]>
Commit | Line | Data |
---|---|---|
d94d71cb | 1 | // SPDX-License-Identifier: GPL-2.0-only |
749cf76c CD |
2 | /* |
3 | * Copyright (C) 2012 - Virtual Open Systems and Columbia University | |
4 | * Author: Christoffer Dall <c.dall@virtualopensystems.com> | |
749cf76c CD |
5 | */ |
6 | ||
85acda3b | 7 | #include <linux/bug.h> |
1fcf7ce0 | 8 | #include <linux/cpu_pm.h> |
749cf76c CD |
9 | #include <linux/errno.h> |
10 | #include <linux/err.h> | |
11 | #include <linux/kvm_host.h> | |
1085fdc6 | 12 | #include <linux/list.h> |
749cf76c CD |
13 | #include <linux/module.h> |
14 | #include <linux/vmalloc.h> | |
15 | #include <linux/fs.h> | |
16 | #include <linux/mman.h> | |
17 | #include <linux/sched.h> | |
86ce8535 | 18 | #include <linux/kvm.h> |
2412405b EA |
19 | #include <linux/kvm_irqfd.h> |
20 | #include <linux/irqbypass.h> | |
de737089 | 21 | #include <linux/sched/stat.h> |
749cf76c | 22 | #include <trace/events/kvm.h> |
b02386eb | 23 | #include <kvm/arm_pmu.h> |
1a2fb94e | 24 | #include <kvm/arm_psci.h> |
749cf76c CD |
25 | |
26 | #define CREATE_TRACE_POINTS | |
27 | #include "trace.h" | |
28 | ||
7c0f6ba6 | 29 | #include <linux/uaccess.h> |
749cf76c CD |
30 | #include <asm/ptrace.h> |
31 | #include <asm/mman.h> | |
342cd0ab | 32 | #include <asm/tlbflush.h> |
5b3e5e5b | 33 | #include <asm/cacheflush.h> |
85acda3b | 34 | #include <asm/cpufeature.h> |
342cd0ab CD |
35 | #include <asm/virt.h> |
36 | #include <asm/kvm_arm.h> | |
37 | #include <asm/kvm_asm.h> | |
38 | #include <asm/kvm_mmu.h> | |
f7ed45be | 39 | #include <asm/kvm_emulate.h> |
5b3e5e5b | 40 | #include <asm/kvm_coproc.h> |
910917bb | 41 | #include <asm/sections.h> |
749cf76c CD |
42 | |
43 | #ifdef REQUIRES_VIRT | |
44 | __asm__(".arch_extension virt"); | |
45 | #endif | |
46 | ||
630a1685 | 47 | DEFINE_PER_CPU(kvm_host_data_t, kvm_host_data); |
342cd0ab | 48 | static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); |
342cd0ab | 49 | |
1638a12d MZ |
50 | /* Per-CPU variable containing the currently running vcpu. */ |
51 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu); | |
52 | ||
f7ed45be CD |
53 | /* The VMID used in the VTTBR */ |
54 | static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1); | |
20475f78 | 55 | static u32 kvm_next_vmid; |
fb544d1c | 56 | static DEFINE_SPINLOCK(kvm_vmid_lock); |
342cd0ab | 57 | |
c7da6fa4 PF |
58 | static bool vgic_present; |
59 | ||
67f69197 AT |
60 | static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled); |
61 | ||
1638a12d MZ |
62 | static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu) |
63 | { | |
1436c1aa | 64 | __this_cpu_write(kvm_arm_running_vcpu, vcpu); |
1638a12d MZ |
65 | } |
66 | ||
61bbe380 CD |
67 | DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use); |
68 | ||
1638a12d MZ |
69 | /** |
70 | * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU. | |
71 | * Must be called from non-preemptible context | |
72 | */ | |
73 | struct kvm_vcpu *kvm_arm_get_running_vcpu(void) | |
74 | { | |
1436c1aa | 75 | return __this_cpu_read(kvm_arm_running_vcpu); |
1638a12d MZ |
76 | } |
77 | ||
78 | /** | |
79 | * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
80 | */ | |
4000be42 | 81 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) |
1638a12d MZ |
82 | { |
83 | return &kvm_arm_running_vcpu; | |
84 | } | |
85 | ||
749cf76c CD |
86 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
87 | { | |
88 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; | |
89 | } | |
90 | ||
749cf76c CD |
91 | int kvm_arch_hardware_setup(void) |
92 | { | |
93 | return 0; | |
94 | } | |
95 | ||
f257d6dc | 96 | int kvm_arch_check_processor_compat(void) |
749cf76c | 97 | { |
f257d6dc | 98 | return 0; |
749cf76c CD |
99 | } |
100 | ||
749cf76c | 101 | |
d5d8184d CD |
102 | /** |
103 | * kvm_arch_init_vm - initializes a VM data structure | |
104 | * @kvm: pointer to the KVM struct | |
105 | */ | |
749cf76c CD |
106 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
107 | { | |
94d0e598 | 108 | int ret, cpu; |
d5d8184d | 109 | |
bca607eb | 110 | ret = kvm_arm_setup_stage2(kvm, type); |
5b6c6742 SP |
111 | if (ret) |
112 | return ret; | |
749cf76c | 113 | |
94d0e598 MZ |
114 | kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran)); |
115 | if (!kvm->arch.last_vcpu_ran) | |
116 | return -ENOMEM; | |
117 | ||
118 | for_each_possible_cpu(cpu) | |
119 | *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1; | |
120 | ||
d5d8184d CD |
121 | ret = kvm_alloc_stage2_pgd(kvm); |
122 | if (ret) | |
123 | goto out_fail_alloc; | |
124 | ||
c8dddecd | 125 | ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP); |
d5d8184d CD |
126 | if (ret) |
127 | goto out_free_stage2_pgd; | |
128 | ||
6c3d63c9 | 129 | kvm_vgic_early_init(kvm); |
a1a64387 | 130 | |
d5d8184d | 131 | /* Mark the initial VMID generation invalid */ |
e329fb75 | 132 | kvm->arch.vmid.vmid_gen = 0; |
d5d8184d | 133 | |
3caa2d8c | 134 | /* The maximum number of VCPUs is limited by the host's GIC model */ |
c7da6fa4 PF |
135 | kvm->arch.max_vcpus = vgic_present ? |
136 | kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; | |
3caa2d8c | 137 | |
d5d8184d CD |
138 | return ret; |
139 | out_free_stage2_pgd: | |
140 | kvm_free_stage2_pgd(kvm); | |
141 | out_fail_alloc: | |
94d0e598 MZ |
142 | free_percpu(kvm->arch.last_vcpu_ran); |
143 | kvm->arch.last_vcpu_ran = NULL; | |
d5d8184d | 144 | return ret; |
749cf76c CD |
145 | } |
146 | ||
235539b4 LC |
147 | int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
148 | { | |
149 | return 0; | |
150 | } | |
151 | ||
1499fa80 | 152 | vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
749cf76c CD |
153 | { |
154 | return VM_FAULT_SIGBUS; | |
155 | } | |
156 | ||
749cf76c | 157 | |
d5d8184d CD |
158 | /** |
159 | * kvm_arch_destroy_vm - destroy the VM data structure | |
160 | * @kvm: pointer to the KVM struct | |
161 | */ | |
749cf76c CD |
162 | void kvm_arch_destroy_vm(struct kvm *kvm) |
163 | { | |
164 | int i; | |
165 | ||
b2c9a85d MZ |
166 | kvm_vgic_destroy(kvm); |
167 | ||
94d0e598 MZ |
168 | free_percpu(kvm->arch.last_vcpu_ran); |
169 | kvm->arch.last_vcpu_ran = NULL; | |
170 | ||
749cf76c CD |
171 | for (i = 0; i < KVM_MAX_VCPUS; ++i) { |
172 | if (kvm->vcpus[i]) { | |
173 | kvm_arch_vcpu_free(kvm->vcpus[i]); | |
174 | kvm->vcpus[i] = NULL; | |
175 | } | |
176 | } | |
6b2ad81b | 177 | atomic_set(&kvm->online_vcpus, 0); |
749cf76c CD |
178 | } |
179 | ||
784aa3d7 | 180 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
749cf76c CD |
181 | { |
182 | int r; | |
183 | switch (ext) { | |
1a89dd91 | 184 | case KVM_CAP_IRQCHIP: |
c7da6fa4 PF |
185 | r = vgic_present; |
186 | break; | |
d44758c0 | 187 | case KVM_CAP_IOEVENTFD: |
7330672b | 188 | case KVM_CAP_DEVICE_CTRL: |
749cf76c CD |
189 | case KVM_CAP_USER_MEMORY: |
190 | case KVM_CAP_SYNC_MMU: | |
191 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
192 | case KVM_CAP_ONE_REG: | |
aa024c2f | 193 | case KVM_CAP_ARM_PSCI: |
4447a208 | 194 | case KVM_CAP_ARM_PSCI_0_2: |
98047888 | 195 | case KVM_CAP_READONLY_MEM: |
ecccf0cc | 196 | case KVM_CAP_MP_STATE: |
460df4c1 | 197 | case KVM_CAP_IMMEDIATE_EXIT: |
58bf437f | 198 | case KVM_CAP_VCPU_EVENTS: |
92f35b75 | 199 | case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2: |
749cf76c CD |
200 | r = 1; |
201 | break; | |
3401d546 CD |
202 | case KVM_CAP_ARM_SET_DEVICE_ADDR: |
203 | r = 1; | |
ca46e10f | 204 | break; |
749cf76c CD |
205 | case KVM_CAP_NR_VCPUS: |
206 | r = num_online_cpus(); | |
207 | break; | |
208 | case KVM_CAP_MAX_VCPUS: | |
209 | r = KVM_MAX_VCPUS; | |
210 | break; | |
a86cb413 TH |
211 | case KVM_CAP_MAX_VCPU_ID: |
212 | r = KVM_MAX_VCPU_ID; | |
213 | break; | |
2988509d VM |
214 | case KVM_CAP_MSI_DEVID: |
215 | if (!kvm) | |
216 | r = -EINVAL; | |
217 | else | |
218 | r = kvm->arch.vgic.msis_require_devid; | |
219 | break; | |
f7214e60 CD |
220 | case KVM_CAP_ARM_USER_IRQ: |
221 | /* | |
222 | * 1: EL1_VTIMER, EL1_PTIMER, and PMU. | |
223 | * (bump this number if adding more devices) | |
224 | */ | |
225 | r = 1; | |
226 | break; | |
749cf76c | 227 | default: |
375bdd3b | 228 | r = kvm_arch_vm_ioctl_check_extension(kvm, ext); |
749cf76c CD |
229 | break; |
230 | } | |
231 | return r; | |
232 | } | |
233 | ||
234 | long kvm_arch_dev_ioctl(struct file *filp, | |
235 | unsigned int ioctl, unsigned long arg) | |
236 | { | |
237 | return -EINVAL; | |
238 | } | |
239 | ||
d1e5b0e9 MO |
240 | struct kvm *kvm_arch_alloc_vm(void) |
241 | { | |
242 | if (!has_vhe()) | |
243 | return kzalloc(sizeof(struct kvm), GFP_KERNEL); | |
244 | ||
245 | return vzalloc(sizeof(struct kvm)); | |
246 | } | |
247 | ||
248 | void kvm_arch_free_vm(struct kvm *kvm) | |
249 | { | |
250 | if (!has_vhe()) | |
251 | kfree(kvm); | |
252 | else | |
253 | vfree(kvm); | |
254 | } | |
749cf76c CD |
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 | ||
716139df CD |
261 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) { |
262 | err = -EBUSY; | |
263 | goto out; | |
264 | } | |
265 | ||
3caa2d8c AP |
266 | if (id >= kvm->arch.max_vcpus) { |
267 | err = -EINVAL; | |
268 | goto out; | |
269 | } | |
270 | ||
749cf76c CD |
271 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
272 | if (!vcpu) { | |
273 | err = -ENOMEM; | |
274 | goto out; | |
275 | } | |
276 | ||
277 | err = kvm_vcpu_init(vcpu, kvm, id); | |
278 | if (err) | |
279 | goto free_vcpu; | |
280 | ||
c8dddecd | 281 | err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP); |
d5d8184d CD |
282 | if (err) |
283 | goto vcpu_uninit; | |
284 | ||
749cf76c | 285 | return vcpu; |
d5d8184d CD |
286 | vcpu_uninit: |
287 | kvm_vcpu_uninit(vcpu); | |
749cf76c CD |
288 | free_vcpu: |
289 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
290 | out: | |
291 | return ERR_PTR(err); | |
292 | } | |
293 | ||
31928aa5 | 294 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
749cf76c | 295 | { |
749cf76c CD |
296 | } |
297 | ||
298 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
299 | { | |
f1d7231c CD |
300 | if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm))) |
301 | static_branch_dec(&userspace_irqchip_in_use); | |
302 | ||
d5d8184d | 303 | kvm_mmu_free_memory_caches(vcpu); |
967f8427 | 304 | kvm_timer_vcpu_terminate(vcpu); |
5f0a714a | 305 | kvm_pmu_vcpu_destroy(vcpu); |
591d215a | 306 | kvm_vcpu_uninit(vcpu); |
d5d8184d | 307 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
749cf76c CD |
308 | } |
309 | ||
310 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) | |
311 | { | |
312 | kvm_arch_vcpu_free(vcpu); | |
313 | } | |
314 | ||
315 | int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) | |
316 | { | |
1c88ab7e | 317 | return kvm_timer_is_pending(vcpu); |
749cf76c CD |
318 | } |
319 | ||
d35268da CD |
320 | void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) |
321 | { | |
5eeaf10e MZ |
322 | /* |
323 | * If we're about to block (most likely because we've just hit a | |
324 | * WFI), we need to sync back the state of the GIC CPU interface | |
325 | * so that we have the lastest PMR and group enables. This ensures | |
326 | * that kvm_arch_vcpu_runnable has up-to-date data to decide | |
327 | * whether we have pending interrupts. | |
328 | */ | |
329 | preempt_disable(); | |
330 | kvm_vgic_vmcr_sync(vcpu); | |
331 | preempt_enable(); | |
332 | ||
df9ba959 | 333 | kvm_vgic_v4_enable_doorbell(vcpu); |
d35268da CD |
334 | } |
335 | ||
336 | void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
337 | { | |
df9ba959 | 338 | kvm_vgic_v4_disable_doorbell(vcpu); |
d35268da CD |
339 | } |
340 | ||
749cf76c CD |
341 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
342 | { | |
f7ed45be CD |
343 | /* Force users to call KVM_ARM_VCPU_INIT */ |
344 | vcpu->arch.target = -1; | |
f7fa034d | 345 | bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); |
1a89dd91 | 346 | |
967f8427 MZ |
347 | /* Set up the timer */ |
348 | kvm_timer_vcpu_init(vcpu); | |
349 | ||
bca031e2 ZY |
350 | kvm_pmu_vcpu_init(vcpu); |
351 | ||
84e690bf AB |
352 | kvm_arm_reset_debug_ptr(vcpu); |
353 | ||
1aab6f46 | 354 | return kvm_vgic_vcpu_init(vcpu); |
749cf76c CD |
355 | } |
356 | ||
749cf76c CD |
357 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
358 | { | |
94d0e598 | 359 | int *last_ran; |
630a1685 | 360 | kvm_host_data_t *cpu_data; |
94d0e598 MZ |
361 | |
362 | last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran); | |
630a1685 | 363 | cpu_data = this_cpu_ptr(&kvm_host_data); |
94d0e598 MZ |
364 | |
365 | /* | |
366 | * We might get preempted before the vCPU actually runs, but | |
367 | * over-invalidation doesn't affect correctness. | |
368 | */ | |
369 | if (*last_ran != vcpu->vcpu_id) { | |
370 | kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu); | |
371 | *last_ran = vcpu->vcpu_id; | |
372 | } | |
373 | ||
86ce8535 | 374 | vcpu->cpu = cpu; |
630a1685 | 375 | vcpu->arch.host_cpu_context = &cpu_data->host_ctxt; |
5b3e5e5b | 376 | |
1638a12d | 377 | kvm_arm_set_running_vcpu(vcpu); |
328e5664 | 378 | kvm_vgic_load(vcpu); |
b103cc3f | 379 | kvm_timer_vcpu_load(vcpu); |
bc192cee | 380 | kvm_vcpu_load_sysregs(vcpu); |
e6b673b7 | 381 | kvm_arch_vcpu_load_fp(vcpu); |
435e53fb | 382 | kvm_vcpu_pmu_restore_guest(vcpu); |
de737089 MZ |
383 | |
384 | if (single_task_running()) | |
385 | vcpu_clear_wfe_traps(vcpu); | |
386 | else | |
387 | vcpu_set_wfe_traps(vcpu); | |
384b40ca MR |
388 | |
389 | vcpu_ptrauth_setup_lazy(vcpu); | |
749cf76c CD |
390 | } |
391 | ||
392 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
393 | { | |
e6b673b7 | 394 | kvm_arch_vcpu_put_fp(vcpu); |
bc192cee | 395 | kvm_vcpu_put_sysregs(vcpu); |
b103cc3f | 396 | kvm_timer_vcpu_put(vcpu); |
328e5664 | 397 | kvm_vgic_put(vcpu); |
435e53fb | 398 | kvm_vcpu_pmu_restore_host(vcpu); |
328e5664 | 399 | |
e9b152cb CD |
400 | vcpu->cpu = -1; |
401 | ||
1638a12d | 402 | kvm_arm_set_running_vcpu(NULL); |
749cf76c CD |
403 | } |
404 | ||
424c989b AJ |
405 | static void vcpu_power_off(struct kvm_vcpu *vcpu) |
406 | { | |
407 | vcpu->arch.power_off = true; | |
7b244e2b | 408 | kvm_make_request(KVM_REQ_SLEEP, vcpu); |
424c989b AJ |
409 | kvm_vcpu_kick(vcpu); |
410 | } | |
411 | ||
749cf76c CD |
412 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
413 | struct kvm_mp_state *mp_state) | |
414 | { | |
3781528e | 415 | if (vcpu->arch.power_off) |
ecccf0cc AB |
416 | mp_state->mp_state = KVM_MP_STATE_STOPPED; |
417 | else | |
418 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
419 | ||
420 | return 0; | |
749cf76c CD |
421 | } |
422 | ||
423 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
424 | struct kvm_mp_state *mp_state) | |
425 | { | |
e83dff5e CD |
426 | int ret = 0; |
427 | ||
ecccf0cc AB |
428 | switch (mp_state->mp_state) { |
429 | case KVM_MP_STATE_RUNNABLE: | |
3781528e | 430 | vcpu->arch.power_off = false; |
ecccf0cc AB |
431 | break; |
432 | case KVM_MP_STATE_STOPPED: | |
424c989b | 433 | vcpu_power_off(vcpu); |
ecccf0cc AB |
434 | break; |
435 | default: | |
e83dff5e | 436 | ret = -EINVAL; |
ecccf0cc AB |
437 | } |
438 | ||
e83dff5e | 439 | return ret; |
749cf76c CD |
440 | } |
441 | ||
5b3e5e5b CD |
442 | /** |
443 | * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled | |
444 | * @v: The VCPU pointer | |
445 | * | |
446 | * If the guest CPU is not waiting for interrupts or an interrupt line is | |
447 | * asserted, the CPU is by definition runnable. | |
448 | */ | |
749cf76c CD |
449 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) |
450 | { | |
3df59d8d CD |
451 | bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF); |
452 | return ((irq_lines || kvm_vgic_vcpu_pending_irq(v)) | |
3b92830a | 453 | && !v->arch.power_off && !v->arch.pause); |
749cf76c CD |
454 | } |
455 | ||
199b5763 LM |
456 | bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) |
457 | { | |
f01fbd2f | 458 | return vcpu_mode_priv(vcpu); |
199b5763 LM |
459 | } |
460 | ||
f7ed45be CD |
461 | /* Just ensure a guest exit from a particular CPU */ |
462 | static void exit_vm_noop(void *info) | |
463 | { | |
464 | } | |
465 | ||
466 | void force_vm_exit(const cpumask_t *mask) | |
467 | { | |
898f949f | 468 | preempt_disable(); |
f7ed45be | 469 | smp_call_function_many(mask, exit_vm_noop, NULL, true); |
898f949f | 470 | preempt_enable(); |
f7ed45be CD |
471 | } |
472 | ||
473 | /** | |
474 | * need_new_vmid_gen - check that the VMID is still valid | |
e329fb75 | 475 | * @vmid: The VMID to check |
f7ed45be CD |
476 | * |
477 | * return true if there is a new generation of VMIDs being used | |
478 | * | |
e329fb75 CD |
479 | * The hardware supports a limited set of values with the value zero reserved |
480 | * for the host, so we check if an assigned value belongs to a previous | |
481 | * generation, which which requires us to assign a new value. If we're the | |
482 | * first to use a VMID for the new generation, we must flush necessary caches | |
483 | * and TLBs on all CPUs. | |
f7ed45be | 484 | */ |
e329fb75 | 485 | static bool need_new_vmid_gen(struct kvm_vmid *vmid) |
f7ed45be | 486 | { |
fb544d1c CD |
487 | u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen); |
488 | smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */ | |
e329fb75 | 489 | return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen); |
f7ed45be CD |
490 | } |
491 | ||
492 | /** | |
e329fb75 CD |
493 | * update_vmid - Update the vmid with a valid VMID for the current generation |
494 | * @kvm: The guest that struct vmid belongs to | |
495 | * @vmid: The stage-2 VMID information struct | |
f7ed45be | 496 | */ |
e329fb75 | 497 | static void update_vmid(struct kvm_vmid *vmid) |
f7ed45be | 498 | { |
e329fb75 | 499 | if (!need_new_vmid_gen(vmid)) |
f7ed45be CD |
500 | return; |
501 | ||
fb544d1c | 502 | spin_lock(&kvm_vmid_lock); |
f7ed45be CD |
503 | |
504 | /* | |
505 | * We need to re-check the vmid_gen here to ensure that if another vcpu | |
506 | * already allocated a valid vmid for this vm, then this vcpu should | |
507 | * use the same vmid. | |
508 | */ | |
e329fb75 | 509 | if (!need_new_vmid_gen(vmid)) { |
fb544d1c | 510 | spin_unlock(&kvm_vmid_lock); |
f7ed45be CD |
511 | return; |
512 | } | |
513 | ||
514 | /* First user of a new VMID generation? */ | |
515 | if (unlikely(kvm_next_vmid == 0)) { | |
516 | atomic64_inc(&kvm_vmid_gen); | |
517 | kvm_next_vmid = 1; | |
518 | ||
519 | /* | |
520 | * On SMP we know no other CPUs can use this CPU's or each | |
521 | * other's VMID after force_vm_exit returns since the | |
522 | * kvm_vmid_lock blocks them from reentry to the guest. | |
523 | */ | |
524 | force_vm_exit(cpu_all_mask); | |
525 | /* | |
526 | * Now broadcast TLB + ICACHE invalidation over the inner | |
527 | * shareable domain to make sure all data structures are | |
528 | * clean. | |
529 | */ | |
530 | kvm_call_hyp(__kvm_flush_vm_context); | |
531 | } | |
532 | ||
e329fb75 | 533 | vmid->vmid = kvm_next_vmid; |
f7ed45be | 534 | kvm_next_vmid++; |
e329fb75 | 535 | kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1; |
f7ed45be | 536 | |
fb544d1c | 537 | smp_wmb(); |
e329fb75 | 538 | WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen)); |
fb544d1c CD |
539 | |
540 | spin_unlock(&kvm_vmid_lock); | |
f7ed45be CD |
541 | } |
542 | ||
f7ed45be CD |
543 | static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) |
544 | { | |
05971120 | 545 | struct kvm *kvm = vcpu->kvm; |
41a54482 | 546 | int ret = 0; |
e1ba0207 | 547 | |
f7ed45be CD |
548 | if (likely(vcpu->arch.has_run_once)) |
549 | return 0; | |
550 | ||
7dd32a0d DM |
551 | if (!kvm_arm_vcpu_is_finalized(vcpu)) |
552 | return -EPERM; | |
553 | ||
f7ed45be | 554 | vcpu->arch.has_run_once = true; |
aa024c2f | 555 | |
61bbe380 CD |
556 | if (likely(irqchip_in_kernel(kvm))) { |
557 | /* | |
558 | * Map the VGIC hardware resources before running a vcpu the | |
559 | * first time on this VM. | |
560 | */ | |
561 | if (unlikely(!vgic_ready(kvm))) { | |
562 | ret = kvm_vgic_map_resources(kvm); | |
563 | if (ret) | |
564 | return ret; | |
565 | } | |
566 | } else { | |
567 | /* | |
568 | * Tell the rest of the code that there are userspace irqchip | |
569 | * VMs in the wild. | |
570 | */ | |
571 | static_branch_inc(&userspace_irqchip_in_use); | |
01ac5e34 MZ |
572 | } |
573 | ||
d9e13977 | 574 | ret = kvm_timer_enable(vcpu); |
a2befacf CD |
575 | if (ret) |
576 | return ret; | |
577 | ||
578 | ret = kvm_arm_pmu_v3_enable(vcpu); | |
05971120 | 579 | |
41a54482 | 580 | return ret; |
f7ed45be CD |
581 | } |
582 | ||
c1426e4c EA |
583 | bool kvm_arch_intc_initialized(struct kvm *kvm) |
584 | { | |
585 | return vgic_initialized(kvm); | |
586 | } | |
587 | ||
b13216cf | 588 | void kvm_arm_halt_guest(struct kvm *kvm) |
3b92830a EA |
589 | { |
590 | int i; | |
591 | struct kvm_vcpu *vcpu; | |
592 | ||
593 | kvm_for_each_vcpu(i, vcpu, kvm) | |
594 | vcpu->arch.pause = true; | |
7b244e2b | 595 | kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP); |
3b92830a EA |
596 | } |
597 | ||
b13216cf | 598 | void kvm_arm_resume_guest(struct kvm *kvm) |
3b92830a EA |
599 | { |
600 | int i; | |
601 | struct kvm_vcpu *vcpu; | |
602 | ||
abd72296 CD |
603 | kvm_for_each_vcpu(i, vcpu, kvm) { |
604 | vcpu->arch.pause = false; | |
b3dae109 | 605 | swake_up_one(kvm_arch_vcpu_wq(vcpu)); |
abd72296 | 606 | } |
3b92830a EA |
607 | } |
608 | ||
7b244e2b | 609 | static void vcpu_req_sleep(struct kvm_vcpu *vcpu) |
aa024c2f | 610 | { |
8577370f | 611 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); |
aa024c2f | 612 | |
b3dae109 | 613 | swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) && |
3b92830a | 614 | (!vcpu->arch.pause))); |
0592c005 | 615 | |
424c989b | 616 | if (vcpu->arch.power_off || vcpu->arch.pause) { |
0592c005 | 617 | /* Awaken to handle a signal, request we sleep again later. */ |
7b244e2b | 618 | kvm_make_request(KVM_REQ_SLEEP, vcpu); |
0592c005 | 619 | } |
358b28f0 MZ |
620 | |
621 | /* | |
622 | * Make sure we will observe a potential reset request if we've | |
623 | * observed a change to the power state. Pairs with the smp_wmb() in | |
624 | * kvm_psci_vcpu_on(). | |
625 | */ | |
626 | smp_rmb(); | |
aa024c2f MZ |
627 | } |
628 | ||
e8180dca AP |
629 | static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) |
630 | { | |
631 | return vcpu->arch.target >= 0; | |
632 | } | |
633 | ||
0592c005 AJ |
634 | static void check_vcpu_requests(struct kvm_vcpu *vcpu) |
635 | { | |
636 | if (kvm_request_pending(vcpu)) { | |
7b244e2b AJ |
637 | if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) |
638 | vcpu_req_sleep(vcpu); | |
325f9c64 | 639 | |
358b28f0 MZ |
640 | if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu)) |
641 | kvm_reset_vcpu(vcpu); | |
642 | ||
325f9c64 AJ |
643 | /* |
644 | * Clear IRQ_PENDING requests that were made to guarantee | |
645 | * that a VCPU sees new virtual interrupts. | |
646 | */ | |
647 | kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu); | |
0592c005 AJ |
648 | } |
649 | } | |
650 | ||
f7ed45be CD |
651 | /** |
652 | * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code | |
653 | * @vcpu: The VCPU pointer | |
654 | * @run: The kvm_run structure pointer used for userspace state exchange | |
655 | * | |
656 | * This function is called through the VCPU_RUN ioctl called from user space. It | |
657 | * will execute VM code in a loop until the time slice for the process is used | |
658 | * or some emulation is needed from user space in which case the function will | |
659 | * return with return value 0 and with the kvm_run structure filled in with the | |
660 | * required data for the requested emulation. | |
661 | */ | |
749cf76c CD |
662 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) |
663 | { | |
f7ed45be | 664 | int ret; |
f7ed45be | 665 | |
e8180dca | 666 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
f7ed45be CD |
667 | return -ENOEXEC; |
668 | ||
669 | ret = kvm_vcpu_first_run_init(vcpu); | |
670 | if (ret) | |
829a5863 | 671 | return ret; |
f7ed45be | 672 | |
45e96ea6 CD |
673 | if (run->exit_reason == KVM_EXIT_MMIO) { |
674 | ret = kvm_handle_mmio_return(vcpu, vcpu->run); | |
675 | if (ret) | |
829a5863 | 676 | return ret; |
45e96ea6 CD |
677 | } |
678 | ||
829a5863 CD |
679 | if (run->immediate_exit) |
680 | return -EINTR; | |
681 | ||
682 | vcpu_load(vcpu); | |
460df4c1 | 683 | |
20b7035c | 684 | kvm_sigset_activate(vcpu); |
f7ed45be CD |
685 | |
686 | ret = 1; | |
687 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
688 | while (ret > 0) { | |
689 | /* | |
690 | * Check conditions before entering the guest | |
691 | */ | |
692 | cond_resched(); | |
693 | ||
e329fb75 | 694 | update_vmid(&vcpu->kvm->arch.vmid); |
f7ed45be | 695 | |
0592c005 AJ |
696 | check_vcpu_requests(vcpu); |
697 | ||
abdf5843 MZ |
698 | /* |
699 | * Preparing the interrupts to be injected also | |
700 | * involves poking the GIC, which must be done in a | |
701 | * non-preemptible context. | |
702 | */ | |
1b3d546d | 703 | preempt_disable(); |
328e5664 | 704 | |
b02386eb | 705 | kvm_pmu_flush_hwstate(vcpu); |
328e5664 | 706 | |
f7ed45be CD |
707 | local_irq_disable(); |
708 | ||
abdf5843 MZ |
709 | kvm_vgic_flush_hwstate(vcpu); |
710 | ||
f7ed45be | 711 | /* |
61bbe380 CD |
712 | * Exit if we have a signal pending so that we can deliver the |
713 | * signal to user space. | |
f7ed45be | 714 | */ |
61bbe380 | 715 | if (signal_pending(current)) { |
f7ed45be CD |
716 | ret = -EINTR; |
717 | run->exit_reason = KVM_EXIT_INTR; | |
718 | } | |
719 | ||
61bbe380 CD |
720 | /* |
721 | * If we're using a userspace irqchip, then check if we need | |
722 | * to tell a userspace irqchip about timer or PMU level | |
723 | * changes and if so, exit to userspace (the actual level | |
724 | * state gets updated in kvm_timer_update_run and | |
725 | * kvm_pmu_update_run below). | |
726 | */ | |
727 | if (static_branch_unlikely(&userspace_irqchip_in_use)) { | |
728 | if (kvm_timer_should_notify_user(vcpu) || | |
729 | kvm_pmu_should_notify_user(vcpu)) { | |
730 | ret = -EINTR; | |
731 | run->exit_reason = KVM_EXIT_INTR; | |
732 | } | |
733 | } | |
734 | ||
6a6d73be AJ |
735 | /* |
736 | * Ensure we set mode to IN_GUEST_MODE after we disable | |
737 | * interrupts and before the final VCPU requests check. | |
738 | * See the comment in kvm_vcpu_exiting_guest_mode() and | |
2f5947df | 739 | * Documentation/virt/kvm/vcpu-requests.rst |
6a6d73be AJ |
740 | */ |
741 | smp_store_mb(vcpu->mode, IN_GUEST_MODE); | |
742 | ||
e329fb75 | 743 | if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) || |
424c989b | 744 | kvm_request_pending(vcpu)) { |
6a6d73be | 745 | vcpu->mode = OUTSIDE_GUEST_MODE; |
771621b0 | 746 | isb(); /* Ensure work in x_flush_hwstate is committed */ |
b02386eb | 747 | kvm_pmu_sync_hwstate(vcpu); |
61bbe380 CD |
748 | if (static_branch_unlikely(&userspace_irqchip_in_use)) |
749 | kvm_timer_sync_hwstate(vcpu); | |
1a89dd91 | 750 | kvm_vgic_sync_hwstate(vcpu); |
ee9bb9a1 | 751 | local_irq_enable(); |
abdf5843 | 752 | preempt_enable(); |
f7ed45be CD |
753 | continue; |
754 | } | |
755 | ||
56c7f5e7 AB |
756 | kvm_arm_setup_debug(vcpu); |
757 | ||
f7ed45be CD |
758 | /************************************************************** |
759 | * Enter the guest | |
760 | */ | |
761 | trace_kvm_entry(*vcpu_pc(vcpu)); | |
6edaa530 | 762 | guest_enter_irqoff(); |
f7ed45be | 763 | |
3f5c90b8 CD |
764 | if (has_vhe()) { |
765 | kvm_arm_vhe_guest_enter(); | |
766 | ret = kvm_vcpu_run_vhe(vcpu); | |
4f5abad9 | 767 | kvm_arm_vhe_guest_exit(); |
3f5c90b8 | 768 | } else { |
7aa8d146 | 769 | ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu); |
3f5c90b8 CD |
770 | } |
771 | ||
f7ed45be | 772 | vcpu->mode = OUTSIDE_GUEST_MODE; |
b19e6892 | 773 | vcpu->stat.exits++; |
1b3d546d CD |
774 | /* |
775 | * Back from guest | |
776 | *************************************************************/ | |
777 | ||
56c7f5e7 AB |
778 | kvm_arm_clear_debug(vcpu); |
779 | ||
ee9bb9a1 | 780 | /* |
b103cc3f | 781 | * We must sync the PMU state before the vgic state so |
ee9bb9a1 CD |
782 | * that the vgic can properly sample the updated state of the |
783 | * interrupt line. | |
784 | */ | |
785 | kvm_pmu_sync_hwstate(vcpu); | |
ee9bb9a1 | 786 | |
b103cc3f CD |
787 | /* |
788 | * Sync the vgic state before syncing the timer state because | |
789 | * the timer code needs to know if the virtual timer | |
790 | * interrupts are active. | |
791 | */ | |
ee9bb9a1 CD |
792 | kvm_vgic_sync_hwstate(vcpu); |
793 | ||
b103cc3f CD |
794 | /* |
795 | * Sync the timer hardware state before enabling interrupts as | |
796 | * we don't want vtimer interrupts to race with syncing the | |
797 | * timer virtual interrupt state. | |
798 | */ | |
61bbe380 CD |
799 | if (static_branch_unlikely(&userspace_irqchip_in_use)) |
800 | kvm_timer_sync_hwstate(vcpu); | |
b103cc3f | 801 | |
e6b673b7 DM |
802 | kvm_arch_vcpu_ctxsync_fp(vcpu); |
803 | ||
f7ed45be CD |
804 | /* |
805 | * We may have taken a host interrupt in HYP mode (ie | |
806 | * while executing the guest). This interrupt is still | |
807 | * pending, as we haven't serviced it yet! | |
808 | * | |
809 | * We're now back in SVC mode, with interrupts | |
810 | * disabled. Enabling the interrupts now will have | |
811 | * the effect of taking the interrupt again, in SVC | |
812 | * mode this time. | |
813 | */ | |
814 | local_irq_enable(); | |
815 | ||
816 | /* | |
6edaa530 | 817 | * We do local_irq_enable() before calling guest_exit() so |
1b3d546d CD |
818 | * that if a timer interrupt hits while running the guest we |
819 | * account that tick as being spent in the guest. We enable | |
6edaa530 | 820 | * preemption after calling guest_exit() so that if we get |
1b3d546d CD |
821 | * preempted we make sure ticks after that is not counted as |
822 | * guest time. | |
823 | */ | |
6edaa530 | 824 | guest_exit(); |
b5905dc1 | 825 | trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); |
1b3d546d | 826 | |
3368bd80 JM |
827 | /* Exit types that need handling before we can be preempted */ |
828 | handle_exit_early(vcpu, run, ret); | |
829 | ||
abdf5843 MZ |
830 | preempt_enable(); |
831 | ||
f7ed45be CD |
832 | ret = handle_exit(vcpu, run, ret); |
833 | } | |
834 | ||
d9e13977 | 835 | /* Tell userspace about in-kernel device output levels */ |
3dbbdf78 CD |
836 | if (unlikely(!irqchip_in_kernel(vcpu->kvm))) { |
837 | kvm_timer_update_run(vcpu); | |
838 | kvm_pmu_update_run(vcpu); | |
839 | } | |
d9e13977 | 840 | |
20b7035c JS |
841 | kvm_sigset_deactivate(vcpu); |
842 | ||
accb757d | 843 | vcpu_put(vcpu); |
f7ed45be | 844 | return ret; |
749cf76c CD |
845 | } |
846 | ||
86ce8535 CD |
847 | static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) |
848 | { | |
849 | int bit_index; | |
850 | bool set; | |
3df59d8d | 851 | unsigned long *hcr; |
86ce8535 CD |
852 | |
853 | if (number == KVM_ARM_IRQ_CPU_IRQ) | |
854 | bit_index = __ffs(HCR_VI); | |
855 | else /* KVM_ARM_IRQ_CPU_FIQ */ | |
856 | bit_index = __ffs(HCR_VF); | |
857 | ||
3df59d8d | 858 | hcr = vcpu_hcr(vcpu); |
86ce8535 | 859 | if (level) |
3df59d8d | 860 | set = test_and_set_bit(bit_index, hcr); |
86ce8535 | 861 | else |
3df59d8d | 862 | set = test_and_clear_bit(bit_index, hcr); |
86ce8535 CD |
863 | |
864 | /* | |
865 | * If we didn't change anything, no need to wake up or kick other CPUs | |
866 | */ | |
867 | if (set == level) | |
868 | return 0; | |
869 | ||
870 | /* | |
871 | * The vcpu irq_lines field was updated, wake up sleeping VCPUs and | |
872 | * trigger a world-switch round on the running physical CPU to set the | |
873 | * virtual IRQ/FIQ fields in the HCR appropriately. | |
874 | */ | |
325f9c64 | 875 | kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu); |
86ce8535 CD |
876 | kvm_vcpu_kick(vcpu); |
877 | ||
878 | return 0; | |
879 | } | |
880 | ||
79558f11 AG |
881 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, |
882 | bool line_status) | |
86ce8535 CD |
883 | { |
884 | u32 irq = irq_level->irq; | |
885 | unsigned int irq_type, vcpu_idx, irq_num; | |
886 | int nrcpus = atomic_read(&kvm->online_vcpus); | |
887 | struct kvm_vcpu *vcpu = NULL; | |
888 | bool level = irq_level->level; | |
889 | ||
890 | irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; | |
891 | vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; | |
92f35b75 | 892 | vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1); |
86ce8535 CD |
893 | irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; |
894 | ||
895 | trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); | |
896 | ||
5863c2ce MZ |
897 | switch (irq_type) { |
898 | case KVM_ARM_IRQ_TYPE_CPU: | |
899 | if (irqchip_in_kernel(kvm)) | |
900 | return -ENXIO; | |
86ce8535 | 901 | |
5863c2ce MZ |
902 | if (vcpu_idx >= nrcpus) |
903 | return -EINVAL; | |
86ce8535 | 904 | |
5863c2ce MZ |
905 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); |
906 | if (!vcpu) | |
907 | return -EINVAL; | |
86ce8535 | 908 | |
5863c2ce MZ |
909 | if (irq_num > KVM_ARM_IRQ_CPU_FIQ) |
910 | return -EINVAL; | |
911 | ||
912 | return vcpu_interrupt_line(vcpu, irq_num, level); | |
913 | case KVM_ARM_IRQ_TYPE_PPI: | |
914 | if (!irqchip_in_kernel(kvm)) | |
915 | return -ENXIO; | |
916 | ||
917 | if (vcpu_idx >= nrcpus) | |
918 | return -EINVAL; | |
919 | ||
920 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); | |
921 | if (!vcpu) | |
922 | return -EINVAL; | |
923 | ||
924 | if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) | |
925 | return -EINVAL; | |
86ce8535 | 926 | |
cb3f0ad8 | 927 | return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL); |
5863c2ce MZ |
928 | case KVM_ARM_IRQ_TYPE_SPI: |
929 | if (!irqchip_in_kernel(kvm)) | |
930 | return -ENXIO; | |
931 | ||
fd1d0ddf | 932 | if (irq_num < VGIC_NR_PRIVATE_IRQS) |
5863c2ce MZ |
933 | return -EINVAL; |
934 | ||
cb3f0ad8 | 935 | return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL); |
5863c2ce MZ |
936 | } |
937 | ||
938 | return -EINVAL; | |
86ce8535 CD |
939 | } |
940 | ||
f7fa034d CD |
941 | static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, |
942 | const struct kvm_vcpu_init *init) | |
943 | { | |
811328fc | 944 | unsigned int i, ret; |
f7fa034d CD |
945 | int phys_target = kvm_target_cpu(); |
946 | ||
947 | if (init->target != phys_target) | |
948 | return -EINVAL; | |
949 | ||
950 | /* | |
951 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
952 | * use the same target. | |
953 | */ | |
954 | if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) | |
955 | return -EINVAL; | |
956 | ||
957 | /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ | |
958 | for (i = 0; i < sizeof(init->features) * 8; i++) { | |
959 | bool set = (init->features[i / 32] & (1 << (i % 32))); | |
960 | ||
961 | if (set && i >= KVM_VCPU_MAX_FEATURES) | |
962 | return -ENOENT; | |
963 | ||
964 | /* | |
965 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
966 | * use the same feature set. | |
967 | */ | |
968 | if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && | |
969 | test_bit(i, vcpu->arch.features) != set) | |
970 | return -EINVAL; | |
971 | ||
972 | if (set) | |
973 | set_bit(i, vcpu->arch.features); | |
974 | } | |
975 | ||
976 | vcpu->arch.target = phys_target; | |
977 | ||
978 | /* Now we know what it is, we can reset it. */ | |
811328fc AJ |
979 | ret = kvm_reset_vcpu(vcpu); |
980 | if (ret) { | |
981 | vcpu->arch.target = -1; | |
982 | bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); | |
983 | } | |
f7fa034d | 984 | |
811328fc AJ |
985 | return ret; |
986 | } | |
f7fa034d | 987 | |
478a8237 CD |
988 | static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, |
989 | struct kvm_vcpu_init *init) | |
990 | { | |
991 | int ret; | |
992 | ||
993 | ret = kvm_vcpu_set_target(vcpu, init); | |
994 | if (ret) | |
995 | return ret; | |
996 | ||
957db105 CD |
997 | /* |
998 | * Ensure a rebooted VM will fault in RAM pages and detect if the | |
999 | * guest MMU is turned off and flush the caches as needed. | |
1000 | */ | |
1001 | if (vcpu->arch.has_run_once) | |
1002 | stage2_unmap_vm(vcpu->kvm); | |
1003 | ||
b856a591 CD |
1004 | vcpu_reset_hcr(vcpu); |
1005 | ||
478a8237 | 1006 | /* |
3781528e | 1007 | * Handle the "start in power-off" case. |
478a8237 | 1008 | */ |
03f1d4c1 | 1009 | if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) |
424c989b | 1010 | vcpu_power_off(vcpu); |
3ad8b3de | 1011 | else |
3781528e | 1012 | vcpu->arch.power_off = false; |
478a8237 CD |
1013 | |
1014 | return 0; | |
1015 | } | |
1016 | ||
f577f6c2 SZ |
1017 | static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, |
1018 | struct kvm_device_attr *attr) | |
1019 | { | |
1020 | int ret = -ENXIO; | |
1021 | ||
1022 | switch (attr->group) { | |
1023 | default: | |
bb0c70bc | 1024 | ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); |
f577f6c2 SZ |
1025 | break; |
1026 | } | |
1027 | ||
1028 | return ret; | |
1029 | } | |
1030 | ||
1031 | static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, | |
1032 | struct kvm_device_attr *attr) | |
1033 | { | |
1034 | int ret = -ENXIO; | |
1035 | ||
1036 | switch (attr->group) { | |
1037 | default: | |
bb0c70bc | 1038 | ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); |
f577f6c2 SZ |
1039 | break; |
1040 | } | |
1041 | ||
1042 | return ret; | |
1043 | } | |
1044 | ||
1045 | static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, | |
1046 | struct kvm_device_attr *attr) | |
1047 | { | |
1048 | int ret = -ENXIO; | |
1049 | ||
1050 | switch (attr->group) { | |
1051 | default: | |
bb0c70bc | 1052 | ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); |
f577f6c2 SZ |
1053 | break; |
1054 | } | |
1055 | ||
1056 | return ret; | |
1057 | } | |
1058 | ||
539aee0e JM |
1059 | static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu, |
1060 | struct kvm_vcpu_events *events) | |
1061 | { | |
1062 | memset(events, 0, sizeof(*events)); | |
1063 | ||
1064 | return __kvm_arm_vcpu_get_events(vcpu, events); | |
1065 | } | |
1066 | ||
1067 | static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu, | |
1068 | struct kvm_vcpu_events *events) | |
1069 | { | |
1070 | int i; | |
1071 | ||
1072 | /* check whether the reserved field is zero */ | |
1073 | for (i = 0; i < ARRAY_SIZE(events->reserved); i++) | |
1074 | if (events->reserved[i]) | |
1075 | return -EINVAL; | |
1076 | ||
1077 | /* check whether the pad field is zero */ | |
1078 | for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++) | |
1079 | if (events->exception.pad[i]) | |
1080 | return -EINVAL; | |
1081 | ||
1082 | return __kvm_arm_vcpu_set_events(vcpu, events); | |
1083 | } | |
539aee0e | 1084 | |
749cf76c CD |
1085 | long kvm_arch_vcpu_ioctl(struct file *filp, |
1086 | unsigned int ioctl, unsigned long arg) | |
1087 | { | |
1088 | struct kvm_vcpu *vcpu = filp->private_data; | |
1089 | void __user *argp = (void __user *)arg; | |
f577f6c2 | 1090 | struct kvm_device_attr attr; |
9b062471 CD |
1091 | long r; |
1092 | ||
749cf76c CD |
1093 | switch (ioctl) { |
1094 | case KVM_ARM_VCPU_INIT: { | |
1095 | struct kvm_vcpu_init init; | |
1096 | ||
9b062471 | 1097 | r = -EFAULT; |
749cf76c | 1098 | if (copy_from_user(&init, argp, sizeof(init))) |
9b062471 | 1099 | break; |
749cf76c | 1100 | |
9b062471 CD |
1101 | r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); |
1102 | break; | |
749cf76c CD |
1103 | } |
1104 | case KVM_SET_ONE_REG: | |
1105 | case KVM_GET_ONE_REG: { | |
1106 | struct kvm_one_reg reg; | |
e8180dca | 1107 | |
9b062471 | 1108 | r = -ENOEXEC; |
e8180dca | 1109 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
9b062471 | 1110 | break; |
e8180dca | 1111 | |
9b062471 | 1112 | r = -EFAULT; |
749cf76c | 1113 | if (copy_from_user(®, argp, sizeof(reg))) |
9b062471 CD |
1114 | break; |
1115 | ||
749cf76c | 1116 | if (ioctl == KVM_SET_ONE_REG) |
9b062471 | 1117 | r = kvm_arm_set_reg(vcpu, ®); |
749cf76c | 1118 | else |
9b062471 CD |
1119 | r = kvm_arm_get_reg(vcpu, ®); |
1120 | break; | |
749cf76c CD |
1121 | } |
1122 | case KVM_GET_REG_LIST: { | |
1123 | struct kvm_reg_list __user *user_list = argp; | |
1124 | struct kvm_reg_list reg_list; | |
1125 | unsigned n; | |
1126 | ||
9b062471 | 1127 | r = -ENOEXEC; |
e8180dca | 1128 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
9b062471 | 1129 | break; |
e8180dca | 1130 | |
7dd32a0d DM |
1131 | r = -EPERM; |
1132 | if (!kvm_arm_vcpu_is_finalized(vcpu)) | |
1133 | break; | |
1134 | ||
9b062471 | 1135 | r = -EFAULT; |
749cf76c | 1136 | if (copy_from_user(®_list, user_list, sizeof(reg_list))) |
9b062471 | 1137 | break; |
749cf76c CD |
1138 | n = reg_list.n; |
1139 | reg_list.n = kvm_arm_num_regs(vcpu); | |
1140 | if (copy_to_user(user_list, ®_list, sizeof(reg_list))) | |
9b062471 CD |
1141 | break; |
1142 | r = -E2BIG; | |
749cf76c | 1143 | if (n < reg_list.n) |
9b062471 CD |
1144 | break; |
1145 | r = kvm_arm_copy_reg_indices(vcpu, user_list->reg); | |
1146 | break; | |
749cf76c | 1147 | } |
f577f6c2 | 1148 | case KVM_SET_DEVICE_ATTR: { |
9b062471 | 1149 | r = -EFAULT; |
f577f6c2 | 1150 | if (copy_from_user(&attr, argp, sizeof(attr))) |
9b062471 CD |
1151 | break; |
1152 | r = kvm_arm_vcpu_set_attr(vcpu, &attr); | |
1153 | break; | |
f577f6c2 SZ |
1154 | } |
1155 | case KVM_GET_DEVICE_ATTR: { | |
9b062471 | 1156 | r = -EFAULT; |
f577f6c2 | 1157 | if (copy_from_user(&attr, argp, sizeof(attr))) |
9b062471 CD |
1158 | break; |
1159 | r = kvm_arm_vcpu_get_attr(vcpu, &attr); | |
1160 | break; | |
f577f6c2 SZ |
1161 | } |
1162 | case KVM_HAS_DEVICE_ATTR: { | |
9b062471 | 1163 | r = -EFAULT; |
f577f6c2 | 1164 | if (copy_from_user(&attr, argp, sizeof(attr))) |
9b062471 CD |
1165 | break; |
1166 | r = kvm_arm_vcpu_has_attr(vcpu, &attr); | |
1167 | break; | |
f577f6c2 | 1168 | } |
b7b27fac DG |
1169 | case KVM_GET_VCPU_EVENTS: { |
1170 | struct kvm_vcpu_events events; | |
1171 | ||
1172 | if (kvm_arm_vcpu_get_events(vcpu, &events)) | |
1173 | return -EINVAL; | |
1174 | ||
1175 | if (copy_to_user(argp, &events, sizeof(events))) | |
1176 | return -EFAULT; | |
1177 | ||
1178 | return 0; | |
1179 | } | |
1180 | case KVM_SET_VCPU_EVENTS: { | |
1181 | struct kvm_vcpu_events events; | |
1182 | ||
1183 | if (copy_from_user(&events, argp, sizeof(events))) | |
1184 | return -EFAULT; | |
1185 | ||
1186 | return kvm_arm_vcpu_set_events(vcpu, &events); | |
1187 | } | |
7dd32a0d DM |
1188 | case KVM_ARM_VCPU_FINALIZE: { |
1189 | int what; | |
1190 | ||
1191 | if (!kvm_vcpu_initialized(vcpu)) | |
1192 | return -ENOEXEC; | |
1193 | ||
1194 | if (get_user(what, (const int __user *)argp)) | |
1195 | return -EFAULT; | |
1196 | ||
1197 | return kvm_arm_vcpu_finalize(vcpu, what); | |
1198 | } | |
749cf76c | 1199 | default: |
9b062471 | 1200 | r = -EINVAL; |
749cf76c | 1201 | } |
9b062471 | 1202 | |
9b062471 | 1203 | return r; |
749cf76c CD |
1204 | } |
1205 | ||
53c810c3 MS |
1206 | /** |
1207 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
1208 | * @kvm: kvm instance | |
1209 | * @log: slot id and address to which we copy the log | |
1210 | * | |
1211 | * Steps 1-4 below provide general overview of dirty page logging. See | |
1212 | * kvm_get_dirty_log_protect() function description for additional details. | |
1213 | * | |
1214 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
1215 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
1216 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
1217 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
1218 | * writes will be marked dirty for next log read. | |
1219 | * | |
1220 | * 1. Take a snapshot of the bit and clear it if needed. | |
1221 | * 2. Write protect the corresponding page. | |
1222 | * 3. Copy the snapshot to the userspace. | |
1223 | * 4. Flush TLB's if needed. | |
1224 | */ | |
749cf76c CD |
1225 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
1226 | { | |
8fe65a82 | 1227 | bool flush = false; |
53c810c3 MS |
1228 | int r; |
1229 | ||
1230 | mutex_lock(&kvm->slots_lock); | |
1231 | ||
8fe65a82 | 1232 | r = kvm_get_dirty_log_protect(kvm, log, &flush); |
53c810c3 | 1233 | |
8fe65a82 | 1234 | if (flush) |
53c810c3 MS |
1235 | kvm_flush_remote_tlbs(kvm); |
1236 | ||
1237 | mutex_unlock(&kvm->slots_lock); | |
1238 | return r; | |
749cf76c CD |
1239 | } |
1240 | ||
2a31b9db PB |
1241 | int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log) |
1242 | { | |
1243 | bool flush = false; | |
53c810c3 MS |
1244 | int r; |
1245 | ||
1246 | mutex_lock(&kvm->slots_lock); | |
1247 | ||
2a31b9db | 1248 | r = kvm_clear_dirty_log_protect(kvm, log, &flush); |
53c810c3 | 1249 | |
2a31b9db | 1250 | if (flush) |
53c810c3 MS |
1251 | kvm_flush_remote_tlbs(kvm); |
1252 | ||
1253 | mutex_unlock(&kvm->slots_lock); | |
1254 | return r; | |
749cf76c CD |
1255 | } |
1256 | ||
3401d546 CD |
1257 | static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, |
1258 | struct kvm_arm_device_addr *dev_addr) | |
1259 | { | |
330690cd CD |
1260 | unsigned long dev_id, type; |
1261 | ||
1262 | dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> | |
1263 | KVM_ARM_DEVICE_ID_SHIFT; | |
1264 | type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> | |
1265 | KVM_ARM_DEVICE_TYPE_SHIFT; | |
1266 | ||
1267 | switch (dev_id) { | |
1268 | case KVM_ARM_DEVICE_VGIC_V2: | |
c7da6fa4 PF |
1269 | if (!vgic_present) |
1270 | return -ENXIO; | |
ce01e4e8 | 1271 | return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); |
330690cd CD |
1272 | default: |
1273 | return -ENODEV; | |
1274 | } | |
3401d546 CD |
1275 | } |
1276 | ||
749cf76c CD |
1277 | long kvm_arch_vm_ioctl(struct file *filp, |
1278 | unsigned int ioctl, unsigned long arg) | |
1279 | { | |
3401d546 CD |
1280 | struct kvm *kvm = filp->private_data; |
1281 | void __user *argp = (void __user *)arg; | |
1282 | ||
1283 | switch (ioctl) { | |
5863c2ce | 1284 | case KVM_CREATE_IRQCHIP: { |
a28ebea2 | 1285 | int ret; |
c7da6fa4 PF |
1286 | if (!vgic_present) |
1287 | return -ENXIO; | |
a28ebea2 CD |
1288 | mutex_lock(&kvm->lock); |
1289 | ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); | |
1290 | mutex_unlock(&kvm->lock); | |
1291 | return ret; | |
5863c2ce | 1292 | } |
3401d546 CD |
1293 | case KVM_ARM_SET_DEVICE_ADDR: { |
1294 | struct kvm_arm_device_addr dev_addr; | |
1295 | ||
1296 | if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) | |
1297 | return -EFAULT; | |
1298 | return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); | |
1299 | } | |
42c4e0c7 AP |
1300 | case KVM_ARM_PREFERRED_TARGET: { |
1301 | int err; | |
1302 | struct kvm_vcpu_init init; | |
1303 | ||
1304 | err = kvm_vcpu_preferred_target(&init); | |
1305 | if (err) | |
1306 | return err; | |
1307 | ||
1308 | if (copy_to_user(argp, &init, sizeof(init))) | |
1309 | return -EFAULT; | |
1310 | ||
1311 | return 0; | |
1312 | } | |
3401d546 CD |
1313 | default: |
1314 | return -EINVAL; | |
1315 | } | |
749cf76c CD |
1316 | } |
1317 | ||
d157f4a5 | 1318 | static void cpu_init_hyp_mode(void *dummy) |
342cd0ab | 1319 | { |
dac288f7 | 1320 | phys_addr_t pgd_ptr; |
342cd0ab CD |
1321 | unsigned long hyp_stack_ptr; |
1322 | unsigned long stack_page; | |
1323 | unsigned long vector_ptr; | |
1324 | ||
1325 | /* Switch from the HYP stub to our own HYP init vector */ | |
5a677ce0 | 1326 | __hyp_set_vectors(kvm_get_idmap_vector()); |
342cd0ab | 1327 | |
dac288f7 | 1328 | pgd_ptr = kvm_mmu_get_httbr(); |
1436c1aa | 1329 | stack_page = __this_cpu_read(kvm_arm_hyp_stack_page); |
342cd0ab | 1330 | hyp_stack_ptr = stack_page + PAGE_SIZE; |
6840bdd7 | 1331 | vector_ptr = (unsigned long)kvm_get_hyp_vector(); |
342cd0ab | 1332 | |
12fda812 | 1333 | __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr); |
35a2491a | 1334 | __cpu_init_stage2(); |
342cd0ab CD |
1335 | } |
1336 | ||
47eb3cba MZ |
1337 | static void cpu_hyp_reset(void) |
1338 | { | |
1339 | if (!is_kernel_in_hyp_mode()) | |
1340 | __hyp_reset_vectors(); | |
1341 | } | |
1342 | ||
5f5560b1 JM |
1343 | static void cpu_hyp_reinit(void) |
1344 | { | |
1e0cf16c MZ |
1345 | kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt); |
1346 | ||
47eb3cba MZ |
1347 | cpu_hyp_reset(); |
1348 | ||
9d47bb0d | 1349 | if (is_kernel_in_hyp_mode()) |
02d50cda | 1350 | kvm_timer_init_vhe(); |
9d47bb0d | 1351 | else |
47eb3cba | 1352 | cpu_init_hyp_mode(NULL); |
5b0d2cc2 | 1353 | |
da5a3ce6 | 1354 | kvm_arm_init_debug(); |
5b0d2cc2 CD |
1355 | |
1356 | if (vgic_present) | |
1357 | kvm_vgic_init_cpu_hardware(); | |
5f5560b1 JM |
1358 | } |
1359 | ||
67f69197 AT |
1360 | static void _kvm_arch_hardware_enable(void *discard) |
1361 | { | |
1362 | if (!__this_cpu_read(kvm_arm_hardware_enabled)) { | |
5f5560b1 | 1363 | cpu_hyp_reinit(); |
67f69197 | 1364 | __this_cpu_write(kvm_arm_hardware_enabled, 1); |
d157f4a5 | 1365 | } |
67f69197 | 1366 | } |
d157f4a5 | 1367 | |
67f69197 AT |
1368 | int kvm_arch_hardware_enable(void) |
1369 | { | |
1370 | _kvm_arch_hardware_enable(NULL); | |
1371 | return 0; | |
342cd0ab CD |
1372 | } |
1373 | ||
67f69197 AT |
1374 | static void _kvm_arch_hardware_disable(void *discard) |
1375 | { | |
1376 | if (__this_cpu_read(kvm_arm_hardware_enabled)) { | |
1377 | cpu_hyp_reset(); | |
1378 | __this_cpu_write(kvm_arm_hardware_enabled, 0); | |
1379 | } | |
1380 | } | |
1381 | ||
1382 | void kvm_arch_hardware_disable(void) | |
1383 | { | |
1384 | _kvm_arch_hardware_disable(NULL); | |
1385 | } | |
d157f4a5 | 1386 | |
1fcf7ce0 LP |
1387 | #ifdef CONFIG_CPU_PM |
1388 | static int hyp_init_cpu_pm_notifier(struct notifier_block *self, | |
1389 | unsigned long cmd, | |
1390 | void *v) | |
1391 | { | |
67f69197 AT |
1392 | /* |
1393 | * kvm_arm_hardware_enabled is left with its old value over | |
1394 | * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should | |
1395 | * re-enable hyp. | |
1396 | */ | |
1397 | switch (cmd) { | |
1398 | case CPU_PM_ENTER: | |
1399 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1400 | /* | |
1401 | * don't update kvm_arm_hardware_enabled here | |
1402 | * so that the hardware will be re-enabled | |
1403 | * when we resume. See below. | |
1404 | */ | |
1405 | cpu_hyp_reset(); | |
1406 | ||
1fcf7ce0 | 1407 | return NOTIFY_OK; |
58d6b15e | 1408 | case CPU_PM_ENTER_FAILED: |
67f69197 AT |
1409 | case CPU_PM_EXIT: |
1410 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1411 | /* The hardware was enabled before suspend. */ | |
1412 | cpu_hyp_reinit(); | |
1fcf7ce0 | 1413 | |
67f69197 AT |
1414 | return NOTIFY_OK; |
1415 | ||
1416 | default: | |
1417 | return NOTIFY_DONE; | |
1418 | } | |
1fcf7ce0 LP |
1419 | } |
1420 | ||
1421 | static struct notifier_block hyp_init_cpu_pm_nb = { | |
1422 | .notifier_call = hyp_init_cpu_pm_notifier, | |
1423 | }; | |
1424 | ||
1425 | static void __init hyp_cpu_pm_init(void) | |
1426 | { | |
1427 | cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); | |
1428 | } | |
06a71a24 SH |
1429 | static void __init hyp_cpu_pm_exit(void) |
1430 | { | |
1431 | cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); | |
1432 | } | |
1fcf7ce0 LP |
1433 | #else |
1434 | static inline void hyp_cpu_pm_init(void) | |
1435 | { | |
1436 | } | |
06a71a24 SH |
1437 | static inline void hyp_cpu_pm_exit(void) |
1438 | { | |
1439 | } | |
1fcf7ce0 LP |
1440 | #endif |
1441 | ||
1e947bad MZ |
1442 | static int init_common_resources(void) |
1443 | { | |
0f62f0e9 SP |
1444 | kvm_set_ipa_limit(); |
1445 | ||
1e947bad MZ |
1446 | return 0; |
1447 | } | |
1448 | ||
1449 | static int init_subsystems(void) | |
1450 | { | |
67f69197 | 1451 | int err = 0; |
1e947bad | 1452 | |
5f5560b1 | 1453 | /* |
67f69197 | 1454 | * Enable hardware so that subsystem initialisation can access EL2. |
5f5560b1 | 1455 | */ |
67f69197 | 1456 | on_each_cpu(_kvm_arch_hardware_enable, NULL, 1); |
5f5560b1 JM |
1457 | |
1458 | /* | |
1459 | * Register CPU lower-power notifier | |
1460 | */ | |
1461 | hyp_cpu_pm_init(); | |
1462 | ||
1e947bad MZ |
1463 | /* |
1464 | * Init HYP view of VGIC | |
1465 | */ | |
1466 | err = kvm_vgic_hyp_init(); | |
1467 | switch (err) { | |
1468 | case 0: | |
1469 | vgic_present = true; | |
1470 | break; | |
1471 | case -ENODEV: | |
1472 | case -ENXIO: | |
1473 | vgic_present = false; | |
67f69197 | 1474 | err = 0; |
1e947bad MZ |
1475 | break; |
1476 | default: | |
67f69197 | 1477 | goto out; |
1e947bad MZ |
1478 | } |
1479 | ||
1480 | /* | |
1481 | * Init HYP architected timer support | |
1482 | */ | |
f384dcfe | 1483 | err = kvm_timer_hyp_init(vgic_present); |
1e947bad | 1484 | if (err) |
67f69197 | 1485 | goto out; |
1e947bad MZ |
1486 | |
1487 | kvm_perf_init(); | |
1488 | kvm_coproc_table_init(); | |
1489 | ||
67f69197 AT |
1490 | out: |
1491 | on_each_cpu(_kvm_arch_hardware_disable, NULL, 1); | |
1492 | ||
1493 | return err; | |
1e947bad MZ |
1494 | } |
1495 | ||
1496 | static void teardown_hyp_mode(void) | |
1497 | { | |
1498 | int cpu; | |
1499 | ||
1e947bad MZ |
1500 | free_hyp_pgds(); |
1501 | for_each_possible_cpu(cpu) | |
1502 | free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); | |
06a71a24 | 1503 | hyp_cpu_pm_exit(); |
1e947bad MZ |
1504 | } |
1505 | ||
342cd0ab CD |
1506 | /** |
1507 | * Inits Hyp-mode on all online CPUs | |
1508 | */ | |
1509 | static int init_hyp_mode(void) | |
1510 | { | |
342cd0ab CD |
1511 | int cpu; |
1512 | int err = 0; | |
1513 | ||
1514 | /* | |
1515 | * Allocate Hyp PGD and setup Hyp identity mapping | |
1516 | */ | |
1517 | err = kvm_mmu_init(); | |
1518 | if (err) | |
1519 | goto out_err; | |
1520 | ||
342cd0ab CD |
1521 | /* |
1522 | * Allocate stack pages for Hypervisor-mode | |
1523 | */ | |
1524 | for_each_possible_cpu(cpu) { | |
1525 | unsigned long stack_page; | |
1526 | ||
1527 | stack_page = __get_free_page(GFP_KERNEL); | |
1528 | if (!stack_page) { | |
1529 | err = -ENOMEM; | |
1e947bad | 1530 | goto out_err; |
342cd0ab CD |
1531 | } |
1532 | ||
1533 | per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; | |
1534 | } | |
1535 | ||
342cd0ab CD |
1536 | /* |
1537 | * Map the Hyp-code called directly from the host | |
1538 | */ | |
588ab3f9 | 1539 | err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), |
59002705 | 1540 | kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC); |
342cd0ab CD |
1541 | if (err) { |
1542 | kvm_err("Cannot map world-switch code\n"); | |
1e947bad | 1543 | goto out_err; |
342cd0ab CD |
1544 | } |
1545 | ||
a0bf9776 | 1546 | err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), |
74a6b888 | 1547 | kvm_ksym_ref(__end_rodata), PAGE_HYP_RO); |
910917bb MZ |
1548 | if (err) { |
1549 | kvm_err("Cannot map rodata section\n"); | |
c8ea0395 MZ |
1550 | goto out_err; |
1551 | } | |
1552 | ||
1553 | err = create_hyp_mappings(kvm_ksym_ref(__bss_start), | |
1554 | kvm_ksym_ref(__bss_stop), PAGE_HYP_RO); | |
1555 | if (err) { | |
1556 | kvm_err("Cannot map bss section\n"); | |
1e947bad | 1557 | goto out_err; |
910917bb MZ |
1558 | } |
1559 | ||
6840bdd7 MZ |
1560 | err = kvm_map_vectors(); |
1561 | if (err) { | |
1562 | kvm_err("Cannot map vectors\n"); | |
1563 | goto out_err; | |
1564 | } | |
1565 | ||
342cd0ab CD |
1566 | /* |
1567 | * Map the Hyp stack pages | |
1568 | */ | |
1569 | for_each_possible_cpu(cpu) { | |
1570 | char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); | |
c8dddecd MZ |
1571 | err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE, |
1572 | PAGE_HYP); | |
342cd0ab CD |
1573 | |
1574 | if (err) { | |
1575 | kvm_err("Cannot map hyp stack\n"); | |
1e947bad | 1576 | goto out_err; |
342cd0ab CD |
1577 | } |
1578 | } | |
1579 | ||
342cd0ab | 1580 | for_each_possible_cpu(cpu) { |
630a1685 | 1581 | kvm_host_data_t *cpu_data; |
342cd0ab | 1582 | |
630a1685 | 1583 | cpu_data = per_cpu_ptr(&kvm_host_data, cpu); |
630a1685 | 1584 | err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP); |
342cd0ab CD |
1585 | |
1586 | if (err) { | |
3de50da6 | 1587 | kvm_err("Cannot map host CPU state: %d\n", err); |
1e947bad | 1588 | goto out_err; |
342cd0ab CD |
1589 | } |
1590 | } | |
1591 | ||
55e3748e MZ |
1592 | err = hyp_map_aux_data(); |
1593 | if (err) | |
a37f0c3c | 1594 | kvm_err("Cannot map host auxiliary data: %d\n", err); |
55e3748e | 1595 | |
342cd0ab | 1596 | return 0; |
1e947bad | 1597 | |
342cd0ab | 1598 | out_err: |
1e947bad | 1599 | teardown_hyp_mode(); |
342cd0ab CD |
1600 | kvm_err("error initializing Hyp mode: %d\n", err); |
1601 | return err; | |
1602 | } | |
1603 | ||
d4e071ce AP |
1604 | static void check_kvm_target_cpu(void *ret) |
1605 | { | |
1606 | *(int *)ret = kvm_target_cpu(); | |
1607 | } | |
1608 | ||
4429fc64 AP |
1609 | struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) |
1610 | { | |
1611 | struct kvm_vcpu *vcpu; | |
1612 | int i; | |
1613 | ||
1614 | mpidr &= MPIDR_HWID_BITMASK; | |
1615 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
1616 | if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) | |
1617 | return vcpu; | |
1618 | } | |
1619 | return NULL; | |
1620 | } | |
1621 | ||
2412405b EA |
1622 | bool kvm_arch_has_irq_bypass(void) |
1623 | { | |
1624 | return true; | |
1625 | } | |
1626 | ||
1627 | int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, | |
1628 | struct irq_bypass_producer *prod) | |
1629 | { | |
1630 | struct kvm_kernel_irqfd *irqfd = | |
1631 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1632 | ||
196b1364 MZ |
1633 | return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq, |
1634 | &irqfd->irq_entry); | |
2412405b EA |
1635 | } |
1636 | void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, | |
1637 | struct irq_bypass_producer *prod) | |
1638 | { | |
1639 | struct kvm_kernel_irqfd *irqfd = | |
1640 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1641 | ||
196b1364 MZ |
1642 | kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq, |
1643 | &irqfd->irq_entry); | |
2412405b EA |
1644 | } |
1645 | ||
1646 | void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons) | |
1647 | { | |
1648 | struct kvm_kernel_irqfd *irqfd = | |
1649 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1650 | ||
1651 | kvm_arm_halt_guest(irqfd->kvm); | |
1652 | } | |
1653 | ||
1654 | void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons) | |
1655 | { | |
1656 | struct kvm_kernel_irqfd *irqfd = | |
1657 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1658 | ||
1659 | kvm_arm_resume_guest(irqfd->kvm); | |
1660 | } | |
1661 | ||
342cd0ab CD |
1662 | /** |
1663 | * Initialize Hyp-mode and memory mappings on all CPUs. | |
1664 | */ | |
749cf76c CD |
1665 | int kvm_arch_init(void *opaque) |
1666 | { | |
342cd0ab | 1667 | int err; |
d4e071ce | 1668 | int ret, cpu; |
fe7d7b03 | 1669 | bool in_hyp_mode; |
342cd0ab CD |
1670 | |
1671 | if (!is_hyp_mode_available()) { | |
58d0d19a | 1672 | kvm_info("HYP mode not available\n"); |
342cd0ab CD |
1673 | return -ENODEV; |
1674 | } | |
1675 | ||
33e5f4e5 MZ |
1676 | in_hyp_mode = is_kernel_in_hyp_mode(); |
1677 | ||
1678 | if (!in_hyp_mode && kvm_arch_requires_vhe()) { | |
1679 | kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n"); | |
85acda3b DM |
1680 | return -ENODEV; |
1681 | } | |
1682 | ||
d4e071ce AP |
1683 | for_each_online_cpu(cpu) { |
1684 | smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); | |
1685 | if (ret < 0) { | |
1686 | kvm_err("Error, CPU %d not supported!\n", cpu); | |
1687 | return -ENODEV; | |
1688 | } | |
342cd0ab CD |
1689 | } |
1690 | ||
1e947bad | 1691 | err = init_common_resources(); |
342cd0ab | 1692 | if (err) |
1e947bad | 1693 | return err; |
342cd0ab | 1694 | |
a3be836d | 1695 | err = kvm_arm_init_sve(); |
0f062bfe DM |
1696 | if (err) |
1697 | return err; | |
1698 | ||
fe7d7b03 | 1699 | if (!in_hyp_mode) { |
1e947bad | 1700 | err = init_hyp_mode(); |
fe7d7b03 JT |
1701 | if (err) |
1702 | goto out_err; | |
1703 | } | |
8146875d | 1704 | |
1e947bad MZ |
1705 | err = init_subsystems(); |
1706 | if (err) | |
1707 | goto out_hyp; | |
1fcf7ce0 | 1708 | |
fe7d7b03 JT |
1709 | if (in_hyp_mode) |
1710 | kvm_info("VHE mode initialized successfully\n"); | |
1711 | else | |
1712 | kvm_info("Hyp mode initialized successfully\n"); | |
1713 | ||
749cf76c | 1714 | return 0; |
1e947bad MZ |
1715 | |
1716 | out_hyp: | |
fe7d7b03 JT |
1717 | if (!in_hyp_mode) |
1718 | teardown_hyp_mode(); | |
342cd0ab CD |
1719 | out_err: |
1720 | return err; | |
749cf76c CD |
1721 | } |
1722 | ||
1723 | /* NOP: Compiling as a module not supported */ | |
1724 | void kvm_arch_exit(void) | |
1725 | { | |
210552c1 | 1726 | kvm_perf_teardown(); |
749cf76c CD |
1727 | } |
1728 | ||
1729 | static int arm_init(void) | |
1730 | { | |
1731 | int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1732 | return rc; | |
1733 | } | |
1734 | ||
1735 | module_init(arm_init); |