
198 }
199 return r;
200 }
201
kvm_arch_dev_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)202 long kvm_arch_dev_ioctl(struct file *filp,
203 unsigned int ioctl, unsigned long arg)
204 {
205 return -EINVAL;
206 }
207
208
kvm_arch_vcpu_create(struct kvm * kvm,unsigned int id)209 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
210 {
211 int err;
212 struct kvm_vcpu *vcpu;
213
214 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
215 err = -EBUSY;
216 goto out;
217 }
218
219 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
220 if (!vcpu) {
221 err = -ENOMEM;
222 goto out;
223 }
224
225 err = kvm_vcpu_init(vcpu, kvm, id);
226 if (err)
227 goto free_vcpu;
228
229 err = create_hyp_mappings(vcpu, vcpu + 1);
230 if (err)
231 goto vcpu_uninit;
232
233 return vcpu;
234 vcpu_uninit:
235 kvm_vcpu_uninit(vcpu);
236 free_vcpu:
237 kmem_cache_free(kvm_vcpu_cache, vcpu);
238 out:
239 return ERR_PTR(err);
240 }
241
kvm_arch_vcpu_postcreate(struct kvm_vcpu * vcpu)242 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
243 {
244 return 0;
245 }
246
kvm_arch_vcpu_free(struct kvm_vcpu * vcpu)247 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
248 {
249 kvm_mmu_free_memory_caches(vcpu);
250 kvm_timer_vcpu_terminate(vcpu);
251 kvm_vgic_vcpu_destroy(vcpu);
252 kvm_vcpu_uninit(vcpu);
253 kmem_cache_free(kvm_vcpu_cache, vcpu);
254 }
255
kvm_arch_vcpu_destroy(struct kvm_vcpu * vcpu)256 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
257 {
258 kvm_arch_vcpu_free(vcpu);
259 }
260
kvm_cpu_has_pending_timer(struct kvm_vcpu * vcpu)261 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
262 {
263 return 0;
264 }
265
kvm_arch_vcpu_init(struct kvm_vcpu * vcpu)266 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
267 {
268 /* Force users to call KVM_ARM_VCPU_INIT */
269 vcpu->arch.target = -1;
270
271 /* Set up the timer */
272 kvm_timer_vcpu_init(vcpu);
273
274 return 0;
275 }
276
kvm_arch_vcpu_load(struct kvm_vcpu * vcpu,int cpu)277 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
278 {
279 vcpu->cpu = cpu;
280 vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
281
282 kvm_arm_set_running_vcpu(vcpu);
283 }
284
kvm_arch_vcpu_put(struct kvm_vcpu * vcpu)285 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
286 {
287 /*
288 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
289 * if the vcpu is no longer assigned to a cpu. This is used for the
290 * optimized make_all_cpus_request path.
291 */
292 vcpu->cpu = -1;
293
294 kvm_arm_set_running_vcpu(NULL);
295 }
296
kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu * vcpu,struct kvm_guest_debug * dbg)297 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
298 struct kvm_guest_debug *dbg)
299 {
300 return -EINVAL;
301 }
302
303
kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu * vcpu,struct kvm_mp_state * mp_state)304 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
305 struct kvm_mp_state *mp_state)
306 {
307 return -EINVAL;
308 }
309
kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu * vcpu,struct kvm_mp_state * mp_state)310 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
311 struct kvm_mp_state *mp_state)
312 {
313 return -EINVAL;
314 }
315
316 /**
317 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
318 * @v: The VCPU pointer
319 *
320 * If the guest CPU is not waiting for interrupts or an interrupt line is
321 * asserted, the CPU is by definition runnable.
322 */
kvm_arch_vcpu_runnable(struct kvm_vcpu * v)323 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
324 {
325 return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
326 }
327
328 /* Just ensure a guest exit from a particular CPU */
exit_vm_noop(void * info)329 static void exit_vm_noop(void *info)
330 {
331 }
332
force_vm_exit(const cpumask_t * mask)333 void force_vm_exit(const cpumask_t *mask)
334 {
335 smp_call_function_many(mask, exit_vm_noop, NULL, true);
336 }
337
338 /**
339 * need_new_vmid_gen - check that the VMID is still valid
340 * @kvm: The VM's VMID to checkt
341 *
342 * return true if there is a new generation of VMIDs being used
343 *
344 * The hardware supports only 256 values with the value zero reserved for the
345 * host, so we check if an assigned value belongs to a previous generation,
346 * which which requires us to assign a new value. If we're the first to use a
347 * VMID for the new generation, we must flush necessary caches and TLBs on all
348 * CPUs.
349 */
need_new_vmid_gen(struct kvm * kvm)350 static bool need_new_vmid_gen(struct kvm *kvm)
351 {
352 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
353 }
354
355 /**
356 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
357 * @kvm The guest that we are about to run
358 *
359 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
360 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
361 * caches and TLBs.
362 */
update_vttbr(struct kvm * kvm)363 static void update_vttbr(struct kvm *kvm)
364 {
365 phys_addr_t pgd_phys;
366 u64 vmid;
367
368 if (!need_new_vmid_gen(kvm))
369 return;
370
371 spin_lock(&kvm_vmid_lock);
372
373 /*
374 * We need to re-check the vmid_gen here to ensure that if another vcpu
375 * already allocated a valid vmid for this vm, then this vcpu should
376 * use the same vmid.
377 */
378 if (!need_new_vmid_gen(kvm)) {
379 spin_unlock(&kvm_vmid_lock);
380 return;
381 }
382
383 /* First user of a new VMID generation? */
384 if (unlikely(kvm_next_vmid == 0)) {
385 atomic64_inc(&kvm_vmid_gen);
386 kvm_next_vmid = 1;
387
388 /*
389 * On SMP we know no other CPUs can use this CPU's or each
390 * other's VMID after force_vm_exit returns since the
391 * kvm_vmid_lock blocks them from reentry to the guest.
392 */
393 force_vm_exit(cpu_all_mask);
394 /*
395 * Now broadcast TLB + ICACHE invalidation over the inner
396 * shareable domain to make sure all data structures are
397 * clean.
398 */
399 kvm_call_hyp(__kvm_flush_vm_context);
400 }
401
402 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
403 kvm->arch.vmid = kvm_next_vmid;
404 kvm_next_vmid++;
405
406 /* update vttbr to be used with the new vmid */
407 pgd_phys = virt_to_phys(kvm_get_hwpgd(kvm));
408 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
409 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
410 kvm->arch.vttbr = pgd_phys | vmid;
411
412 spin_unlock(&kvm_vmid_lock);
413 }
414
kvm_vcpu_first_run_init(struct kvm_vcpu * vcpu)415 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
416 {
417 struct kvm *kvm = vcpu->kvm;
418 int ret;
419
420 if (likely(vcpu->arch.has_run_once))
421 return 0;
422
423 vcpu->arch.has_run_once = true;
424
425 /*
426 * Map the VGIC hardware resources before running a vcpu the first
427 * time on this VM.
428 */
429 if (unlikely(!vgic_initialized(kvm))) {
430 ret = kvm_vgic_map_resources(kvm);
431 if (ret)
432 return ret;
433 }
434
435 /*
436 * Enable the arch timers only if we have an in-kernel VGIC
437 * and it has been properly initialized, since we cannot handle
438 * interrupts from the virtual timer with a userspace gic.
439 */
440 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
441 kvm_timer_enable(kvm);
442
443 return 0;
444 }
445
vcpu_pause(struct kvm_vcpu * vcpu)446 static void vcpu_pause(struct kvm_vcpu *vcpu)
447 {
448 wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
449
450 wait_event_interruptible(*wq, !vcpu->arch.pause);
451 }
452
kvm_vcpu_initialized(struct kvm_vcpu * vcpu)453 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
454 {
455 return vcpu->arch.target >= 0;
456 }
457
458 /**
459 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
460 * @vcpu: The VCPU pointer
461 * @run: The kvm_run structure pointer used for userspace state exchange
462 *
463 * This function is called through the VCPU_RUN ioctl called from user space. It
464 * will execute VM code in a loop until the time slice for the process is used
465 * or some emulation is needed from user space in which case the function will
466 * return with return value 0 and with the kvm_run structure filled in with the
467 * required data for the requested emulation.
468 */
kvm_arch_vcpu_ioctl_run(struct kvm_vcpu * vcpu,struct kvm_run * run)469 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
470 {
471 int ret;
472 sigset_t sigsaved;
473
474 if (unlikely(!kvm_vcpu_initialized(vcpu)))
475 return -ENOEXEC;
476
477 ret = kvm_vcpu_first_run_init(vcpu);
478 if (ret)
479 return ret;
480
481 if (run->exit_reason == KVM_EXIT_MMIO) {
482 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
483 if (ret)
484 return ret;
485 }
486
487 if (vcpu->sigset_active)
488 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
489
490 ret = 1;
491 run->exit_reason = KVM_EXIT_UNKNOWN;
492 while (ret > 0) {
493 /*
494 * Check conditions before entering the guest
495 */
496 cond_resched();
497
498 update_vttbr(vcpu->kvm);
499
500 if (vcpu->arch.pause)
501 vcpu_pause(vcpu);
502
503 kvm_vgic_flush_hwstate(vcpu);
504 kvm_timer_flush_hwstate(vcpu);
505
506 local_irq_disable();
507
508 /*
509 * Re-check atomic conditions
510 */
511 if (signal_pending(current)) {
512 ret = -EINTR;
513 run->exit_reason = KVM_EXIT_INTR;
514 }
515
516 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
517 local_irq_enable();
518 kvm_timer_sync_hwstate(vcpu);
519 kvm_vgic_sync_hwstate(vcpu);
520 continue;
521 }
522
523 /**************************************************************
524 * Enter the guest
525 */
526 trace_kvm_entry(*vcpu_pc(vcpu));
527 kvm_guest_enter();
528 vcpu->mode = IN_GUEST_MODE;
529
530 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
531
532 vcpu->mode = OUTSIDE_GUEST_MODE;
533 kvm_guest_exit();
534 trace_kvm_exit(*vcpu_pc(vcpu));
535 /*
536 * We may have taken a host interrupt in HYP mode (ie
537 * while executing the guest). This interrupt is still
538 * pending, as we haven't serviced it yet!
539 *
540 * We're now back in SVC mode, with interrupts
541 * disabled. Enabling the interrupts now will have
542 * the effect of taking the interrupt again, in SVC
543 * mode this time.
544 */
545 local_irq_enable();
546
547 /*
548 * Back from guest
549 *************************************************************/
550
551 kvm_timer_sync_hwstate(vcpu);
552 kvm_vgic_sync_hwstate(vcpu);
553
554 ret = handle_exit(vcpu, run, ret);
555 }
556
557 if (vcpu->sigset_active)
558 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
559 return ret;
560 }
561
vcpu_interrupt_line(struct kvm_vcpu * vcpu,int number,bool level)562 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
563 {
564 int bit_index;
565 bool set;
566 unsigned long *ptr;
567
568 if (number == KVM_ARM_IRQ_CPU_IRQ)
569 bit_index = __ffs(HCR_VI);
570 else /* KVM_ARM_IRQ_CPU_FIQ */
571 bit_index = __ffs(HCR_VF);
572
573 ptr = (unsigned long *)&vcpu->arch.irq_lines;
574 if (level)
575 set = test_and_set_bit(bit_index, ptr);
576 else
577 set = test_and_clear_bit(bit_index, ptr);
578
579 /*
580 * If we didn't change anything, no need to wake up or kick other CPUs
581 */
582 if (set == level)
583 return 0;
584
585 /*
586 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
587 * trigger a world-switch round on the running physical CPU to set the
588 * virtual IRQ/FIQ fields in the HCR appropriately.
589 */
590 kvm_vcpu_kick(vcpu);
591
592 return 0;
593 }
594
kvm_vm_ioctl_irq_line(struct kvm * kvm,struct kvm_irq_level * irq_level,bool line_status)595 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
596 bool line_status)
597 {
598 u32 irq = irq_level->irq;
599 unsigned int irq_type, vcpu_idx, irq_num;
600 int nrcpus = atomic_read(&kvm->online_vcpus);
601 struct kvm_vcpu *vcpu = NULL;
602 bool level = irq_level->level;
603
604 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
605 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
606 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
607
608 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
609
610 switch (irq_type) {
611 case KVM_ARM_IRQ_TYPE_CPU:
612 if (irqchip_in_kernel(kvm))
613 return -ENXIO;
614
615 if (vcpu_idx >= nrcpus)
616 return -EINVAL;
617
618 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
619 if (!vcpu)
620 return -EINVAL;
621
622 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
623 return -EINVAL;
624
625 return vcpu_interrupt_line(vcpu, irq_num, level);
626 case KVM_ARM_IRQ_TYPE_PPI:
627 if (!irqchip_in_kernel(kvm))
628 return -ENXIO;
629
630 if (vcpu_idx >= nrcpus)
631 return -EINVAL;
632
633 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
634 if (!vcpu)
635 return -EINVAL;
636
637 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
638 return -EINVAL;
639
640 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
641 case KVM_ARM_IRQ_TYPE_SPI:
642 if (!irqchip_in_kernel(kvm))
643 return -ENXIO;
644
645 if (irq_num < VGIC_NR_PRIVATE_IRQS)
646 return -EINVAL;
647
648 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
649 }
650
651 return -EINVAL;
652 }
653
kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu * vcpu,struct kvm_vcpu_init * init)654 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
655 struct kvm_vcpu_init *init)
656 {
657 int ret;
658
659 ret = kvm_vcpu_set_target(vcpu, init);
660 if (ret)
661 return ret;
662
663 vcpu_reset_hcr(vcpu);
664
665 /*
666 * Ensure a rebooted VM will fault in RAM pages and detect if the
667 * guest MMU is turned off and flush the caches as needed.
668 */
669 if (vcpu->arch.has_run_once)
670 stage2_unmap_vm(vcpu->kvm);
671
672 vcpu_reset_hcr(vcpu);
673
674 /*
675 * Handle the "start in power-off" case by marking the VCPU as paused.
676 */
677 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
678 vcpu->arch.pause = true;
679 else
680 vcpu->arch.pause = false;
681
682 return 0;
683 }
684
kvm_arch_vcpu_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)685 long kvm_arch_vcpu_ioctl(struct file *filp,
686 unsigned int ioctl, unsigned long arg)
687 {
688 struct kvm_vcpu *vcpu = filp->private_data;
689 void __user *argp = (void __user *)arg;
690
691 switch (ioctl) {
692 case KVM_ARM_VCPU_INIT: {
693 struct kvm_vcpu_init init;
694
695 if (copy_from_user(&init, argp, sizeof(init)))
696 return -EFAULT;
697
698 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
699 }
700 case KVM_SET_ONE_REG:
701 case KVM_GET_ONE_REG: {
702 struct kvm_one_reg reg;
703
704 if (unlikely(!kvm_vcpu_initialized(vcpu)))
705 return -ENOEXEC;
706
707 if (copy_from_user(®, argp, sizeof(reg)))
708 return -EFAULT;
709 if (ioctl == KVM_SET_ONE_REG)
710 return kvm_arm_set_reg(vcpu, ®);
711 else
712 return kvm_arm_get_reg(vcpu, ®);
713 }
714 case KVM_GET_REG_LIST: {
715 struct kvm_reg_list __user *user_list = argp;
716 struct kvm_reg_list reg_list;
717 unsigned n;
718
719 if (unlikely(!kvm_vcpu_initialized(vcpu)))
720 return -ENOEXEC;
721
722 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
723 return -EFAULT;
724 n = reg_list.n;
725 reg_list.n = kvm_arm_num_regs(vcpu);
726 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
727 return -EFAULT;
728 if (n < reg_list.n)
729 return -E2BIG;
730 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
731 }
732 default:
733 return -EINVAL;
734 }
735 }
736
kvm_vm_ioctl_get_dirty_log(struct kvm * kvm,struct kvm_dirty_log * log)737 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
738 {
739 return -EINVAL;
740 }
741
kvm_vm_ioctl_set_device_addr(struct kvm * kvm,struct kvm_arm_device_addr * dev_addr)742 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
743 struct kvm_arm_device_addr *dev_addr)
744 {
745 unsigned long dev_id, type;
746
747 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
748 KVM_ARM_DEVICE_ID_SHIFT;
749 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
750 KVM_ARM_DEVICE_TYPE_SHIFT;
751
752 switch (dev_id) {
753 case KVM_ARM_DEVICE_VGIC_V2:
754 if (!vgic_present)
755 return -ENXIO;
756 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
757 default:
758 return -ENODEV;
759 }
760 }
761
kvm_arch_vm_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)762 long kvm_arch_vm_ioctl(struct file *filp,
763 unsigned int ioctl, unsigned long arg)
764 {
765 struct kvm *kvm = filp->private_data;
766 void __user *argp = (void __user *)arg;
767
768 switch (ioctl) {
769 case KVM_CREATE_IRQCHIP: {
770 if (vgic_present)
771 return kvm_vgic_create(kvm);
772 else
773 return -ENXIO;
774 }
775 case KVM_ARM_SET_DEVICE_ADDR: {
776 struct kvm_arm_device_addr dev_addr;
777
778 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
779 return -EFAULT;
780 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
781 }
782 case KVM_ARM_PREFERRED_TARGET: {
783 int err;
784 struct kvm_vcpu_init init;
785
786 err = kvm_vcpu_preferred_target(&init);
787 if (err)
788 return err;
789
790 if (copy_to_user(argp, &init, sizeof(init)))
791 return -EFAULT;
792
793 return 0;
794 }
795 default:
796 return -EINVAL;
797 }
798 }
799
cpu_init_hyp_mode(void * dummy)800 static void cpu_init_hyp_mode(void *dummy)
801 {
802 phys_addr_t boot_pgd_ptr;
803 phys_addr_t pgd_ptr;
804 unsigned long hyp_stack_ptr;
805 unsigned long stack_page;
806 unsigned long vector_ptr;
807
808 /* Switch from the HYP stub to our own HYP init vector */
809 __hyp_set_vectors(kvm_get_idmap_vector());
810
811 boot_pgd_ptr = kvm_mmu_get_boot_httbr();
812 pgd_ptr = kvm_mmu_get_httbr();
813 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
814 hyp_stack_ptr = stack_page + PAGE_SIZE;
815 vector_ptr = (unsigned long)__kvm_hyp_vector;
816
817 __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
818 }
819
hyp_init_cpu_notify(struct notifier_block * self,unsigned long action,void * cpu)820 static int hyp_init_cpu_notify(struct notifier_block *self,
821 unsigned long action, void *cpu)
822 {
823 switch (action) {
824 case CPU_STARTING:
825 case CPU_STARTING_FROZEN:
826 if (__hyp_get_vectors() == hyp_default_vectors)
827 cpu_init_hyp_mode(NULL);
828 break;
829 }
830
831 return NOTIFY_OK;
832 }
833
834 static struct notifier_block hyp_init_cpu_nb = {
835 .notifier_call = hyp_init_cpu_notify,
836 };
837
838 #ifdef CONFIG_CPU_PM
hyp_init_cpu_pm_notifier(struct notifier_block * self,unsigned long cmd,void * v)839 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
840 unsigned long cmd,
841 void *v)
842 {
843 if (cmd == CPU_PM_EXIT &&
844 __hyp_get_vectors() == hyp_default_vectors) {
845 cpu_init_hyp_mode(NULL);
846 return NOTIFY_OK;
847 }
848
849 return NOTIFY_DONE;
850 }
851
852 static struct notifier_block hyp_init_cpu_pm_nb = {
853 .notifier_call = hyp_init_cpu_pm_notifier,
854 };
855
hyp_cpu_pm_init(void)856 static void __init hyp_cpu_pm_init(void)
857 {
858 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
859 }
860 #else
hyp_cpu_pm_init(void)861 static inline void hyp_cpu_pm_init(void)
862 {
863 }
864 #endif
865
866 /**
867 * Inits Hyp-mode on all online CPUs
868 */
init_hyp_mode(void)869 static int init_hyp_mode(void)
870 {
871 int cpu;
872 int err = 0;
873
874 /*
875 * Allocate Hyp PGD and setup Hyp identity mapping
876 */
877 err = kvm_mmu_init();
878 if (err)
879 goto out_err;
880
881 /*
882 * It is probably enough to obtain the default on one
883 * CPU. It's unlikely to be different on the others.
884 */
885 hyp_default_vectors = __hyp_get_vectors();
886
887 /*
888 * Allocate stack pages for Hypervisor-mode
889 */
890 for_each_possible_cpu(cpu) {
891 unsigned long stack_page;
892
893 stack_page = __get_free_page(GFP_KERNEL);
894 if (!stack_page) {
895 err = -ENOMEM;
896 goto out_free_stack_pages;
897 }
898
899 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
900 }
901
902 /*
903 * Map the Hyp-code called directly from the host
904 */
905 err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
906 if (err) {
907 kvm_err("Cannot map world-switch code\n");
908 goto out_free_mappings;
909 }
910
911 /*
912 * Map the Hyp stack pages
913 */
914 for_each_possible_cpu(cpu) {
915 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
916 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
917
918 if (err) {
919 kvm_err("Cannot map hyp stack\n");
920 goto out_free_mappings;
921 }
922 }
923
924 /*
925 * Map the host CPU structures
926 */
927 kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
928 if (!kvm_host_cpu_state) {
929 err = -ENOMEM;
930 kvm_err("Cannot allocate host CPU state\n");
931 goto out_free_mappings;
932 }
933
934 for_each_possible_cpu(cpu) {
935 kvm_cpu_context_t *cpu_ctxt;
936
937 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
938 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
939
940 if (err) {
941 kvm_err("Cannot map host CPU state: %d\n", err);
942 goto out_free_context;
943 }
944 }
945
946 /*
947 * Execute the init code on each CPU.
948 */
949 on_each_cpu(cpu_init_hyp_mode, NULL, 1);
950
951 /*
952 * Init HYP view of VGIC
953 */
954 err = kvm_vgic_hyp_init();
955 if (err)
956 goto out_free_context;
957
958 #ifdef CONFIG_KVM_ARM_VGIC
959 vgic_present = true;
960 #endif
961
962 /*
963 * Init HYP architected timer support
964 */
965 err = kvm_timer_hyp_init();
966 if (err)
967 goto out_free_mappings;
968
969 #ifndef CONFIG_HOTPLUG_CPU
970 free_boot_hyp_pgd();
971 #endif
972
973 kvm_perf_init();
974
975 kvm_info("Hyp mode initialized successfully\n");
976
977 return 0;
978 out_free_context:
979 free_percpu(kvm_host_cpu_state);
980 out_free_mappings:
981 free_hyp_pgds();
982 out_free_stack_pages:
983 for_each_possible_cpu(cpu)
984 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
985 out_err:
986 kvm_err("error initializing Hyp mode: %d\n", err);
987 return err;
988 }
989
check_kvm_target_cpu(void * ret)990 static void check_kvm_target_cpu(void *ret)
991 {
992 *(int *)ret = kvm_target_cpu();
993 }
994
995 /**
996 * Initialize Hyp-mode and memory mappings on all CPUs.
997 */
kvm_arch_init(void * opaque)998 int kvm_arch_init(void *opaque)
999 {
1000 int err;
1001 int ret, cpu;
1002
1003 if (!is_hyp_mode_available()) {
1004 kvm_err("HYP mode not available\n");
1005 return -ENODEV;
1006 }
1007
1008 for_each_online_cpu(cpu) {
1009 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1010 if (ret < 0) {
1011 kvm_err("Error, CPU %d not supported!\n", cpu);
1012 return -ENODEV;
1013 }
1014 }
1015
1016 cpu_notifier_register_begin();
1017
1018 err = init_hyp_mode();
1019 if (err)
1020 goto out_err;
1021
1022 err = __register_cpu_notifier(&hyp_init_cpu_nb);
1023 if (err) {
1024 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1025 goto out_err;
1026 }
1027
1028 cpu_notifier_register_done();
1029
1030 hyp_cpu_pm_init();
1031
1032 kvm_coproc_table_init();
1033 return 0;
1034 out_err:
1035 cpu_notifier_register_done();
1036 return err;
1037 }
1038
1039 /* NOP: Compiling as a module not supported */
kvm_arch_exit(void)1040 void kvm_arch_exit(void)
1041 {
1042 kvm_perf_teardown();
1043 }
1044
arm_init(void)1045 static int arm_init(void)
1046 {
1047 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1048 return rc;
1049 }
1050
1051 module_init(arm_init);
1052