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 | 42 | |
8564d637 SP |
43 | #include <kvm/arm_hypercalls.h> |
44 | #include <kvm/arm_pmu.h> | |
45 | #include <kvm/arm_psci.h> | |
46 | ||
749cf76c CD |
47 | #ifdef REQUIRES_VIRT |
48 | __asm__(".arch_extension virt"); | |
49 | #endif | |
50 | ||
630a1685 | 51 | DEFINE_PER_CPU(kvm_host_data_t, kvm_host_data); |
342cd0ab | 52 | static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); |
342cd0ab | 53 | |
1638a12d MZ |
54 | /* Per-CPU variable containing the currently running vcpu. */ |
55 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu); | |
56 | ||
f7ed45be CD |
57 | /* The VMID used in the VTTBR */ |
58 | static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1); | |
20475f78 | 59 | static u32 kvm_next_vmid; |
fb544d1c | 60 | static DEFINE_SPINLOCK(kvm_vmid_lock); |
342cd0ab | 61 | |
c7da6fa4 PF |
62 | static bool vgic_present; |
63 | ||
67f69197 AT |
64 | static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled); |
65 | ||
1638a12d MZ |
66 | static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu) |
67 | { | |
1436c1aa | 68 | __this_cpu_write(kvm_arm_running_vcpu, vcpu); |
1638a12d MZ |
69 | } |
70 | ||
61bbe380 CD |
71 | DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use); |
72 | ||
1638a12d MZ |
73 | /** |
74 | * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU. | |
75 | * Must be called from non-preemptible context | |
76 | */ | |
77 | struct kvm_vcpu *kvm_arm_get_running_vcpu(void) | |
78 | { | |
1436c1aa | 79 | return __this_cpu_read(kvm_arm_running_vcpu); |
1638a12d MZ |
80 | } |
81 | ||
82 | /** | |
83 | * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
84 | */ | |
4000be42 | 85 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) |
1638a12d MZ |
86 | { |
87 | return &kvm_arm_running_vcpu; | |
88 | } | |
89 | ||
749cf76c CD |
90 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
91 | { | |
92 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; | |
93 | } | |
94 | ||
749cf76c CD |
95 | int kvm_arch_hardware_setup(void) |
96 | { | |
97 | return 0; | |
98 | } | |
99 | ||
f257d6dc | 100 | int kvm_arch_check_processor_compat(void) |
749cf76c | 101 | { |
f257d6dc | 102 | return 0; |
749cf76c CD |
103 | } |
104 | ||
c726200d CD |
105 | int kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
106 | struct kvm_enable_cap *cap) | |
107 | { | |
108 | int r; | |
109 | ||
110 | if (cap->flags) | |
111 | return -EINVAL; | |
112 | ||
113 | switch (cap->cap) { | |
114 | case KVM_CAP_ARM_NISV_TO_USER: | |
115 | r = 0; | |
116 | kvm->arch.return_nisv_io_abort_to_user = true; | |
117 | break; | |
118 | default: | |
119 | r = -EINVAL; | |
120 | break; | |
121 | } | |
122 | ||
123 | return r; | |
124 | } | |
749cf76c | 125 | |
d5d8184d CD |
126 | /** |
127 | * kvm_arch_init_vm - initializes a VM data structure | |
128 | * @kvm: pointer to the KVM struct | |
129 | */ | |
749cf76c CD |
130 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
131 | { | |
94d0e598 | 132 | int ret, cpu; |
d5d8184d | 133 | |
bca607eb | 134 | ret = kvm_arm_setup_stage2(kvm, type); |
5b6c6742 SP |
135 | if (ret) |
136 | return ret; | |
749cf76c | 137 | |
94d0e598 MZ |
138 | kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran)); |
139 | if (!kvm->arch.last_vcpu_ran) | |
140 | return -ENOMEM; | |
141 | ||
142 | for_each_possible_cpu(cpu) | |
143 | *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1; | |
144 | ||
d5d8184d CD |
145 | ret = kvm_alloc_stage2_pgd(kvm); |
146 | if (ret) | |
147 | goto out_fail_alloc; | |
148 | ||
c8dddecd | 149 | ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP); |
d5d8184d CD |
150 | if (ret) |
151 | goto out_free_stage2_pgd; | |
152 | ||
6c3d63c9 | 153 | kvm_vgic_early_init(kvm); |
a1a64387 | 154 | |
d5d8184d | 155 | /* Mark the initial VMID generation invalid */ |
e329fb75 | 156 | kvm->arch.vmid.vmid_gen = 0; |
d5d8184d | 157 | |
3caa2d8c | 158 | /* The maximum number of VCPUs is limited by the host's GIC model */ |
c7da6fa4 PF |
159 | kvm->arch.max_vcpus = vgic_present ? |
160 | kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; | |
3caa2d8c | 161 | |
d5d8184d CD |
162 | return ret; |
163 | out_free_stage2_pgd: | |
164 | kvm_free_stage2_pgd(kvm); | |
165 | out_fail_alloc: | |
94d0e598 MZ |
166 | free_percpu(kvm->arch.last_vcpu_ran); |
167 | kvm->arch.last_vcpu_ran = NULL; | |
d5d8184d | 168 | return ret; |
749cf76c CD |
169 | } |
170 | ||
235539b4 LC |
171 | int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
172 | { | |
173 | return 0; | |
174 | } | |
175 | ||
1499fa80 | 176 | vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
749cf76c CD |
177 | { |
178 | return VM_FAULT_SIGBUS; | |
179 | } | |
180 | ||
749cf76c | 181 | |
d5d8184d CD |
182 | /** |
183 | * kvm_arch_destroy_vm - destroy the VM data structure | |
184 | * @kvm: pointer to the KVM struct | |
185 | */ | |
749cf76c CD |
186 | void kvm_arch_destroy_vm(struct kvm *kvm) |
187 | { | |
188 | int i; | |
189 | ||
b2c9a85d MZ |
190 | kvm_vgic_destroy(kvm); |
191 | ||
94d0e598 MZ |
192 | free_percpu(kvm->arch.last_vcpu_ran); |
193 | kvm->arch.last_vcpu_ran = NULL; | |
194 | ||
749cf76c CD |
195 | for (i = 0; i < KVM_MAX_VCPUS; ++i) { |
196 | if (kvm->vcpus[i]) { | |
4b8fff78 | 197 | kvm_arch_vcpu_destroy(kvm->vcpus[i]); |
749cf76c CD |
198 | kvm->vcpus[i] = NULL; |
199 | } | |
200 | } | |
6b2ad81b | 201 | atomic_set(&kvm->online_vcpus, 0); |
749cf76c CD |
202 | } |
203 | ||
784aa3d7 | 204 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
749cf76c CD |
205 | { |
206 | int r; | |
207 | switch (ext) { | |
1a89dd91 | 208 | case KVM_CAP_IRQCHIP: |
c7da6fa4 PF |
209 | r = vgic_present; |
210 | break; | |
d44758c0 | 211 | case KVM_CAP_IOEVENTFD: |
7330672b | 212 | case KVM_CAP_DEVICE_CTRL: |
749cf76c CD |
213 | case KVM_CAP_USER_MEMORY: |
214 | case KVM_CAP_SYNC_MMU: | |
215 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
216 | case KVM_CAP_ONE_REG: | |
aa024c2f | 217 | case KVM_CAP_ARM_PSCI: |
4447a208 | 218 | case KVM_CAP_ARM_PSCI_0_2: |
98047888 | 219 | case KVM_CAP_READONLY_MEM: |
ecccf0cc | 220 | case KVM_CAP_MP_STATE: |
460df4c1 | 221 | case KVM_CAP_IMMEDIATE_EXIT: |
58bf437f | 222 | case KVM_CAP_VCPU_EVENTS: |
92f35b75 | 223 | case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2: |
c726200d | 224 | case KVM_CAP_ARM_NISV_TO_USER: |
da345174 | 225 | case KVM_CAP_ARM_INJECT_EXT_DABT: |
749cf76c CD |
226 | r = 1; |
227 | break; | |
3401d546 CD |
228 | case KVM_CAP_ARM_SET_DEVICE_ADDR: |
229 | r = 1; | |
ca46e10f | 230 | break; |
749cf76c CD |
231 | case KVM_CAP_NR_VCPUS: |
232 | r = num_online_cpus(); | |
233 | break; | |
234 | case KVM_CAP_MAX_VCPUS: | |
235 | r = KVM_MAX_VCPUS; | |
236 | break; | |
a86cb413 TH |
237 | case KVM_CAP_MAX_VCPU_ID: |
238 | r = KVM_MAX_VCPU_ID; | |
239 | break; | |
2988509d VM |
240 | case KVM_CAP_MSI_DEVID: |
241 | if (!kvm) | |
242 | r = -EINVAL; | |
243 | else | |
244 | r = kvm->arch.vgic.msis_require_devid; | |
245 | break; | |
f7214e60 CD |
246 | case KVM_CAP_ARM_USER_IRQ: |
247 | /* | |
248 | * 1: EL1_VTIMER, EL1_PTIMER, and PMU. | |
249 | * (bump this number if adding more devices) | |
250 | */ | |
251 | r = 1; | |
252 | break; | |
749cf76c | 253 | default: |
375bdd3b | 254 | r = kvm_arch_vm_ioctl_check_extension(kvm, ext); |
749cf76c CD |
255 | break; |
256 | } | |
257 | return r; | |
258 | } | |
259 | ||
260 | long kvm_arch_dev_ioctl(struct file *filp, | |
261 | unsigned int ioctl, unsigned long arg) | |
262 | { | |
263 | return -EINVAL; | |
264 | } | |
265 | ||
d1e5b0e9 MO |
266 | struct kvm *kvm_arch_alloc_vm(void) |
267 | { | |
268 | if (!has_vhe()) | |
269 | return kzalloc(sizeof(struct kvm), GFP_KERNEL); | |
270 | ||
271 | return vzalloc(sizeof(struct kvm)); | |
272 | } | |
273 | ||
274 | void kvm_arch_free_vm(struct kvm *kvm) | |
275 | { | |
276 | if (!has_vhe()) | |
277 | kfree(kvm); | |
278 | else | |
279 | vfree(kvm); | |
280 | } | |
749cf76c CD |
281 | |
282 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) | |
283 | { | |
284 | int err; | |
285 | struct kvm_vcpu *vcpu; | |
286 | ||
716139df CD |
287 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) { |
288 | err = -EBUSY; | |
289 | goto out; | |
290 | } | |
291 | ||
3caa2d8c AP |
292 | if (id >= kvm->arch.max_vcpus) { |
293 | err = -EINVAL; | |
294 | goto out; | |
295 | } | |
296 | ||
749cf76c CD |
297 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
298 | if (!vcpu) { | |
299 | err = -ENOMEM; | |
300 | goto out; | |
301 | } | |
302 | ||
303 | err = kvm_vcpu_init(vcpu, kvm, id); | |
304 | if (err) | |
305 | goto free_vcpu; | |
306 | ||
c8dddecd | 307 | err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP); |
d5d8184d CD |
308 | if (err) |
309 | goto vcpu_uninit; | |
310 | ||
749cf76c | 311 | return vcpu; |
d5d8184d CD |
312 | vcpu_uninit: |
313 | kvm_vcpu_uninit(vcpu); | |
749cf76c CD |
314 | free_vcpu: |
315 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
316 | out: | |
317 | return ERR_PTR(err); | |
318 | } | |
319 | ||
31928aa5 | 320 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
749cf76c | 321 | { |
749cf76c CD |
322 | } |
323 | ||
4b8fff78 | 324 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) |
749cf76c | 325 | { |
f1d7231c CD |
326 | if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm))) |
327 | static_branch_dec(&userspace_irqchip_in_use); | |
328 | ||
d5d8184d | 329 | kvm_mmu_free_memory_caches(vcpu); |
967f8427 | 330 | kvm_timer_vcpu_terminate(vcpu); |
5f0a714a | 331 | kvm_pmu_vcpu_destroy(vcpu); |
591d215a | 332 | kvm_vcpu_uninit(vcpu); |
d5d8184d | 333 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
749cf76c CD |
334 | } |
335 | ||
749cf76c CD |
336 | int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) |
337 | { | |
1c88ab7e | 338 | return kvm_timer_is_pending(vcpu); |
749cf76c CD |
339 | } |
340 | ||
d35268da CD |
341 | void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) |
342 | { | |
5eeaf10e MZ |
343 | /* |
344 | * If we're about to block (most likely because we've just hit a | |
345 | * WFI), we need to sync back the state of the GIC CPU interface | |
8e01d9a3 | 346 | * so that we have the latest PMR and group enables. This ensures |
5eeaf10e MZ |
347 | * that kvm_arch_vcpu_runnable has up-to-date data to decide |
348 | * whether we have pending interrupts. | |
8e01d9a3 MZ |
349 | * |
350 | * For the same reason, we want to tell GICv4 that we need | |
351 | * doorbells to be signalled, should an interrupt become pending. | |
5eeaf10e MZ |
352 | */ |
353 | preempt_disable(); | |
354 | kvm_vgic_vmcr_sync(vcpu); | |
8e01d9a3 | 355 | vgic_v4_put(vcpu, true); |
5eeaf10e | 356 | preempt_enable(); |
d35268da CD |
357 | } |
358 | ||
359 | void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
360 | { | |
8e01d9a3 MZ |
361 | preempt_disable(); |
362 | vgic_v4_load(vcpu); | |
363 | preempt_enable(); | |
d35268da CD |
364 | } |
365 | ||
749cf76c CD |
366 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
367 | { | |
f7ed45be CD |
368 | /* Force users to call KVM_ARM_VCPU_INIT */ |
369 | vcpu->arch.target = -1; | |
f7fa034d | 370 | bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); |
1a89dd91 | 371 | |
967f8427 MZ |
372 | /* Set up the timer */ |
373 | kvm_timer_vcpu_init(vcpu); | |
374 | ||
bca031e2 ZY |
375 | kvm_pmu_vcpu_init(vcpu); |
376 | ||
84e690bf AB |
377 | kvm_arm_reset_debug_ptr(vcpu); |
378 | ||
8564d637 SP |
379 | kvm_arm_pvtime_vcpu_init(&vcpu->arch); |
380 | ||
1aab6f46 | 381 | return kvm_vgic_vcpu_init(vcpu); |
749cf76c CD |
382 | } |
383 | ||
749cf76c CD |
384 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
385 | { | |
94d0e598 | 386 | int *last_ran; |
630a1685 | 387 | kvm_host_data_t *cpu_data; |
94d0e598 MZ |
388 | |
389 | last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran); | |
630a1685 | 390 | cpu_data = this_cpu_ptr(&kvm_host_data); |
94d0e598 MZ |
391 | |
392 | /* | |
393 | * We might get preempted before the vCPU actually runs, but | |
394 | * over-invalidation doesn't affect correctness. | |
395 | */ | |
396 | if (*last_ran != vcpu->vcpu_id) { | |
397 | kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu); | |
398 | *last_ran = vcpu->vcpu_id; | |
399 | } | |
400 | ||
86ce8535 | 401 | vcpu->cpu = cpu; |
630a1685 | 402 | vcpu->arch.host_cpu_context = &cpu_data->host_ctxt; |
5b3e5e5b | 403 | |
1638a12d | 404 | kvm_arm_set_running_vcpu(vcpu); |
328e5664 | 405 | kvm_vgic_load(vcpu); |
b103cc3f | 406 | kvm_timer_vcpu_load(vcpu); |
bc192cee | 407 | kvm_vcpu_load_sysregs(vcpu); |
e6b673b7 | 408 | kvm_arch_vcpu_load_fp(vcpu); |
435e53fb | 409 | kvm_vcpu_pmu_restore_guest(vcpu); |
8564d637 SP |
410 | if (kvm_arm_is_pvtime_enabled(&vcpu->arch)) |
411 | kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu); | |
de737089 MZ |
412 | |
413 | if (single_task_running()) | |
ef2e78dd | 414 | vcpu_clear_wfx_traps(vcpu); |
de737089 | 415 | else |
ef2e78dd | 416 | vcpu_set_wfx_traps(vcpu); |
384b40ca MR |
417 | |
418 | vcpu_ptrauth_setup_lazy(vcpu); | |
749cf76c CD |
419 | } |
420 | ||
421 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
422 | { | |
e6b673b7 | 423 | kvm_arch_vcpu_put_fp(vcpu); |
bc192cee | 424 | kvm_vcpu_put_sysregs(vcpu); |
b103cc3f | 425 | kvm_timer_vcpu_put(vcpu); |
328e5664 | 426 | kvm_vgic_put(vcpu); |
435e53fb | 427 | kvm_vcpu_pmu_restore_host(vcpu); |
328e5664 | 428 | |
e9b152cb CD |
429 | vcpu->cpu = -1; |
430 | ||
1638a12d | 431 | kvm_arm_set_running_vcpu(NULL); |
749cf76c CD |
432 | } |
433 | ||
424c989b AJ |
434 | static void vcpu_power_off(struct kvm_vcpu *vcpu) |
435 | { | |
436 | vcpu->arch.power_off = true; | |
7b244e2b | 437 | kvm_make_request(KVM_REQ_SLEEP, vcpu); |
424c989b AJ |
438 | kvm_vcpu_kick(vcpu); |
439 | } | |
440 | ||
749cf76c CD |
441 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
442 | struct kvm_mp_state *mp_state) | |
443 | { | |
3781528e | 444 | if (vcpu->arch.power_off) |
ecccf0cc AB |
445 | mp_state->mp_state = KVM_MP_STATE_STOPPED; |
446 | else | |
447 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
448 | ||
449 | return 0; | |
749cf76c CD |
450 | } |
451 | ||
452 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
453 | struct kvm_mp_state *mp_state) | |
454 | { | |
e83dff5e CD |
455 | int ret = 0; |
456 | ||
ecccf0cc AB |
457 | switch (mp_state->mp_state) { |
458 | case KVM_MP_STATE_RUNNABLE: | |
3781528e | 459 | vcpu->arch.power_off = false; |
ecccf0cc AB |
460 | break; |
461 | case KVM_MP_STATE_STOPPED: | |
424c989b | 462 | vcpu_power_off(vcpu); |
ecccf0cc AB |
463 | break; |
464 | default: | |
e83dff5e | 465 | ret = -EINVAL; |
ecccf0cc AB |
466 | } |
467 | ||
e83dff5e | 468 | return ret; |
749cf76c CD |
469 | } |
470 | ||
5b3e5e5b CD |
471 | /** |
472 | * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled | |
473 | * @v: The VCPU pointer | |
474 | * | |
475 | * If the guest CPU is not waiting for interrupts or an interrupt line is | |
476 | * asserted, the CPU is by definition runnable. | |
477 | */ | |
749cf76c CD |
478 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) |
479 | { | |
3df59d8d CD |
480 | bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF); |
481 | return ((irq_lines || kvm_vgic_vcpu_pending_irq(v)) | |
3b92830a | 482 | && !v->arch.power_off && !v->arch.pause); |
749cf76c CD |
483 | } |
484 | ||
199b5763 LM |
485 | bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) |
486 | { | |
f01fbd2f | 487 | return vcpu_mode_priv(vcpu); |
199b5763 LM |
488 | } |
489 | ||
f7ed45be CD |
490 | /* Just ensure a guest exit from a particular CPU */ |
491 | static void exit_vm_noop(void *info) | |
492 | { | |
493 | } | |
494 | ||
495 | void force_vm_exit(const cpumask_t *mask) | |
496 | { | |
898f949f | 497 | preempt_disable(); |
f7ed45be | 498 | smp_call_function_many(mask, exit_vm_noop, NULL, true); |
898f949f | 499 | preempt_enable(); |
f7ed45be CD |
500 | } |
501 | ||
502 | /** | |
503 | * need_new_vmid_gen - check that the VMID is still valid | |
e329fb75 | 504 | * @vmid: The VMID to check |
f7ed45be CD |
505 | * |
506 | * return true if there is a new generation of VMIDs being used | |
507 | * | |
e329fb75 CD |
508 | * The hardware supports a limited set of values with the value zero reserved |
509 | * for the host, so we check if an assigned value belongs to a previous | |
510 | * generation, which which requires us to assign a new value. If we're the | |
511 | * first to use a VMID for the new generation, we must flush necessary caches | |
512 | * and TLBs on all CPUs. | |
f7ed45be | 513 | */ |
e329fb75 | 514 | static bool need_new_vmid_gen(struct kvm_vmid *vmid) |
f7ed45be | 515 | { |
fb544d1c CD |
516 | u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen); |
517 | smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */ | |
e329fb75 | 518 | return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen); |
f7ed45be CD |
519 | } |
520 | ||
521 | /** | |
e329fb75 CD |
522 | * update_vmid - Update the vmid with a valid VMID for the current generation |
523 | * @kvm: The guest that struct vmid belongs to | |
524 | * @vmid: The stage-2 VMID information struct | |
f7ed45be | 525 | */ |
e329fb75 | 526 | static void update_vmid(struct kvm_vmid *vmid) |
f7ed45be | 527 | { |
e329fb75 | 528 | if (!need_new_vmid_gen(vmid)) |
f7ed45be CD |
529 | return; |
530 | ||
fb544d1c | 531 | spin_lock(&kvm_vmid_lock); |
f7ed45be CD |
532 | |
533 | /* | |
534 | * We need to re-check the vmid_gen here to ensure that if another vcpu | |
535 | * already allocated a valid vmid for this vm, then this vcpu should | |
536 | * use the same vmid. | |
537 | */ | |
e329fb75 | 538 | if (!need_new_vmid_gen(vmid)) { |
fb544d1c | 539 | spin_unlock(&kvm_vmid_lock); |
f7ed45be CD |
540 | return; |
541 | } | |
542 | ||
543 | /* First user of a new VMID generation? */ | |
544 | if (unlikely(kvm_next_vmid == 0)) { | |
545 | atomic64_inc(&kvm_vmid_gen); | |
546 | kvm_next_vmid = 1; | |
547 | ||
548 | /* | |
549 | * On SMP we know no other CPUs can use this CPU's or each | |
550 | * other's VMID after force_vm_exit returns since the | |
551 | * kvm_vmid_lock blocks them from reentry to the guest. | |
552 | */ | |
553 | force_vm_exit(cpu_all_mask); | |
554 | /* | |
555 | * Now broadcast TLB + ICACHE invalidation over the inner | |
556 | * shareable domain to make sure all data structures are | |
557 | * clean. | |
558 | */ | |
559 | kvm_call_hyp(__kvm_flush_vm_context); | |
560 | } | |
561 | ||
e329fb75 | 562 | vmid->vmid = kvm_next_vmid; |
f7ed45be | 563 | kvm_next_vmid++; |
e329fb75 | 564 | kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1; |
f7ed45be | 565 | |
fb544d1c | 566 | smp_wmb(); |
e329fb75 | 567 | WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen)); |
fb544d1c CD |
568 | |
569 | spin_unlock(&kvm_vmid_lock); | |
f7ed45be CD |
570 | } |
571 | ||
f7ed45be CD |
572 | static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) |
573 | { | |
05971120 | 574 | struct kvm *kvm = vcpu->kvm; |
41a54482 | 575 | int ret = 0; |
e1ba0207 | 576 | |
f7ed45be CD |
577 | if (likely(vcpu->arch.has_run_once)) |
578 | return 0; | |
579 | ||
7dd32a0d DM |
580 | if (!kvm_arm_vcpu_is_finalized(vcpu)) |
581 | return -EPERM; | |
582 | ||
f7ed45be | 583 | vcpu->arch.has_run_once = true; |
aa024c2f | 584 | |
61bbe380 CD |
585 | if (likely(irqchip_in_kernel(kvm))) { |
586 | /* | |
587 | * Map the VGIC hardware resources before running a vcpu the | |
588 | * first time on this VM. | |
589 | */ | |
590 | if (unlikely(!vgic_ready(kvm))) { | |
591 | ret = kvm_vgic_map_resources(kvm); | |
592 | if (ret) | |
593 | return ret; | |
594 | } | |
595 | } else { | |
596 | /* | |
597 | * Tell the rest of the code that there are userspace irqchip | |
598 | * VMs in the wild. | |
599 | */ | |
600 | static_branch_inc(&userspace_irqchip_in_use); | |
01ac5e34 MZ |
601 | } |
602 | ||
d9e13977 | 603 | ret = kvm_timer_enable(vcpu); |
a2befacf CD |
604 | if (ret) |
605 | return ret; | |
606 | ||
607 | ret = kvm_arm_pmu_v3_enable(vcpu); | |
05971120 | 608 | |
41a54482 | 609 | return ret; |
f7ed45be CD |
610 | } |
611 | ||
c1426e4c EA |
612 | bool kvm_arch_intc_initialized(struct kvm *kvm) |
613 | { | |
614 | return vgic_initialized(kvm); | |
615 | } | |
616 | ||
b13216cf | 617 | void kvm_arm_halt_guest(struct kvm *kvm) |
3b92830a EA |
618 | { |
619 | int i; | |
620 | struct kvm_vcpu *vcpu; | |
621 | ||
622 | kvm_for_each_vcpu(i, vcpu, kvm) | |
623 | vcpu->arch.pause = true; | |
7b244e2b | 624 | kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP); |
3b92830a EA |
625 | } |
626 | ||
b13216cf | 627 | void kvm_arm_resume_guest(struct kvm *kvm) |
3b92830a EA |
628 | { |
629 | int i; | |
630 | struct kvm_vcpu *vcpu; | |
631 | ||
abd72296 CD |
632 | kvm_for_each_vcpu(i, vcpu, kvm) { |
633 | vcpu->arch.pause = false; | |
b3dae109 | 634 | swake_up_one(kvm_arch_vcpu_wq(vcpu)); |
abd72296 | 635 | } |
3b92830a EA |
636 | } |
637 | ||
7b244e2b | 638 | static void vcpu_req_sleep(struct kvm_vcpu *vcpu) |
aa024c2f | 639 | { |
8577370f | 640 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); |
aa024c2f | 641 | |
b3dae109 | 642 | swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) && |
3b92830a | 643 | (!vcpu->arch.pause))); |
0592c005 | 644 | |
424c989b | 645 | if (vcpu->arch.power_off || vcpu->arch.pause) { |
0592c005 | 646 | /* Awaken to handle a signal, request we sleep again later. */ |
7b244e2b | 647 | kvm_make_request(KVM_REQ_SLEEP, vcpu); |
0592c005 | 648 | } |
358b28f0 MZ |
649 | |
650 | /* | |
651 | * Make sure we will observe a potential reset request if we've | |
652 | * observed a change to the power state. Pairs with the smp_wmb() in | |
653 | * kvm_psci_vcpu_on(). | |
654 | */ | |
655 | smp_rmb(); | |
aa024c2f MZ |
656 | } |
657 | ||
e8180dca AP |
658 | static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) |
659 | { | |
660 | return vcpu->arch.target >= 0; | |
661 | } | |
662 | ||
0592c005 AJ |
663 | static void check_vcpu_requests(struct kvm_vcpu *vcpu) |
664 | { | |
665 | if (kvm_request_pending(vcpu)) { | |
7b244e2b AJ |
666 | if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) |
667 | vcpu_req_sleep(vcpu); | |
325f9c64 | 668 | |
358b28f0 MZ |
669 | if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu)) |
670 | kvm_reset_vcpu(vcpu); | |
671 | ||
325f9c64 AJ |
672 | /* |
673 | * Clear IRQ_PENDING requests that were made to guarantee | |
674 | * that a VCPU sees new virtual interrupts. | |
675 | */ | |
676 | kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu); | |
8564d637 SP |
677 | |
678 | if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu)) | |
679 | kvm_update_stolen_time(vcpu); | |
0592c005 AJ |
680 | } |
681 | } | |
682 | ||
f7ed45be CD |
683 | /** |
684 | * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code | |
685 | * @vcpu: The VCPU pointer | |
686 | * @run: The kvm_run structure pointer used for userspace state exchange | |
687 | * | |
688 | * This function is called through the VCPU_RUN ioctl called from user space. It | |
689 | * will execute VM code in a loop until the time slice for the process is used | |
690 | * or some emulation is needed from user space in which case the function will | |
691 | * return with return value 0 and with the kvm_run structure filled in with the | |
692 | * required data for the requested emulation. | |
693 | */ | |
749cf76c CD |
694 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) |
695 | { | |
f7ed45be | 696 | int ret; |
f7ed45be | 697 | |
e8180dca | 698 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
f7ed45be CD |
699 | return -ENOEXEC; |
700 | ||
701 | ret = kvm_vcpu_first_run_init(vcpu); | |
702 | if (ret) | |
829a5863 | 703 | return ret; |
f7ed45be | 704 | |
45e96ea6 CD |
705 | if (run->exit_reason == KVM_EXIT_MMIO) { |
706 | ret = kvm_handle_mmio_return(vcpu, vcpu->run); | |
707 | if (ret) | |
829a5863 | 708 | return ret; |
45e96ea6 CD |
709 | } |
710 | ||
829a5863 CD |
711 | if (run->immediate_exit) |
712 | return -EINTR; | |
713 | ||
714 | vcpu_load(vcpu); | |
460df4c1 | 715 | |
20b7035c | 716 | kvm_sigset_activate(vcpu); |
f7ed45be CD |
717 | |
718 | ret = 1; | |
719 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
720 | while (ret > 0) { | |
721 | /* | |
722 | * Check conditions before entering the guest | |
723 | */ | |
724 | cond_resched(); | |
725 | ||
e329fb75 | 726 | update_vmid(&vcpu->kvm->arch.vmid); |
f7ed45be | 727 | |
0592c005 AJ |
728 | check_vcpu_requests(vcpu); |
729 | ||
abdf5843 MZ |
730 | /* |
731 | * Preparing the interrupts to be injected also | |
732 | * involves poking the GIC, which must be done in a | |
733 | * non-preemptible context. | |
734 | */ | |
1b3d546d | 735 | preempt_disable(); |
328e5664 | 736 | |
b02386eb | 737 | kvm_pmu_flush_hwstate(vcpu); |
328e5664 | 738 | |
f7ed45be CD |
739 | local_irq_disable(); |
740 | ||
abdf5843 MZ |
741 | kvm_vgic_flush_hwstate(vcpu); |
742 | ||
f7ed45be | 743 | /* |
61bbe380 CD |
744 | * Exit if we have a signal pending so that we can deliver the |
745 | * signal to user space. | |
f7ed45be | 746 | */ |
61bbe380 | 747 | if (signal_pending(current)) { |
f7ed45be CD |
748 | ret = -EINTR; |
749 | run->exit_reason = KVM_EXIT_INTR; | |
750 | } | |
751 | ||
61bbe380 CD |
752 | /* |
753 | * If we're using a userspace irqchip, then check if we need | |
754 | * to tell a userspace irqchip about timer or PMU level | |
755 | * changes and if so, exit to userspace (the actual level | |
756 | * state gets updated in kvm_timer_update_run and | |
757 | * kvm_pmu_update_run below). | |
758 | */ | |
759 | if (static_branch_unlikely(&userspace_irqchip_in_use)) { | |
760 | if (kvm_timer_should_notify_user(vcpu) || | |
761 | kvm_pmu_should_notify_user(vcpu)) { | |
762 | ret = -EINTR; | |
763 | run->exit_reason = KVM_EXIT_INTR; | |
764 | } | |
765 | } | |
766 | ||
6a6d73be AJ |
767 | /* |
768 | * Ensure we set mode to IN_GUEST_MODE after we disable | |
769 | * interrupts and before the final VCPU requests check. | |
770 | * See the comment in kvm_vcpu_exiting_guest_mode() and | |
2f5947df | 771 | * Documentation/virt/kvm/vcpu-requests.rst |
6a6d73be AJ |
772 | */ |
773 | smp_store_mb(vcpu->mode, IN_GUEST_MODE); | |
774 | ||
e329fb75 | 775 | if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) || |
424c989b | 776 | kvm_request_pending(vcpu)) { |
6a6d73be | 777 | vcpu->mode = OUTSIDE_GUEST_MODE; |
771621b0 | 778 | isb(); /* Ensure work in x_flush_hwstate is committed */ |
b02386eb | 779 | kvm_pmu_sync_hwstate(vcpu); |
61bbe380 CD |
780 | if (static_branch_unlikely(&userspace_irqchip_in_use)) |
781 | kvm_timer_sync_hwstate(vcpu); | |
1a89dd91 | 782 | kvm_vgic_sync_hwstate(vcpu); |
ee9bb9a1 | 783 | local_irq_enable(); |
abdf5843 | 784 | preempt_enable(); |
f7ed45be CD |
785 | continue; |
786 | } | |
787 | ||
56c7f5e7 AB |
788 | kvm_arm_setup_debug(vcpu); |
789 | ||
f7ed45be CD |
790 | /************************************************************** |
791 | * Enter the guest | |
792 | */ | |
793 | trace_kvm_entry(*vcpu_pc(vcpu)); | |
6edaa530 | 794 | guest_enter_irqoff(); |
f7ed45be | 795 | |
3f5c90b8 CD |
796 | if (has_vhe()) { |
797 | kvm_arm_vhe_guest_enter(); | |
798 | ret = kvm_vcpu_run_vhe(vcpu); | |
4f5abad9 | 799 | kvm_arm_vhe_guest_exit(); |
3f5c90b8 | 800 | } else { |
7aa8d146 | 801 | ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu); |
3f5c90b8 CD |
802 | } |
803 | ||
f7ed45be | 804 | vcpu->mode = OUTSIDE_GUEST_MODE; |
b19e6892 | 805 | vcpu->stat.exits++; |
1b3d546d CD |
806 | /* |
807 | * Back from guest | |
808 | *************************************************************/ | |
809 | ||
56c7f5e7 AB |
810 | kvm_arm_clear_debug(vcpu); |
811 | ||
ee9bb9a1 | 812 | /* |
b103cc3f | 813 | * We must sync the PMU state before the vgic state so |
ee9bb9a1 CD |
814 | * that the vgic can properly sample the updated state of the |
815 | * interrupt line. | |
816 | */ | |
817 | kvm_pmu_sync_hwstate(vcpu); | |
ee9bb9a1 | 818 | |
b103cc3f CD |
819 | /* |
820 | * Sync the vgic state before syncing the timer state because | |
821 | * the timer code needs to know if the virtual timer | |
822 | * interrupts are active. | |
823 | */ | |
ee9bb9a1 CD |
824 | kvm_vgic_sync_hwstate(vcpu); |
825 | ||
b103cc3f CD |
826 | /* |
827 | * Sync the timer hardware state before enabling interrupts as | |
828 | * we don't want vtimer interrupts to race with syncing the | |
829 | * timer virtual interrupt state. | |
830 | */ | |
61bbe380 CD |
831 | if (static_branch_unlikely(&userspace_irqchip_in_use)) |
832 | kvm_timer_sync_hwstate(vcpu); | |
b103cc3f | 833 | |
e6b673b7 DM |
834 | kvm_arch_vcpu_ctxsync_fp(vcpu); |
835 | ||
f7ed45be CD |
836 | /* |
837 | * We may have taken a host interrupt in HYP mode (ie | |
838 | * while executing the guest). This interrupt is still | |
839 | * pending, as we haven't serviced it yet! | |
840 | * | |
841 | * We're now back in SVC mode, with interrupts | |
842 | * disabled. Enabling the interrupts now will have | |
843 | * the effect of taking the interrupt again, in SVC | |
844 | * mode this time. | |
845 | */ | |
846 | local_irq_enable(); | |
847 | ||
848 | /* | |
6edaa530 | 849 | * We do local_irq_enable() before calling guest_exit() so |
1b3d546d CD |
850 | * that if a timer interrupt hits while running the guest we |
851 | * account that tick as being spent in the guest. We enable | |
6edaa530 | 852 | * preemption after calling guest_exit() so that if we get |
1b3d546d CD |
853 | * preempted we make sure ticks after that is not counted as |
854 | * guest time. | |
855 | */ | |
6edaa530 | 856 | guest_exit(); |
b5905dc1 | 857 | trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); |
1b3d546d | 858 | |
3368bd80 JM |
859 | /* Exit types that need handling before we can be preempted */ |
860 | handle_exit_early(vcpu, run, ret); | |
861 | ||
abdf5843 MZ |
862 | preempt_enable(); |
863 | ||
f7ed45be CD |
864 | ret = handle_exit(vcpu, run, ret); |
865 | } | |
866 | ||
d9e13977 | 867 | /* Tell userspace about in-kernel device output levels */ |
3dbbdf78 CD |
868 | if (unlikely(!irqchip_in_kernel(vcpu->kvm))) { |
869 | kvm_timer_update_run(vcpu); | |
870 | kvm_pmu_update_run(vcpu); | |
871 | } | |
d9e13977 | 872 | |
20b7035c JS |
873 | kvm_sigset_deactivate(vcpu); |
874 | ||
accb757d | 875 | vcpu_put(vcpu); |
f7ed45be | 876 | return ret; |
749cf76c CD |
877 | } |
878 | ||
86ce8535 CD |
879 | static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) |
880 | { | |
881 | int bit_index; | |
882 | bool set; | |
3df59d8d | 883 | unsigned long *hcr; |
86ce8535 CD |
884 | |
885 | if (number == KVM_ARM_IRQ_CPU_IRQ) | |
886 | bit_index = __ffs(HCR_VI); | |
887 | else /* KVM_ARM_IRQ_CPU_FIQ */ | |
888 | bit_index = __ffs(HCR_VF); | |
889 | ||
3df59d8d | 890 | hcr = vcpu_hcr(vcpu); |
86ce8535 | 891 | if (level) |
3df59d8d | 892 | set = test_and_set_bit(bit_index, hcr); |
86ce8535 | 893 | else |
3df59d8d | 894 | set = test_and_clear_bit(bit_index, hcr); |
86ce8535 CD |
895 | |
896 | /* | |
897 | * If we didn't change anything, no need to wake up or kick other CPUs | |
898 | */ | |
899 | if (set == level) | |
900 | return 0; | |
901 | ||
902 | /* | |
903 | * The vcpu irq_lines field was updated, wake up sleeping VCPUs and | |
904 | * trigger a world-switch round on the running physical CPU to set the | |
905 | * virtual IRQ/FIQ fields in the HCR appropriately. | |
906 | */ | |
325f9c64 | 907 | kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu); |
86ce8535 CD |
908 | kvm_vcpu_kick(vcpu); |
909 | ||
910 | return 0; | |
911 | } | |
912 | ||
79558f11 AG |
913 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, |
914 | bool line_status) | |
86ce8535 CD |
915 | { |
916 | u32 irq = irq_level->irq; | |
917 | unsigned int irq_type, vcpu_idx, irq_num; | |
918 | int nrcpus = atomic_read(&kvm->online_vcpus); | |
919 | struct kvm_vcpu *vcpu = NULL; | |
920 | bool level = irq_level->level; | |
921 | ||
922 | irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; | |
923 | vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; | |
92f35b75 | 924 | vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1); |
86ce8535 CD |
925 | irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; |
926 | ||
927 | trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); | |
928 | ||
5863c2ce MZ |
929 | switch (irq_type) { |
930 | case KVM_ARM_IRQ_TYPE_CPU: | |
931 | if (irqchip_in_kernel(kvm)) | |
932 | return -ENXIO; | |
86ce8535 | 933 | |
5863c2ce MZ |
934 | if (vcpu_idx >= nrcpus) |
935 | return -EINVAL; | |
86ce8535 | 936 | |
5863c2ce MZ |
937 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); |
938 | if (!vcpu) | |
939 | return -EINVAL; | |
86ce8535 | 940 | |
5863c2ce MZ |
941 | if (irq_num > KVM_ARM_IRQ_CPU_FIQ) |
942 | return -EINVAL; | |
943 | ||
944 | return vcpu_interrupt_line(vcpu, irq_num, level); | |
945 | case KVM_ARM_IRQ_TYPE_PPI: | |
946 | if (!irqchip_in_kernel(kvm)) | |
947 | return -ENXIO; | |
948 | ||
949 | if (vcpu_idx >= nrcpus) | |
950 | return -EINVAL; | |
951 | ||
952 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); | |
953 | if (!vcpu) | |
954 | return -EINVAL; | |
955 | ||
956 | if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) | |
957 | return -EINVAL; | |
86ce8535 | 958 | |
cb3f0ad8 | 959 | return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL); |
5863c2ce MZ |
960 | case KVM_ARM_IRQ_TYPE_SPI: |
961 | if (!irqchip_in_kernel(kvm)) | |
962 | return -ENXIO; | |
963 | ||
fd1d0ddf | 964 | if (irq_num < VGIC_NR_PRIVATE_IRQS) |
5863c2ce MZ |
965 | return -EINVAL; |
966 | ||
cb3f0ad8 | 967 | return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL); |
5863c2ce MZ |
968 | } |
969 | ||
970 | return -EINVAL; | |
86ce8535 CD |
971 | } |
972 | ||
f7fa034d CD |
973 | static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, |
974 | const struct kvm_vcpu_init *init) | |
975 | { | |
811328fc | 976 | unsigned int i, ret; |
f7fa034d CD |
977 | int phys_target = kvm_target_cpu(); |
978 | ||
979 | if (init->target != phys_target) | |
980 | return -EINVAL; | |
981 | ||
982 | /* | |
983 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
984 | * use the same target. | |
985 | */ | |
986 | if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) | |
987 | return -EINVAL; | |
988 | ||
989 | /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ | |
990 | for (i = 0; i < sizeof(init->features) * 8; i++) { | |
991 | bool set = (init->features[i / 32] & (1 << (i % 32))); | |
992 | ||
993 | if (set && i >= KVM_VCPU_MAX_FEATURES) | |
994 | return -ENOENT; | |
995 | ||
996 | /* | |
997 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
998 | * use the same feature set. | |
999 | */ | |
1000 | if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && | |
1001 | test_bit(i, vcpu->arch.features) != set) | |
1002 | return -EINVAL; | |
1003 | ||
1004 | if (set) | |
1005 | set_bit(i, vcpu->arch.features); | |
1006 | } | |
1007 | ||
1008 | vcpu->arch.target = phys_target; | |
1009 | ||
1010 | /* Now we know what it is, we can reset it. */ | |
811328fc AJ |
1011 | ret = kvm_reset_vcpu(vcpu); |
1012 | if (ret) { | |
1013 | vcpu->arch.target = -1; | |
1014 | bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); | |
1015 | } | |
f7fa034d | 1016 | |
811328fc AJ |
1017 | return ret; |
1018 | } | |
f7fa034d | 1019 | |
478a8237 CD |
1020 | static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, |
1021 | struct kvm_vcpu_init *init) | |
1022 | { | |
1023 | int ret; | |
1024 | ||
1025 | ret = kvm_vcpu_set_target(vcpu, init); | |
1026 | if (ret) | |
1027 | return ret; | |
1028 | ||
957db105 CD |
1029 | /* |
1030 | * Ensure a rebooted VM will fault in RAM pages and detect if the | |
1031 | * guest MMU is turned off and flush the caches as needed. | |
1032 | */ | |
1033 | if (vcpu->arch.has_run_once) | |
1034 | stage2_unmap_vm(vcpu->kvm); | |
1035 | ||
b856a591 CD |
1036 | vcpu_reset_hcr(vcpu); |
1037 | ||
478a8237 | 1038 | /* |
3781528e | 1039 | * Handle the "start in power-off" case. |
478a8237 | 1040 | */ |
03f1d4c1 | 1041 | if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) |
424c989b | 1042 | vcpu_power_off(vcpu); |
3ad8b3de | 1043 | else |
3781528e | 1044 | vcpu->arch.power_off = false; |
478a8237 CD |
1045 | |
1046 | return 0; | |
1047 | } | |
1048 | ||
f577f6c2 SZ |
1049 | static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, |
1050 | struct kvm_device_attr *attr) | |
1051 | { | |
1052 | int ret = -ENXIO; | |
1053 | ||
1054 | switch (attr->group) { | |
1055 | default: | |
bb0c70bc | 1056 | ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); |
f577f6c2 SZ |
1057 | break; |
1058 | } | |
1059 | ||
1060 | return ret; | |
1061 | } | |
1062 | ||
1063 | static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, | |
1064 | struct kvm_device_attr *attr) | |
1065 | { | |
1066 | int ret = -ENXIO; | |
1067 | ||
1068 | switch (attr->group) { | |
1069 | default: | |
bb0c70bc | 1070 | ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); |
f577f6c2 SZ |
1071 | break; |
1072 | } | |
1073 | ||
1074 | return ret; | |
1075 | } | |
1076 | ||
1077 | static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, | |
1078 | struct kvm_device_attr *attr) | |
1079 | { | |
1080 | int ret = -ENXIO; | |
1081 | ||
1082 | switch (attr->group) { | |
1083 | default: | |
bb0c70bc | 1084 | ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); |
f577f6c2 SZ |
1085 | break; |
1086 | } | |
1087 | ||
1088 | return ret; | |
1089 | } | |
1090 | ||
539aee0e JM |
1091 | static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu, |
1092 | struct kvm_vcpu_events *events) | |
1093 | { | |
1094 | memset(events, 0, sizeof(*events)); | |
1095 | ||
1096 | return __kvm_arm_vcpu_get_events(vcpu, events); | |
1097 | } | |
1098 | ||
1099 | static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu, | |
1100 | struct kvm_vcpu_events *events) | |
1101 | { | |
1102 | int i; | |
1103 | ||
1104 | /* check whether the reserved field is zero */ | |
1105 | for (i = 0; i < ARRAY_SIZE(events->reserved); i++) | |
1106 | if (events->reserved[i]) | |
1107 | return -EINVAL; | |
1108 | ||
1109 | /* check whether the pad field is zero */ | |
1110 | for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++) | |
1111 | if (events->exception.pad[i]) | |
1112 | return -EINVAL; | |
1113 | ||
1114 | return __kvm_arm_vcpu_set_events(vcpu, events); | |
1115 | } | |
539aee0e | 1116 | |
749cf76c CD |
1117 | long kvm_arch_vcpu_ioctl(struct file *filp, |
1118 | unsigned int ioctl, unsigned long arg) | |
1119 | { | |
1120 | struct kvm_vcpu *vcpu = filp->private_data; | |
1121 | void __user *argp = (void __user *)arg; | |
f577f6c2 | 1122 | struct kvm_device_attr attr; |
9b062471 CD |
1123 | long r; |
1124 | ||
749cf76c CD |
1125 | switch (ioctl) { |
1126 | case KVM_ARM_VCPU_INIT: { | |
1127 | struct kvm_vcpu_init init; | |
1128 | ||
9b062471 | 1129 | r = -EFAULT; |
749cf76c | 1130 | if (copy_from_user(&init, argp, sizeof(init))) |
9b062471 | 1131 | break; |
749cf76c | 1132 | |
9b062471 CD |
1133 | r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); |
1134 | break; | |
749cf76c CD |
1135 | } |
1136 | case KVM_SET_ONE_REG: | |
1137 | case KVM_GET_ONE_REG: { | |
1138 | struct kvm_one_reg reg; | |
e8180dca | 1139 | |
9b062471 | 1140 | r = -ENOEXEC; |
e8180dca | 1141 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
9b062471 | 1142 | break; |
e8180dca | 1143 | |
9b062471 | 1144 | r = -EFAULT; |
749cf76c | 1145 | if (copy_from_user(®, argp, sizeof(reg))) |
9b062471 CD |
1146 | break; |
1147 | ||
749cf76c | 1148 | if (ioctl == KVM_SET_ONE_REG) |
9b062471 | 1149 | r = kvm_arm_set_reg(vcpu, ®); |
749cf76c | 1150 | else |
9b062471 CD |
1151 | r = kvm_arm_get_reg(vcpu, ®); |
1152 | break; | |
749cf76c CD |
1153 | } |
1154 | case KVM_GET_REG_LIST: { | |
1155 | struct kvm_reg_list __user *user_list = argp; | |
1156 | struct kvm_reg_list reg_list; | |
1157 | unsigned n; | |
1158 | ||
9b062471 | 1159 | r = -ENOEXEC; |
e8180dca | 1160 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
9b062471 | 1161 | break; |
e8180dca | 1162 | |
7dd32a0d DM |
1163 | r = -EPERM; |
1164 | if (!kvm_arm_vcpu_is_finalized(vcpu)) | |
1165 | break; | |
1166 | ||
9b062471 | 1167 | r = -EFAULT; |
749cf76c | 1168 | if (copy_from_user(®_list, user_list, sizeof(reg_list))) |
9b062471 | 1169 | break; |
749cf76c CD |
1170 | n = reg_list.n; |
1171 | reg_list.n = kvm_arm_num_regs(vcpu); | |
1172 | if (copy_to_user(user_list, ®_list, sizeof(reg_list))) | |
9b062471 CD |
1173 | break; |
1174 | r = -E2BIG; | |
749cf76c | 1175 | if (n < reg_list.n) |
9b062471 CD |
1176 | break; |
1177 | r = kvm_arm_copy_reg_indices(vcpu, user_list->reg); | |
1178 | break; | |
749cf76c | 1179 | } |
f577f6c2 | 1180 | case KVM_SET_DEVICE_ATTR: { |
9b062471 | 1181 | r = -EFAULT; |
f577f6c2 | 1182 | if (copy_from_user(&attr, argp, sizeof(attr))) |
9b062471 CD |
1183 | break; |
1184 | r = kvm_arm_vcpu_set_attr(vcpu, &attr); | |
1185 | break; | |
f577f6c2 SZ |
1186 | } |
1187 | case KVM_GET_DEVICE_ATTR: { | |
9b062471 | 1188 | r = -EFAULT; |
f577f6c2 | 1189 | if (copy_from_user(&attr, argp, sizeof(attr))) |
9b062471 CD |
1190 | break; |
1191 | r = kvm_arm_vcpu_get_attr(vcpu, &attr); | |
1192 | break; | |
f577f6c2 SZ |
1193 | } |
1194 | case KVM_HAS_DEVICE_ATTR: { | |
9b062471 | 1195 | r = -EFAULT; |
f577f6c2 | 1196 | if (copy_from_user(&attr, argp, sizeof(attr))) |
9b062471 CD |
1197 | break; |
1198 | r = kvm_arm_vcpu_has_attr(vcpu, &attr); | |
1199 | break; | |
f577f6c2 | 1200 | } |
b7b27fac DG |
1201 | case KVM_GET_VCPU_EVENTS: { |
1202 | struct kvm_vcpu_events events; | |
1203 | ||
1204 | if (kvm_arm_vcpu_get_events(vcpu, &events)) | |
1205 | return -EINVAL; | |
1206 | ||
1207 | if (copy_to_user(argp, &events, sizeof(events))) | |
1208 | return -EFAULT; | |
1209 | ||
1210 | return 0; | |
1211 | } | |
1212 | case KVM_SET_VCPU_EVENTS: { | |
1213 | struct kvm_vcpu_events events; | |
1214 | ||
1215 | if (copy_from_user(&events, argp, sizeof(events))) | |
1216 | return -EFAULT; | |
1217 | ||
1218 | return kvm_arm_vcpu_set_events(vcpu, &events); | |
1219 | } | |
7dd32a0d DM |
1220 | case KVM_ARM_VCPU_FINALIZE: { |
1221 | int what; | |
1222 | ||
1223 | if (!kvm_vcpu_initialized(vcpu)) | |
1224 | return -ENOEXEC; | |
1225 | ||
1226 | if (get_user(what, (const int __user *)argp)) | |
1227 | return -EFAULT; | |
1228 | ||
1229 | return kvm_arm_vcpu_finalize(vcpu, what); | |
1230 | } | |
749cf76c | 1231 | default: |
9b062471 | 1232 | r = -EINVAL; |
749cf76c | 1233 | } |
9b062471 | 1234 | |
9b062471 | 1235 | return r; |
749cf76c CD |
1236 | } |
1237 | ||
53c810c3 MS |
1238 | /** |
1239 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
1240 | * @kvm: kvm instance | |
1241 | * @log: slot id and address to which we copy the log | |
1242 | * | |
1243 | * Steps 1-4 below provide general overview of dirty page logging. See | |
1244 | * kvm_get_dirty_log_protect() function description for additional details. | |
1245 | * | |
1246 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
1247 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
1248 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
1249 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
1250 | * writes will be marked dirty for next log read. | |
1251 | * | |
1252 | * 1. Take a snapshot of the bit and clear it if needed. | |
1253 | * 2. Write protect the corresponding page. | |
1254 | * 3. Copy the snapshot to the userspace. | |
1255 | * 4. Flush TLB's if needed. | |
1256 | */ | |
749cf76c CD |
1257 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
1258 | { | |
8fe65a82 | 1259 | bool flush = false; |
53c810c3 MS |
1260 | int r; |
1261 | ||
1262 | mutex_lock(&kvm->slots_lock); | |
1263 | ||
8fe65a82 | 1264 | r = kvm_get_dirty_log_protect(kvm, log, &flush); |
53c810c3 | 1265 | |
8fe65a82 | 1266 | if (flush) |
53c810c3 MS |
1267 | kvm_flush_remote_tlbs(kvm); |
1268 | ||
1269 | mutex_unlock(&kvm->slots_lock); | |
1270 | return r; | |
749cf76c CD |
1271 | } |
1272 | ||
2a31b9db PB |
1273 | int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log) |
1274 | { | |
1275 | bool flush = false; | |
53c810c3 MS |
1276 | int r; |
1277 | ||
1278 | mutex_lock(&kvm->slots_lock); | |
1279 | ||
2a31b9db | 1280 | r = kvm_clear_dirty_log_protect(kvm, log, &flush); |
53c810c3 | 1281 | |
2a31b9db | 1282 | if (flush) |
53c810c3 MS |
1283 | kvm_flush_remote_tlbs(kvm); |
1284 | ||
1285 | mutex_unlock(&kvm->slots_lock); | |
1286 | return r; | |
749cf76c CD |
1287 | } |
1288 | ||
3401d546 CD |
1289 | static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, |
1290 | struct kvm_arm_device_addr *dev_addr) | |
1291 | { | |
330690cd CD |
1292 | unsigned long dev_id, type; |
1293 | ||
1294 | dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> | |
1295 | KVM_ARM_DEVICE_ID_SHIFT; | |
1296 | type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> | |
1297 | KVM_ARM_DEVICE_TYPE_SHIFT; | |
1298 | ||
1299 | switch (dev_id) { | |
1300 | case KVM_ARM_DEVICE_VGIC_V2: | |
c7da6fa4 PF |
1301 | if (!vgic_present) |
1302 | return -ENXIO; | |
ce01e4e8 | 1303 | return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); |
330690cd CD |
1304 | default: |
1305 | return -ENODEV; | |
1306 | } | |
3401d546 CD |
1307 | } |
1308 | ||
749cf76c CD |
1309 | long kvm_arch_vm_ioctl(struct file *filp, |
1310 | unsigned int ioctl, unsigned long arg) | |
1311 | { | |
3401d546 CD |
1312 | struct kvm *kvm = filp->private_data; |
1313 | void __user *argp = (void __user *)arg; | |
1314 | ||
1315 | switch (ioctl) { | |
5863c2ce | 1316 | case KVM_CREATE_IRQCHIP: { |
a28ebea2 | 1317 | int ret; |
c7da6fa4 PF |
1318 | if (!vgic_present) |
1319 | return -ENXIO; | |
a28ebea2 CD |
1320 | mutex_lock(&kvm->lock); |
1321 | ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); | |
1322 | mutex_unlock(&kvm->lock); | |
1323 | return ret; | |
5863c2ce | 1324 | } |
3401d546 CD |
1325 | case KVM_ARM_SET_DEVICE_ADDR: { |
1326 | struct kvm_arm_device_addr dev_addr; | |
1327 | ||
1328 | if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) | |
1329 | return -EFAULT; | |
1330 | return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); | |
1331 | } | |
42c4e0c7 AP |
1332 | case KVM_ARM_PREFERRED_TARGET: { |
1333 | int err; | |
1334 | struct kvm_vcpu_init init; | |
1335 | ||
1336 | err = kvm_vcpu_preferred_target(&init); | |
1337 | if (err) | |
1338 | return err; | |
1339 | ||
1340 | if (copy_to_user(argp, &init, sizeof(init))) | |
1341 | return -EFAULT; | |
1342 | ||
1343 | return 0; | |
1344 | } | |
3401d546 CD |
1345 | default: |
1346 | return -EINVAL; | |
1347 | } | |
749cf76c CD |
1348 | } |
1349 | ||
7e0befd5 | 1350 | static void cpu_init_hyp_mode(void) |
342cd0ab | 1351 | { |
dac288f7 | 1352 | phys_addr_t pgd_ptr; |
342cd0ab CD |
1353 | unsigned long hyp_stack_ptr; |
1354 | unsigned long stack_page; | |
1355 | unsigned long vector_ptr; | |
1356 | ||
1357 | /* Switch from the HYP stub to our own HYP init vector */ | |
5a677ce0 | 1358 | __hyp_set_vectors(kvm_get_idmap_vector()); |
342cd0ab | 1359 | |
dac288f7 | 1360 | pgd_ptr = kvm_mmu_get_httbr(); |
1436c1aa | 1361 | stack_page = __this_cpu_read(kvm_arm_hyp_stack_page); |
342cd0ab | 1362 | hyp_stack_ptr = stack_page + PAGE_SIZE; |
6840bdd7 | 1363 | vector_ptr = (unsigned long)kvm_get_hyp_vector(); |
342cd0ab | 1364 | |
12fda812 | 1365 | __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr); |
35a2491a | 1366 | __cpu_init_stage2(); |
342cd0ab CD |
1367 | } |
1368 | ||
47eb3cba MZ |
1369 | static void cpu_hyp_reset(void) |
1370 | { | |
1371 | if (!is_kernel_in_hyp_mode()) | |
1372 | __hyp_reset_vectors(); | |
1373 | } | |
1374 | ||
5f5560b1 JM |
1375 | static void cpu_hyp_reinit(void) |
1376 | { | |
1e0cf16c MZ |
1377 | kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt); |
1378 | ||
47eb3cba MZ |
1379 | cpu_hyp_reset(); |
1380 | ||
9d47bb0d | 1381 | if (is_kernel_in_hyp_mode()) |
02d50cda | 1382 | kvm_timer_init_vhe(); |
9d47bb0d | 1383 | else |
7e0befd5 | 1384 | cpu_init_hyp_mode(); |
5b0d2cc2 | 1385 | |
da5a3ce6 | 1386 | kvm_arm_init_debug(); |
5b0d2cc2 CD |
1387 | |
1388 | if (vgic_present) | |
1389 | kvm_vgic_init_cpu_hardware(); | |
5f5560b1 JM |
1390 | } |
1391 | ||
67f69197 AT |
1392 | static void _kvm_arch_hardware_enable(void *discard) |
1393 | { | |
1394 | if (!__this_cpu_read(kvm_arm_hardware_enabled)) { | |
5f5560b1 | 1395 | cpu_hyp_reinit(); |
67f69197 | 1396 | __this_cpu_write(kvm_arm_hardware_enabled, 1); |
d157f4a5 | 1397 | } |
67f69197 | 1398 | } |
d157f4a5 | 1399 | |
67f69197 AT |
1400 | int kvm_arch_hardware_enable(void) |
1401 | { | |
1402 | _kvm_arch_hardware_enable(NULL); | |
1403 | return 0; | |
342cd0ab CD |
1404 | } |
1405 | ||
67f69197 AT |
1406 | static void _kvm_arch_hardware_disable(void *discard) |
1407 | { | |
1408 | if (__this_cpu_read(kvm_arm_hardware_enabled)) { | |
1409 | cpu_hyp_reset(); | |
1410 | __this_cpu_write(kvm_arm_hardware_enabled, 0); | |
1411 | } | |
1412 | } | |
1413 | ||
1414 | void kvm_arch_hardware_disable(void) | |
1415 | { | |
1416 | _kvm_arch_hardware_disable(NULL); | |
1417 | } | |
d157f4a5 | 1418 | |
1fcf7ce0 LP |
1419 | #ifdef CONFIG_CPU_PM |
1420 | static int hyp_init_cpu_pm_notifier(struct notifier_block *self, | |
1421 | unsigned long cmd, | |
1422 | void *v) | |
1423 | { | |
67f69197 AT |
1424 | /* |
1425 | * kvm_arm_hardware_enabled is left with its old value over | |
1426 | * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should | |
1427 | * re-enable hyp. | |
1428 | */ | |
1429 | switch (cmd) { | |
1430 | case CPU_PM_ENTER: | |
1431 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1432 | /* | |
1433 | * don't update kvm_arm_hardware_enabled here | |
1434 | * so that the hardware will be re-enabled | |
1435 | * when we resume. See below. | |
1436 | */ | |
1437 | cpu_hyp_reset(); | |
1438 | ||
1fcf7ce0 | 1439 | return NOTIFY_OK; |
58d6b15e | 1440 | case CPU_PM_ENTER_FAILED: |
67f69197 AT |
1441 | case CPU_PM_EXIT: |
1442 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1443 | /* The hardware was enabled before suspend. */ | |
1444 | cpu_hyp_reinit(); | |
1fcf7ce0 | 1445 | |
67f69197 AT |
1446 | return NOTIFY_OK; |
1447 | ||
1448 | default: | |
1449 | return NOTIFY_DONE; | |
1450 | } | |
1fcf7ce0 LP |
1451 | } |
1452 | ||
1453 | static struct notifier_block hyp_init_cpu_pm_nb = { | |
1454 | .notifier_call = hyp_init_cpu_pm_notifier, | |
1455 | }; | |
1456 | ||
1457 | static void __init hyp_cpu_pm_init(void) | |
1458 | { | |
1459 | cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); | |
1460 | } | |
06a71a24 SH |
1461 | static void __init hyp_cpu_pm_exit(void) |
1462 | { | |
1463 | cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); | |
1464 | } | |
1fcf7ce0 LP |
1465 | #else |
1466 | static inline void hyp_cpu_pm_init(void) | |
1467 | { | |
1468 | } | |
06a71a24 SH |
1469 | static inline void hyp_cpu_pm_exit(void) |
1470 | { | |
1471 | } | |
1fcf7ce0 LP |
1472 | #endif |
1473 | ||
1e947bad MZ |
1474 | static int init_common_resources(void) |
1475 | { | |
0f62f0e9 SP |
1476 | kvm_set_ipa_limit(); |
1477 | ||
1e947bad MZ |
1478 | return 0; |
1479 | } | |
1480 | ||
1481 | static int init_subsystems(void) | |
1482 | { | |
67f69197 | 1483 | int err = 0; |
1e947bad | 1484 | |
5f5560b1 | 1485 | /* |
67f69197 | 1486 | * Enable hardware so that subsystem initialisation can access EL2. |
5f5560b1 | 1487 | */ |
67f69197 | 1488 | on_each_cpu(_kvm_arch_hardware_enable, NULL, 1); |
5f5560b1 JM |
1489 | |
1490 | /* | |
1491 | * Register CPU lower-power notifier | |
1492 | */ | |
1493 | hyp_cpu_pm_init(); | |
1494 | ||
1e947bad MZ |
1495 | /* |
1496 | * Init HYP view of VGIC | |
1497 | */ | |
1498 | err = kvm_vgic_hyp_init(); | |
1499 | switch (err) { | |
1500 | case 0: | |
1501 | vgic_present = true; | |
1502 | break; | |
1503 | case -ENODEV: | |
1504 | case -ENXIO: | |
1505 | vgic_present = false; | |
67f69197 | 1506 | err = 0; |
1e947bad MZ |
1507 | break; |
1508 | default: | |
67f69197 | 1509 | goto out; |
1e947bad MZ |
1510 | } |
1511 | ||
1512 | /* | |
1513 | * Init HYP architected timer support | |
1514 | */ | |
f384dcfe | 1515 | err = kvm_timer_hyp_init(vgic_present); |
1e947bad | 1516 | if (err) |
67f69197 | 1517 | goto out; |
1e947bad MZ |
1518 | |
1519 | kvm_perf_init(); | |
1520 | kvm_coproc_table_init(); | |
1521 | ||
67f69197 AT |
1522 | out: |
1523 | on_each_cpu(_kvm_arch_hardware_disable, NULL, 1); | |
1524 | ||
1525 | return err; | |
1e947bad MZ |
1526 | } |
1527 | ||
1528 | static void teardown_hyp_mode(void) | |
1529 | { | |
1530 | int cpu; | |
1531 | ||
1e947bad MZ |
1532 | free_hyp_pgds(); |
1533 | for_each_possible_cpu(cpu) | |
1534 | free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); | |
06a71a24 | 1535 | hyp_cpu_pm_exit(); |
1e947bad MZ |
1536 | } |
1537 | ||
342cd0ab CD |
1538 | /** |
1539 | * Inits Hyp-mode on all online CPUs | |
1540 | */ | |
1541 | static int init_hyp_mode(void) | |
1542 | { | |
342cd0ab CD |
1543 | int cpu; |
1544 | int err = 0; | |
1545 | ||
1546 | /* | |
1547 | * Allocate Hyp PGD and setup Hyp identity mapping | |
1548 | */ | |
1549 | err = kvm_mmu_init(); | |
1550 | if (err) | |
1551 | goto out_err; | |
1552 | ||
342cd0ab CD |
1553 | /* |
1554 | * Allocate stack pages for Hypervisor-mode | |
1555 | */ | |
1556 | for_each_possible_cpu(cpu) { | |
1557 | unsigned long stack_page; | |
1558 | ||
1559 | stack_page = __get_free_page(GFP_KERNEL); | |
1560 | if (!stack_page) { | |
1561 | err = -ENOMEM; | |
1e947bad | 1562 | goto out_err; |
342cd0ab CD |
1563 | } |
1564 | ||
1565 | per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; | |
1566 | } | |
1567 | ||
342cd0ab CD |
1568 | /* |
1569 | * Map the Hyp-code called directly from the host | |
1570 | */ | |
588ab3f9 | 1571 | err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), |
59002705 | 1572 | kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC); |
342cd0ab CD |
1573 | if (err) { |
1574 | kvm_err("Cannot map world-switch code\n"); | |
1e947bad | 1575 | goto out_err; |
342cd0ab CD |
1576 | } |
1577 | ||
a0bf9776 | 1578 | err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), |
74a6b888 | 1579 | kvm_ksym_ref(__end_rodata), PAGE_HYP_RO); |
910917bb MZ |
1580 | if (err) { |
1581 | kvm_err("Cannot map rodata section\n"); | |
c8ea0395 MZ |
1582 | goto out_err; |
1583 | } | |
1584 | ||
1585 | err = create_hyp_mappings(kvm_ksym_ref(__bss_start), | |
1586 | kvm_ksym_ref(__bss_stop), PAGE_HYP_RO); | |
1587 | if (err) { | |
1588 | kvm_err("Cannot map bss section\n"); | |
1e947bad | 1589 | goto out_err; |
910917bb MZ |
1590 | } |
1591 | ||
6840bdd7 MZ |
1592 | err = kvm_map_vectors(); |
1593 | if (err) { | |
1594 | kvm_err("Cannot map vectors\n"); | |
1595 | goto out_err; | |
1596 | } | |
1597 | ||
342cd0ab CD |
1598 | /* |
1599 | * Map the Hyp stack pages | |
1600 | */ | |
1601 | for_each_possible_cpu(cpu) { | |
1602 | char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); | |
c8dddecd MZ |
1603 | err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE, |
1604 | PAGE_HYP); | |
342cd0ab CD |
1605 | |
1606 | if (err) { | |
1607 | kvm_err("Cannot map hyp stack\n"); | |
1e947bad | 1608 | goto out_err; |
342cd0ab CD |
1609 | } |
1610 | } | |
1611 | ||
342cd0ab | 1612 | for_each_possible_cpu(cpu) { |
630a1685 | 1613 | kvm_host_data_t *cpu_data; |
342cd0ab | 1614 | |
630a1685 | 1615 | cpu_data = per_cpu_ptr(&kvm_host_data, cpu); |
630a1685 | 1616 | err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP); |
342cd0ab CD |
1617 | |
1618 | if (err) { | |
3de50da6 | 1619 | kvm_err("Cannot map host CPU state: %d\n", err); |
1e947bad | 1620 | goto out_err; |
342cd0ab CD |
1621 | } |
1622 | } | |
1623 | ||
55e3748e MZ |
1624 | err = hyp_map_aux_data(); |
1625 | if (err) | |
a37f0c3c | 1626 | kvm_err("Cannot map host auxiliary data: %d\n", err); |
55e3748e | 1627 | |
342cd0ab | 1628 | return 0; |
1e947bad | 1629 | |
342cd0ab | 1630 | out_err: |
1e947bad | 1631 | teardown_hyp_mode(); |
342cd0ab CD |
1632 | kvm_err("error initializing Hyp mode: %d\n", err); |
1633 | return err; | |
1634 | } | |
1635 | ||
d4e071ce AP |
1636 | static void check_kvm_target_cpu(void *ret) |
1637 | { | |
1638 | *(int *)ret = kvm_target_cpu(); | |
1639 | } | |
1640 | ||
4429fc64 AP |
1641 | struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) |
1642 | { | |
1643 | struct kvm_vcpu *vcpu; | |
1644 | int i; | |
1645 | ||
1646 | mpidr &= MPIDR_HWID_BITMASK; | |
1647 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
1648 | if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) | |
1649 | return vcpu; | |
1650 | } | |
1651 | return NULL; | |
1652 | } | |
1653 | ||
2412405b EA |
1654 | bool kvm_arch_has_irq_bypass(void) |
1655 | { | |
1656 | return true; | |
1657 | } | |
1658 | ||
1659 | int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, | |
1660 | struct irq_bypass_producer *prod) | |
1661 | { | |
1662 | struct kvm_kernel_irqfd *irqfd = | |
1663 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1664 | ||
196b1364 MZ |
1665 | return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq, |
1666 | &irqfd->irq_entry); | |
2412405b EA |
1667 | } |
1668 | void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, | |
1669 | struct irq_bypass_producer *prod) | |
1670 | { | |
1671 | struct kvm_kernel_irqfd *irqfd = | |
1672 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1673 | ||
196b1364 MZ |
1674 | kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq, |
1675 | &irqfd->irq_entry); | |
2412405b EA |
1676 | } |
1677 | ||
1678 | void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons) | |
1679 | { | |
1680 | struct kvm_kernel_irqfd *irqfd = | |
1681 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1682 | ||
1683 | kvm_arm_halt_guest(irqfd->kvm); | |
1684 | } | |
1685 | ||
1686 | void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons) | |
1687 | { | |
1688 | struct kvm_kernel_irqfd *irqfd = | |
1689 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1690 | ||
1691 | kvm_arm_resume_guest(irqfd->kvm); | |
1692 | } | |
1693 | ||
342cd0ab CD |
1694 | /** |
1695 | * Initialize Hyp-mode and memory mappings on all CPUs. | |
1696 | */ | |
749cf76c CD |
1697 | int kvm_arch_init(void *opaque) |
1698 | { | |
342cd0ab | 1699 | int err; |
d4e071ce | 1700 | int ret, cpu; |
fe7d7b03 | 1701 | bool in_hyp_mode; |
342cd0ab CD |
1702 | |
1703 | if (!is_hyp_mode_available()) { | |
58d0d19a | 1704 | kvm_info("HYP mode not available\n"); |
342cd0ab CD |
1705 | return -ENODEV; |
1706 | } | |
1707 | ||
33e5f4e5 MZ |
1708 | in_hyp_mode = is_kernel_in_hyp_mode(); |
1709 | ||
1710 | if (!in_hyp_mode && kvm_arch_requires_vhe()) { | |
1711 | kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n"); | |
85acda3b DM |
1712 | return -ENODEV; |
1713 | } | |
1714 | ||
d4e071ce AP |
1715 | for_each_online_cpu(cpu) { |
1716 | smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); | |
1717 | if (ret < 0) { | |
1718 | kvm_err("Error, CPU %d not supported!\n", cpu); | |
1719 | return -ENODEV; | |
1720 | } | |
342cd0ab CD |
1721 | } |
1722 | ||
1e947bad | 1723 | err = init_common_resources(); |
342cd0ab | 1724 | if (err) |
1e947bad | 1725 | return err; |
342cd0ab | 1726 | |
a3be836d | 1727 | err = kvm_arm_init_sve(); |
0f062bfe DM |
1728 | if (err) |
1729 | return err; | |
1730 | ||
fe7d7b03 | 1731 | if (!in_hyp_mode) { |
1e947bad | 1732 | err = init_hyp_mode(); |
fe7d7b03 JT |
1733 | if (err) |
1734 | goto out_err; | |
1735 | } | |
8146875d | 1736 | |
1e947bad MZ |
1737 | err = init_subsystems(); |
1738 | if (err) | |
1739 | goto out_hyp; | |
1fcf7ce0 | 1740 | |
fe7d7b03 JT |
1741 | if (in_hyp_mode) |
1742 | kvm_info("VHE mode initialized successfully\n"); | |
1743 | else | |
1744 | kvm_info("Hyp mode initialized successfully\n"); | |
1745 | ||
749cf76c | 1746 | return 0; |
1e947bad MZ |
1747 | |
1748 | out_hyp: | |
fe7d7b03 JT |
1749 | if (!in_hyp_mode) |
1750 | teardown_hyp_mode(); | |
342cd0ab CD |
1751 | out_err: |
1752 | return err; | |
749cf76c CD |
1753 | } |
1754 | ||
1755 | /* NOP: Compiling as a module not supported */ | |
1756 | void kvm_arch_exit(void) | |
1757 | { | |
210552c1 | 1758 | kvm_perf_teardown(); |
749cf76c CD |
1759 | } |
1760 | ||
1761 | static int arm_init(void) | |
1762 | { | |
1763 | int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1764 | return rc; | |
1765 | } | |
1766 | ||
1767 | module_init(arm_init); |