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 | ||
d157f4a5 | 19 | #include <linux/cpu.h> |
749cf76c CD |
20 | #include <linux/errno.h> |
21 | #include <linux/err.h> | |
22 | #include <linux/kvm_host.h> | |
23 | #include <linux/module.h> | |
24 | #include <linux/vmalloc.h> | |
25 | #include <linux/fs.h> | |
26 | #include <linux/mman.h> | |
27 | #include <linux/sched.h> | |
86ce8535 | 28 | #include <linux/kvm.h> |
749cf76c CD |
29 | #include <trace/events/kvm.h> |
30 | ||
31 | #define CREATE_TRACE_POINTS | |
32 | #include "trace.h" | |
33 | ||
749cf76c CD |
34 | #include <asm/uaccess.h> |
35 | #include <asm/ptrace.h> | |
36 | #include <asm/mman.h> | |
342cd0ab | 37 | #include <asm/tlbflush.h> |
5b3e5e5b | 38 | #include <asm/cacheflush.h> |
342cd0ab CD |
39 | #include <asm/virt.h> |
40 | #include <asm/kvm_arm.h> | |
41 | #include <asm/kvm_asm.h> | |
42 | #include <asm/kvm_mmu.h> | |
f7ed45be | 43 | #include <asm/kvm_emulate.h> |
5b3e5e5b | 44 | #include <asm/kvm_coproc.h> |
aa024c2f | 45 | #include <asm/kvm_psci.h> |
749cf76c CD |
46 | |
47 | #ifdef REQUIRES_VIRT | |
48 | __asm__(".arch_extension virt"); | |
49 | #endif | |
50 | ||
342cd0ab | 51 | static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); |
3de50da6 | 52 | static kvm_cpu_context_t __percpu *kvm_host_cpu_state; |
342cd0ab CD |
53 | static unsigned long hyp_default_vectors; |
54 | ||
1638a12d MZ |
55 | /* Per-CPU variable containing the currently running vcpu. */ |
56 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu); | |
57 | ||
f7ed45be CD |
58 | /* The VMID used in the VTTBR */ |
59 | static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1); | |
60 | static u8 kvm_next_vmid; | |
61 | static DEFINE_SPINLOCK(kvm_vmid_lock); | |
342cd0ab | 62 | |
1a89dd91 MZ |
63 | static bool vgic_present; |
64 | ||
1638a12d MZ |
65 | static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu) |
66 | { | |
67 | BUG_ON(preemptible()); | |
1436c1aa | 68 | __this_cpu_write(kvm_arm_running_vcpu, vcpu); |
1638a12d MZ |
69 | } |
70 | ||
71 | /** | |
72 | * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU. | |
73 | * Must be called from non-preemptible context | |
74 | */ | |
75 | struct kvm_vcpu *kvm_arm_get_running_vcpu(void) | |
76 | { | |
77 | BUG_ON(preemptible()); | |
1436c1aa | 78 | return __this_cpu_read(kvm_arm_running_vcpu); |
1638a12d MZ |
79 | } |
80 | ||
81 | /** | |
82 | * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
83 | */ | |
84 | struct kvm_vcpu __percpu **kvm_get_running_vcpus(void) | |
85 | { | |
86 | return &kvm_arm_running_vcpu; | |
87 | } | |
88 | ||
749cf76c CD |
89 | int kvm_arch_hardware_enable(void *garbage) |
90 | { | |
91 | return 0; | |
92 | } | |
93 | ||
94 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) | |
95 | { | |
96 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; | |
97 | } | |
98 | ||
99 | void kvm_arch_hardware_disable(void *garbage) | |
100 | { | |
101 | } | |
102 | ||
103 | int kvm_arch_hardware_setup(void) | |
104 | { | |
105 | return 0; | |
106 | } | |
107 | ||
108 | void kvm_arch_hardware_unsetup(void) | |
109 | { | |
110 | } | |
111 | ||
112 | void kvm_arch_check_processor_compat(void *rtn) | |
113 | { | |
114 | *(int *)rtn = 0; | |
115 | } | |
116 | ||
117 | void kvm_arch_sync_events(struct kvm *kvm) | |
118 | { | |
119 | } | |
120 | ||
d5d8184d CD |
121 | /** |
122 | * kvm_arch_init_vm - initializes a VM data structure | |
123 | * @kvm: pointer to the KVM struct | |
124 | */ | |
749cf76c CD |
125 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
126 | { | |
d5d8184d CD |
127 | int ret = 0; |
128 | ||
749cf76c CD |
129 | if (type) |
130 | return -EINVAL; | |
131 | ||
d5d8184d CD |
132 | ret = kvm_alloc_stage2_pgd(kvm); |
133 | if (ret) | |
134 | goto out_fail_alloc; | |
135 | ||
136 | ret = create_hyp_mappings(kvm, kvm + 1); | |
137 | if (ret) | |
138 | goto out_free_stage2_pgd; | |
139 | ||
a1a64387 CD |
140 | kvm_timer_init(kvm); |
141 | ||
d5d8184d CD |
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 | ||
5587027c | 157 | void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, |
749cf76c CD |
158 | struct kvm_memory_slot *dont) |
159 | { | |
160 | } | |
161 | ||
5587027c AK |
162 | int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, |
163 | unsigned long npages) | |
749cf76c CD |
164 | { |
165 | return 0; | |
166 | } | |
167 | ||
d5d8184d CD |
168 | /** |
169 | * kvm_arch_destroy_vm - destroy the VM data structure | |
170 | * @kvm: pointer to the KVM struct | |
171 | */ | |
749cf76c CD |
172 | void kvm_arch_destroy_vm(struct kvm *kvm) |
173 | { | |
174 | int i; | |
175 | ||
d5d8184d CD |
176 | kvm_free_stage2_pgd(kvm); |
177 | ||
749cf76c CD |
178 | for (i = 0; i < KVM_MAX_VCPUS; ++i) { |
179 | if (kvm->vcpus[i]) { | |
180 | kvm_arch_vcpu_free(kvm->vcpus[i]); | |
181 | kvm->vcpus[i] = NULL; | |
182 | } | |
183 | } | |
184 | } | |
185 | ||
186 | int kvm_dev_ioctl_check_extension(long ext) | |
187 | { | |
188 | int r; | |
189 | switch (ext) { | |
1a89dd91 MZ |
190 | case KVM_CAP_IRQCHIP: |
191 | r = vgic_present; | |
192 | break; | |
7330672b | 193 | case KVM_CAP_DEVICE_CTRL: |
749cf76c CD |
194 | case KVM_CAP_USER_MEMORY: |
195 | case KVM_CAP_SYNC_MMU: | |
196 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
197 | case KVM_CAP_ONE_REG: | |
aa024c2f | 198 | case KVM_CAP_ARM_PSCI: |
749cf76c CD |
199 | r = 1; |
200 | break; | |
201 | case KVM_CAP_COALESCED_MMIO: | |
202 | r = KVM_COALESCED_MMIO_PAGE_OFFSET; | |
203 | break; | |
3401d546 CD |
204 | case KVM_CAP_ARM_SET_DEVICE_ADDR: |
205 | r = 1; | |
ca46e10f | 206 | break; |
749cf76c CD |
207 | case KVM_CAP_NR_VCPUS: |
208 | r = num_online_cpus(); | |
209 | break; | |
210 | case KVM_CAP_MAX_VCPUS: | |
211 | r = KVM_MAX_VCPUS; | |
212 | break; | |
213 | default: | |
17b1e31f | 214 | r = kvm_arch_dev_ioctl_check_extension(ext); |
749cf76c CD |
215 | break; |
216 | } | |
217 | return r; | |
218 | } | |
219 | ||
220 | long kvm_arch_dev_ioctl(struct file *filp, | |
221 | unsigned int ioctl, unsigned long arg) | |
222 | { | |
223 | return -EINVAL; | |
224 | } | |
225 | ||
e59dbe09 TY |
226 | void kvm_arch_memslots_updated(struct kvm *kvm) |
227 | { | |
228 | } | |
229 | ||
749cf76c CD |
230 | int kvm_arch_prepare_memory_region(struct kvm *kvm, |
231 | struct kvm_memory_slot *memslot, | |
749cf76c | 232 | struct kvm_userspace_memory_region *mem, |
7b6195a9 | 233 | enum kvm_mr_change change) |
749cf76c CD |
234 | { |
235 | return 0; | |
236 | } | |
237 | ||
238 | void kvm_arch_commit_memory_region(struct kvm *kvm, | |
239 | struct kvm_userspace_memory_region *mem, | |
8482644a TY |
240 | const struct kvm_memory_slot *old, |
241 | enum kvm_mr_change change) | |
749cf76c CD |
242 | { |
243 | } | |
244 | ||
245 | void kvm_arch_flush_shadow_all(struct kvm *kvm) | |
246 | { | |
247 | } | |
248 | ||
249 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, | |
250 | struct kvm_memory_slot *slot) | |
251 | { | |
252 | } | |
253 | ||
254 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) | |
255 | { | |
256 | int err; | |
257 | struct kvm_vcpu *vcpu; | |
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 | ||
d5d8184d CD |
269 | err = create_hyp_mappings(vcpu, vcpu + 1); |
270 | if (err) | |
271 | goto vcpu_uninit; | |
272 | ||
749cf76c | 273 | return vcpu; |
d5d8184d CD |
274 | vcpu_uninit: |
275 | kvm_vcpu_uninit(vcpu); | |
749cf76c CD |
276 | free_vcpu: |
277 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
278 | out: | |
279 | return ERR_PTR(err); | |
280 | } | |
281 | ||
282 | int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) | |
283 | { | |
284 | return 0; | |
285 | } | |
286 | ||
287 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
288 | { | |
d5d8184d | 289 | kvm_mmu_free_memory_caches(vcpu); |
967f8427 | 290 | kvm_timer_vcpu_terminate(vcpu); |
d5d8184d | 291 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
749cf76c CD |
292 | } |
293 | ||
294 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) | |
295 | { | |
296 | kvm_arch_vcpu_free(vcpu); | |
297 | } | |
298 | ||
299 | int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) | |
300 | { | |
301 | return 0; | |
302 | } | |
303 | ||
749cf76c CD |
304 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
305 | { | |
1a89dd91 MZ |
306 | int ret; |
307 | ||
f7ed45be CD |
308 | /* Force users to call KVM_ARM_VCPU_INIT */ |
309 | vcpu->arch.target = -1; | |
1a89dd91 MZ |
310 | |
311 | /* Set up VGIC */ | |
312 | ret = kvm_vgic_vcpu_init(vcpu); | |
313 | if (ret) | |
314 | return ret; | |
315 | ||
967f8427 MZ |
316 | /* Set up the timer */ |
317 | kvm_timer_vcpu_init(vcpu); | |
318 | ||
749cf76c CD |
319 | return 0; |
320 | } | |
321 | ||
322 | void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) | |
323 | { | |
324 | } | |
325 | ||
326 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) | |
327 | { | |
86ce8535 | 328 | vcpu->cpu = cpu; |
3de50da6 | 329 | vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state); |
5b3e5e5b CD |
330 | |
331 | /* | |
332 | * Check whether this vcpu requires the cache to be flushed on | |
333 | * this physical CPU. This is a consequence of doing dcache | |
334 | * operations by set/way on this vcpu. We do it here to be in | |
335 | * a non-preemptible section. | |
336 | */ | |
337 | if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush)) | |
338 | flush_cache_all(); /* We'd really want v7_flush_dcache_all() */ | |
1638a12d MZ |
339 | |
340 | kvm_arm_set_running_vcpu(vcpu); | |
749cf76c CD |
341 | } |
342 | ||
343 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
344 | { | |
1638a12d | 345 | kvm_arm_set_running_vcpu(NULL); |
749cf76c CD |
346 | } |
347 | ||
348 | int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, | |
349 | struct kvm_guest_debug *dbg) | |
350 | { | |
351 | return -EINVAL; | |
352 | } | |
353 | ||
354 | ||
355 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, | |
356 | struct kvm_mp_state *mp_state) | |
357 | { | |
358 | return -EINVAL; | |
359 | } | |
360 | ||
361 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
362 | struct kvm_mp_state *mp_state) | |
363 | { | |
364 | return -EINVAL; | |
365 | } | |
366 | ||
5b3e5e5b CD |
367 | /** |
368 | * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled | |
369 | * @v: The VCPU pointer | |
370 | * | |
371 | * If the guest CPU is not waiting for interrupts or an interrupt line is | |
372 | * asserted, the CPU is by definition runnable. | |
373 | */ | |
749cf76c CD |
374 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) |
375 | { | |
1a89dd91 | 376 | return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v); |
749cf76c CD |
377 | } |
378 | ||
f7ed45be CD |
379 | /* Just ensure a guest exit from a particular CPU */ |
380 | static void exit_vm_noop(void *info) | |
381 | { | |
382 | } | |
383 | ||
384 | void force_vm_exit(const cpumask_t *mask) | |
385 | { | |
386 | smp_call_function_many(mask, exit_vm_noop, NULL, true); | |
387 | } | |
388 | ||
389 | /** | |
390 | * need_new_vmid_gen - check that the VMID is still valid | |
391 | * @kvm: The VM's VMID to checkt | |
392 | * | |
393 | * return true if there is a new generation of VMIDs being used | |
394 | * | |
395 | * The hardware supports only 256 values with the value zero reserved for the | |
396 | * host, so we check if an assigned value belongs to a previous generation, | |
397 | * which which requires us to assign a new value. If we're the first to use a | |
398 | * VMID for the new generation, we must flush necessary caches and TLBs on all | |
399 | * CPUs. | |
400 | */ | |
401 | static bool need_new_vmid_gen(struct kvm *kvm) | |
402 | { | |
403 | return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen)); | |
404 | } | |
405 | ||
406 | /** | |
407 | * update_vttbr - Update the VTTBR with a valid VMID before the guest runs | |
408 | * @kvm The guest that we are about to run | |
409 | * | |
410 | * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the | |
411 | * VM has a valid VMID, otherwise assigns a new one and flushes corresponding | |
412 | * caches and TLBs. | |
413 | */ | |
414 | static void update_vttbr(struct kvm *kvm) | |
415 | { | |
416 | phys_addr_t pgd_phys; | |
417 | u64 vmid; | |
418 | ||
419 | if (!need_new_vmid_gen(kvm)) | |
420 | return; | |
421 | ||
422 | spin_lock(&kvm_vmid_lock); | |
423 | ||
424 | /* | |
425 | * We need to re-check the vmid_gen here to ensure that if another vcpu | |
426 | * already allocated a valid vmid for this vm, then this vcpu should | |
427 | * use the same vmid. | |
428 | */ | |
429 | if (!need_new_vmid_gen(kvm)) { | |
430 | spin_unlock(&kvm_vmid_lock); | |
431 | return; | |
432 | } | |
433 | ||
434 | /* First user of a new VMID generation? */ | |
435 | if (unlikely(kvm_next_vmid == 0)) { | |
436 | atomic64_inc(&kvm_vmid_gen); | |
437 | kvm_next_vmid = 1; | |
438 | ||
439 | /* | |
440 | * On SMP we know no other CPUs can use this CPU's or each | |
441 | * other's VMID after force_vm_exit returns since the | |
442 | * kvm_vmid_lock blocks them from reentry to the guest. | |
443 | */ | |
444 | force_vm_exit(cpu_all_mask); | |
445 | /* | |
446 | * Now broadcast TLB + ICACHE invalidation over the inner | |
447 | * shareable domain to make sure all data structures are | |
448 | * clean. | |
449 | */ | |
450 | kvm_call_hyp(__kvm_flush_vm_context); | |
451 | } | |
452 | ||
453 | kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen); | |
454 | kvm->arch.vmid = kvm_next_vmid; | |
455 | kvm_next_vmid++; | |
456 | ||
457 | /* update vttbr to be used with the new vmid */ | |
458 | pgd_phys = virt_to_phys(kvm->arch.pgd); | |
459 | vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK; | |
460 | kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK; | |
461 | kvm->arch.vttbr |= vmid; | |
462 | ||
463 | spin_unlock(&kvm_vmid_lock); | |
464 | } | |
465 | ||
f7ed45be CD |
466 | static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) |
467 | { | |
e1ba0207 CD |
468 | int ret; |
469 | ||
f7ed45be CD |
470 | if (likely(vcpu->arch.has_run_once)) |
471 | return 0; | |
472 | ||
473 | vcpu->arch.has_run_once = true; | |
aa024c2f | 474 | |
01ac5e34 MZ |
475 | /* |
476 | * Initialize the VGIC before running a vcpu the first time on | |
477 | * this VM. | |
478 | */ | |
e1ba0207 CD |
479 | if (unlikely(!vgic_initialized(vcpu->kvm))) { |
480 | ret = kvm_vgic_init(vcpu->kvm); | |
01ac5e34 MZ |
481 | if (ret) |
482 | return ret; | |
483 | } | |
484 | ||
aa024c2f MZ |
485 | /* |
486 | * Handle the "start in power-off" case by calling into the | |
487 | * PSCI code. | |
488 | */ | |
489 | if (test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) { | |
490 | *vcpu_reg(vcpu, 0) = KVM_PSCI_FN_CPU_OFF; | |
491 | kvm_psci_call(vcpu); | |
492 | } | |
493 | ||
f7ed45be CD |
494 | return 0; |
495 | } | |
496 | ||
aa024c2f MZ |
497 | static void vcpu_pause(struct kvm_vcpu *vcpu) |
498 | { | |
499 | wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu); | |
500 | ||
501 | wait_event_interruptible(*wq, !vcpu->arch.pause); | |
502 | } | |
503 | ||
e8180dca AP |
504 | static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) |
505 | { | |
506 | return vcpu->arch.target >= 0; | |
507 | } | |
508 | ||
f7ed45be CD |
509 | /** |
510 | * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code | |
511 | * @vcpu: The VCPU pointer | |
512 | * @run: The kvm_run structure pointer used for userspace state exchange | |
513 | * | |
514 | * This function is called through the VCPU_RUN ioctl called from user space. It | |
515 | * will execute VM code in a loop until the time slice for the process is used | |
516 | * or some emulation is needed from user space in which case the function will | |
517 | * return with return value 0 and with the kvm_run structure filled in with the | |
518 | * required data for the requested emulation. | |
519 | */ | |
749cf76c CD |
520 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) |
521 | { | |
f7ed45be CD |
522 | int ret; |
523 | sigset_t sigsaved; | |
524 | ||
e8180dca | 525 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
f7ed45be CD |
526 | return -ENOEXEC; |
527 | ||
528 | ret = kvm_vcpu_first_run_init(vcpu); | |
529 | if (ret) | |
530 | return ret; | |
531 | ||
45e96ea6 CD |
532 | if (run->exit_reason == KVM_EXIT_MMIO) { |
533 | ret = kvm_handle_mmio_return(vcpu, vcpu->run); | |
534 | if (ret) | |
535 | return ret; | |
536 | } | |
537 | ||
f7ed45be CD |
538 | if (vcpu->sigset_active) |
539 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
540 | ||
541 | ret = 1; | |
542 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
543 | while (ret > 0) { | |
544 | /* | |
545 | * Check conditions before entering the guest | |
546 | */ | |
547 | cond_resched(); | |
548 | ||
549 | update_vttbr(vcpu->kvm); | |
550 | ||
aa024c2f MZ |
551 | if (vcpu->arch.pause) |
552 | vcpu_pause(vcpu); | |
553 | ||
1a89dd91 | 554 | kvm_vgic_flush_hwstate(vcpu); |
c7e3ba64 | 555 | kvm_timer_flush_hwstate(vcpu); |
1a89dd91 | 556 | |
f7ed45be CD |
557 | local_irq_disable(); |
558 | ||
559 | /* | |
560 | * Re-check atomic conditions | |
561 | */ | |
562 | if (signal_pending(current)) { | |
563 | ret = -EINTR; | |
564 | run->exit_reason = KVM_EXIT_INTR; | |
565 | } | |
566 | ||
567 | if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) { | |
568 | local_irq_enable(); | |
c7e3ba64 | 569 | kvm_timer_sync_hwstate(vcpu); |
1a89dd91 | 570 | kvm_vgic_sync_hwstate(vcpu); |
f7ed45be CD |
571 | continue; |
572 | } | |
573 | ||
574 | /************************************************************** | |
575 | * Enter the guest | |
576 | */ | |
577 | trace_kvm_entry(*vcpu_pc(vcpu)); | |
578 | kvm_guest_enter(); | |
579 | vcpu->mode = IN_GUEST_MODE; | |
580 | ||
581 | ret = kvm_call_hyp(__kvm_vcpu_run, vcpu); | |
582 | ||
583 | vcpu->mode = OUTSIDE_GUEST_MODE; | |
5b3e5e5b | 584 | vcpu->arch.last_pcpu = smp_processor_id(); |
f7ed45be CD |
585 | kvm_guest_exit(); |
586 | trace_kvm_exit(*vcpu_pc(vcpu)); | |
587 | /* | |
588 | * We may have taken a host interrupt in HYP mode (ie | |
589 | * while executing the guest). This interrupt is still | |
590 | * pending, as we haven't serviced it yet! | |
591 | * | |
592 | * We're now back in SVC mode, with interrupts | |
593 | * disabled. Enabling the interrupts now will have | |
594 | * the effect of taking the interrupt again, in SVC | |
595 | * mode this time. | |
596 | */ | |
597 | local_irq_enable(); | |
598 | ||
599 | /* | |
600 | * Back from guest | |
601 | *************************************************************/ | |
602 | ||
c7e3ba64 | 603 | kvm_timer_sync_hwstate(vcpu); |
1a89dd91 MZ |
604 | kvm_vgic_sync_hwstate(vcpu); |
605 | ||
f7ed45be CD |
606 | ret = handle_exit(vcpu, run, ret); |
607 | } | |
608 | ||
609 | if (vcpu->sigset_active) | |
610 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
611 | return ret; | |
749cf76c CD |
612 | } |
613 | ||
86ce8535 CD |
614 | static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) |
615 | { | |
616 | int bit_index; | |
617 | bool set; | |
618 | unsigned long *ptr; | |
619 | ||
620 | if (number == KVM_ARM_IRQ_CPU_IRQ) | |
621 | bit_index = __ffs(HCR_VI); | |
622 | else /* KVM_ARM_IRQ_CPU_FIQ */ | |
623 | bit_index = __ffs(HCR_VF); | |
624 | ||
625 | ptr = (unsigned long *)&vcpu->arch.irq_lines; | |
626 | if (level) | |
627 | set = test_and_set_bit(bit_index, ptr); | |
628 | else | |
629 | set = test_and_clear_bit(bit_index, ptr); | |
630 | ||
631 | /* | |
632 | * If we didn't change anything, no need to wake up or kick other CPUs | |
633 | */ | |
634 | if (set == level) | |
635 | return 0; | |
636 | ||
637 | /* | |
638 | * The vcpu irq_lines field was updated, wake up sleeping VCPUs and | |
639 | * trigger a world-switch round on the running physical CPU to set the | |
640 | * virtual IRQ/FIQ fields in the HCR appropriately. | |
641 | */ | |
642 | kvm_vcpu_kick(vcpu); | |
643 | ||
644 | return 0; | |
645 | } | |
646 | ||
79558f11 AG |
647 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, |
648 | bool line_status) | |
86ce8535 CD |
649 | { |
650 | u32 irq = irq_level->irq; | |
651 | unsigned int irq_type, vcpu_idx, irq_num; | |
652 | int nrcpus = atomic_read(&kvm->online_vcpus); | |
653 | struct kvm_vcpu *vcpu = NULL; | |
654 | bool level = irq_level->level; | |
655 | ||
656 | irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; | |
657 | vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; | |
658 | irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; | |
659 | ||
660 | trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); | |
661 | ||
5863c2ce MZ |
662 | switch (irq_type) { |
663 | case KVM_ARM_IRQ_TYPE_CPU: | |
664 | if (irqchip_in_kernel(kvm)) | |
665 | return -ENXIO; | |
86ce8535 | 666 | |
5863c2ce MZ |
667 | if (vcpu_idx >= nrcpus) |
668 | return -EINVAL; | |
86ce8535 | 669 | |
5863c2ce MZ |
670 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); |
671 | if (!vcpu) | |
672 | return -EINVAL; | |
86ce8535 | 673 | |
5863c2ce MZ |
674 | if (irq_num > KVM_ARM_IRQ_CPU_FIQ) |
675 | return -EINVAL; | |
676 | ||
677 | return vcpu_interrupt_line(vcpu, irq_num, level); | |
678 | case KVM_ARM_IRQ_TYPE_PPI: | |
679 | if (!irqchip_in_kernel(kvm)) | |
680 | return -ENXIO; | |
681 | ||
682 | if (vcpu_idx >= nrcpus) | |
683 | return -EINVAL; | |
684 | ||
685 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); | |
686 | if (!vcpu) | |
687 | return -EINVAL; | |
688 | ||
689 | if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) | |
690 | return -EINVAL; | |
86ce8535 | 691 | |
5863c2ce MZ |
692 | return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level); |
693 | case KVM_ARM_IRQ_TYPE_SPI: | |
694 | if (!irqchip_in_kernel(kvm)) | |
695 | return -ENXIO; | |
696 | ||
697 | if (irq_num < VGIC_NR_PRIVATE_IRQS || | |
698 | irq_num > KVM_ARM_IRQ_GIC_MAX) | |
699 | return -EINVAL; | |
700 | ||
701 | return kvm_vgic_inject_irq(kvm, 0, irq_num, level); | |
702 | } | |
703 | ||
704 | return -EINVAL; | |
86ce8535 CD |
705 | } |
706 | ||
749cf76c CD |
707 | long kvm_arch_vcpu_ioctl(struct file *filp, |
708 | unsigned int ioctl, unsigned long arg) | |
709 | { | |
710 | struct kvm_vcpu *vcpu = filp->private_data; | |
711 | void __user *argp = (void __user *)arg; | |
712 | ||
713 | switch (ioctl) { | |
714 | case KVM_ARM_VCPU_INIT: { | |
715 | struct kvm_vcpu_init init; | |
716 | ||
717 | if (copy_from_user(&init, argp, sizeof(init))) | |
718 | return -EFAULT; | |
719 | ||
720 | return kvm_vcpu_set_target(vcpu, &init); | |
721 | ||
722 | } | |
723 | case KVM_SET_ONE_REG: | |
724 | case KVM_GET_ONE_REG: { | |
725 | struct kvm_one_reg reg; | |
e8180dca AP |
726 | |
727 | if (unlikely(!kvm_vcpu_initialized(vcpu))) | |
728 | return -ENOEXEC; | |
729 | ||
749cf76c CD |
730 | if (copy_from_user(®, argp, sizeof(reg))) |
731 | return -EFAULT; | |
732 | if (ioctl == KVM_SET_ONE_REG) | |
733 | return kvm_arm_set_reg(vcpu, ®); | |
734 | else | |
735 | return kvm_arm_get_reg(vcpu, ®); | |
736 | } | |
737 | case KVM_GET_REG_LIST: { | |
738 | struct kvm_reg_list __user *user_list = argp; | |
739 | struct kvm_reg_list reg_list; | |
740 | unsigned n; | |
741 | ||
e8180dca AP |
742 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
743 | return -ENOEXEC; | |
744 | ||
749cf76c CD |
745 | if (copy_from_user(®_list, user_list, sizeof(reg_list))) |
746 | return -EFAULT; | |
747 | n = reg_list.n; | |
748 | reg_list.n = kvm_arm_num_regs(vcpu); | |
749 | if (copy_to_user(user_list, ®_list, sizeof(reg_list))) | |
750 | return -EFAULT; | |
751 | if (n < reg_list.n) | |
752 | return -E2BIG; | |
753 | return kvm_arm_copy_reg_indices(vcpu, user_list->reg); | |
754 | } | |
755 | default: | |
756 | return -EINVAL; | |
757 | } | |
758 | } | |
759 | ||
760 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) | |
761 | { | |
762 | return -EINVAL; | |
763 | } | |
764 | ||
3401d546 CD |
765 | static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, |
766 | struct kvm_arm_device_addr *dev_addr) | |
767 | { | |
330690cd CD |
768 | unsigned long dev_id, type; |
769 | ||
770 | dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> | |
771 | KVM_ARM_DEVICE_ID_SHIFT; | |
772 | type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> | |
773 | KVM_ARM_DEVICE_TYPE_SHIFT; | |
774 | ||
775 | switch (dev_id) { | |
776 | case KVM_ARM_DEVICE_VGIC_V2: | |
777 | if (!vgic_present) | |
778 | return -ENXIO; | |
ce01e4e8 | 779 | return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); |
330690cd CD |
780 | default: |
781 | return -ENODEV; | |
782 | } | |
3401d546 CD |
783 | } |
784 | ||
749cf76c CD |
785 | long kvm_arch_vm_ioctl(struct file *filp, |
786 | unsigned int ioctl, unsigned long arg) | |
787 | { | |
3401d546 CD |
788 | struct kvm *kvm = filp->private_data; |
789 | void __user *argp = (void __user *)arg; | |
790 | ||
791 | switch (ioctl) { | |
5863c2ce MZ |
792 | case KVM_CREATE_IRQCHIP: { |
793 | if (vgic_present) | |
794 | return kvm_vgic_create(kvm); | |
795 | else | |
796 | return -ENXIO; | |
797 | } | |
3401d546 CD |
798 | case KVM_ARM_SET_DEVICE_ADDR: { |
799 | struct kvm_arm_device_addr dev_addr; | |
800 | ||
801 | if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) | |
802 | return -EFAULT; | |
803 | return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); | |
804 | } | |
42c4e0c7 AP |
805 | case KVM_ARM_PREFERRED_TARGET: { |
806 | int err; | |
807 | struct kvm_vcpu_init init; | |
808 | ||
809 | err = kvm_vcpu_preferred_target(&init); | |
810 | if (err) | |
811 | return err; | |
812 | ||
813 | if (copy_to_user(argp, &init, sizeof(init))) | |
814 | return -EFAULT; | |
815 | ||
816 | return 0; | |
817 | } | |
3401d546 CD |
818 | default: |
819 | return -EINVAL; | |
820 | } | |
749cf76c CD |
821 | } |
822 | ||
d157f4a5 | 823 | static void cpu_init_hyp_mode(void *dummy) |
342cd0ab | 824 | { |
dac288f7 MZ |
825 | phys_addr_t boot_pgd_ptr; |
826 | phys_addr_t pgd_ptr; | |
342cd0ab CD |
827 | unsigned long hyp_stack_ptr; |
828 | unsigned long stack_page; | |
829 | unsigned long vector_ptr; | |
830 | ||
831 | /* Switch from the HYP stub to our own HYP init vector */ | |
5a677ce0 | 832 | __hyp_set_vectors(kvm_get_idmap_vector()); |
342cd0ab | 833 | |
dac288f7 MZ |
834 | boot_pgd_ptr = kvm_mmu_get_boot_httbr(); |
835 | pgd_ptr = kvm_mmu_get_httbr(); | |
1436c1aa | 836 | stack_page = __this_cpu_read(kvm_arm_hyp_stack_page); |
342cd0ab CD |
837 | hyp_stack_ptr = stack_page + PAGE_SIZE; |
838 | vector_ptr = (unsigned long)__kvm_hyp_vector; | |
839 | ||
5a677ce0 | 840 | __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr); |
342cd0ab CD |
841 | } |
842 | ||
d157f4a5 MZ |
843 | static int hyp_init_cpu_notify(struct notifier_block *self, |
844 | unsigned long action, void *cpu) | |
845 | { | |
846 | switch (action) { | |
847 | case CPU_STARTING: | |
848 | case CPU_STARTING_FROZEN: | |
849 | cpu_init_hyp_mode(NULL); | |
850 | break; | |
851 | } | |
852 | ||
853 | return NOTIFY_OK; | |
342cd0ab CD |
854 | } |
855 | ||
d157f4a5 MZ |
856 | static struct notifier_block hyp_init_cpu_nb = { |
857 | .notifier_call = hyp_init_cpu_notify, | |
858 | }; | |
859 | ||
342cd0ab CD |
860 | /** |
861 | * Inits Hyp-mode on all online CPUs | |
862 | */ | |
863 | static int init_hyp_mode(void) | |
864 | { | |
342cd0ab CD |
865 | int cpu; |
866 | int err = 0; | |
867 | ||
868 | /* | |
869 | * Allocate Hyp PGD and setup Hyp identity mapping | |
870 | */ | |
871 | err = kvm_mmu_init(); | |
872 | if (err) | |
873 | goto out_err; | |
874 | ||
875 | /* | |
876 | * It is probably enough to obtain the default on one | |
877 | * CPU. It's unlikely to be different on the others. | |
878 | */ | |
879 | hyp_default_vectors = __hyp_get_vectors(); | |
880 | ||
881 | /* | |
882 | * Allocate stack pages for Hypervisor-mode | |
883 | */ | |
884 | for_each_possible_cpu(cpu) { | |
885 | unsigned long stack_page; | |
886 | ||
887 | stack_page = __get_free_page(GFP_KERNEL); | |
888 | if (!stack_page) { | |
889 | err = -ENOMEM; | |
890 | goto out_free_stack_pages; | |
891 | } | |
892 | ||
893 | per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; | |
894 | } | |
895 | ||
342cd0ab CD |
896 | /* |
897 | * Map the Hyp-code called directly from the host | |
898 | */ | |
899 | err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end); | |
900 | if (err) { | |
901 | kvm_err("Cannot map world-switch code\n"); | |
902 | goto out_free_mappings; | |
903 | } | |
904 | ||
905 | /* | |
906 | * Map the Hyp stack pages | |
907 | */ | |
908 | for_each_possible_cpu(cpu) { | |
909 | char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); | |
910 | err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE); | |
911 | ||
912 | if (err) { | |
913 | kvm_err("Cannot map hyp stack\n"); | |
914 | goto out_free_mappings; | |
915 | } | |
916 | } | |
917 | ||
918 | /* | |
3de50da6 | 919 | * Map the host CPU structures |
342cd0ab | 920 | */ |
3de50da6 MZ |
921 | kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t); |
922 | if (!kvm_host_cpu_state) { | |
342cd0ab | 923 | err = -ENOMEM; |
3de50da6 | 924 | kvm_err("Cannot allocate host CPU state\n"); |
342cd0ab CD |
925 | goto out_free_mappings; |
926 | } | |
927 | ||
928 | for_each_possible_cpu(cpu) { | |
3de50da6 | 929 | kvm_cpu_context_t *cpu_ctxt; |
342cd0ab | 930 | |
3de50da6 MZ |
931 | cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu); |
932 | err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1); | |
342cd0ab CD |
933 | |
934 | if (err) { | |
3de50da6 MZ |
935 | kvm_err("Cannot map host CPU state: %d\n", err); |
936 | goto out_free_context; | |
342cd0ab CD |
937 | } |
938 | } | |
939 | ||
d157f4a5 MZ |
940 | /* |
941 | * Execute the init code on each CPU. | |
942 | */ | |
943 | on_each_cpu(cpu_init_hyp_mode, NULL, 1); | |
944 | ||
1a89dd91 MZ |
945 | /* |
946 | * Init HYP view of VGIC | |
947 | */ | |
948 | err = kvm_vgic_hyp_init(); | |
949 | if (err) | |
3de50da6 | 950 | goto out_free_context; |
1a89dd91 | 951 | |
01ac5e34 MZ |
952 | #ifdef CONFIG_KVM_ARM_VGIC |
953 | vgic_present = true; | |
954 | #endif | |
955 | ||
967f8427 MZ |
956 | /* |
957 | * Init HYP architected timer support | |
958 | */ | |
959 | err = kvm_timer_hyp_init(); | |
960 | if (err) | |
961 | goto out_free_mappings; | |
962 | ||
d157f4a5 MZ |
963 | #ifndef CONFIG_HOTPLUG_CPU |
964 | free_boot_hyp_pgd(); | |
965 | #endif | |
966 | ||
210552c1 MZ |
967 | kvm_perf_init(); |
968 | ||
342cd0ab | 969 | kvm_info("Hyp mode initialized successfully\n"); |
210552c1 | 970 | |
342cd0ab | 971 | return 0; |
3de50da6 MZ |
972 | out_free_context: |
973 | free_percpu(kvm_host_cpu_state); | |
342cd0ab | 974 | out_free_mappings: |
4f728276 | 975 | free_hyp_pgds(); |
342cd0ab CD |
976 | out_free_stack_pages: |
977 | for_each_possible_cpu(cpu) | |
978 | free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); | |
979 | out_err: | |
980 | kvm_err("error initializing Hyp mode: %d\n", err); | |
981 | return err; | |
982 | } | |
983 | ||
d4e071ce AP |
984 | static void check_kvm_target_cpu(void *ret) |
985 | { | |
986 | *(int *)ret = kvm_target_cpu(); | |
987 | } | |
988 | ||
342cd0ab CD |
989 | /** |
990 | * Initialize Hyp-mode and memory mappings on all CPUs. | |
991 | */ | |
749cf76c CD |
992 | int kvm_arch_init(void *opaque) |
993 | { | |
342cd0ab | 994 | int err; |
d4e071ce | 995 | int ret, cpu; |
342cd0ab CD |
996 | |
997 | if (!is_hyp_mode_available()) { | |
998 | kvm_err("HYP mode not available\n"); | |
999 | return -ENODEV; | |
1000 | } | |
1001 | ||
d4e071ce AP |
1002 | for_each_online_cpu(cpu) { |
1003 | smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); | |
1004 | if (ret < 0) { | |
1005 | kvm_err("Error, CPU %d not supported!\n", cpu); | |
1006 | return -ENODEV; | |
1007 | } | |
342cd0ab CD |
1008 | } |
1009 | ||
1010 | err = init_hyp_mode(); | |
1011 | if (err) | |
1012 | goto out_err; | |
1013 | ||
d157f4a5 MZ |
1014 | err = register_cpu_notifier(&hyp_init_cpu_nb); |
1015 | if (err) { | |
1016 | kvm_err("Cannot register HYP init CPU notifier (%d)\n", err); | |
1017 | goto out_err; | |
1018 | } | |
1019 | ||
5b3e5e5b | 1020 | kvm_coproc_table_init(); |
749cf76c | 1021 | return 0; |
342cd0ab CD |
1022 | out_err: |
1023 | return err; | |
749cf76c CD |
1024 | } |
1025 | ||
1026 | /* NOP: Compiling as a module not supported */ | |
1027 | void kvm_arch_exit(void) | |
1028 | { | |
210552c1 | 1029 | kvm_perf_teardown(); |
749cf76c CD |
1030 | } |
1031 | ||
1032 | static int arm_init(void) | |
1033 | { | |
1034 | int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1035 | return rc; | |
1036 | } | |
1037 | ||
1038 | module_init(arm_init); |