]>
Commit | Line | Data |
---|---|---|
749cf76c CD |
1 | /* |
2 | * Copyright (C) 2012 - Virtual Open Systems and Columbia University | |
3 | * Author: Christoffer Dall <c.dall@virtualopensystems.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License, version 2, as | |
7 | * published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. | |
17 | */ | |
18 | ||
d157f4a5 | 19 | #include <linux/cpu.h> |
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> | |
24 | #include <linux/module.h> | |
25 | #include <linux/vmalloc.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/mman.h> | |
28 | #include <linux/sched.h> | |
86ce8535 | 29 | #include <linux/kvm.h> |
749cf76c CD |
30 | #include <trace/events/kvm.h> |
31 | ||
32 | #define CREATE_TRACE_POINTS | |
33 | #include "trace.h" | |
34 | ||
749cf76c CD |
35 | #include <asm/uaccess.h> |
36 | #include <asm/ptrace.h> | |
37 | #include <asm/mman.h> | |
342cd0ab | 38 | #include <asm/tlbflush.h> |
5b3e5e5b | 39 | #include <asm/cacheflush.h> |
342cd0ab CD |
40 | #include <asm/virt.h> |
41 | #include <asm/kvm_arm.h> | |
42 | #include <asm/kvm_asm.h> | |
43 | #include <asm/kvm_mmu.h> | |
f7ed45be | 44 | #include <asm/kvm_emulate.h> |
5b3e5e5b | 45 | #include <asm/kvm_coproc.h> |
aa024c2f | 46 | #include <asm/kvm_psci.h> |
910917bb | 47 | #include <asm/sections.h> |
749cf76c CD |
48 | |
49 | #ifdef REQUIRES_VIRT | |
50 | __asm__(".arch_extension virt"); | |
51 | #endif | |
52 | ||
342cd0ab | 53 | static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); |
3de50da6 | 54 | static kvm_cpu_context_t __percpu *kvm_host_cpu_state; |
342cd0ab CD |
55 | static unsigned long hyp_default_vectors; |
56 | ||
1638a12d MZ |
57 | /* Per-CPU variable containing the currently running vcpu. */ |
58 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu); | |
59 | ||
f7ed45be CD |
60 | /* The VMID used in the VTTBR */ |
61 | static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1); | |
20475f78 VM |
62 | static u32 kvm_next_vmid; |
63 | static unsigned int kvm_vmid_bits __read_mostly; | |
f7ed45be | 64 | static DEFINE_SPINLOCK(kvm_vmid_lock); |
342cd0ab | 65 | |
c7da6fa4 PF |
66 | static bool vgic_present; |
67 | ||
1638a12d MZ |
68 | static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu) |
69 | { | |
70 | BUG_ON(preemptible()); | |
1436c1aa | 71 | __this_cpu_write(kvm_arm_running_vcpu, vcpu); |
1638a12d MZ |
72 | } |
73 | ||
74 | /** | |
75 | * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU. | |
76 | * Must be called from non-preemptible context | |
77 | */ | |
78 | struct kvm_vcpu *kvm_arm_get_running_vcpu(void) | |
79 | { | |
80 | BUG_ON(preemptible()); | |
1436c1aa | 81 | return __this_cpu_read(kvm_arm_running_vcpu); |
1638a12d MZ |
82 | } |
83 | ||
84 | /** | |
85 | * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
86 | */ | |
4000be42 | 87 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) |
1638a12d MZ |
88 | { |
89 | return &kvm_arm_running_vcpu; | |
90 | } | |
91 | ||
13a34e06 | 92 | int kvm_arch_hardware_enable(void) |
749cf76c CD |
93 | { |
94 | return 0; | |
95 | } | |
96 | ||
97 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) | |
98 | { | |
99 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; | |
100 | } | |
101 | ||
749cf76c CD |
102 | int kvm_arch_hardware_setup(void) |
103 | { | |
104 | return 0; | |
105 | } | |
106 | ||
749cf76c CD |
107 | void kvm_arch_check_processor_compat(void *rtn) |
108 | { | |
109 | *(int *)rtn = 0; | |
110 | } | |
111 | ||
749cf76c | 112 | |
d5d8184d CD |
113 | /** |
114 | * kvm_arch_init_vm - initializes a VM data structure | |
115 | * @kvm: pointer to the KVM struct | |
116 | */ | |
749cf76c CD |
117 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
118 | { | |
d5d8184d CD |
119 | int ret = 0; |
120 | ||
749cf76c CD |
121 | if (type) |
122 | return -EINVAL; | |
123 | ||
d5d8184d CD |
124 | ret = kvm_alloc_stage2_pgd(kvm); |
125 | if (ret) | |
126 | goto out_fail_alloc; | |
127 | ||
128 | ret = create_hyp_mappings(kvm, kvm + 1); | |
129 | if (ret) | |
130 | goto out_free_stage2_pgd; | |
131 | ||
6c3d63c9 | 132 | kvm_vgic_early_init(kvm); |
a1a64387 CD |
133 | kvm_timer_init(kvm); |
134 | ||
d5d8184d CD |
135 | /* Mark the initial VMID generation invalid */ |
136 | kvm->arch.vmid_gen = 0; | |
137 | ||
3caa2d8c | 138 | /* The maximum number of VCPUs is limited by the host's GIC model */ |
c7da6fa4 PF |
139 | kvm->arch.max_vcpus = vgic_present ? |
140 | kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; | |
3caa2d8c | 141 | |
d5d8184d CD |
142 | return ret; |
143 | out_free_stage2_pgd: | |
144 | kvm_free_stage2_pgd(kvm); | |
145 | out_fail_alloc: | |
146 | return ret; | |
749cf76c CD |
147 | } |
148 | ||
149 | int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) | |
150 | { | |
151 | return VM_FAULT_SIGBUS; | |
152 | } | |
153 | ||
749cf76c | 154 | |
d5d8184d CD |
155 | /** |
156 | * kvm_arch_destroy_vm - destroy the VM data structure | |
157 | * @kvm: pointer to the KVM struct | |
158 | */ | |
749cf76c CD |
159 | void kvm_arch_destroy_vm(struct kvm *kvm) |
160 | { | |
161 | int i; | |
162 | ||
d5d8184d CD |
163 | kvm_free_stage2_pgd(kvm); |
164 | ||
749cf76c CD |
165 | for (i = 0; i < KVM_MAX_VCPUS; ++i) { |
166 | if (kvm->vcpus[i]) { | |
167 | kvm_arch_vcpu_free(kvm->vcpus[i]); | |
168 | kvm->vcpus[i] = NULL; | |
169 | } | |
170 | } | |
c1bfb577 MZ |
171 | |
172 | kvm_vgic_destroy(kvm); | |
749cf76c CD |
173 | } |
174 | ||
784aa3d7 | 175 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
749cf76c CD |
176 | { |
177 | int r; | |
178 | switch (ext) { | |
1a89dd91 | 179 | case KVM_CAP_IRQCHIP: |
c7da6fa4 PF |
180 | r = vgic_present; |
181 | break; | |
d44758c0 | 182 | case KVM_CAP_IOEVENTFD: |
7330672b | 183 | case KVM_CAP_DEVICE_CTRL: |
749cf76c CD |
184 | case KVM_CAP_USER_MEMORY: |
185 | case KVM_CAP_SYNC_MMU: | |
186 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
187 | case KVM_CAP_ONE_REG: | |
aa024c2f | 188 | case KVM_CAP_ARM_PSCI: |
4447a208 | 189 | case KVM_CAP_ARM_PSCI_0_2: |
98047888 | 190 | case KVM_CAP_READONLY_MEM: |
ecccf0cc | 191 | case KVM_CAP_MP_STATE: |
749cf76c CD |
192 | r = 1; |
193 | break; | |
194 | case KVM_CAP_COALESCED_MMIO: | |
195 | r = KVM_COALESCED_MMIO_PAGE_OFFSET; | |
196 | break; | |
3401d546 CD |
197 | case KVM_CAP_ARM_SET_DEVICE_ADDR: |
198 | r = 1; | |
ca46e10f | 199 | break; |
749cf76c CD |
200 | case KVM_CAP_NR_VCPUS: |
201 | r = num_online_cpus(); | |
202 | break; | |
203 | case KVM_CAP_MAX_VCPUS: | |
204 | r = KVM_MAX_VCPUS; | |
205 | break; | |
206 | default: | |
17b1e31f | 207 | r = kvm_arch_dev_ioctl_check_extension(ext); |
749cf76c CD |
208 | break; |
209 | } | |
210 | return r; | |
211 | } | |
212 | ||
213 | long kvm_arch_dev_ioctl(struct file *filp, | |
214 | unsigned int ioctl, unsigned long arg) | |
215 | { | |
216 | return -EINVAL; | |
217 | } | |
218 | ||
749cf76c CD |
219 | |
220 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) | |
221 | { | |
222 | int err; | |
223 | struct kvm_vcpu *vcpu; | |
224 | ||
716139df CD |
225 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) { |
226 | err = -EBUSY; | |
227 | goto out; | |
228 | } | |
229 | ||
3caa2d8c AP |
230 | if (id >= kvm->arch.max_vcpus) { |
231 | err = -EINVAL; | |
232 | goto out; | |
233 | } | |
234 | ||
749cf76c CD |
235 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
236 | if (!vcpu) { | |
237 | err = -ENOMEM; | |
238 | goto out; | |
239 | } | |
240 | ||
241 | err = kvm_vcpu_init(vcpu, kvm, id); | |
242 | if (err) | |
243 | goto free_vcpu; | |
244 | ||
d5d8184d CD |
245 | err = create_hyp_mappings(vcpu, vcpu + 1); |
246 | if (err) | |
247 | goto vcpu_uninit; | |
248 | ||
749cf76c | 249 | return vcpu; |
d5d8184d CD |
250 | vcpu_uninit: |
251 | kvm_vcpu_uninit(vcpu); | |
749cf76c CD |
252 | free_vcpu: |
253 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
254 | out: | |
255 | return ERR_PTR(err); | |
256 | } | |
257 | ||
31928aa5 | 258 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
749cf76c | 259 | { |
6c3d63c9 | 260 | kvm_vgic_vcpu_early_init(vcpu); |
749cf76c CD |
261 | } |
262 | ||
263 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
264 | { | |
d5d8184d | 265 | kvm_mmu_free_memory_caches(vcpu); |
967f8427 | 266 | kvm_timer_vcpu_terminate(vcpu); |
c1bfb577 | 267 | kvm_vgic_vcpu_destroy(vcpu); |
d5d8184d | 268 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
749cf76c CD |
269 | } |
270 | ||
271 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) | |
272 | { | |
273 | kvm_arch_vcpu_free(vcpu); | |
274 | } | |
275 | ||
276 | int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) | |
277 | { | |
1a748478 | 278 | return kvm_timer_should_fire(vcpu); |
749cf76c CD |
279 | } |
280 | ||
d35268da CD |
281 | void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) |
282 | { | |
283 | kvm_timer_schedule(vcpu); | |
284 | } | |
285 | ||
286 | void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
287 | { | |
288 | kvm_timer_unschedule(vcpu); | |
289 | } | |
290 | ||
749cf76c CD |
291 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
292 | { | |
f7ed45be CD |
293 | /* Force users to call KVM_ARM_VCPU_INIT */ |
294 | vcpu->arch.target = -1; | |
f7fa034d | 295 | bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); |
1a89dd91 | 296 | |
967f8427 MZ |
297 | /* Set up the timer */ |
298 | kvm_timer_vcpu_init(vcpu); | |
299 | ||
84e690bf AB |
300 | kvm_arm_reset_debug_ptr(vcpu); |
301 | ||
749cf76c CD |
302 | return 0; |
303 | } | |
304 | ||
749cf76c CD |
305 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
306 | { | |
86ce8535 | 307 | vcpu->cpu = cpu; |
3de50da6 | 308 | vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state); |
5b3e5e5b | 309 | |
1638a12d | 310 | kvm_arm_set_running_vcpu(vcpu); |
749cf76c CD |
311 | } |
312 | ||
313 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
314 | { | |
e9b152cb CD |
315 | /* |
316 | * The arch-generic KVM code expects the cpu field of a vcpu to be -1 | |
317 | * if the vcpu is no longer assigned to a cpu. This is used for the | |
318 | * optimized make_all_cpus_request path. | |
319 | */ | |
320 | vcpu->cpu = -1; | |
321 | ||
1638a12d | 322 | kvm_arm_set_running_vcpu(NULL); |
749cf76c CD |
323 | } |
324 | ||
749cf76c CD |
325 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
326 | struct kvm_mp_state *mp_state) | |
327 | { | |
3781528e | 328 | if (vcpu->arch.power_off) |
ecccf0cc AB |
329 | mp_state->mp_state = KVM_MP_STATE_STOPPED; |
330 | else | |
331 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
332 | ||
333 | return 0; | |
749cf76c CD |
334 | } |
335 | ||
336 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
337 | struct kvm_mp_state *mp_state) | |
338 | { | |
ecccf0cc AB |
339 | switch (mp_state->mp_state) { |
340 | case KVM_MP_STATE_RUNNABLE: | |
3781528e | 341 | vcpu->arch.power_off = false; |
ecccf0cc AB |
342 | break; |
343 | case KVM_MP_STATE_STOPPED: | |
3781528e | 344 | vcpu->arch.power_off = true; |
ecccf0cc AB |
345 | break; |
346 | default: | |
347 | return -EINVAL; | |
348 | } | |
349 | ||
350 | return 0; | |
749cf76c CD |
351 | } |
352 | ||
5b3e5e5b CD |
353 | /** |
354 | * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled | |
355 | * @v: The VCPU pointer | |
356 | * | |
357 | * If the guest CPU is not waiting for interrupts or an interrupt line is | |
358 | * asserted, the CPU is by definition runnable. | |
359 | */ | |
749cf76c CD |
360 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) |
361 | { | |
4f5f1dc0 | 362 | return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v)) |
3b92830a | 363 | && !v->arch.power_off && !v->arch.pause); |
749cf76c CD |
364 | } |
365 | ||
f7ed45be CD |
366 | /* Just ensure a guest exit from a particular CPU */ |
367 | static void exit_vm_noop(void *info) | |
368 | { | |
369 | } | |
370 | ||
371 | void force_vm_exit(const cpumask_t *mask) | |
372 | { | |
373 | smp_call_function_many(mask, exit_vm_noop, NULL, true); | |
374 | } | |
375 | ||
376 | /** | |
377 | * need_new_vmid_gen - check that the VMID is still valid | |
378 | * @kvm: The VM's VMID to checkt | |
379 | * | |
380 | * return true if there is a new generation of VMIDs being used | |
381 | * | |
382 | * The hardware supports only 256 values with the value zero reserved for the | |
383 | * host, so we check if an assigned value belongs to a previous generation, | |
384 | * which which requires us to assign a new value. If we're the first to use a | |
385 | * VMID for the new generation, we must flush necessary caches and TLBs on all | |
386 | * CPUs. | |
387 | */ | |
388 | static bool need_new_vmid_gen(struct kvm *kvm) | |
389 | { | |
390 | return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen)); | |
391 | } | |
392 | ||
393 | /** | |
394 | * update_vttbr - Update the VTTBR with a valid VMID before the guest runs | |
395 | * @kvm The guest that we are about to run | |
396 | * | |
397 | * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the | |
398 | * VM has a valid VMID, otherwise assigns a new one and flushes corresponding | |
399 | * caches and TLBs. | |
400 | */ | |
401 | static void update_vttbr(struct kvm *kvm) | |
402 | { | |
403 | phys_addr_t pgd_phys; | |
404 | u64 vmid; | |
405 | ||
406 | if (!need_new_vmid_gen(kvm)) | |
407 | return; | |
408 | ||
409 | spin_lock(&kvm_vmid_lock); | |
410 | ||
411 | /* | |
412 | * We need to re-check the vmid_gen here to ensure that if another vcpu | |
413 | * already allocated a valid vmid for this vm, then this vcpu should | |
414 | * use the same vmid. | |
415 | */ | |
416 | if (!need_new_vmid_gen(kvm)) { | |
417 | spin_unlock(&kvm_vmid_lock); | |
418 | return; | |
419 | } | |
420 | ||
421 | /* First user of a new VMID generation? */ | |
422 | if (unlikely(kvm_next_vmid == 0)) { | |
423 | atomic64_inc(&kvm_vmid_gen); | |
424 | kvm_next_vmid = 1; | |
425 | ||
426 | /* | |
427 | * On SMP we know no other CPUs can use this CPU's or each | |
428 | * other's VMID after force_vm_exit returns since the | |
429 | * kvm_vmid_lock blocks them from reentry to the guest. | |
430 | */ | |
431 | force_vm_exit(cpu_all_mask); | |
432 | /* | |
433 | * Now broadcast TLB + ICACHE invalidation over the inner | |
434 | * shareable domain to make sure all data structures are | |
435 | * clean. | |
436 | */ | |
437 | kvm_call_hyp(__kvm_flush_vm_context); | |
438 | } | |
439 | ||
440 | kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen); | |
441 | kvm->arch.vmid = kvm_next_vmid; | |
442 | kvm_next_vmid++; | |
20475f78 | 443 | kvm_next_vmid &= (1 << kvm_vmid_bits) - 1; |
f7ed45be CD |
444 | |
445 | /* update vttbr to be used with the new vmid */ | |
38f791a4 | 446 | pgd_phys = virt_to_phys(kvm_get_hwpgd(kvm)); |
dbff124e | 447 | BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK); |
20475f78 | 448 | vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits); |
dbff124e | 449 | kvm->arch.vttbr = pgd_phys | vmid; |
f7ed45be CD |
450 | |
451 | spin_unlock(&kvm_vmid_lock); | |
452 | } | |
453 | ||
f7ed45be CD |
454 | static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) |
455 | { | |
05971120 | 456 | struct kvm *kvm = vcpu->kvm; |
e1ba0207 CD |
457 | int ret; |
458 | ||
f7ed45be CD |
459 | if (likely(vcpu->arch.has_run_once)) |
460 | return 0; | |
461 | ||
462 | vcpu->arch.has_run_once = true; | |
aa024c2f | 463 | |
01ac5e34 | 464 | /* |
6d3cfbe2 PM |
465 | * Map the VGIC hardware resources before running a vcpu the first |
466 | * time on this VM. | |
01ac5e34 | 467 | */ |
c2f58514 | 468 | if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) { |
05971120 | 469 | ret = kvm_vgic_map_resources(kvm); |
01ac5e34 MZ |
470 | if (ret) |
471 | return ret; | |
472 | } | |
473 | ||
05971120 CD |
474 | /* |
475 | * Enable the arch timers only if we have an in-kernel VGIC | |
476 | * and it has been properly initialized, since we cannot handle | |
477 | * interrupts from the virtual timer with a userspace gic. | |
478 | */ | |
479 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) | |
480 | kvm_timer_enable(kvm); | |
481 | ||
f7ed45be CD |
482 | return 0; |
483 | } | |
484 | ||
c1426e4c EA |
485 | bool kvm_arch_intc_initialized(struct kvm *kvm) |
486 | { | |
487 | return vgic_initialized(kvm); | |
488 | } | |
489 | ||
3b92830a EA |
490 | static void kvm_arm_halt_guest(struct kvm *kvm) __maybe_unused; |
491 | static void kvm_arm_resume_guest(struct kvm *kvm) __maybe_unused; | |
492 | ||
493 | static void kvm_arm_halt_guest(struct kvm *kvm) | |
494 | { | |
495 | int i; | |
496 | struct kvm_vcpu *vcpu; | |
497 | ||
498 | kvm_for_each_vcpu(i, vcpu, kvm) | |
499 | vcpu->arch.pause = true; | |
500 | force_vm_exit(cpu_all_mask); | |
501 | } | |
502 | ||
503 | static void kvm_arm_resume_guest(struct kvm *kvm) | |
504 | { | |
505 | int i; | |
506 | struct kvm_vcpu *vcpu; | |
507 | ||
508 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
509 | wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu); | |
510 | ||
511 | vcpu->arch.pause = false; | |
512 | wake_up_interruptible(wq); | |
513 | } | |
514 | } | |
515 | ||
3781528e | 516 | static void vcpu_sleep(struct kvm_vcpu *vcpu) |
aa024c2f MZ |
517 | { |
518 | wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu); | |
519 | ||
3b92830a EA |
520 | wait_event_interruptible(*wq, ((!vcpu->arch.power_off) && |
521 | (!vcpu->arch.pause))); | |
aa024c2f MZ |
522 | } |
523 | ||
e8180dca AP |
524 | static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) |
525 | { | |
526 | return vcpu->arch.target >= 0; | |
527 | } | |
528 | ||
f7ed45be CD |
529 | /** |
530 | * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code | |
531 | * @vcpu: The VCPU pointer | |
532 | * @run: The kvm_run structure pointer used for userspace state exchange | |
533 | * | |
534 | * This function is called through the VCPU_RUN ioctl called from user space. It | |
535 | * will execute VM code in a loop until the time slice for the process is used | |
536 | * or some emulation is needed from user space in which case the function will | |
537 | * return with return value 0 and with the kvm_run structure filled in with the | |
538 | * required data for the requested emulation. | |
539 | */ | |
749cf76c CD |
540 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) |
541 | { | |
f7ed45be CD |
542 | int ret; |
543 | sigset_t sigsaved; | |
544 | ||
e8180dca | 545 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
f7ed45be CD |
546 | return -ENOEXEC; |
547 | ||
548 | ret = kvm_vcpu_first_run_init(vcpu); | |
549 | if (ret) | |
550 | return ret; | |
551 | ||
45e96ea6 CD |
552 | if (run->exit_reason == KVM_EXIT_MMIO) { |
553 | ret = kvm_handle_mmio_return(vcpu, vcpu->run); | |
554 | if (ret) | |
555 | return ret; | |
556 | } | |
557 | ||
f7ed45be CD |
558 | if (vcpu->sigset_active) |
559 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
560 | ||
561 | ret = 1; | |
562 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
563 | while (ret > 0) { | |
564 | /* | |
565 | * Check conditions before entering the guest | |
566 | */ | |
567 | cond_resched(); | |
568 | ||
569 | update_vttbr(vcpu->kvm); | |
570 | ||
3b92830a | 571 | if (vcpu->arch.power_off || vcpu->arch.pause) |
3781528e | 572 | vcpu_sleep(vcpu); |
aa024c2f | 573 | |
abdf5843 MZ |
574 | /* |
575 | * Preparing the interrupts to be injected also | |
576 | * involves poking the GIC, which must be done in a | |
577 | * non-preemptible context. | |
578 | */ | |
1b3d546d | 579 | preempt_disable(); |
7e16aa81 | 580 | kvm_timer_flush_hwstate(vcpu); |
abdf5843 MZ |
581 | kvm_vgic_flush_hwstate(vcpu); |
582 | ||
f7ed45be CD |
583 | local_irq_disable(); |
584 | ||
585 | /* | |
586 | * Re-check atomic conditions | |
587 | */ | |
588 | if (signal_pending(current)) { | |
589 | ret = -EINTR; | |
590 | run->exit_reason = KVM_EXIT_INTR; | |
591 | } | |
592 | ||
101d3da0 | 593 | if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) || |
3b92830a | 594 | vcpu->arch.power_off || vcpu->arch.pause) { |
f7ed45be | 595 | local_irq_enable(); |
4b4b4512 | 596 | kvm_timer_sync_hwstate(vcpu); |
1a89dd91 | 597 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 | 598 | preempt_enable(); |
f7ed45be CD |
599 | continue; |
600 | } | |
601 | ||
56c7f5e7 AB |
602 | kvm_arm_setup_debug(vcpu); |
603 | ||
f7ed45be CD |
604 | /************************************************************** |
605 | * Enter the guest | |
606 | */ | |
607 | trace_kvm_entry(*vcpu_pc(vcpu)); | |
ccf73aaf | 608 | __kvm_guest_enter(); |
f7ed45be CD |
609 | vcpu->mode = IN_GUEST_MODE; |
610 | ||
611 | ret = kvm_call_hyp(__kvm_vcpu_run, vcpu); | |
612 | ||
613 | vcpu->mode = OUTSIDE_GUEST_MODE; | |
b19e6892 | 614 | vcpu->stat.exits++; |
1b3d546d CD |
615 | /* |
616 | * Back from guest | |
617 | *************************************************************/ | |
618 | ||
56c7f5e7 AB |
619 | kvm_arm_clear_debug(vcpu); |
620 | ||
f7ed45be CD |
621 | /* |
622 | * We may have taken a host interrupt in HYP mode (ie | |
623 | * while executing the guest). This interrupt is still | |
624 | * pending, as we haven't serviced it yet! | |
625 | * | |
626 | * We're now back in SVC mode, with interrupts | |
627 | * disabled. Enabling the interrupts now will have | |
628 | * the effect of taking the interrupt again, in SVC | |
629 | * mode this time. | |
630 | */ | |
631 | local_irq_enable(); | |
632 | ||
633 | /* | |
1b3d546d CD |
634 | * We do local_irq_enable() before calling kvm_guest_exit() so |
635 | * that if a timer interrupt hits while running the guest we | |
636 | * account that tick as being spent in the guest. We enable | |
637 | * preemption after calling kvm_guest_exit() so that if we get | |
638 | * preempted we make sure ticks after that is not counted as | |
639 | * guest time. | |
640 | */ | |
641 | kvm_guest_exit(); | |
b5905dc1 | 642 | trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); |
1b3d546d | 643 | |
4b4b4512 CD |
644 | /* |
645 | * We must sync the timer state before the vgic state so that | |
646 | * the vgic can properly sample the updated state of the | |
647 | * interrupt line. | |
648 | */ | |
649 | kvm_timer_sync_hwstate(vcpu); | |
650 | ||
1a89dd91 | 651 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 MZ |
652 | |
653 | preempt_enable(); | |
654 | ||
f7ed45be CD |
655 | ret = handle_exit(vcpu, run, ret); |
656 | } | |
657 | ||
658 | if (vcpu->sigset_active) | |
659 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
660 | return ret; | |
749cf76c CD |
661 | } |
662 | ||
86ce8535 CD |
663 | static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) |
664 | { | |
665 | int bit_index; | |
666 | bool set; | |
667 | unsigned long *ptr; | |
668 | ||
669 | if (number == KVM_ARM_IRQ_CPU_IRQ) | |
670 | bit_index = __ffs(HCR_VI); | |
671 | else /* KVM_ARM_IRQ_CPU_FIQ */ | |
672 | bit_index = __ffs(HCR_VF); | |
673 | ||
674 | ptr = (unsigned long *)&vcpu->arch.irq_lines; | |
675 | if (level) | |
676 | set = test_and_set_bit(bit_index, ptr); | |
677 | else | |
678 | set = test_and_clear_bit(bit_index, ptr); | |
679 | ||
680 | /* | |
681 | * If we didn't change anything, no need to wake up or kick other CPUs | |
682 | */ | |
683 | if (set == level) | |
684 | return 0; | |
685 | ||
686 | /* | |
687 | * The vcpu irq_lines field was updated, wake up sleeping VCPUs and | |
688 | * trigger a world-switch round on the running physical CPU to set the | |
689 | * virtual IRQ/FIQ fields in the HCR appropriately. | |
690 | */ | |
691 | kvm_vcpu_kick(vcpu); | |
692 | ||
693 | return 0; | |
694 | } | |
695 | ||
79558f11 AG |
696 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, |
697 | bool line_status) | |
86ce8535 CD |
698 | { |
699 | u32 irq = irq_level->irq; | |
700 | unsigned int irq_type, vcpu_idx, irq_num; | |
701 | int nrcpus = atomic_read(&kvm->online_vcpus); | |
702 | struct kvm_vcpu *vcpu = NULL; | |
703 | bool level = irq_level->level; | |
704 | ||
705 | irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; | |
706 | vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; | |
707 | irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; | |
708 | ||
709 | trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); | |
710 | ||
5863c2ce MZ |
711 | switch (irq_type) { |
712 | case KVM_ARM_IRQ_TYPE_CPU: | |
713 | if (irqchip_in_kernel(kvm)) | |
714 | return -ENXIO; | |
86ce8535 | 715 | |
5863c2ce MZ |
716 | if (vcpu_idx >= nrcpus) |
717 | return -EINVAL; | |
86ce8535 | 718 | |
5863c2ce MZ |
719 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); |
720 | if (!vcpu) | |
721 | return -EINVAL; | |
86ce8535 | 722 | |
5863c2ce MZ |
723 | if (irq_num > KVM_ARM_IRQ_CPU_FIQ) |
724 | return -EINVAL; | |
725 | ||
726 | return vcpu_interrupt_line(vcpu, irq_num, level); | |
727 | case KVM_ARM_IRQ_TYPE_PPI: | |
728 | if (!irqchip_in_kernel(kvm)) | |
729 | return -ENXIO; | |
730 | ||
731 | if (vcpu_idx >= nrcpus) | |
732 | return -EINVAL; | |
733 | ||
734 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); | |
735 | if (!vcpu) | |
736 | return -EINVAL; | |
737 | ||
738 | if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) | |
739 | return -EINVAL; | |
86ce8535 | 740 | |
5863c2ce MZ |
741 | return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level); |
742 | case KVM_ARM_IRQ_TYPE_SPI: | |
743 | if (!irqchip_in_kernel(kvm)) | |
744 | return -ENXIO; | |
745 | ||
fd1d0ddf | 746 | if (irq_num < VGIC_NR_PRIVATE_IRQS) |
5863c2ce MZ |
747 | return -EINVAL; |
748 | ||
749 | return kvm_vgic_inject_irq(kvm, 0, irq_num, level); | |
750 | } | |
751 | ||
752 | return -EINVAL; | |
86ce8535 CD |
753 | } |
754 | ||
f7fa034d CD |
755 | static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, |
756 | const struct kvm_vcpu_init *init) | |
757 | { | |
758 | unsigned int i; | |
759 | int phys_target = kvm_target_cpu(); | |
760 | ||
761 | if (init->target != phys_target) | |
762 | return -EINVAL; | |
763 | ||
764 | /* | |
765 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
766 | * use the same target. | |
767 | */ | |
768 | if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) | |
769 | return -EINVAL; | |
770 | ||
771 | /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ | |
772 | for (i = 0; i < sizeof(init->features) * 8; i++) { | |
773 | bool set = (init->features[i / 32] & (1 << (i % 32))); | |
774 | ||
775 | if (set && i >= KVM_VCPU_MAX_FEATURES) | |
776 | return -ENOENT; | |
777 | ||
778 | /* | |
779 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
780 | * use the same feature set. | |
781 | */ | |
782 | if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && | |
783 | test_bit(i, vcpu->arch.features) != set) | |
784 | return -EINVAL; | |
785 | ||
786 | if (set) | |
787 | set_bit(i, vcpu->arch.features); | |
788 | } | |
789 | ||
790 | vcpu->arch.target = phys_target; | |
791 | ||
792 | /* Now we know what it is, we can reset it. */ | |
793 | return kvm_reset_vcpu(vcpu); | |
794 | } | |
795 | ||
796 | ||
478a8237 CD |
797 | static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, |
798 | struct kvm_vcpu_init *init) | |
799 | { | |
800 | int ret; | |
801 | ||
802 | ret = kvm_vcpu_set_target(vcpu, init); | |
803 | if (ret) | |
804 | return ret; | |
805 | ||
957db105 CD |
806 | /* |
807 | * Ensure a rebooted VM will fault in RAM pages and detect if the | |
808 | * guest MMU is turned off and flush the caches as needed. | |
809 | */ | |
810 | if (vcpu->arch.has_run_once) | |
811 | stage2_unmap_vm(vcpu->kvm); | |
812 | ||
b856a591 CD |
813 | vcpu_reset_hcr(vcpu); |
814 | ||
478a8237 | 815 | /* |
3781528e | 816 | * Handle the "start in power-off" case. |
478a8237 | 817 | */ |
03f1d4c1 | 818 | if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) |
3781528e | 819 | vcpu->arch.power_off = true; |
3ad8b3de | 820 | else |
3781528e | 821 | vcpu->arch.power_off = false; |
478a8237 CD |
822 | |
823 | return 0; | |
824 | } | |
825 | ||
749cf76c CD |
826 | long kvm_arch_vcpu_ioctl(struct file *filp, |
827 | unsigned int ioctl, unsigned long arg) | |
828 | { | |
829 | struct kvm_vcpu *vcpu = filp->private_data; | |
830 | void __user *argp = (void __user *)arg; | |
831 | ||
832 | switch (ioctl) { | |
833 | case KVM_ARM_VCPU_INIT: { | |
834 | struct kvm_vcpu_init init; | |
835 | ||
836 | if (copy_from_user(&init, argp, sizeof(init))) | |
837 | return -EFAULT; | |
838 | ||
478a8237 | 839 | return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); |
749cf76c CD |
840 | } |
841 | case KVM_SET_ONE_REG: | |
842 | case KVM_GET_ONE_REG: { | |
843 | struct kvm_one_reg reg; | |
e8180dca AP |
844 | |
845 | if (unlikely(!kvm_vcpu_initialized(vcpu))) | |
846 | return -ENOEXEC; | |
847 | ||
749cf76c CD |
848 | if (copy_from_user(®, argp, sizeof(reg))) |
849 | return -EFAULT; | |
850 | if (ioctl == KVM_SET_ONE_REG) | |
851 | return kvm_arm_set_reg(vcpu, ®); | |
852 | else | |
853 | return kvm_arm_get_reg(vcpu, ®); | |
854 | } | |
855 | case KVM_GET_REG_LIST: { | |
856 | struct kvm_reg_list __user *user_list = argp; | |
857 | struct kvm_reg_list reg_list; | |
858 | unsigned n; | |
859 | ||
e8180dca AP |
860 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
861 | return -ENOEXEC; | |
862 | ||
749cf76c CD |
863 | if (copy_from_user(®_list, user_list, sizeof(reg_list))) |
864 | return -EFAULT; | |
865 | n = reg_list.n; | |
866 | reg_list.n = kvm_arm_num_regs(vcpu); | |
867 | if (copy_to_user(user_list, ®_list, sizeof(reg_list))) | |
868 | return -EFAULT; | |
869 | if (n < reg_list.n) | |
870 | return -E2BIG; | |
871 | return kvm_arm_copy_reg_indices(vcpu, user_list->reg); | |
872 | } | |
873 | default: | |
874 | return -EINVAL; | |
875 | } | |
876 | } | |
877 | ||
53c810c3 MS |
878 | /** |
879 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
880 | * @kvm: kvm instance | |
881 | * @log: slot id and address to which we copy the log | |
882 | * | |
883 | * Steps 1-4 below provide general overview of dirty page logging. See | |
884 | * kvm_get_dirty_log_protect() function description for additional details. | |
885 | * | |
886 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
887 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
888 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
889 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
890 | * writes will be marked dirty for next log read. | |
891 | * | |
892 | * 1. Take a snapshot of the bit and clear it if needed. | |
893 | * 2. Write protect the corresponding page. | |
894 | * 3. Copy the snapshot to the userspace. | |
895 | * 4. Flush TLB's if needed. | |
896 | */ | |
749cf76c CD |
897 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
898 | { | |
53c810c3 MS |
899 | bool is_dirty = false; |
900 | int r; | |
901 | ||
902 | mutex_lock(&kvm->slots_lock); | |
903 | ||
904 | r = kvm_get_dirty_log_protect(kvm, log, &is_dirty); | |
905 | ||
906 | if (is_dirty) | |
907 | kvm_flush_remote_tlbs(kvm); | |
908 | ||
909 | mutex_unlock(&kvm->slots_lock); | |
910 | return r; | |
749cf76c CD |
911 | } |
912 | ||
3401d546 CD |
913 | static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, |
914 | struct kvm_arm_device_addr *dev_addr) | |
915 | { | |
330690cd CD |
916 | unsigned long dev_id, type; |
917 | ||
918 | dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> | |
919 | KVM_ARM_DEVICE_ID_SHIFT; | |
920 | type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> | |
921 | KVM_ARM_DEVICE_TYPE_SHIFT; | |
922 | ||
923 | switch (dev_id) { | |
924 | case KVM_ARM_DEVICE_VGIC_V2: | |
c7da6fa4 PF |
925 | if (!vgic_present) |
926 | return -ENXIO; | |
ce01e4e8 | 927 | return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); |
330690cd CD |
928 | default: |
929 | return -ENODEV; | |
930 | } | |
3401d546 CD |
931 | } |
932 | ||
749cf76c CD |
933 | long kvm_arch_vm_ioctl(struct file *filp, |
934 | unsigned int ioctl, unsigned long arg) | |
935 | { | |
3401d546 CD |
936 | struct kvm *kvm = filp->private_data; |
937 | void __user *argp = (void __user *)arg; | |
938 | ||
939 | switch (ioctl) { | |
5863c2ce | 940 | case KVM_CREATE_IRQCHIP: { |
c7da6fa4 PF |
941 | if (!vgic_present) |
942 | return -ENXIO; | |
69ff5c61 | 943 | return kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); |
5863c2ce | 944 | } |
3401d546 CD |
945 | case KVM_ARM_SET_DEVICE_ADDR: { |
946 | struct kvm_arm_device_addr dev_addr; | |
947 | ||
948 | if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) | |
949 | return -EFAULT; | |
950 | return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); | |
951 | } | |
42c4e0c7 AP |
952 | case KVM_ARM_PREFERRED_TARGET: { |
953 | int err; | |
954 | struct kvm_vcpu_init init; | |
955 | ||
956 | err = kvm_vcpu_preferred_target(&init); | |
957 | if (err) | |
958 | return err; | |
959 | ||
960 | if (copy_to_user(argp, &init, sizeof(init))) | |
961 | return -EFAULT; | |
962 | ||
963 | return 0; | |
964 | } | |
3401d546 CD |
965 | default: |
966 | return -EINVAL; | |
967 | } | |
749cf76c CD |
968 | } |
969 | ||
d157f4a5 | 970 | static void cpu_init_hyp_mode(void *dummy) |
342cd0ab | 971 | { |
dac288f7 MZ |
972 | phys_addr_t boot_pgd_ptr; |
973 | phys_addr_t pgd_ptr; | |
342cd0ab CD |
974 | unsigned long hyp_stack_ptr; |
975 | unsigned long stack_page; | |
976 | unsigned long vector_ptr; | |
977 | ||
978 | /* Switch from the HYP stub to our own HYP init vector */ | |
5a677ce0 | 979 | __hyp_set_vectors(kvm_get_idmap_vector()); |
342cd0ab | 980 | |
dac288f7 MZ |
981 | boot_pgd_ptr = kvm_mmu_get_boot_httbr(); |
982 | pgd_ptr = kvm_mmu_get_httbr(); | |
1436c1aa | 983 | stack_page = __this_cpu_read(kvm_arm_hyp_stack_page); |
342cd0ab CD |
984 | hyp_stack_ptr = stack_page + PAGE_SIZE; |
985 | vector_ptr = (unsigned long)__kvm_hyp_vector; | |
986 | ||
5a677ce0 | 987 | __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr); |
56c7f5e7 AB |
988 | |
989 | kvm_arm_init_debug(); | |
342cd0ab CD |
990 | } |
991 | ||
d157f4a5 MZ |
992 | static int hyp_init_cpu_notify(struct notifier_block *self, |
993 | unsigned long action, void *cpu) | |
994 | { | |
995 | switch (action) { | |
996 | case CPU_STARTING: | |
997 | case CPU_STARTING_FROZEN: | |
37a34ac1 VM |
998 | if (__hyp_get_vectors() == hyp_default_vectors) |
999 | cpu_init_hyp_mode(NULL); | |
d157f4a5 MZ |
1000 | break; |
1001 | } | |
1002 | ||
1003 | return NOTIFY_OK; | |
342cd0ab CD |
1004 | } |
1005 | ||
d157f4a5 MZ |
1006 | static struct notifier_block hyp_init_cpu_nb = { |
1007 | .notifier_call = hyp_init_cpu_notify, | |
1008 | }; | |
1009 | ||
1fcf7ce0 LP |
1010 | #ifdef CONFIG_CPU_PM |
1011 | static int hyp_init_cpu_pm_notifier(struct notifier_block *self, | |
1012 | unsigned long cmd, | |
1013 | void *v) | |
1014 | { | |
b20c9f29 MZ |
1015 | if (cmd == CPU_PM_EXIT && |
1016 | __hyp_get_vectors() == hyp_default_vectors) { | |
1fcf7ce0 LP |
1017 | cpu_init_hyp_mode(NULL); |
1018 | return NOTIFY_OK; | |
1019 | } | |
1020 | ||
1021 | return NOTIFY_DONE; | |
1022 | } | |
1023 | ||
1024 | static struct notifier_block hyp_init_cpu_pm_nb = { | |
1025 | .notifier_call = hyp_init_cpu_pm_notifier, | |
1026 | }; | |
1027 | ||
1028 | static void __init hyp_cpu_pm_init(void) | |
1029 | { | |
1030 | cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); | |
1031 | } | |
1032 | #else | |
1033 | static inline void hyp_cpu_pm_init(void) | |
1034 | { | |
1035 | } | |
1036 | #endif | |
1037 | ||
342cd0ab CD |
1038 | /** |
1039 | * Inits Hyp-mode on all online CPUs | |
1040 | */ | |
1041 | static int init_hyp_mode(void) | |
1042 | { | |
342cd0ab CD |
1043 | int cpu; |
1044 | int err = 0; | |
1045 | ||
1046 | /* | |
1047 | * Allocate Hyp PGD and setup Hyp identity mapping | |
1048 | */ | |
1049 | err = kvm_mmu_init(); | |
1050 | if (err) | |
1051 | goto out_err; | |
1052 | ||
1053 | /* | |
1054 | * It is probably enough to obtain the default on one | |
1055 | * CPU. It's unlikely to be different on the others. | |
1056 | */ | |
1057 | hyp_default_vectors = __hyp_get_vectors(); | |
1058 | ||
1059 | /* | |
1060 | * Allocate stack pages for Hypervisor-mode | |
1061 | */ | |
1062 | for_each_possible_cpu(cpu) { | |
1063 | unsigned long stack_page; | |
1064 | ||
1065 | stack_page = __get_free_page(GFP_KERNEL); | |
1066 | if (!stack_page) { | |
1067 | err = -ENOMEM; | |
1068 | goto out_free_stack_pages; | |
1069 | } | |
1070 | ||
1071 | per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; | |
1072 | } | |
1073 | ||
342cd0ab CD |
1074 | /* |
1075 | * Map the Hyp-code called directly from the host | |
1076 | */ | |
1077 | err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end); | |
1078 | if (err) { | |
1079 | kvm_err("Cannot map world-switch code\n"); | |
1080 | goto out_free_mappings; | |
1081 | } | |
1082 | ||
910917bb MZ |
1083 | err = create_hyp_mappings(__start_rodata, __end_rodata); |
1084 | if (err) { | |
1085 | kvm_err("Cannot map rodata section\n"); | |
1086 | goto out_free_mappings; | |
1087 | } | |
1088 | ||
342cd0ab CD |
1089 | /* |
1090 | * Map the Hyp stack pages | |
1091 | */ | |
1092 | for_each_possible_cpu(cpu) { | |
1093 | char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); | |
1094 | err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE); | |
1095 | ||
1096 | if (err) { | |
1097 | kvm_err("Cannot map hyp stack\n"); | |
1098 | goto out_free_mappings; | |
1099 | } | |
1100 | } | |
1101 | ||
1102 | /* | |
3de50da6 | 1103 | * Map the host CPU structures |
342cd0ab | 1104 | */ |
3de50da6 MZ |
1105 | kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t); |
1106 | if (!kvm_host_cpu_state) { | |
342cd0ab | 1107 | err = -ENOMEM; |
3de50da6 | 1108 | kvm_err("Cannot allocate host CPU state\n"); |
342cd0ab CD |
1109 | goto out_free_mappings; |
1110 | } | |
1111 | ||
1112 | for_each_possible_cpu(cpu) { | |
3de50da6 | 1113 | kvm_cpu_context_t *cpu_ctxt; |
342cd0ab | 1114 | |
3de50da6 MZ |
1115 | cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu); |
1116 | err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1); | |
342cd0ab CD |
1117 | |
1118 | if (err) { | |
3de50da6 MZ |
1119 | kvm_err("Cannot map host CPU state: %d\n", err); |
1120 | goto out_free_context; | |
342cd0ab CD |
1121 | } |
1122 | } | |
1123 | ||
d157f4a5 MZ |
1124 | /* |
1125 | * Execute the init code on each CPU. | |
1126 | */ | |
1127 | on_each_cpu(cpu_init_hyp_mode, NULL, 1); | |
1128 | ||
1a89dd91 MZ |
1129 | /* |
1130 | * Init HYP view of VGIC | |
1131 | */ | |
1132 | err = kvm_vgic_hyp_init(); | |
c7da6fa4 PF |
1133 | switch (err) { |
1134 | case 0: | |
1135 | vgic_present = true; | |
1136 | break; | |
1137 | case -ENODEV: | |
1138 | case -ENXIO: | |
1139 | vgic_present = false; | |
1140 | break; | |
1141 | default: | |
3de50da6 | 1142 | goto out_free_context; |
c7da6fa4 | 1143 | } |
1a89dd91 | 1144 | |
967f8427 MZ |
1145 | /* |
1146 | * Init HYP architected timer support | |
1147 | */ | |
1148 | err = kvm_timer_hyp_init(); | |
1149 | if (err) | |
399ea0f6 | 1150 | goto out_free_context; |
967f8427 | 1151 | |
d157f4a5 MZ |
1152 | #ifndef CONFIG_HOTPLUG_CPU |
1153 | free_boot_hyp_pgd(); | |
1154 | #endif | |
1155 | ||
210552c1 MZ |
1156 | kvm_perf_init(); |
1157 | ||
20475f78 VM |
1158 | /* set size of VMID supported by CPU */ |
1159 | kvm_vmid_bits = kvm_get_vmid_bits(); | |
1160 | kvm_info("%d-bit VMID\n", kvm_vmid_bits); | |
1161 | ||
342cd0ab | 1162 | kvm_info("Hyp mode initialized successfully\n"); |
210552c1 | 1163 | |
342cd0ab | 1164 | return 0; |
3de50da6 MZ |
1165 | out_free_context: |
1166 | free_percpu(kvm_host_cpu_state); | |
342cd0ab | 1167 | out_free_mappings: |
4f728276 | 1168 | free_hyp_pgds(); |
342cd0ab CD |
1169 | out_free_stack_pages: |
1170 | for_each_possible_cpu(cpu) | |
1171 | free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); | |
1172 | out_err: | |
1173 | kvm_err("error initializing Hyp mode: %d\n", err); | |
1174 | return err; | |
1175 | } | |
1176 | ||
d4e071ce AP |
1177 | static void check_kvm_target_cpu(void *ret) |
1178 | { | |
1179 | *(int *)ret = kvm_target_cpu(); | |
1180 | } | |
1181 | ||
4429fc64 AP |
1182 | struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) |
1183 | { | |
1184 | struct kvm_vcpu *vcpu; | |
1185 | int i; | |
1186 | ||
1187 | mpidr &= MPIDR_HWID_BITMASK; | |
1188 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
1189 | if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) | |
1190 | return vcpu; | |
1191 | } | |
1192 | return NULL; | |
1193 | } | |
1194 | ||
342cd0ab CD |
1195 | /** |
1196 | * Initialize Hyp-mode and memory mappings on all CPUs. | |
1197 | */ | |
749cf76c CD |
1198 | int kvm_arch_init(void *opaque) |
1199 | { | |
342cd0ab | 1200 | int err; |
d4e071ce | 1201 | int ret, cpu; |
342cd0ab CD |
1202 | |
1203 | if (!is_hyp_mode_available()) { | |
1204 | kvm_err("HYP mode not available\n"); | |
1205 | return -ENODEV; | |
1206 | } | |
1207 | ||
d4e071ce AP |
1208 | for_each_online_cpu(cpu) { |
1209 | smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); | |
1210 | if (ret < 0) { | |
1211 | kvm_err("Error, CPU %d not supported!\n", cpu); | |
1212 | return -ENODEV; | |
1213 | } | |
342cd0ab CD |
1214 | } |
1215 | ||
8146875d SB |
1216 | cpu_notifier_register_begin(); |
1217 | ||
342cd0ab CD |
1218 | err = init_hyp_mode(); |
1219 | if (err) | |
1220 | goto out_err; | |
1221 | ||
8146875d | 1222 | err = __register_cpu_notifier(&hyp_init_cpu_nb); |
d157f4a5 MZ |
1223 | if (err) { |
1224 | kvm_err("Cannot register HYP init CPU notifier (%d)\n", err); | |
1225 | goto out_err; | |
1226 | } | |
1227 | ||
8146875d SB |
1228 | cpu_notifier_register_done(); |
1229 | ||
1fcf7ce0 LP |
1230 | hyp_cpu_pm_init(); |
1231 | ||
5b3e5e5b | 1232 | kvm_coproc_table_init(); |
749cf76c | 1233 | return 0; |
342cd0ab | 1234 | out_err: |
8146875d | 1235 | cpu_notifier_register_done(); |
342cd0ab | 1236 | return err; |
749cf76c CD |
1237 | } |
1238 | ||
1239 | /* NOP: Compiling as a module not supported */ | |
1240 | void kvm_arch_exit(void) | |
1241 | { | |
210552c1 | 1242 | kvm_perf_teardown(); |
749cf76c CD |
1243 | } |
1244 | ||
1245 | static int arm_init(void) | |
1246 | { | |
1247 | int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1248 | return rc; | |
1249 | } | |
1250 | ||
1251 | module_init(arm_init); |