]>
Commit | Line | Data |
---|---|---|
749cf76c CD |
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/errno.h> | |
20 | #include <linux/err.h> | |
21 | #include <linux/kvm_host.h> | |
22 | #include <linux/module.h> | |
23 | #include <linux/vmalloc.h> | |
24 | #include <linux/fs.h> | |
25 | #include <linux/mman.h> | |
26 | #include <linux/sched.h> | |
86ce8535 | 27 | #include <linux/kvm.h> |
749cf76c CD |
28 | #include <trace/events/kvm.h> |
29 | ||
30 | #define CREATE_TRACE_POINTS | |
31 | #include "trace.h" | |
32 | ||
33 | #include <asm/unified.h> | |
34 | #include <asm/uaccess.h> | |
35 | #include <asm/ptrace.h> | |
36 | #include <asm/mman.h> | |
37 | #include <asm/cputype.h> | |
342cd0ab | 38 | #include <asm/tlbflush.h> |
5b3e5e5b | 39 | #include <asm/cacheflush.h> |
342cd0ab CD |
40 | #include <asm/virt.h> |
41 | #include <asm/kvm_arm.h> | |
42 | #include <asm/kvm_asm.h> | |
43 | #include <asm/kvm_mmu.h> | |
f7ed45be | 44 | #include <asm/kvm_emulate.h> |
5b3e5e5b | 45 | #include <asm/kvm_coproc.h> |
aa024c2f | 46 | #include <asm/kvm_psci.h> |
5b3e5e5b | 47 | #include <asm/opcodes.h> |
749cf76c CD |
48 | |
49 | #ifdef REQUIRES_VIRT | |
50 | __asm__(".arch_extension virt"); | |
51 | #endif | |
52 | ||
342cd0ab CD |
53 | static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); |
54 | static struct vfp_hard_struct __percpu *kvm_host_vfp_state; | |
55 | static unsigned long hyp_default_vectors; | |
56 | ||
1638a12d MZ |
57 | /* Per-CPU variable containing the currently running vcpu. */ |
58 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu); | |
59 | ||
f7ed45be CD |
60 | /* The VMID used in the VTTBR */ |
61 | static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1); | |
62 | static u8 kvm_next_vmid; | |
63 | static DEFINE_SPINLOCK(kvm_vmid_lock); | |
342cd0ab | 64 | |
1a89dd91 MZ |
65 | static bool vgic_present; |
66 | ||
1638a12d MZ |
67 | static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu) |
68 | { | |
69 | BUG_ON(preemptible()); | |
70 | __get_cpu_var(kvm_arm_running_vcpu) = vcpu; | |
71 | } | |
72 | ||
73 | /** | |
74 | * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU. | |
75 | * Must be called from non-preemptible context | |
76 | */ | |
77 | struct kvm_vcpu *kvm_arm_get_running_vcpu(void) | |
78 | { | |
79 | BUG_ON(preemptible()); | |
80 | return __get_cpu_var(kvm_arm_running_vcpu); | |
81 | } | |
82 | ||
83 | /** | |
84 | * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
85 | */ | |
86 | struct kvm_vcpu __percpu **kvm_get_running_vcpus(void) | |
87 | { | |
88 | return &kvm_arm_running_vcpu; | |
89 | } | |
90 | ||
749cf76c CD |
91 | int kvm_arch_hardware_enable(void *garbage) |
92 | { | |
93 | return 0; | |
94 | } | |
95 | ||
96 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) | |
97 | { | |
98 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; | |
99 | } | |
100 | ||
101 | void kvm_arch_hardware_disable(void *garbage) | |
102 | { | |
103 | } | |
104 | ||
105 | int kvm_arch_hardware_setup(void) | |
106 | { | |
107 | return 0; | |
108 | } | |
109 | ||
110 | void kvm_arch_hardware_unsetup(void) | |
111 | { | |
112 | } | |
113 | ||
114 | void kvm_arch_check_processor_compat(void *rtn) | |
115 | { | |
116 | *(int *)rtn = 0; | |
117 | } | |
118 | ||
119 | void kvm_arch_sync_events(struct kvm *kvm) | |
120 | { | |
121 | } | |
122 | ||
d5d8184d CD |
123 | /** |
124 | * kvm_arch_init_vm - initializes a VM data structure | |
125 | * @kvm: pointer to the KVM struct | |
126 | */ | |
749cf76c CD |
127 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
128 | { | |
d5d8184d CD |
129 | int ret = 0; |
130 | ||
749cf76c CD |
131 | if (type) |
132 | return -EINVAL; | |
133 | ||
d5d8184d CD |
134 | ret = kvm_alloc_stage2_pgd(kvm); |
135 | if (ret) | |
136 | goto out_fail_alloc; | |
137 | ||
138 | ret = create_hyp_mappings(kvm, kvm + 1); | |
139 | if (ret) | |
140 | goto out_free_stage2_pgd; | |
141 | ||
142 | /* Mark the initial VMID generation invalid */ | |
143 | kvm->arch.vmid_gen = 0; | |
144 | ||
145 | return ret; | |
146 | out_free_stage2_pgd: | |
147 | kvm_free_stage2_pgd(kvm); | |
148 | out_fail_alloc: | |
149 | return ret; | |
749cf76c CD |
150 | } |
151 | ||
152 | int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) | |
153 | { | |
154 | return VM_FAULT_SIGBUS; | |
155 | } | |
156 | ||
157 | void kvm_arch_free_memslot(struct kvm_memory_slot *free, | |
158 | struct kvm_memory_slot *dont) | |
159 | { | |
160 | } | |
161 | ||
162 | int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages) | |
163 | { | |
164 | return 0; | |
165 | } | |
166 | ||
d5d8184d CD |
167 | /** |
168 | * kvm_arch_destroy_vm - destroy the VM data structure | |
169 | * @kvm: pointer to the KVM struct | |
170 | */ | |
749cf76c CD |
171 | void kvm_arch_destroy_vm(struct kvm *kvm) |
172 | { | |
173 | int i; | |
174 | ||
d5d8184d CD |
175 | kvm_free_stage2_pgd(kvm); |
176 | ||
749cf76c CD |
177 | for (i = 0; i < KVM_MAX_VCPUS; ++i) { |
178 | if (kvm->vcpus[i]) { | |
179 | kvm_arch_vcpu_free(kvm->vcpus[i]); | |
180 | kvm->vcpus[i] = NULL; | |
181 | } | |
182 | } | |
183 | } | |
184 | ||
185 | int kvm_dev_ioctl_check_extension(long ext) | |
186 | { | |
187 | int r; | |
188 | switch (ext) { | |
1a89dd91 MZ |
189 | case KVM_CAP_IRQCHIP: |
190 | r = vgic_present; | |
191 | break; | |
749cf76c CD |
192 | case KVM_CAP_USER_MEMORY: |
193 | case KVM_CAP_SYNC_MMU: | |
194 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
195 | case KVM_CAP_ONE_REG: | |
aa024c2f | 196 | case KVM_CAP_ARM_PSCI: |
749cf76c CD |
197 | r = 1; |
198 | break; | |
199 | case KVM_CAP_COALESCED_MMIO: | |
200 | r = KVM_COALESCED_MMIO_PAGE_OFFSET; | |
201 | break; | |
3401d546 CD |
202 | case KVM_CAP_ARM_SET_DEVICE_ADDR: |
203 | r = 1; | |
749cf76c CD |
204 | case KVM_CAP_NR_VCPUS: |
205 | r = num_online_cpus(); | |
206 | break; | |
207 | case KVM_CAP_MAX_VCPUS: | |
208 | r = KVM_MAX_VCPUS; | |
209 | break; | |
210 | default: | |
211 | r = 0; | |
212 | break; | |
213 | } | |
214 | return r; | |
215 | } | |
216 | ||
217 | long kvm_arch_dev_ioctl(struct file *filp, | |
218 | unsigned int ioctl, unsigned long arg) | |
219 | { | |
220 | return -EINVAL; | |
221 | } | |
222 | ||
223 | int kvm_arch_set_memory_region(struct kvm *kvm, | |
224 | struct kvm_userspace_memory_region *mem, | |
225 | struct kvm_memory_slot old, | |
226 | int user_alloc) | |
227 | { | |
228 | return 0; | |
229 | } | |
230 | ||
231 | int kvm_arch_prepare_memory_region(struct kvm *kvm, | |
232 | struct kvm_memory_slot *memslot, | |
233 | struct kvm_memory_slot old, | |
234 | struct kvm_userspace_memory_region *mem, | |
235 | int user_alloc) | |
236 | { | |
237 | return 0; | |
238 | } | |
239 | ||
240 | void kvm_arch_commit_memory_region(struct kvm *kvm, | |
241 | struct kvm_userspace_memory_region *mem, | |
242 | struct kvm_memory_slot old, | |
243 | int user_alloc) | |
244 | { | |
245 | } | |
246 | ||
247 | void kvm_arch_flush_shadow_all(struct kvm *kvm) | |
248 | { | |
249 | } | |
250 | ||
251 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, | |
252 | struct kvm_memory_slot *slot) | |
253 | { | |
254 | } | |
255 | ||
256 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) | |
257 | { | |
258 | int err; | |
259 | struct kvm_vcpu *vcpu; | |
260 | ||
261 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); | |
262 | if (!vcpu) { | |
263 | err = -ENOMEM; | |
264 | goto out; | |
265 | } | |
266 | ||
267 | err = kvm_vcpu_init(vcpu, kvm, id); | |
268 | if (err) | |
269 | goto free_vcpu; | |
270 | ||
d5d8184d CD |
271 | err = create_hyp_mappings(vcpu, vcpu + 1); |
272 | if (err) | |
273 | goto vcpu_uninit; | |
274 | ||
749cf76c | 275 | return vcpu; |
d5d8184d CD |
276 | vcpu_uninit: |
277 | kvm_vcpu_uninit(vcpu); | |
749cf76c CD |
278 | free_vcpu: |
279 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
280 | out: | |
281 | return ERR_PTR(err); | |
282 | } | |
283 | ||
284 | int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) | |
285 | { | |
286 | return 0; | |
287 | } | |
288 | ||
289 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
290 | { | |
d5d8184d CD |
291 | kvm_mmu_free_memory_caches(vcpu); |
292 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
749cf76c CD |
293 | } |
294 | ||
295 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) | |
296 | { | |
297 | kvm_arch_vcpu_free(vcpu); | |
298 | } | |
299 | ||
300 | int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) | |
301 | { | |
302 | return 0; | |
303 | } | |
304 | ||
305 | int __attribute_const__ kvm_target_cpu(void) | |
306 | { | |
307 | unsigned long implementor = read_cpuid_implementor(); | |
308 | unsigned long part_number = read_cpuid_part_number(); | |
309 | ||
310 | if (implementor != ARM_CPU_IMP_ARM) | |
311 | return -EINVAL; | |
312 | ||
313 | switch (part_number) { | |
314 | case ARM_CPU_PART_CORTEX_A15: | |
315 | return KVM_ARM_TARGET_CORTEX_A15; | |
316 | default: | |
317 | return -EINVAL; | |
318 | } | |
319 | } | |
320 | ||
321 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) | |
322 | { | |
1a89dd91 MZ |
323 | int ret; |
324 | ||
f7ed45be CD |
325 | /* Force users to call KVM_ARM_VCPU_INIT */ |
326 | vcpu->arch.target = -1; | |
1a89dd91 MZ |
327 | |
328 | /* Set up VGIC */ | |
329 | ret = kvm_vgic_vcpu_init(vcpu); | |
330 | if (ret) | |
331 | return ret; | |
332 | ||
749cf76c CD |
333 | return 0; |
334 | } | |
335 | ||
336 | void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) | |
337 | { | |
338 | } | |
339 | ||
340 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) | |
341 | { | |
86ce8535 | 342 | vcpu->cpu = cpu; |
f7ed45be | 343 | vcpu->arch.vfp_host = this_cpu_ptr(kvm_host_vfp_state); |
5b3e5e5b CD |
344 | |
345 | /* | |
346 | * Check whether this vcpu requires the cache to be flushed on | |
347 | * this physical CPU. This is a consequence of doing dcache | |
348 | * operations by set/way on this vcpu. We do it here to be in | |
349 | * a non-preemptible section. | |
350 | */ | |
351 | if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush)) | |
352 | flush_cache_all(); /* We'd really want v7_flush_dcache_all() */ | |
1638a12d MZ |
353 | |
354 | kvm_arm_set_running_vcpu(vcpu); | |
749cf76c CD |
355 | } |
356 | ||
357 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
358 | { | |
1638a12d | 359 | kvm_arm_set_running_vcpu(NULL); |
749cf76c CD |
360 | } |
361 | ||
362 | int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, | |
363 | struct kvm_guest_debug *dbg) | |
364 | { | |
365 | return -EINVAL; | |
366 | } | |
367 | ||
368 | ||
369 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, | |
370 | struct kvm_mp_state *mp_state) | |
371 | { | |
372 | return -EINVAL; | |
373 | } | |
374 | ||
375 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
376 | struct kvm_mp_state *mp_state) | |
377 | { | |
378 | return -EINVAL; | |
379 | } | |
380 | ||
5b3e5e5b CD |
381 | /** |
382 | * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled | |
383 | * @v: The VCPU pointer | |
384 | * | |
385 | * If the guest CPU is not waiting for interrupts or an interrupt line is | |
386 | * asserted, the CPU is by definition runnable. | |
387 | */ | |
749cf76c CD |
388 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) |
389 | { | |
1a89dd91 | 390 | return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v); |
749cf76c CD |
391 | } |
392 | ||
f7ed45be CD |
393 | /* Just ensure a guest exit from a particular CPU */ |
394 | static void exit_vm_noop(void *info) | |
395 | { | |
396 | } | |
397 | ||
398 | void force_vm_exit(const cpumask_t *mask) | |
399 | { | |
400 | smp_call_function_many(mask, exit_vm_noop, NULL, true); | |
401 | } | |
402 | ||
403 | /** | |
404 | * need_new_vmid_gen - check that the VMID is still valid | |
405 | * @kvm: The VM's VMID to checkt | |
406 | * | |
407 | * return true if there is a new generation of VMIDs being used | |
408 | * | |
409 | * The hardware supports only 256 values with the value zero reserved for the | |
410 | * host, so we check if an assigned value belongs to a previous generation, | |
411 | * which which requires us to assign a new value. If we're the first to use a | |
412 | * VMID for the new generation, we must flush necessary caches and TLBs on all | |
413 | * CPUs. | |
414 | */ | |
415 | static bool need_new_vmid_gen(struct kvm *kvm) | |
416 | { | |
417 | return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen)); | |
418 | } | |
419 | ||
420 | /** | |
421 | * update_vttbr - Update the VTTBR with a valid VMID before the guest runs | |
422 | * @kvm The guest that we are about to run | |
423 | * | |
424 | * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the | |
425 | * VM has a valid VMID, otherwise assigns a new one and flushes corresponding | |
426 | * caches and TLBs. | |
427 | */ | |
428 | static void update_vttbr(struct kvm *kvm) | |
429 | { | |
430 | phys_addr_t pgd_phys; | |
431 | u64 vmid; | |
432 | ||
433 | if (!need_new_vmid_gen(kvm)) | |
434 | return; | |
435 | ||
436 | spin_lock(&kvm_vmid_lock); | |
437 | ||
438 | /* | |
439 | * We need to re-check the vmid_gen here to ensure that if another vcpu | |
440 | * already allocated a valid vmid for this vm, then this vcpu should | |
441 | * use the same vmid. | |
442 | */ | |
443 | if (!need_new_vmid_gen(kvm)) { | |
444 | spin_unlock(&kvm_vmid_lock); | |
445 | return; | |
446 | } | |
447 | ||
448 | /* First user of a new VMID generation? */ | |
449 | if (unlikely(kvm_next_vmid == 0)) { | |
450 | atomic64_inc(&kvm_vmid_gen); | |
451 | kvm_next_vmid = 1; | |
452 | ||
453 | /* | |
454 | * On SMP we know no other CPUs can use this CPU's or each | |
455 | * other's VMID after force_vm_exit returns since the | |
456 | * kvm_vmid_lock blocks them from reentry to the guest. | |
457 | */ | |
458 | force_vm_exit(cpu_all_mask); | |
459 | /* | |
460 | * Now broadcast TLB + ICACHE invalidation over the inner | |
461 | * shareable domain to make sure all data structures are | |
462 | * clean. | |
463 | */ | |
464 | kvm_call_hyp(__kvm_flush_vm_context); | |
465 | } | |
466 | ||
467 | kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen); | |
468 | kvm->arch.vmid = kvm_next_vmid; | |
469 | kvm_next_vmid++; | |
470 | ||
471 | /* update vttbr to be used with the new vmid */ | |
472 | pgd_phys = virt_to_phys(kvm->arch.pgd); | |
473 | vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK; | |
474 | kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK; | |
475 | kvm->arch.vttbr |= vmid; | |
476 | ||
477 | spin_unlock(&kvm_vmid_lock); | |
478 | } | |
479 | ||
5b3e5e5b CD |
480 | static int handle_svc_hyp(struct kvm_vcpu *vcpu, struct kvm_run *run) |
481 | { | |
482 | /* SVC called from Hyp mode should never get here */ | |
483 | kvm_debug("SVC called from Hyp mode shouldn't go here\n"); | |
484 | BUG(); | |
485 | return -EINVAL; /* Squash warning */ | |
486 | } | |
487 | ||
488 | static int handle_hvc(struct kvm_vcpu *vcpu, struct kvm_run *run) | |
489 | { | |
490 | trace_kvm_hvc(*vcpu_pc(vcpu), *vcpu_reg(vcpu, 0), | |
491 | vcpu->arch.hsr & HSR_HVC_IMM_MASK); | |
492 | ||
aa024c2f MZ |
493 | if (kvm_psci_call(vcpu)) |
494 | return 1; | |
495 | ||
5b3e5e5b CD |
496 | kvm_inject_undefined(vcpu); |
497 | return 1; | |
498 | } | |
499 | ||
500 | static int handle_smc(struct kvm_vcpu *vcpu, struct kvm_run *run) | |
501 | { | |
aa024c2f MZ |
502 | if (kvm_psci_call(vcpu)) |
503 | return 1; | |
504 | ||
5b3e5e5b CD |
505 | kvm_inject_undefined(vcpu); |
506 | return 1; | |
507 | } | |
508 | ||
509 | static int handle_pabt_hyp(struct kvm_vcpu *vcpu, struct kvm_run *run) | |
510 | { | |
511 | /* The hypervisor should never cause aborts */ | |
512 | kvm_err("Prefetch Abort taken from Hyp mode at %#08x (HSR: %#08x)\n", | |
513 | vcpu->arch.hxfar, vcpu->arch.hsr); | |
514 | return -EFAULT; | |
515 | } | |
516 | ||
517 | static int handle_dabt_hyp(struct kvm_vcpu *vcpu, struct kvm_run *run) | |
518 | { | |
519 | /* This is either an error in the ws. code or an external abort */ | |
520 | kvm_err("Data Abort taken from Hyp mode at %#08x (HSR: %#08x)\n", | |
521 | vcpu->arch.hxfar, vcpu->arch.hsr); | |
522 | return -EFAULT; | |
523 | } | |
524 | ||
525 | typedef int (*exit_handle_fn)(struct kvm_vcpu *, struct kvm_run *); | |
526 | static exit_handle_fn arm_exit_handlers[] = { | |
527 | [HSR_EC_WFI] = kvm_handle_wfi, | |
528 | [HSR_EC_CP15_32] = kvm_handle_cp15_32, | |
529 | [HSR_EC_CP15_64] = kvm_handle_cp15_64, | |
530 | [HSR_EC_CP14_MR] = kvm_handle_cp14_access, | |
531 | [HSR_EC_CP14_LS] = kvm_handle_cp14_load_store, | |
532 | [HSR_EC_CP14_64] = kvm_handle_cp14_access, | |
533 | [HSR_EC_CP_0_13] = kvm_handle_cp_0_13_access, | |
534 | [HSR_EC_CP10_ID] = kvm_handle_cp10_id, | |
535 | [HSR_EC_SVC_HYP] = handle_svc_hyp, | |
536 | [HSR_EC_HVC] = handle_hvc, | |
537 | [HSR_EC_SMC] = handle_smc, | |
538 | [HSR_EC_IABT] = kvm_handle_guest_abort, | |
539 | [HSR_EC_IABT_HYP] = handle_pabt_hyp, | |
540 | [HSR_EC_DABT] = kvm_handle_guest_abort, | |
541 | [HSR_EC_DABT_HYP] = handle_dabt_hyp, | |
542 | }; | |
543 | ||
544 | /* | |
545 | * A conditional instruction is allowed to trap, even though it | |
546 | * wouldn't be executed. So let's re-implement the hardware, in | |
547 | * software! | |
548 | */ | |
549 | static bool kvm_condition_valid(struct kvm_vcpu *vcpu) | |
550 | { | |
551 | unsigned long cpsr, cond, insn; | |
552 | ||
553 | /* | |
554 | * Exception Code 0 can only happen if we set HCR.TGE to 1, to | |
555 | * catch undefined instructions, and then we won't get past | |
556 | * the arm_exit_handlers test anyway. | |
557 | */ | |
558 | BUG_ON(((vcpu->arch.hsr & HSR_EC) >> HSR_EC_SHIFT) == 0); | |
559 | ||
560 | /* Top two bits non-zero? Unconditional. */ | |
561 | if (vcpu->arch.hsr >> 30) | |
562 | return true; | |
563 | ||
564 | cpsr = *vcpu_cpsr(vcpu); | |
565 | ||
566 | /* Is condition field valid? */ | |
567 | if ((vcpu->arch.hsr & HSR_CV) >> HSR_CV_SHIFT) | |
568 | cond = (vcpu->arch.hsr & HSR_COND) >> HSR_COND_SHIFT; | |
569 | else { | |
570 | /* This can happen in Thumb mode: examine IT state. */ | |
571 | unsigned long it; | |
572 | ||
573 | it = ((cpsr >> 8) & 0xFC) | ((cpsr >> 25) & 0x3); | |
574 | ||
575 | /* it == 0 => unconditional. */ | |
576 | if (it == 0) | |
577 | return true; | |
578 | ||
579 | /* The cond for this insn works out as the top 4 bits. */ | |
580 | cond = (it >> 4); | |
581 | } | |
582 | ||
583 | /* Shift makes it look like an ARM-mode instruction */ | |
584 | insn = cond << 28; | |
585 | return arm_check_condition(insn, cpsr) != ARM_OPCODE_CONDTEST_FAIL; | |
586 | } | |
587 | ||
f7ed45be CD |
588 | /* |
589 | * Return > 0 to return to guest, < 0 on error, 0 (and set exit_reason) on | |
590 | * proper exit to QEMU. | |
591 | */ | |
592 | static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *run, | |
593 | int exception_index) | |
594 | { | |
5b3e5e5b CD |
595 | unsigned long hsr_ec; |
596 | ||
597 | switch (exception_index) { | |
598 | case ARM_EXCEPTION_IRQ: | |
599 | return 1; | |
600 | case ARM_EXCEPTION_UNDEFINED: | |
601 | kvm_err("Undefined exception in Hyp mode at: %#08x\n", | |
602 | vcpu->arch.hyp_pc); | |
603 | BUG(); | |
604 | panic("KVM: Hypervisor undefined exception!\n"); | |
605 | case ARM_EXCEPTION_DATA_ABORT: | |
606 | case ARM_EXCEPTION_PREF_ABORT: | |
607 | case ARM_EXCEPTION_HVC: | |
608 | hsr_ec = (vcpu->arch.hsr & HSR_EC) >> HSR_EC_SHIFT; | |
609 | ||
610 | if (hsr_ec >= ARRAY_SIZE(arm_exit_handlers) | |
611 | || !arm_exit_handlers[hsr_ec]) { | |
612 | kvm_err("Unkown exception class: %#08lx, " | |
613 | "hsr: %#08x\n", hsr_ec, | |
614 | (unsigned int)vcpu->arch.hsr); | |
615 | BUG(); | |
616 | } | |
617 | ||
618 | /* | |
619 | * See ARM ARM B1.14.1: "Hyp traps on instructions | |
620 | * that fail their condition code check" | |
621 | */ | |
622 | if (!kvm_condition_valid(vcpu)) { | |
623 | bool is_wide = vcpu->arch.hsr & HSR_IL; | |
624 | kvm_skip_instr(vcpu, is_wide); | |
625 | return 1; | |
626 | } | |
627 | ||
628 | return arm_exit_handlers[hsr_ec](vcpu, run); | |
629 | default: | |
630 | kvm_pr_unimpl("Unsupported exception type: %d", | |
631 | exception_index); | |
632 | run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
633 | return 0; | |
634 | } | |
f7ed45be CD |
635 | } |
636 | ||
637 | static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) | |
638 | { | |
639 | if (likely(vcpu->arch.has_run_once)) | |
640 | return 0; | |
641 | ||
642 | vcpu->arch.has_run_once = true; | |
aa024c2f MZ |
643 | |
644 | /* | |
645 | * Handle the "start in power-off" case by calling into the | |
646 | * PSCI code. | |
647 | */ | |
648 | if (test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) { | |
649 | *vcpu_reg(vcpu, 0) = KVM_PSCI_FN_CPU_OFF; | |
650 | kvm_psci_call(vcpu); | |
651 | } | |
652 | ||
f7ed45be CD |
653 | return 0; |
654 | } | |
655 | ||
aa024c2f MZ |
656 | static void vcpu_pause(struct kvm_vcpu *vcpu) |
657 | { | |
658 | wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu); | |
659 | ||
660 | wait_event_interruptible(*wq, !vcpu->arch.pause); | |
661 | } | |
662 | ||
f7ed45be CD |
663 | /** |
664 | * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code | |
665 | * @vcpu: The VCPU pointer | |
666 | * @run: The kvm_run structure pointer used for userspace state exchange | |
667 | * | |
668 | * This function is called through the VCPU_RUN ioctl called from user space. It | |
669 | * will execute VM code in a loop until the time slice for the process is used | |
670 | * or some emulation is needed from user space in which case the function will | |
671 | * return with return value 0 and with the kvm_run structure filled in with the | |
672 | * required data for the requested emulation. | |
673 | */ | |
749cf76c CD |
674 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) |
675 | { | |
f7ed45be CD |
676 | int ret; |
677 | sigset_t sigsaved; | |
678 | ||
679 | /* Make sure they initialize the vcpu with KVM_ARM_VCPU_INIT */ | |
680 | if (unlikely(vcpu->arch.target < 0)) | |
681 | return -ENOEXEC; | |
682 | ||
683 | ret = kvm_vcpu_first_run_init(vcpu); | |
684 | if (ret) | |
685 | return ret; | |
686 | ||
45e96ea6 CD |
687 | if (run->exit_reason == KVM_EXIT_MMIO) { |
688 | ret = kvm_handle_mmio_return(vcpu, vcpu->run); | |
689 | if (ret) | |
690 | return ret; | |
691 | } | |
692 | ||
f7ed45be CD |
693 | if (vcpu->sigset_active) |
694 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
695 | ||
696 | ret = 1; | |
697 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
698 | while (ret > 0) { | |
699 | /* | |
700 | * Check conditions before entering the guest | |
701 | */ | |
702 | cond_resched(); | |
703 | ||
704 | update_vttbr(vcpu->kvm); | |
705 | ||
aa024c2f MZ |
706 | if (vcpu->arch.pause) |
707 | vcpu_pause(vcpu); | |
708 | ||
1a89dd91 MZ |
709 | kvm_vgic_flush_hwstate(vcpu); |
710 | ||
f7ed45be CD |
711 | local_irq_disable(); |
712 | ||
713 | /* | |
714 | * Re-check atomic conditions | |
715 | */ | |
716 | if (signal_pending(current)) { | |
717 | ret = -EINTR; | |
718 | run->exit_reason = KVM_EXIT_INTR; | |
719 | } | |
720 | ||
721 | if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) { | |
722 | local_irq_enable(); | |
1a89dd91 | 723 | kvm_vgic_sync_hwstate(vcpu); |
f7ed45be CD |
724 | continue; |
725 | } | |
726 | ||
727 | /************************************************************** | |
728 | * Enter the guest | |
729 | */ | |
730 | trace_kvm_entry(*vcpu_pc(vcpu)); | |
731 | kvm_guest_enter(); | |
732 | vcpu->mode = IN_GUEST_MODE; | |
733 | ||
734 | ret = kvm_call_hyp(__kvm_vcpu_run, vcpu); | |
735 | ||
736 | vcpu->mode = OUTSIDE_GUEST_MODE; | |
5b3e5e5b | 737 | vcpu->arch.last_pcpu = smp_processor_id(); |
f7ed45be CD |
738 | kvm_guest_exit(); |
739 | trace_kvm_exit(*vcpu_pc(vcpu)); | |
740 | /* | |
741 | * We may have taken a host interrupt in HYP mode (ie | |
742 | * while executing the guest). This interrupt is still | |
743 | * pending, as we haven't serviced it yet! | |
744 | * | |
745 | * We're now back in SVC mode, with interrupts | |
746 | * disabled. Enabling the interrupts now will have | |
747 | * the effect of taking the interrupt again, in SVC | |
748 | * mode this time. | |
749 | */ | |
750 | local_irq_enable(); | |
751 | ||
752 | /* | |
753 | * Back from guest | |
754 | *************************************************************/ | |
755 | ||
1a89dd91 MZ |
756 | kvm_vgic_sync_hwstate(vcpu); |
757 | ||
f7ed45be CD |
758 | ret = handle_exit(vcpu, run, ret); |
759 | } | |
760 | ||
761 | if (vcpu->sigset_active) | |
762 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
763 | return ret; | |
749cf76c CD |
764 | } |
765 | ||
86ce8535 CD |
766 | static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) |
767 | { | |
768 | int bit_index; | |
769 | bool set; | |
770 | unsigned long *ptr; | |
771 | ||
772 | if (number == KVM_ARM_IRQ_CPU_IRQ) | |
773 | bit_index = __ffs(HCR_VI); | |
774 | else /* KVM_ARM_IRQ_CPU_FIQ */ | |
775 | bit_index = __ffs(HCR_VF); | |
776 | ||
777 | ptr = (unsigned long *)&vcpu->arch.irq_lines; | |
778 | if (level) | |
779 | set = test_and_set_bit(bit_index, ptr); | |
780 | else | |
781 | set = test_and_clear_bit(bit_index, ptr); | |
782 | ||
783 | /* | |
784 | * If we didn't change anything, no need to wake up or kick other CPUs | |
785 | */ | |
786 | if (set == level) | |
787 | return 0; | |
788 | ||
789 | /* | |
790 | * The vcpu irq_lines field was updated, wake up sleeping VCPUs and | |
791 | * trigger a world-switch round on the running physical CPU to set the | |
792 | * virtual IRQ/FIQ fields in the HCR appropriately. | |
793 | */ | |
794 | kvm_vcpu_kick(vcpu); | |
795 | ||
796 | return 0; | |
797 | } | |
798 | ||
799 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level) | |
800 | { | |
801 | u32 irq = irq_level->irq; | |
802 | unsigned int irq_type, vcpu_idx, irq_num; | |
803 | int nrcpus = atomic_read(&kvm->online_vcpus); | |
804 | struct kvm_vcpu *vcpu = NULL; | |
805 | bool level = irq_level->level; | |
806 | ||
807 | irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; | |
808 | vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; | |
809 | irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; | |
810 | ||
811 | trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); | |
812 | ||
813 | if (irq_type != KVM_ARM_IRQ_TYPE_CPU) | |
814 | return -EINVAL; | |
815 | ||
816 | if (vcpu_idx >= nrcpus) | |
817 | return -EINVAL; | |
818 | ||
819 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); | |
820 | if (!vcpu) | |
821 | return -EINVAL; | |
822 | ||
823 | if (irq_num > KVM_ARM_IRQ_CPU_FIQ) | |
824 | return -EINVAL; | |
825 | ||
826 | return vcpu_interrupt_line(vcpu, irq_num, level); | |
827 | } | |
828 | ||
749cf76c CD |
829 | long kvm_arch_vcpu_ioctl(struct file *filp, |
830 | unsigned int ioctl, unsigned long arg) | |
831 | { | |
832 | struct kvm_vcpu *vcpu = filp->private_data; | |
833 | void __user *argp = (void __user *)arg; | |
834 | ||
835 | switch (ioctl) { | |
836 | case KVM_ARM_VCPU_INIT: { | |
837 | struct kvm_vcpu_init init; | |
838 | ||
839 | if (copy_from_user(&init, argp, sizeof(init))) | |
840 | return -EFAULT; | |
841 | ||
842 | return kvm_vcpu_set_target(vcpu, &init); | |
843 | ||
844 | } | |
845 | case KVM_SET_ONE_REG: | |
846 | case KVM_GET_ONE_REG: { | |
847 | struct kvm_one_reg reg; | |
848 | if (copy_from_user(®, argp, sizeof(reg))) | |
849 | return -EFAULT; | |
850 | if (ioctl == KVM_SET_ONE_REG) | |
851 | return kvm_arm_set_reg(vcpu, ®); | |
852 | else | |
853 | return kvm_arm_get_reg(vcpu, ®); | |
854 | } | |
855 | case KVM_GET_REG_LIST: { | |
856 | struct kvm_reg_list __user *user_list = argp; | |
857 | struct kvm_reg_list reg_list; | |
858 | unsigned n; | |
859 | ||
860 | if (copy_from_user(®_list, user_list, sizeof(reg_list))) | |
861 | return -EFAULT; | |
862 | n = reg_list.n; | |
863 | reg_list.n = kvm_arm_num_regs(vcpu); | |
864 | if (copy_to_user(user_list, ®_list, sizeof(reg_list))) | |
865 | return -EFAULT; | |
866 | if (n < reg_list.n) | |
867 | return -E2BIG; | |
868 | return kvm_arm_copy_reg_indices(vcpu, user_list->reg); | |
869 | } | |
870 | default: | |
871 | return -EINVAL; | |
872 | } | |
873 | } | |
874 | ||
875 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) | |
876 | { | |
877 | return -EINVAL; | |
878 | } | |
879 | ||
3401d546 CD |
880 | static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, |
881 | struct kvm_arm_device_addr *dev_addr) | |
882 | { | |
883 | return -ENODEV; | |
884 | } | |
885 | ||
749cf76c CD |
886 | long kvm_arch_vm_ioctl(struct file *filp, |
887 | unsigned int ioctl, unsigned long arg) | |
888 | { | |
3401d546 CD |
889 | struct kvm *kvm = filp->private_data; |
890 | void __user *argp = (void __user *)arg; | |
891 | ||
892 | switch (ioctl) { | |
893 | case KVM_ARM_SET_DEVICE_ADDR: { | |
894 | struct kvm_arm_device_addr dev_addr; | |
895 | ||
896 | if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) | |
897 | return -EFAULT; | |
898 | return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); | |
899 | } | |
900 | default: | |
901 | return -EINVAL; | |
902 | } | |
749cf76c CD |
903 | } |
904 | ||
342cd0ab CD |
905 | static void cpu_init_hyp_mode(void *vector) |
906 | { | |
907 | unsigned long long pgd_ptr; | |
908 | unsigned long pgd_low, pgd_high; | |
909 | unsigned long hyp_stack_ptr; | |
910 | unsigned long stack_page; | |
911 | unsigned long vector_ptr; | |
912 | ||
913 | /* Switch from the HYP stub to our own HYP init vector */ | |
914 | __hyp_set_vectors((unsigned long)vector); | |
915 | ||
916 | pgd_ptr = (unsigned long long)kvm_mmu_get_httbr(); | |
917 | pgd_low = (pgd_ptr & ((1ULL << 32) - 1)); | |
918 | pgd_high = (pgd_ptr >> 32ULL); | |
919 | stack_page = __get_cpu_var(kvm_arm_hyp_stack_page); | |
920 | hyp_stack_ptr = stack_page + PAGE_SIZE; | |
921 | vector_ptr = (unsigned long)__kvm_hyp_vector; | |
922 | ||
923 | /* | |
924 | * Call initialization code, and switch to the full blown | |
925 | * HYP code. The init code doesn't need to preserve these registers as | |
926 | * r1-r3 and r12 are already callee save according to the AAPCS. | |
927 | * Note that we slightly misuse the prototype by casing the pgd_low to | |
928 | * a void *. | |
929 | */ | |
930 | kvm_call_hyp((void *)pgd_low, pgd_high, hyp_stack_ptr, vector_ptr); | |
931 | } | |
932 | ||
933 | /** | |
934 | * Inits Hyp-mode on all online CPUs | |
935 | */ | |
936 | static int init_hyp_mode(void) | |
937 | { | |
938 | phys_addr_t init_phys_addr; | |
939 | int cpu; | |
940 | int err = 0; | |
941 | ||
942 | /* | |
943 | * Allocate Hyp PGD and setup Hyp identity mapping | |
944 | */ | |
945 | err = kvm_mmu_init(); | |
946 | if (err) | |
947 | goto out_err; | |
948 | ||
949 | /* | |
950 | * It is probably enough to obtain the default on one | |
951 | * CPU. It's unlikely to be different on the others. | |
952 | */ | |
953 | hyp_default_vectors = __hyp_get_vectors(); | |
954 | ||
955 | /* | |
956 | * Allocate stack pages for Hypervisor-mode | |
957 | */ | |
958 | for_each_possible_cpu(cpu) { | |
959 | unsigned long stack_page; | |
960 | ||
961 | stack_page = __get_free_page(GFP_KERNEL); | |
962 | if (!stack_page) { | |
963 | err = -ENOMEM; | |
964 | goto out_free_stack_pages; | |
965 | } | |
966 | ||
967 | per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; | |
968 | } | |
969 | ||
970 | /* | |
971 | * Execute the init code on each CPU. | |
972 | * | |
973 | * Note: The stack is not mapped yet, so don't do anything else than | |
974 | * initializing the hypervisor mode on each CPU using a local stack | |
975 | * space for temporary storage. | |
976 | */ | |
977 | init_phys_addr = virt_to_phys(__kvm_hyp_init); | |
978 | for_each_online_cpu(cpu) { | |
979 | smp_call_function_single(cpu, cpu_init_hyp_mode, | |
980 | (void *)(long)init_phys_addr, 1); | |
981 | } | |
982 | ||
983 | /* | |
984 | * Unmap the identity mapping | |
985 | */ | |
986 | kvm_clear_hyp_idmap(); | |
987 | ||
988 | /* | |
989 | * Map the Hyp-code called directly from the host | |
990 | */ | |
991 | err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end); | |
992 | if (err) { | |
993 | kvm_err("Cannot map world-switch code\n"); | |
994 | goto out_free_mappings; | |
995 | } | |
996 | ||
997 | /* | |
998 | * Map the Hyp stack pages | |
999 | */ | |
1000 | for_each_possible_cpu(cpu) { | |
1001 | char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); | |
1002 | err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE); | |
1003 | ||
1004 | if (err) { | |
1005 | kvm_err("Cannot map hyp stack\n"); | |
1006 | goto out_free_mappings; | |
1007 | } | |
1008 | } | |
1009 | ||
1010 | /* | |
1011 | * Map the host VFP structures | |
1012 | */ | |
1013 | kvm_host_vfp_state = alloc_percpu(struct vfp_hard_struct); | |
1014 | if (!kvm_host_vfp_state) { | |
1015 | err = -ENOMEM; | |
1016 | kvm_err("Cannot allocate host VFP state\n"); | |
1017 | goto out_free_mappings; | |
1018 | } | |
1019 | ||
1020 | for_each_possible_cpu(cpu) { | |
1021 | struct vfp_hard_struct *vfp; | |
1022 | ||
1023 | vfp = per_cpu_ptr(kvm_host_vfp_state, cpu); | |
1024 | err = create_hyp_mappings(vfp, vfp + 1); | |
1025 | ||
1026 | if (err) { | |
1027 | kvm_err("Cannot map host VFP state: %d\n", err); | |
1028 | goto out_free_vfp; | |
1029 | } | |
1030 | } | |
1031 | ||
1a89dd91 MZ |
1032 | /* |
1033 | * Init HYP view of VGIC | |
1034 | */ | |
1035 | err = kvm_vgic_hyp_init(); | |
1036 | if (err) | |
1037 | goto out_free_vfp; | |
1038 | ||
342cd0ab CD |
1039 | kvm_info("Hyp mode initialized successfully\n"); |
1040 | return 0; | |
1041 | out_free_vfp: | |
1042 | free_percpu(kvm_host_vfp_state); | |
1043 | out_free_mappings: | |
1044 | free_hyp_pmds(); | |
1045 | out_free_stack_pages: | |
1046 | for_each_possible_cpu(cpu) | |
1047 | free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); | |
1048 | out_err: | |
1049 | kvm_err("error initializing Hyp mode: %d\n", err); | |
1050 | return err; | |
1051 | } | |
1052 | ||
1053 | /** | |
1054 | * Initialize Hyp-mode and memory mappings on all CPUs. | |
1055 | */ | |
749cf76c CD |
1056 | int kvm_arch_init(void *opaque) |
1057 | { | |
342cd0ab CD |
1058 | int err; |
1059 | ||
1060 | if (!is_hyp_mode_available()) { | |
1061 | kvm_err("HYP mode not available\n"); | |
1062 | return -ENODEV; | |
1063 | } | |
1064 | ||
1065 | if (kvm_target_cpu() < 0) { | |
1066 | kvm_err("Target CPU not supported!\n"); | |
1067 | return -ENODEV; | |
1068 | } | |
1069 | ||
1070 | err = init_hyp_mode(); | |
1071 | if (err) | |
1072 | goto out_err; | |
1073 | ||
5b3e5e5b | 1074 | kvm_coproc_table_init(); |
749cf76c | 1075 | return 0; |
342cd0ab CD |
1076 | out_err: |
1077 | return err; | |
749cf76c CD |
1078 | } |
1079 | ||
1080 | /* NOP: Compiling as a module not supported */ | |
1081 | void kvm_arch_exit(void) | |
1082 | { | |
1083 | } | |
1084 | ||
1085 | static int arm_init(void) | |
1086 | { | |
1087 | int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1088 | return rc; | |
1089 | } | |
1090 | ||
1091 | module_init(arm_init); |