]>
Commit | Line | Data |
---|---|---|
749cf76c CD |
1 | /* |
2 | * Copyright (C) 2012 - Virtual Open Systems and Columbia University | |
3 | * Author: Christoffer Dall <c.dall@virtualopensystems.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License, version 2, as | |
7 | * published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. | |
17 | */ | |
18 | ||
85acda3b | 19 | #include <linux/bug.h> |
1fcf7ce0 | 20 | #include <linux/cpu_pm.h> |
749cf76c CD |
21 | #include <linux/errno.h> |
22 | #include <linux/err.h> | |
23 | #include <linux/kvm_host.h> | |
1085fdc6 | 24 | #include <linux/list.h> |
749cf76c CD |
25 | #include <linux/module.h> |
26 | #include <linux/vmalloc.h> | |
27 | #include <linux/fs.h> | |
28 | #include <linux/mman.h> | |
29 | #include <linux/sched.h> | |
86ce8535 | 30 | #include <linux/kvm.h> |
2412405b EA |
31 | #include <linux/kvm_irqfd.h> |
32 | #include <linux/irqbypass.h> | |
de737089 | 33 | #include <linux/sched/stat.h> |
749cf76c | 34 | #include <trace/events/kvm.h> |
b02386eb | 35 | #include <kvm/arm_pmu.h> |
1a2fb94e | 36 | #include <kvm/arm_psci.h> |
749cf76c CD |
37 | |
38 | #define CREATE_TRACE_POINTS | |
39 | #include "trace.h" | |
40 | ||
7c0f6ba6 | 41 | #include <linux/uaccess.h> |
749cf76c CD |
42 | #include <asm/ptrace.h> |
43 | #include <asm/mman.h> | |
342cd0ab | 44 | #include <asm/tlbflush.h> |
5b3e5e5b | 45 | #include <asm/cacheflush.h> |
85acda3b | 46 | #include <asm/cpufeature.h> |
342cd0ab CD |
47 | #include <asm/virt.h> |
48 | #include <asm/kvm_arm.h> | |
49 | #include <asm/kvm_asm.h> | |
50 | #include <asm/kvm_mmu.h> | |
f7ed45be | 51 | #include <asm/kvm_emulate.h> |
5b3e5e5b | 52 | #include <asm/kvm_coproc.h> |
910917bb | 53 | #include <asm/sections.h> |
749cf76c CD |
54 | |
55 | #ifdef REQUIRES_VIRT | |
56 | __asm__(".arch_extension virt"); | |
57 | #endif | |
58 | ||
36989e7f | 59 | DEFINE_PER_CPU(kvm_cpu_context_t, kvm_host_cpu_state); |
342cd0ab | 60 | static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); |
342cd0ab | 61 | |
1638a12d MZ |
62 | /* Per-CPU variable containing the currently running vcpu. */ |
63 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu); | |
64 | ||
f7ed45be CD |
65 | /* The VMID used in the VTTBR */ |
66 | static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1); | |
20475f78 VM |
67 | static u32 kvm_next_vmid; |
68 | static unsigned int kvm_vmid_bits __read_mostly; | |
f0cf47d9 | 69 | static DEFINE_RWLOCK(kvm_vmid_lock); |
342cd0ab | 70 | |
c7da6fa4 PF |
71 | static bool vgic_present; |
72 | ||
67f69197 AT |
73 | static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled); |
74 | ||
1638a12d MZ |
75 | static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu) |
76 | { | |
1436c1aa | 77 | __this_cpu_write(kvm_arm_running_vcpu, vcpu); |
1638a12d MZ |
78 | } |
79 | ||
61bbe380 CD |
80 | DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use); |
81 | ||
1638a12d MZ |
82 | /** |
83 | * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU. | |
84 | * Must be called from non-preemptible context | |
85 | */ | |
86 | struct kvm_vcpu *kvm_arm_get_running_vcpu(void) | |
87 | { | |
1436c1aa | 88 | return __this_cpu_read(kvm_arm_running_vcpu); |
1638a12d MZ |
89 | } |
90 | ||
91 | /** | |
92 | * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
93 | */ | |
4000be42 | 94 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) |
1638a12d MZ |
95 | { |
96 | return &kvm_arm_running_vcpu; | |
97 | } | |
98 | ||
749cf76c CD |
99 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
100 | { | |
101 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; | |
102 | } | |
103 | ||
749cf76c CD |
104 | int kvm_arch_hardware_setup(void) |
105 | { | |
106 | return 0; | |
107 | } | |
108 | ||
749cf76c CD |
109 | void kvm_arch_check_processor_compat(void *rtn) |
110 | { | |
111 | *(int *)rtn = 0; | |
112 | } | |
113 | ||
749cf76c | 114 | |
d5d8184d CD |
115 | /** |
116 | * kvm_arch_init_vm - initializes a VM data structure | |
117 | * @kvm: pointer to the KVM struct | |
118 | */ | |
749cf76c CD |
119 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
120 | { | |
94d0e598 | 121 | int ret, cpu; |
d5d8184d | 122 | |
bca607eb | 123 | ret = kvm_arm_setup_stage2(kvm, type); |
5b6c6742 SP |
124 | if (ret) |
125 | return ret; | |
749cf76c | 126 | |
94d0e598 MZ |
127 | kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran)); |
128 | if (!kvm->arch.last_vcpu_ran) | |
129 | return -ENOMEM; | |
130 | ||
131 | for_each_possible_cpu(cpu) | |
132 | *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1; | |
133 | ||
d5d8184d CD |
134 | ret = kvm_alloc_stage2_pgd(kvm); |
135 | if (ret) | |
136 | goto out_fail_alloc; | |
137 | ||
c8dddecd | 138 | ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP); |
d5d8184d CD |
139 | if (ret) |
140 | goto out_free_stage2_pgd; | |
141 | ||
6c3d63c9 | 142 | kvm_vgic_early_init(kvm); |
a1a64387 | 143 | |
d5d8184d CD |
144 | /* Mark the initial VMID generation invalid */ |
145 | kvm->arch.vmid_gen = 0; | |
146 | ||
3caa2d8c | 147 | /* The maximum number of VCPUs is limited by the host's GIC model */ |
c7da6fa4 PF |
148 | kvm->arch.max_vcpus = vgic_present ? |
149 | kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; | |
3caa2d8c | 150 | |
d5d8184d CD |
151 | return ret; |
152 | out_free_stage2_pgd: | |
153 | kvm_free_stage2_pgd(kvm); | |
154 | out_fail_alloc: | |
94d0e598 MZ |
155 | free_percpu(kvm->arch.last_vcpu_ran); |
156 | kvm->arch.last_vcpu_ran = NULL; | |
d5d8184d | 157 | return ret; |
749cf76c CD |
158 | } |
159 | ||
235539b4 LC |
160 | bool kvm_arch_has_vcpu_debugfs(void) |
161 | { | |
162 | return false; | |
163 | } | |
164 | ||
165 | int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu) | |
166 | { | |
167 | return 0; | |
168 | } | |
169 | ||
1499fa80 | 170 | vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
749cf76c CD |
171 | { |
172 | return VM_FAULT_SIGBUS; | |
173 | } | |
174 | ||
749cf76c | 175 | |
d5d8184d CD |
176 | /** |
177 | * kvm_arch_destroy_vm - destroy the VM data structure | |
178 | * @kvm: pointer to the KVM struct | |
179 | */ | |
749cf76c CD |
180 | void kvm_arch_destroy_vm(struct kvm *kvm) |
181 | { | |
182 | int i; | |
183 | ||
b2c9a85d MZ |
184 | kvm_vgic_destroy(kvm); |
185 | ||
94d0e598 MZ |
186 | free_percpu(kvm->arch.last_vcpu_ran); |
187 | kvm->arch.last_vcpu_ran = NULL; | |
188 | ||
749cf76c CD |
189 | for (i = 0; i < KVM_MAX_VCPUS; ++i) { |
190 | if (kvm->vcpus[i]) { | |
191 | kvm_arch_vcpu_free(kvm->vcpus[i]); | |
192 | kvm->vcpus[i] = NULL; | |
193 | } | |
194 | } | |
6b2ad81b | 195 | atomic_set(&kvm->online_vcpus, 0); |
749cf76c CD |
196 | } |
197 | ||
784aa3d7 | 198 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
749cf76c CD |
199 | { |
200 | int r; | |
201 | switch (ext) { | |
1a89dd91 | 202 | case KVM_CAP_IRQCHIP: |
c7da6fa4 PF |
203 | r = vgic_present; |
204 | break; | |
d44758c0 | 205 | case KVM_CAP_IOEVENTFD: |
7330672b | 206 | case KVM_CAP_DEVICE_CTRL: |
749cf76c CD |
207 | case KVM_CAP_USER_MEMORY: |
208 | case KVM_CAP_SYNC_MMU: | |
209 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
210 | case KVM_CAP_ONE_REG: | |
aa024c2f | 211 | case KVM_CAP_ARM_PSCI: |
4447a208 | 212 | case KVM_CAP_ARM_PSCI_0_2: |
98047888 | 213 | case KVM_CAP_READONLY_MEM: |
ecccf0cc | 214 | case KVM_CAP_MP_STATE: |
460df4c1 | 215 | case KVM_CAP_IMMEDIATE_EXIT: |
58bf437f | 216 | case KVM_CAP_VCPU_EVENTS: |
749cf76c CD |
217 | r = 1; |
218 | break; | |
3401d546 CD |
219 | case KVM_CAP_ARM_SET_DEVICE_ADDR: |
220 | r = 1; | |
ca46e10f | 221 | break; |
749cf76c CD |
222 | case KVM_CAP_NR_VCPUS: |
223 | r = num_online_cpus(); | |
224 | break; | |
225 | case KVM_CAP_MAX_VCPUS: | |
226 | r = KVM_MAX_VCPUS; | |
227 | break; | |
7af4df85 LC |
228 | case KVM_CAP_NR_MEMSLOTS: |
229 | r = KVM_USER_MEM_SLOTS; | |
230 | break; | |
2988509d VM |
231 | case KVM_CAP_MSI_DEVID: |
232 | if (!kvm) | |
233 | r = -EINVAL; | |
234 | else | |
235 | r = kvm->arch.vgic.msis_require_devid; | |
236 | break; | |
f7214e60 CD |
237 | case KVM_CAP_ARM_USER_IRQ: |
238 | /* | |
239 | * 1: EL1_VTIMER, EL1_PTIMER, and PMU. | |
240 | * (bump this number if adding more devices) | |
241 | */ | |
242 | r = 1; | |
243 | break; | |
749cf76c | 244 | default: |
375bdd3b | 245 | r = kvm_arch_vm_ioctl_check_extension(kvm, ext); |
749cf76c CD |
246 | break; |
247 | } | |
248 | return r; | |
249 | } | |
250 | ||
251 | long kvm_arch_dev_ioctl(struct file *filp, | |
252 | unsigned int ioctl, unsigned long arg) | |
253 | { | |
254 | return -EINVAL; | |
255 | } | |
256 | ||
d1e5b0e9 MO |
257 | struct kvm *kvm_arch_alloc_vm(void) |
258 | { | |
259 | if (!has_vhe()) | |
260 | return kzalloc(sizeof(struct kvm), GFP_KERNEL); | |
261 | ||
262 | return vzalloc(sizeof(struct kvm)); | |
263 | } | |
264 | ||
265 | void kvm_arch_free_vm(struct kvm *kvm) | |
266 | { | |
267 | if (!has_vhe()) | |
268 | kfree(kvm); | |
269 | else | |
270 | vfree(kvm); | |
271 | } | |
749cf76c CD |
272 | |
273 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) | |
274 | { | |
275 | int err; | |
276 | struct kvm_vcpu *vcpu; | |
277 | ||
716139df CD |
278 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) { |
279 | err = -EBUSY; | |
280 | goto out; | |
281 | } | |
282 | ||
3caa2d8c AP |
283 | if (id >= kvm->arch.max_vcpus) { |
284 | err = -EINVAL; | |
285 | goto out; | |
286 | } | |
287 | ||
749cf76c CD |
288 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
289 | if (!vcpu) { | |
290 | err = -ENOMEM; | |
291 | goto out; | |
292 | } | |
293 | ||
294 | err = kvm_vcpu_init(vcpu, kvm, id); | |
295 | if (err) | |
296 | goto free_vcpu; | |
297 | ||
c8dddecd | 298 | err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP); |
d5d8184d CD |
299 | if (err) |
300 | goto vcpu_uninit; | |
301 | ||
749cf76c | 302 | return vcpu; |
d5d8184d CD |
303 | vcpu_uninit: |
304 | kvm_vcpu_uninit(vcpu); | |
749cf76c CD |
305 | free_vcpu: |
306 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
307 | out: | |
308 | return ERR_PTR(err); | |
309 | } | |
310 | ||
31928aa5 | 311 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
749cf76c | 312 | { |
749cf76c CD |
313 | } |
314 | ||
315 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
316 | { | |
f1d7231c CD |
317 | if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm))) |
318 | static_branch_dec(&userspace_irqchip_in_use); | |
319 | ||
d5d8184d | 320 | kvm_mmu_free_memory_caches(vcpu); |
967f8427 | 321 | kvm_timer_vcpu_terminate(vcpu); |
5f0a714a | 322 | kvm_pmu_vcpu_destroy(vcpu); |
591d215a | 323 | kvm_vcpu_uninit(vcpu); |
d5d8184d | 324 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
749cf76c CD |
325 | } |
326 | ||
327 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) | |
328 | { | |
329 | kvm_arch_vcpu_free(vcpu); | |
330 | } | |
331 | ||
332 | int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) | |
333 | { | |
1c88ab7e | 334 | return kvm_timer_is_pending(vcpu); |
749cf76c CD |
335 | } |
336 | ||
d35268da CD |
337 | void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) |
338 | { | |
339 | kvm_timer_schedule(vcpu); | |
df9ba959 | 340 | kvm_vgic_v4_enable_doorbell(vcpu); |
d35268da CD |
341 | } |
342 | ||
343 | void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
344 | { | |
345 | kvm_timer_unschedule(vcpu); | |
df9ba959 | 346 | kvm_vgic_v4_disable_doorbell(vcpu); |
d35268da CD |
347 | } |
348 | ||
749cf76c CD |
349 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
350 | { | |
f7ed45be CD |
351 | /* Force users to call KVM_ARM_VCPU_INIT */ |
352 | vcpu->arch.target = -1; | |
f7fa034d | 353 | bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); |
1a89dd91 | 354 | |
967f8427 MZ |
355 | /* Set up the timer */ |
356 | kvm_timer_vcpu_init(vcpu); | |
357 | ||
84e690bf AB |
358 | kvm_arm_reset_debug_ptr(vcpu); |
359 | ||
1aab6f46 | 360 | return kvm_vgic_vcpu_init(vcpu); |
749cf76c CD |
361 | } |
362 | ||
749cf76c CD |
363 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
364 | { | |
94d0e598 MZ |
365 | int *last_ran; |
366 | ||
367 | last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran); | |
368 | ||
369 | /* | |
370 | * We might get preempted before the vCPU actually runs, but | |
371 | * over-invalidation doesn't affect correctness. | |
372 | */ | |
373 | if (*last_ran != vcpu->vcpu_id) { | |
374 | kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu); | |
375 | *last_ran = vcpu->vcpu_id; | |
376 | } | |
377 | ||
86ce8535 | 378 | vcpu->cpu = cpu; |
36989e7f | 379 | vcpu->arch.host_cpu_context = this_cpu_ptr(&kvm_host_cpu_state); |
5b3e5e5b | 380 | |
1638a12d | 381 | kvm_arm_set_running_vcpu(vcpu); |
328e5664 | 382 | kvm_vgic_load(vcpu); |
b103cc3f | 383 | kvm_timer_vcpu_load(vcpu); |
bc192cee | 384 | kvm_vcpu_load_sysregs(vcpu); |
e6b673b7 | 385 | kvm_arch_vcpu_load_fp(vcpu); |
de737089 MZ |
386 | |
387 | if (single_task_running()) | |
388 | vcpu_clear_wfe_traps(vcpu); | |
389 | else | |
390 | vcpu_set_wfe_traps(vcpu); | |
749cf76c CD |
391 | } |
392 | ||
393 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
394 | { | |
e6b673b7 | 395 | kvm_arch_vcpu_put_fp(vcpu); |
bc192cee | 396 | kvm_vcpu_put_sysregs(vcpu); |
b103cc3f | 397 | kvm_timer_vcpu_put(vcpu); |
328e5664 CD |
398 | kvm_vgic_put(vcpu); |
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 | |
6a727b0b | 475 | * @kvm: The VM's VMID to check |
f7ed45be CD |
476 | * |
477 | * return true if there is a new generation of VMIDs being used | |
478 | * | |
479 | * The hardware supports only 256 values with the value zero reserved for the | |
480 | * host, so we check if an assigned value belongs to a previous generation, | |
481 | * which which requires us to assign a new value. If we're the first to use a | |
482 | * VMID for the new generation, we must flush necessary caches and TLBs on all | |
483 | * CPUs. | |
484 | */ | |
485 | static bool need_new_vmid_gen(struct kvm *kvm) | |
486 | { | |
487 | return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen)); | |
488 | } | |
489 | ||
490 | /** | |
491 | * update_vttbr - Update the VTTBR with a valid VMID before the guest runs | |
492 | * @kvm The guest that we are about to run | |
493 | * | |
494 | * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the | |
495 | * VM has a valid VMID, otherwise assigns a new one and flushes corresponding | |
496 | * caches and TLBs. | |
497 | */ | |
498 | static void update_vttbr(struct kvm *kvm) | |
499 | { | |
500 | phys_addr_t pgd_phys; | |
ab510027 | 501 | u64 vmid, cnp = kvm_cpu_has_cnp() ? VTTBR_CNP_BIT : 0; |
f0cf47d9 | 502 | bool new_gen; |
f7ed45be | 503 | |
f0cf47d9 MZ |
504 | read_lock(&kvm_vmid_lock); |
505 | new_gen = need_new_vmid_gen(kvm); | |
506 | read_unlock(&kvm_vmid_lock); | |
507 | ||
508 | if (!new_gen) | |
f7ed45be CD |
509 | return; |
510 | ||
f0cf47d9 | 511 | write_lock(&kvm_vmid_lock); |
f7ed45be CD |
512 | |
513 | /* | |
514 | * We need to re-check the vmid_gen here to ensure that if another vcpu | |
515 | * already allocated a valid vmid for this vm, then this vcpu should | |
516 | * use the same vmid. | |
517 | */ | |
518 | if (!need_new_vmid_gen(kvm)) { | |
f0cf47d9 | 519 | write_unlock(&kvm_vmid_lock); |
f7ed45be CD |
520 | return; |
521 | } | |
522 | ||
523 | /* First user of a new VMID generation? */ | |
524 | if (unlikely(kvm_next_vmid == 0)) { | |
525 | atomic64_inc(&kvm_vmid_gen); | |
526 | kvm_next_vmid = 1; | |
527 | ||
528 | /* | |
529 | * On SMP we know no other CPUs can use this CPU's or each | |
530 | * other's VMID after force_vm_exit returns since the | |
531 | * kvm_vmid_lock blocks them from reentry to the guest. | |
532 | */ | |
533 | force_vm_exit(cpu_all_mask); | |
534 | /* | |
535 | * Now broadcast TLB + ICACHE invalidation over the inner | |
536 | * shareable domain to make sure all data structures are | |
537 | * clean. | |
538 | */ | |
539 | kvm_call_hyp(__kvm_flush_vm_context); | |
540 | } | |
541 | ||
542 | kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen); | |
543 | kvm->arch.vmid = kvm_next_vmid; | |
544 | kvm_next_vmid++; | |
20475f78 | 545 | kvm_next_vmid &= (1 << kvm_vmid_bits) - 1; |
f7ed45be CD |
546 | |
547 | /* update vttbr to be used with the new vmid */ | |
9163ee23 | 548 | pgd_phys = virt_to_phys(kvm->arch.pgd); |
e55cac5b | 549 | BUG_ON(pgd_phys & ~kvm_vttbr_baddr_mask(kvm)); |
20475f78 | 550 | vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits); |
ab510027 | 551 | kvm->arch.vttbr = kvm_phys_to_vttbr(pgd_phys) | vmid | cnp; |
f7ed45be | 552 | |
f0cf47d9 | 553 | write_unlock(&kvm_vmid_lock); |
f7ed45be CD |
554 | } |
555 | ||
f7ed45be CD |
556 | static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) |
557 | { | |
05971120 | 558 | struct kvm *kvm = vcpu->kvm; |
41a54482 | 559 | int ret = 0; |
e1ba0207 | 560 | |
f7ed45be CD |
561 | if (likely(vcpu->arch.has_run_once)) |
562 | return 0; | |
563 | ||
564 | vcpu->arch.has_run_once = true; | |
aa024c2f | 565 | |
61bbe380 CD |
566 | if (likely(irqchip_in_kernel(kvm))) { |
567 | /* | |
568 | * Map the VGIC hardware resources before running a vcpu the | |
569 | * first time on this VM. | |
570 | */ | |
571 | if (unlikely(!vgic_ready(kvm))) { | |
572 | ret = kvm_vgic_map_resources(kvm); | |
573 | if (ret) | |
574 | return ret; | |
575 | } | |
576 | } else { | |
577 | /* | |
578 | * Tell the rest of the code that there are userspace irqchip | |
579 | * VMs in the wild. | |
580 | */ | |
581 | static_branch_inc(&userspace_irqchip_in_use); | |
01ac5e34 MZ |
582 | } |
583 | ||
d9e13977 | 584 | ret = kvm_timer_enable(vcpu); |
a2befacf CD |
585 | if (ret) |
586 | return ret; | |
587 | ||
588 | ret = kvm_arm_pmu_v3_enable(vcpu); | |
05971120 | 589 | |
41a54482 | 590 | return ret; |
f7ed45be CD |
591 | } |
592 | ||
c1426e4c EA |
593 | bool kvm_arch_intc_initialized(struct kvm *kvm) |
594 | { | |
595 | return vgic_initialized(kvm); | |
596 | } | |
597 | ||
b13216cf | 598 | void kvm_arm_halt_guest(struct kvm *kvm) |
3b92830a EA |
599 | { |
600 | int i; | |
601 | struct kvm_vcpu *vcpu; | |
602 | ||
603 | kvm_for_each_vcpu(i, vcpu, kvm) | |
604 | vcpu->arch.pause = true; | |
7b244e2b | 605 | kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP); |
3b92830a EA |
606 | } |
607 | ||
b13216cf | 608 | void kvm_arm_resume_guest(struct kvm *kvm) |
3b92830a EA |
609 | { |
610 | int i; | |
611 | struct kvm_vcpu *vcpu; | |
612 | ||
abd72296 CD |
613 | kvm_for_each_vcpu(i, vcpu, kvm) { |
614 | vcpu->arch.pause = false; | |
b3dae109 | 615 | swake_up_one(kvm_arch_vcpu_wq(vcpu)); |
abd72296 | 616 | } |
3b92830a EA |
617 | } |
618 | ||
7b244e2b | 619 | static void vcpu_req_sleep(struct kvm_vcpu *vcpu) |
aa024c2f | 620 | { |
8577370f | 621 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); |
aa024c2f | 622 | |
b3dae109 | 623 | swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) && |
3b92830a | 624 | (!vcpu->arch.pause))); |
0592c005 | 625 | |
424c989b | 626 | if (vcpu->arch.power_off || vcpu->arch.pause) { |
0592c005 | 627 | /* Awaken to handle a signal, request we sleep again later. */ |
7b244e2b | 628 | kvm_make_request(KVM_REQ_SLEEP, vcpu); |
0592c005 | 629 | } |
aa024c2f MZ |
630 | } |
631 | ||
e8180dca AP |
632 | static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) |
633 | { | |
634 | return vcpu->arch.target >= 0; | |
635 | } | |
636 | ||
0592c005 AJ |
637 | static void check_vcpu_requests(struct kvm_vcpu *vcpu) |
638 | { | |
639 | if (kvm_request_pending(vcpu)) { | |
7b244e2b AJ |
640 | if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) |
641 | vcpu_req_sleep(vcpu); | |
325f9c64 AJ |
642 | |
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 CD |
676 | return ret; |
677 | if (kvm_arm_handle_step_debug(vcpu, vcpu->run)) | |
678 | return 0; | |
45e96ea6 CD |
679 | } |
680 | ||
829a5863 CD |
681 | if (run->immediate_exit) |
682 | return -EINTR; | |
683 | ||
684 | vcpu_load(vcpu); | |
460df4c1 | 685 | |
20b7035c | 686 | kvm_sigset_activate(vcpu); |
f7ed45be CD |
687 | |
688 | ret = 1; | |
689 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
690 | while (ret > 0) { | |
691 | /* | |
692 | * Check conditions before entering the guest | |
693 | */ | |
694 | cond_resched(); | |
695 | ||
696 | update_vttbr(vcpu->kvm); | |
697 | ||
0592c005 AJ |
698 | check_vcpu_requests(vcpu); |
699 | ||
abdf5843 MZ |
700 | /* |
701 | * Preparing the interrupts to be injected also | |
702 | * involves poking the GIC, which must be done in a | |
703 | * non-preemptible context. | |
704 | */ | |
1b3d546d | 705 | preempt_disable(); |
328e5664 | 706 | |
b02386eb | 707 | kvm_pmu_flush_hwstate(vcpu); |
328e5664 | 708 | |
f7ed45be CD |
709 | local_irq_disable(); |
710 | ||
abdf5843 MZ |
711 | kvm_vgic_flush_hwstate(vcpu); |
712 | ||
f7ed45be | 713 | /* |
61bbe380 CD |
714 | * Exit if we have a signal pending so that we can deliver the |
715 | * signal to user space. | |
f7ed45be | 716 | */ |
61bbe380 | 717 | if (signal_pending(current)) { |
f7ed45be CD |
718 | ret = -EINTR; |
719 | run->exit_reason = KVM_EXIT_INTR; | |
720 | } | |
721 | ||
61bbe380 CD |
722 | /* |
723 | * If we're using a userspace irqchip, then check if we need | |
724 | * to tell a userspace irqchip about timer or PMU level | |
725 | * changes and if so, exit to userspace (the actual level | |
726 | * state gets updated in kvm_timer_update_run and | |
727 | * kvm_pmu_update_run below). | |
728 | */ | |
729 | if (static_branch_unlikely(&userspace_irqchip_in_use)) { | |
730 | if (kvm_timer_should_notify_user(vcpu) || | |
731 | kvm_pmu_should_notify_user(vcpu)) { | |
732 | ret = -EINTR; | |
733 | run->exit_reason = KVM_EXIT_INTR; | |
734 | } | |
735 | } | |
736 | ||
6a6d73be AJ |
737 | /* |
738 | * Ensure we set mode to IN_GUEST_MODE after we disable | |
739 | * interrupts and before the final VCPU requests check. | |
740 | * See the comment in kvm_vcpu_exiting_guest_mode() and | |
741 | * Documentation/virtual/kvm/vcpu-requests.rst | |
742 | */ | |
743 | smp_store_mb(vcpu->mode, IN_GUEST_MODE); | |
744 | ||
101d3da0 | 745 | if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) || |
424c989b | 746 | kvm_request_pending(vcpu)) { |
6a6d73be | 747 | vcpu->mode = OUTSIDE_GUEST_MODE; |
771621b0 | 748 | isb(); /* Ensure work in x_flush_hwstate is committed */ |
b02386eb | 749 | kvm_pmu_sync_hwstate(vcpu); |
61bbe380 CD |
750 | if (static_branch_unlikely(&userspace_irqchip_in_use)) |
751 | kvm_timer_sync_hwstate(vcpu); | |
1a89dd91 | 752 | kvm_vgic_sync_hwstate(vcpu); |
ee9bb9a1 | 753 | local_irq_enable(); |
abdf5843 | 754 | preempt_enable(); |
f7ed45be CD |
755 | continue; |
756 | } | |
757 | ||
56c7f5e7 AB |
758 | kvm_arm_setup_debug(vcpu); |
759 | ||
f7ed45be CD |
760 | /************************************************************** |
761 | * Enter the guest | |
762 | */ | |
763 | trace_kvm_entry(*vcpu_pc(vcpu)); | |
6edaa530 | 764 | guest_enter_irqoff(); |
f7ed45be | 765 | |
3f5c90b8 CD |
766 | if (has_vhe()) { |
767 | kvm_arm_vhe_guest_enter(); | |
768 | ret = kvm_vcpu_run_vhe(vcpu); | |
4f5abad9 | 769 | kvm_arm_vhe_guest_exit(); |
3f5c90b8 CD |
770 | } else { |
771 | ret = kvm_call_hyp(__kvm_vcpu_run_nvhe, vcpu); | |
772 | } | |
773 | ||
f7ed45be | 774 | vcpu->mode = OUTSIDE_GUEST_MODE; |
b19e6892 | 775 | vcpu->stat.exits++; |
1b3d546d CD |
776 | /* |
777 | * Back from guest | |
778 | *************************************************************/ | |
779 | ||
56c7f5e7 AB |
780 | kvm_arm_clear_debug(vcpu); |
781 | ||
ee9bb9a1 | 782 | /* |
b103cc3f | 783 | * We must sync the PMU state before the vgic state so |
ee9bb9a1 CD |
784 | * that the vgic can properly sample the updated state of the |
785 | * interrupt line. | |
786 | */ | |
787 | kvm_pmu_sync_hwstate(vcpu); | |
ee9bb9a1 | 788 | |
b103cc3f CD |
789 | /* |
790 | * Sync the vgic state before syncing the timer state because | |
791 | * the timer code needs to know if the virtual timer | |
792 | * interrupts are active. | |
793 | */ | |
ee9bb9a1 CD |
794 | kvm_vgic_sync_hwstate(vcpu); |
795 | ||
b103cc3f CD |
796 | /* |
797 | * Sync the timer hardware state before enabling interrupts as | |
798 | * we don't want vtimer interrupts to race with syncing the | |
799 | * timer virtual interrupt state. | |
800 | */ | |
61bbe380 CD |
801 | if (static_branch_unlikely(&userspace_irqchip_in_use)) |
802 | kvm_timer_sync_hwstate(vcpu); | |
b103cc3f | 803 | |
e6b673b7 DM |
804 | kvm_arch_vcpu_ctxsync_fp(vcpu); |
805 | ||
f7ed45be CD |
806 | /* |
807 | * We may have taken a host interrupt in HYP mode (ie | |
808 | * while executing the guest). This interrupt is still | |
809 | * pending, as we haven't serviced it yet! | |
810 | * | |
811 | * We're now back in SVC mode, with interrupts | |
812 | * disabled. Enabling the interrupts now will have | |
813 | * the effect of taking the interrupt again, in SVC | |
814 | * mode this time. | |
815 | */ | |
816 | local_irq_enable(); | |
817 | ||
818 | /* | |
6edaa530 | 819 | * We do local_irq_enable() before calling guest_exit() so |
1b3d546d CD |
820 | * that if a timer interrupt hits while running the guest we |
821 | * account that tick as being spent in the guest. We enable | |
6edaa530 | 822 | * preemption after calling guest_exit() so that if we get |
1b3d546d CD |
823 | * preempted we make sure ticks after that is not counted as |
824 | * guest time. | |
825 | */ | |
6edaa530 | 826 | guest_exit(); |
b5905dc1 | 827 | trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); |
1b3d546d | 828 | |
3368bd80 JM |
829 | /* Exit types that need handling before we can be preempted */ |
830 | handle_exit_early(vcpu, run, ret); | |
831 | ||
abdf5843 MZ |
832 | preempt_enable(); |
833 | ||
f7ed45be CD |
834 | ret = handle_exit(vcpu, run, ret); |
835 | } | |
836 | ||
d9e13977 | 837 | /* Tell userspace about in-kernel device output levels */ |
3dbbdf78 CD |
838 | if (unlikely(!irqchip_in_kernel(vcpu->kvm))) { |
839 | kvm_timer_update_run(vcpu); | |
840 | kvm_pmu_update_run(vcpu); | |
841 | } | |
d9e13977 | 842 | |
20b7035c JS |
843 | kvm_sigset_deactivate(vcpu); |
844 | ||
accb757d | 845 | vcpu_put(vcpu); |
f7ed45be | 846 | return ret; |
749cf76c CD |
847 | } |
848 | ||
86ce8535 CD |
849 | static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) |
850 | { | |
851 | int bit_index; | |
852 | bool set; | |
3df59d8d | 853 | unsigned long *hcr; |
86ce8535 CD |
854 | |
855 | if (number == KVM_ARM_IRQ_CPU_IRQ) | |
856 | bit_index = __ffs(HCR_VI); | |
857 | else /* KVM_ARM_IRQ_CPU_FIQ */ | |
858 | bit_index = __ffs(HCR_VF); | |
859 | ||
3df59d8d | 860 | hcr = vcpu_hcr(vcpu); |
86ce8535 | 861 | if (level) |
3df59d8d | 862 | set = test_and_set_bit(bit_index, hcr); |
86ce8535 | 863 | else |
3df59d8d | 864 | set = test_and_clear_bit(bit_index, hcr); |
86ce8535 CD |
865 | |
866 | /* | |
867 | * If we didn't change anything, no need to wake up or kick other CPUs | |
868 | */ | |
869 | if (set == level) | |
870 | return 0; | |
871 | ||
872 | /* | |
873 | * The vcpu irq_lines field was updated, wake up sleeping VCPUs and | |
874 | * trigger a world-switch round on the running physical CPU to set the | |
875 | * virtual IRQ/FIQ fields in the HCR appropriately. | |
876 | */ | |
325f9c64 | 877 | kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu); |
86ce8535 CD |
878 | kvm_vcpu_kick(vcpu); |
879 | ||
880 | return 0; | |
881 | } | |
882 | ||
79558f11 AG |
883 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, |
884 | bool line_status) | |
86ce8535 CD |
885 | { |
886 | u32 irq = irq_level->irq; | |
887 | unsigned int irq_type, vcpu_idx, irq_num; | |
888 | int nrcpus = atomic_read(&kvm->online_vcpus); | |
889 | struct kvm_vcpu *vcpu = NULL; | |
890 | bool level = irq_level->level; | |
891 | ||
892 | irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; | |
893 | vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; | |
894 | irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; | |
895 | ||
896 | trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); | |
897 | ||
5863c2ce MZ |
898 | switch (irq_type) { |
899 | case KVM_ARM_IRQ_TYPE_CPU: | |
900 | if (irqchip_in_kernel(kvm)) | |
901 | return -ENXIO; | |
86ce8535 | 902 | |
5863c2ce MZ |
903 | if (vcpu_idx >= nrcpus) |
904 | return -EINVAL; | |
86ce8535 | 905 | |
5863c2ce MZ |
906 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); |
907 | if (!vcpu) | |
908 | return -EINVAL; | |
86ce8535 | 909 | |
5863c2ce MZ |
910 | if (irq_num > KVM_ARM_IRQ_CPU_FIQ) |
911 | return -EINVAL; | |
912 | ||
913 | return vcpu_interrupt_line(vcpu, irq_num, level); | |
914 | case KVM_ARM_IRQ_TYPE_PPI: | |
915 | if (!irqchip_in_kernel(kvm)) | |
916 | return -ENXIO; | |
917 | ||
918 | if (vcpu_idx >= nrcpus) | |
919 | return -EINVAL; | |
920 | ||
921 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); | |
922 | if (!vcpu) | |
923 | return -EINVAL; | |
924 | ||
925 | if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) | |
926 | return -EINVAL; | |
86ce8535 | 927 | |
cb3f0ad8 | 928 | return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL); |
5863c2ce MZ |
929 | case KVM_ARM_IRQ_TYPE_SPI: |
930 | if (!irqchip_in_kernel(kvm)) | |
931 | return -ENXIO; | |
932 | ||
fd1d0ddf | 933 | if (irq_num < VGIC_NR_PRIVATE_IRQS) |
5863c2ce MZ |
934 | return -EINVAL; |
935 | ||
cb3f0ad8 | 936 | return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL); |
5863c2ce MZ |
937 | } |
938 | ||
939 | return -EINVAL; | |
86ce8535 CD |
940 | } |
941 | ||
f7fa034d CD |
942 | static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, |
943 | const struct kvm_vcpu_init *init) | |
944 | { | |
945 | unsigned int i; | |
946 | int phys_target = kvm_target_cpu(); | |
947 | ||
948 | if (init->target != phys_target) | |
949 | return -EINVAL; | |
950 | ||
951 | /* | |
952 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
953 | * use the same target. | |
954 | */ | |
955 | if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) | |
956 | return -EINVAL; | |
957 | ||
958 | /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ | |
959 | for (i = 0; i < sizeof(init->features) * 8; i++) { | |
960 | bool set = (init->features[i / 32] & (1 << (i % 32))); | |
961 | ||
962 | if (set && i >= KVM_VCPU_MAX_FEATURES) | |
963 | return -ENOENT; | |
964 | ||
965 | /* | |
966 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
967 | * use the same feature set. | |
968 | */ | |
969 | if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && | |
970 | test_bit(i, vcpu->arch.features) != set) | |
971 | return -EINVAL; | |
972 | ||
973 | if (set) | |
974 | set_bit(i, vcpu->arch.features); | |
975 | } | |
976 | ||
977 | vcpu->arch.target = phys_target; | |
978 | ||
979 | /* Now we know what it is, we can reset it. */ | |
980 | return kvm_reset_vcpu(vcpu); | |
981 | } | |
982 | ||
983 | ||
478a8237 CD |
984 | static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, |
985 | struct kvm_vcpu_init *init) | |
986 | { | |
987 | int ret; | |
988 | ||
989 | ret = kvm_vcpu_set_target(vcpu, init); | |
990 | if (ret) | |
991 | return ret; | |
992 | ||
957db105 CD |
993 | /* |
994 | * Ensure a rebooted VM will fault in RAM pages and detect if the | |
995 | * guest MMU is turned off and flush the caches as needed. | |
996 | */ | |
997 | if (vcpu->arch.has_run_once) | |
998 | stage2_unmap_vm(vcpu->kvm); | |
999 | ||
b856a591 CD |
1000 | vcpu_reset_hcr(vcpu); |
1001 | ||
478a8237 | 1002 | /* |
3781528e | 1003 | * Handle the "start in power-off" case. |
478a8237 | 1004 | */ |
03f1d4c1 | 1005 | if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) |
424c989b | 1006 | vcpu_power_off(vcpu); |
3ad8b3de | 1007 | else |
3781528e | 1008 | vcpu->arch.power_off = false; |
478a8237 CD |
1009 | |
1010 | return 0; | |
1011 | } | |
1012 | ||
f577f6c2 SZ |
1013 | static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, |
1014 | struct kvm_device_attr *attr) | |
1015 | { | |
1016 | int ret = -ENXIO; | |
1017 | ||
1018 | switch (attr->group) { | |
1019 | default: | |
bb0c70bc | 1020 | ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); |
f577f6c2 SZ |
1021 | break; |
1022 | } | |
1023 | ||
1024 | return ret; | |
1025 | } | |
1026 | ||
1027 | static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, | |
1028 | struct kvm_device_attr *attr) | |
1029 | { | |
1030 | int ret = -ENXIO; | |
1031 | ||
1032 | switch (attr->group) { | |
1033 | default: | |
bb0c70bc | 1034 | ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); |
f577f6c2 SZ |
1035 | break; |
1036 | } | |
1037 | ||
1038 | return ret; | |
1039 | } | |
1040 | ||
1041 | static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, | |
1042 | struct kvm_device_attr *attr) | |
1043 | { | |
1044 | int ret = -ENXIO; | |
1045 | ||
1046 | switch (attr->group) { | |
1047 | default: | |
bb0c70bc | 1048 | ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); |
f577f6c2 SZ |
1049 | break; |
1050 | } | |
1051 | ||
1052 | return ret; | |
1053 | } | |
1054 | ||
539aee0e JM |
1055 | static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu, |
1056 | struct kvm_vcpu_events *events) | |
1057 | { | |
1058 | memset(events, 0, sizeof(*events)); | |
1059 | ||
1060 | return __kvm_arm_vcpu_get_events(vcpu, events); | |
1061 | } | |
1062 | ||
1063 | static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu, | |
1064 | struct kvm_vcpu_events *events) | |
1065 | { | |
1066 | int i; | |
1067 | ||
1068 | /* check whether the reserved field is zero */ | |
1069 | for (i = 0; i < ARRAY_SIZE(events->reserved); i++) | |
1070 | if (events->reserved[i]) | |
1071 | return -EINVAL; | |
1072 | ||
1073 | /* check whether the pad field is zero */ | |
1074 | for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++) | |
1075 | if (events->exception.pad[i]) | |
1076 | return -EINVAL; | |
1077 | ||
1078 | return __kvm_arm_vcpu_set_events(vcpu, events); | |
1079 | } | |
539aee0e | 1080 | |
749cf76c CD |
1081 | long kvm_arch_vcpu_ioctl(struct file *filp, |
1082 | unsigned int ioctl, unsigned long arg) | |
1083 | { | |
1084 | struct kvm_vcpu *vcpu = filp->private_data; | |
1085 | void __user *argp = (void __user *)arg; | |
f577f6c2 | 1086 | struct kvm_device_attr attr; |
9b062471 CD |
1087 | long r; |
1088 | ||
749cf76c CD |
1089 | switch (ioctl) { |
1090 | case KVM_ARM_VCPU_INIT: { | |
1091 | struct kvm_vcpu_init init; | |
1092 | ||
9b062471 | 1093 | r = -EFAULT; |
749cf76c | 1094 | if (copy_from_user(&init, argp, sizeof(init))) |
9b062471 | 1095 | break; |
749cf76c | 1096 | |
9b062471 CD |
1097 | r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); |
1098 | break; | |
749cf76c CD |
1099 | } |
1100 | case KVM_SET_ONE_REG: | |
1101 | case KVM_GET_ONE_REG: { | |
1102 | struct kvm_one_reg reg; | |
e8180dca | 1103 | |
9b062471 | 1104 | r = -ENOEXEC; |
e8180dca | 1105 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
9b062471 | 1106 | break; |
e8180dca | 1107 | |
9b062471 | 1108 | r = -EFAULT; |
749cf76c | 1109 | if (copy_from_user(®, argp, sizeof(reg))) |
9b062471 CD |
1110 | break; |
1111 | ||
749cf76c | 1112 | if (ioctl == KVM_SET_ONE_REG) |
9b062471 | 1113 | r = kvm_arm_set_reg(vcpu, ®); |
749cf76c | 1114 | else |
9b062471 CD |
1115 | r = kvm_arm_get_reg(vcpu, ®); |
1116 | break; | |
749cf76c CD |
1117 | } |
1118 | case KVM_GET_REG_LIST: { | |
1119 | struct kvm_reg_list __user *user_list = argp; | |
1120 | struct kvm_reg_list reg_list; | |
1121 | unsigned n; | |
1122 | ||
9b062471 | 1123 | r = -ENOEXEC; |
e8180dca | 1124 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
9b062471 | 1125 | break; |
e8180dca | 1126 | |
9b062471 | 1127 | r = -EFAULT; |
749cf76c | 1128 | if (copy_from_user(®_list, user_list, sizeof(reg_list))) |
9b062471 | 1129 | break; |
749cf76c CD |
1130 | n = reg_list.n; |
1131 | reg_list.n = kvm_arm_num_regs(vcpu); | |
1132 | if (copy_to_user(user_list, ®_list, sizeof(reg_list))) | |
9b062471 CD |
1133 | break; |
1134 | r = -E2BIG; | |
749cf76c | 1135 | if (n < reg_list.n) |
9b062471 CD |
1136 | break; |
1137 | r = kvm_arm_copy_reg_indices(vcpu, user_list->reg); | |
1138 | break; | |
749cf76c | 1139 | } |
f577f6c2 | 1140 | case KVM_SET_DEVICE_ATTR: { |
9b062471 | 1141 | r = -EFAULT; |
f577f6c2 | 1142 | if (copy_from_user(&attr, argp, sizeof(attr))) |
9b062471 CD |
1143 | break; |
1144 | r = kvm_arm_vcpu_set_attr(vcpu, &attr); | |
1145 | break; | |
f577f6c2 SZ |
1146 | } |
1147 | case KVM_GET_DEVICE_ATTR: { | |
9b062471 | 1148 | r = -EFAULT; |
f577f6c2 | 1149 | if (copy_from_user(&attr, argp, sizeof(attr))) |
9b062471 CD |
1150 | break; |
1151 | r = kvm_arm_vcpu_get_attr(vcpu, &attr); | |
1152 | break; | |
f577f6c2 SZ |
1153 | } |
1154 | case KVM_HAS_DEVICE_ATTR: { | |
9b062471 | 1155 | r = -EFAULT; |
f577f6c2 | 1156 | if (copy_from_user(&attr, argp, sizeof(attr))) |
9b062471 CD |
1157 | break; |
1158 | r = kvm_arm_vcpu_has_attr(vcpu, &attr); | |
1159 | break; | |
f577f6c2 | 1160 | } |
b7b27fac DG |
1161 | case KVM_GET_VCPU_EVENTS: { |
1162 | struct kvm_vcpu_events events; | |
1163 | ||
1164 | if (kvm_arm_vcpu_get_events(vcpu, &events)) | |
1165 | return -EINVAL; | |
1166 | ||
1167 | if (copy_to_user(argp, &events, sizeof(events))) | |
1168 | return -EFAULT; | |
1169 | ||
1170 | return 0; | |
1171 | } | |
1172 | case KVM_SET_VCPU_EVENTS: { | |
1173 | struct kvm_vcpu_events events; | |
1174 | ||
1175 | if (copy_from_user(&events, argp, sizeof(events))) | |
1176 | return -EFAULT; | |
1177 | ||
1178 | return kvm_arm_vcpu_set_events(vcpu, &events); | |
1179 | } | |
749cf76c | 1180 | default: |
9b062471 | 1181 | r = -EINVAL; |
749cf76c | 1182 | } |
9b062471 | 1183 | |
9b062471 | 1184 | return r; |
749cf76c CD |
1185 | } |
1186 | ||
53c810c3 MS |
1187 | /** |
1188 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
1189 | * @kvm: kvm instance | |
1190 | * @log: slot id and address to which we copy the log | |
1191 | * | |
1192 | * Steps 1-4 below provide general overview of dirty page logging. See | |
1193 | * kvm_get_dirty_log_protect() function description for additional details. | |
1194 | * | |
1195 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
1196 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
1197 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
1198 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
1199 | * writes will be marked dirty for next log read. | |
1200 | * | |
1201 | * 1. Take a snapshot of the bit and clear it if needed. | |
1202 | * 2. Write protect the corresponding page. | |
1203 | * 3. Copy the snapshot to the userspace. | |
1204 | * 4. Flush TLB's if needed. | |
1205 | */ | |
749cf76c CD |
1206 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
1207 | { | |
8fe65a82 | 1208 | bool flush = false; |
53c810c3 MS |
1209 | int r; |
1210 | ||
1211 | mutex_lock(&kvm->slots_lock); | |
1212 | ||
8fe65a82 | 1213 | r = kvm_get_dirty_log_protect(kvm, log, &flush); |
53c810c3 | 1214 | |
8fe65a82 | 1215 | if (flush) |
53c810c3 MS |
1216 | kvm_flush_remote_tlbs(kvm); |
1217 | ||
1218 | mutex_unlock(&kvm->slots_lock); | |
1219 | return r; | |
749cf76c CD |
1220 | } |
1221 | ||
2a31b9db PB |
1222 | int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log) |
1223 | { | |
1224 | bool flush = false; | |
1225 | int r; | |
1226 | ||
1227 | mutex_lock(&kvm->slots_lock); | |
1228 | ||
1229 | r = kvm_clear_dirty_log_protect(kvm, log, &flush); | |
1230 | ||
1231 | if (flush) | |
1232 | kvm_flush_remote_tlbs(kvm); | |
1233 | ||
1234 | mutex_unlock(&kvm->slots_lock); | |
1235 | return r; | |
1236 | } | |
1237 | ||
3401d546 CD |
1238 | static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, |
1239 | struct kvm_arm_device_addr *dev_addr) | |
1240 | { | |
330690cd CD |
1241 | unsigned long dev_id, type; |
1242 | ||
1243 | dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> | |
1244 | KVM_ARM_DEVICE_ID_SHIFT; | |
1245 | type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> | |
1246 | KVM_ARM_DEVICE_TYPE_SHIFT; | |
1247 | ||
1248 | switch (dev_id) { | |
1249 | case KVM_ARM_DEVICE_VGIC_V2: | |
c7da6fa4 PF |
1250 | if (!vgic_present) |
1251 | return -ENXIO; | |
ce01e4e8 | 1252 | return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); |
330690cd CD |
1253 | default: |
1254 | return -ENODEV; | |
1255 | } | |
3401d546 CD |
1256 | } |
1257 | ||
749cf76c CD |
1258 | long kvm_arch_vm_ioctl(struct file *filp, |
1259 | unsigned int ioctl, unsigned long arg) | |
1260 | { | |
3401d546 CD |
1261 | struct kvm *kvm = filp->private_data; |
1262 | void __user *argp = (void __user *)arg; | |
1263 | ||
1264 | switch (ioctl) { | |
5863c2ce | 1265 | case KVM_CREATE_IRQCHIP: { |
a28ebea2 | 1266 | int ret; |
c7da6fa4 PF |
1267 | if (!vgic_present) |
1268 | return -ENXIO; | |
a28ebea2 CD |
1269 | mutex_lock(&kvm->lock); |
1270 | ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); | |
1271 | mutex_unlock(&kvm->lock); | |
1272 | return ret; | |
5863c2ce | 1273 | } |
3401d546 CD |
1274 | case KVM_ARM_SET_DEVICE_ADDR: { |
1275 | struct kvm_arm_device_addr dev_addr; | |
1276 | ||
1277 | if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) | |
1278 | return -EFAULT; | |
1279 | return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); | |
1280 | } | |
42c4e0c7 AP |
1281 | case KVM_ARM_PREFERRED_TARGET: { |
1282 | int err; | |
1283 | struct kvm_vcpu_init init; | |
1284 | ||
1285 | err = kvm_vcpu_preferred_target(&init); | |
1286 | if (err) | |
1287 | return err; | |
1288 | ||
1289 | if (copy_to_user(argp, &init, sizeof(init))) | |
1290 | return -EFAULT; | |
1291 | ||
1292 | return 0; | |
1293 | } | |
3401d546 CD |
1294 | default: |
1295 | return -EINVAL; | |
1296 | } | |
749cf76c CD |
1297 | } |
1298 | ||
d157f4a5 | 1299 | static void cpu_init_hyp_mode(void *dummy) |
342cd0ab | 1300 | { |
dac288f7 | 1301 | phys_addr_t pgd_ptr; |
342cd0ab CD |
1302 | unsigned long hyp_stack_ptr; |
1303 | unsigned long stack_page; | |
1304 | unsigned long vector_ptr; | |
1305 | ||
1306 | /* Switch from the HYP stub to our own HYP init vector */ | |
5a677ce0 | 1307 | __hyp_set_vectors(kvm_get_idmap_vector()); |
342cd0ab | 1308 | |
dac288f7 | 1309 | pgd_ptr = kvm_mmu_get_httbr(); |
1436c1aa | 1310 | stack_page = __this_cpu_read(kvm_arm_hyp_stack_page); |
342cd0ab | 1311 | hyp_stack_ptr = stack_page + PAGE_SIZE; |
6840bdd7 | 1312 | vector_ptr = (unsigned long)kvm_get_hyp_vector(); |
342cd0ab | 1313 | |
12fda812 | 1314 | __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr); |
35a2491a | 1315 | __cpu_init_stage2(); |
342cd0ab CD |
1316 | } |
1317 | ||
47eb3cba MZ |
1318 | static void cpu_hyp_reset(void) |
1319 | { | |
1320 | if (!is_kernel_in_hyp_mode()) | |
1321 | __hyp_reset_vectors(); | |
1322 | } | |
1323 | ||
5f5560b1 JM |
1324 | static void cpu_hyp_reinit(void) |
1325 | { | |
47eb3cba MZ |
1326 | cpu_hyp_reset(); |
1327 | ||
9d47bb0d | 1328 | if (is_kernel_in_hyp_mode()) |
02d50cda | 1329 | kvm_timer_init_vhe(); |
9d47bb0d | 1330 | else |
47eb3cba | 1331 | cpu_init_hyp_mode(NULL); |
5b0d2cc2 | 1332 | |
da5a3ce6 | 1333 | kvm_arm_init_debug(); |
5b0d2cc2 CD |
1334 | |
1335 | if (vgic_present) | |
1336 | kvm_vgic_init_cpu_hardware(); | |
5f5560b1 JM |
1337 | } |
1338 | ||
67f69197 AT |
1339 | static void _kvm_arch_hardware_enable(void *discard) |
1340 | { | |
1341 | if (!__this_cpu_read(kvm_arm_hardware_enabled)) { | |
5f5560b1 | 1342 | cpu_hyp_reinit(); |
67f69197 | 1343 | __this_cpu_write(kvm_arm_hardware_enabled, 1); |
d157f4a5 | 1344 | } |
67f69197 | 1345 | } |
d157f4a5 | 1346 | |
67f69197 AT |
1347 | int kvm_arch_hardware_enable(void) |
1348 | { | |
1349 | _kvm_arch_hardware_enable(NULL); | |
1350 | return 0; | |
342cd0ab CD |
1351 | } |
1352 | ||
67f69197 AT |
1353 | static void _kvm_arch_hardware_disable(void *discard) |
1354 | { | |
1355 | if (__this_cpu_read(kvm_arm_hardware_enabled)) { | |
1356 | cpu_hyp_reset(); | |
1357 | __this_cpu_write(kvm_arm_hardware_enabled, 0); | |
1358 | } | |
1359 | } | |
1360 | ||
1361 | void kvm_arch_hardware_disable(void) | |
1362 | { | |
1363 | _kvm_arch_hardware_disable(NULL); | |
1364 | } | |
d157f4a5 | 1365 | |
1fcf7ce0 LP |
1366 | #ifdef CONFIG_CPU_PM |
1367 | static int hyp_init_cpu_pm_notifier(struct notifier_block *self, | |
1368 | unsigned long cmd, | |
1369 | void *v) | |
1370 | { | |
67f69197 AT |
1371 | /* |
1372 | * kvm_arm_hardware_enabled is left with its old value over | |
1373 | * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should | |
1374 | * re-enable hyp. | |
1375 | */ | |
1376 | switch (cmd) { | |
1377 | case CPU_PM_ENTER: | |
1378 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1379 | /* | |
1380 | * don't update kvm_arm_hardware_enabled here | |
1381 | * so that the hardware will be re-enabled | |
1382 | * when we resume. See below. | |
1383 | */ | |
1384 | cpu_hyp_reset(); | |
1385 | ||
1fcf7ce0 | 1386 | return NOTIFY_OK; |
58d6b15e | 1387 | case CPU_PM_ENTER_FAILED: |
67f69197 AT |
1388 | case CPU_PM_EXIT: |
1389 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1390 | /* The hardware was enabled before suspend. */ | |
1391 | cpu_hyp_reinit(); | |
1fcf7ce0 | 1392 | |
67f69197 AT |
1393 | return NOTIFY_OK; |
1394 | ||
1395 | default: | |
1396 | return NOTIFY_DONE; | |
1397 | } | |
1fcf7ce0 LP |
1398 | } |
1399 | ||
1400 | static struct notifier_block hyp_init_cpu_pm_nb = { | |
1401 | .notifier_call = hyp_init_cpu_pm_notifier, | |
1402 | }; | |
1403 | ||
1404 | static void __init hyp_cpu_pm_init(void) | |
1405 | { | |
1406 | cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); | |
1407 | } | |
06a71a24 SH |
1408 | static void __init hyp_cpu_pm_exit(void) |
1409 | { | |
1410 | cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); | |
1411 | } | |
1fcf7ce0 LP |
1412 | #else |
1413 | static inline void hyp_cpu_pm_init(void) | |
1414 | { | |
1415 | } | |
06a71a24 SH |
1416 | static inline void hyp_cpu_pm_exit(void) |
1417 | { | |
1418 | } | |
1fcf7ce0 LP |
1419 | #endif |
1420 | ||
1e947bad MZ |
1421 | static int init_common_resources(void) |
1422 | { | |
61349937 VM |
1423 | /* set size of VMID supported by CPU */ |
1424 | kvm_vmid_bits = kvm_get_vmid_bits(); | |
1425 | kvm_info("%d-bit VMID\n", kvm_vmid_bits); | |
1426 | ||
0f62f0e9 SP |
1427 | kvm_set_ipa_limit(); |
1428 | ||
1e947bad MZ |
1429 | return 0; |
1430 | } | |
1431 | ||
1432 | static int init_subsystems(void) | |
1433 | { | |
67f69197 | 1434 | int err = 0; |
1e947bad | 1435 | |
5f5560b1 | 1436 | /* |
67f69197 | 1437 | * Enable hardware so that subsystem initialisation can access EL2. |
5f5560b1 | 1438 | */ |
67f69197 | 1439 | on_each_cpu(_kvm_arch_hardware_enable, NULL, 1); |
5f5560b1 JM |
1440 | |
1441 | /* | |
1442 | * Register CPU lower-power notifier | |
1443 | */ | |
1444 | hyp_cpu_pm_init(); | |
1445 | ||
1e947bad MZ |
1446 | /* |
1447 | * Init HYP view of VGIC | |
1448 | */ | |
1449 | err = kvm_vgic_hyp_init(); | |
1450 | switch (err) { | |
1451 | case 0: | |
1452 | vgic_present = true; | |
1453 | break; | |
1454 | case -ENODEV: | |
1455 | case -ENXIO: | |
1456 | vgic_present = false; | |
67f69197 | 1457 | err = 0; |
1e947bad MZ |
1458 | break; |
1459 | default: | |
67f69197 | 1460 | goto out; |
1e947bad MZ |
1461 | } |
1462 | ||
1463 | /* | |
1464 | * Init HYP architected timer support | |
1465 | */ | |
f384dcfe | 1466 | err = kvm_timer_hyp_init(vgic_present); |
1e947bad | 1467 | if (err) |
67f69197 | 1468 | goto out; |
1e947bad MZ |
1469 | |
1470 | kvm_perf_init(); | |
1471 | kvm_coproc_table_init(); | |
1472 | ||
67f69197 AT |
1473 | out: |
1474 | on_each_cpu(_kvm_arch_hardware_disable, NULL, 1); | |
1475 | ||
1476 | return err; | |
1e947bad MZ |
1477 | } |
1478 | ||
1479 | static void teardown_hyp_mode(void) | |
1480 | { | |
1481 | int cpu; | |
1482 | ||
1e947bad MZ |
1483 | free_hyp_pgds(); |
1484 | for_each_possible_cpu(cpu) | |
1485 | free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); | |
06a71a24 | 1486 | hyp_cpu_pm_exit(); |
1e947bad MZ |
1487 | } |
1488 | ||
342cd0ab CD |
1489 | /** |
1490 | * Inits Hyp-mode on all online CPUs | |
1491 | */ | |
1492 | static int init_hyp_mode(void) | |
1493 | { | |
342cd0ab CD |
1494 | int cpu; |
1495 | int err = 0; | |
1496 | ||
1497 | /* | |
1498 | * Allocate Hyp PGD and setup Hyp identity mapping | |
1499 | */ | |
1500 | err = kvm_mmu_init(); | |
1501 | if (err) | |
1502 | goto out_err; | |
1503 | ||
342cd0ab CD |
1504 | /* |
1505 | * Allocate stack pages for Hypervisor-mode | |
1506 | */ | |
1507 | for_each_possible_cpu(cpu) { | |
1508 | unsigned long stack_page; | |
1509 | ||
1510 | stack_page = __get_free_page(GFP_KERNEL); | |
1511 | if (!stack_page) { | |
1512 | err = -ENOMEM; | |
1e947bad | 1513 | goto out_err; |
342cd0ab CD |
1514 | } |
1515 | ||
1516 | per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; | |
1517 | } | |
1518 | ||
342cd0ab CD |
1519 | /* |
1520 | * Map the Hyp-code called directly from the host | |
1521 | */ | |
588ab3f9 | 1522 | err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), |
59002705 | 1523 | kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC); |
342cd0ab CD |
1524 | if (err) { |
1525 | kvm_err("Cannot map world-switch code\n"); | |
1e947bad | 1526 | goto out_err; |
342cd0ab CD |
1527 | } |
1528 | ||
a0bf9776 | 1529 | err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), |
74a6b888 | 1530 | kvm_ksym_ref(__end_rodata), PAGE_HYP_RO); |
910917bb MZ |
1531 | if (err) { |
1532 | kvm_err("Cannot map rodata section\n"); | |
c8ea0395 MZ |
1533 | goto out_err; |
1534 | } | |
1535 | ||
1536 | err = create_hyp_mappings(kvm_ksym_ref(__bss_start), | |
1537 | kvm_ksym_ref(__bss_stop), PAGE_HYP_RO); | |
1538 | if (err) { | |
1539 | kvm_err("Cannot map bss section\n"); | |
1e947bad | 1540 | goto out_err; |
910917bb MZ |
1541 | } |
1542 | ||
6840bdd7 MZ |
1543 | err = kvm_map_vectors(); |
1544 | if (err) { | |
1545 | kvm_err("Cannot map vectors\n"); | |
1546 | goto out_err; | |
1547 | } | |
1548 | ||
342cd0ab CD |
1549 | /* |
1550 | * Map the Hyp stack pages | |
1551 | */ | |
1552 | for_each_possible_cpu(cpu) { | |
1553 | char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); | |
c8dddecd MZ |
1554 | err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE, |
1555 | PAGE_HYP); | |
342cd0ab CD |
1556 | |
1557 | if (err) { | |
1558 | kvm_err("Cannot map hyp stack\n"); | |
1e947bad | 1559 | goto out_err; |
342cd0ab CD |
1560 | } |
1561 | } | |
1562 | ||
342cd0ab | 1563 | for_each_possible_cpu(cpu) { |
3de50da6 | 1564 | kvm_cpu_context_t *cpu_ctxt; |
342cd0ab | 1565 | |
36989e7f | 1566 | cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu); |
c8dddecd | 1567 | err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP); |
342cd0ab CD |
1568 | |
1569 | if (err) { | |
3de50da6 | 1570 | kvm_err("Cannot map host CPU state: %d\n", err); |
1e947bad | 1571 | goto out_err; |
342cd0ab CD |
1572 | } |
1573 | } | |
1574 | ||
55e3748e MZ |
1575 | err = hyp_map_aux_data(); |
1576 | if (err) | |
1577 | kvm_err("Cannot map host auxilary data: %d\n", err); | |
1578 | ||
342cd0ab | 1579 | return 0; |
1e947bad | 1580 | |
342cd0ab | 1581 | out_err: |
1e947bad | 1582 | teardown_hyp_mode(); |
342cd0ab CD |
1583 | kvm_err("error initializing Hyp mode: %d\n", err); |
1584 | return err; | |
1585 | } | |
1586 | ||
d4e071ce AP |
1587 | static void check_kvm_target_cpu(void *ret) |
1588 | { | |
1589 | *(int *)ret = kvm_target_cpu(); | |
1590 | } | |
1591 | ||
4429fc64 AP |
1592 | struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) |
1593 | { | |
1594 | struct kvm_vcpu *vcpu; | |
1595 | int i; | |
1596 | ||
1597 | mpidr &= MPIDR_HWID_BITMASK; | |
1598 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
1599 | if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) | |
1600 | return vcpu; | |
1601 | } | |
1602 | return NULL; | |
1603 | } | |
1604 | ||
2412405b EA |
1605 | bool kvm_arch_has_irq_bypass(void) |
1606 | { | |
1607 | return true; | |
1608 | } | |
1609 | ||
1610 | int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, | |
1611 | struct irq_bypass_producer *prod) | |
1612 | { | |
1613 | struct kvm_kernel_irqfd *irqfd = | |
1614 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1615 | ||
196b1364 MZ |
1616 | return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq, |
1617 | &irqfd->irq_entry); | |
2412405b EA |
1618 | } |
1619 | void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, | |
1620 | struct irq_bypass_producer *prod) | |
1621 | { | |
1622 | struct kvm_kernel_irqfd *irqfd = | |
1623 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1624 | ||
196b1364 MZ |
1625 | kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq, |
1626 | &irqfd->irq_entry); | |
2412405b EA |
1627 | } |
1628 | ||
1629 | void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons) | |
1630 | { | |
1631 | struct kvm_kernel_irqfd *irqfd = | |
1632 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1633 | ||
1634 | kvm_arm_halt_guest(irqfd->kvm); | |
1635 | } | |
1636 | ||
1637 | void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons) | |
1638 | { | |
1639 | struct kvm_kernel_irqfd *irqfd = | |
1640 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
1641 | ||
1642 | kvm_arm_resume_guest(irqfd->kvm); | |
1643 | } | |
1644 | ||
342cd0ab CD |
1645 | /** |
1646 | * Initialize Hyp-mode and memory mappings on all CPUs. | |
1647 | */ | |
749cf76c CD |
1648 | int kvm_arch_init(void *opaque) |
1649 | { | |
342cd0ab | 1650 | int err; |
d4e071ce | 1651 | int ret, cpu; |
fe7d7b03 | 1652 | bool in_hyp_mode; |
342cd0ab CD |
1653 | |
1654 | if (!is_hyp_mode_available()) { | |
58d0d19a | 1655 | kvm_info("HYP mode not available\n"); |
342cd0ab CD |
1656 | return -ENODEV; |
1657 | } | |
1658 | ||
85acda3b DM |
1659 | if (!kvm_arch_check_sve_has_vhe()) { |
1660 | kvm_pr_unimpl("SVE system without VHE unsupported. Broken cpu?"); | |
1661 | return -ENODEV; | |
1662 | } | |
1663 | ||
d4e071ce AP |
1664 | for_each_online_cpu(cpu) { |
1665 | smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); | |
1666 | if (ret < 0) { | |
1667 | kvm_err("Error, CPU %d not supported!\n", cpu); | |
1668 | return -ENODEV; | |
1669 | } | |
342cd0ab CD |
1670 | } |
1671 | ||
1e947bad | 1672 | err = init_common_resources(); |
342cd0ab | 1673 | if (err) |
1e947bad | 1674 | return err; |
342cd0ab | 1675 | |
fe7d7b03 JT |
1676 | in_hyp_mode = is_kernel_in_hyp_mode(); |
1677 | ||
1678 | if (!in_hyp_mode) { | |
1e947bad | 1679 | err = init_hyp_mode(); |
fe7d7b03 JT |
1680 | if (err) |
1681 | goto out_err; | |
1682 | } | |
8146875d | 1683 | |
1e947bad MZ |
1684 | err = init_subsystems(); |
1685 | if (err) | |
1686 | goto out_hyp; | |
1fcf7ce0 | 1687 | |
fe7d7b03 JT |
1688 | if (in_hyp_mode) |
1689 | kvm_info("VHE mode initialized successfully\n"); | |
1690 | else | |
1691 | kvm_info("Hyp mode initialized successfully\n"); | |
1692 | ||
749cf76c | 1693 | return 0; |
1e947bad MZ |
1694 | |
1695 | out_hyp: | |
fe7d7b03 JT |
1696 | if (!in_hyp_mode) |
1697 | teardown_hyp_mode(); | |
342cd0ab CD |
1698 | out_err: |
1699 | return err; | |
749cf76c CD |
1700 | } |
1701 | ||
1702 | /* NOP: Compiling as a module not supported */ | |
1703 | void kvm_arch_exit(void) | |
1704 | { | |
210552c1 | 1705 | kvm_perf_teardown(); |
749cf76c CD |
1706 | } |
1707 | ||
1708 | static int arm_init(void) | |
1709 | { | |
1710 | int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1711 | return rc; | |
1712 | } | |
1713 | ||
1714 | module_init(arm_init); |