]>
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 | ||
1fcf7ce0 | 19 | #include <linux/cpu_pm.h> |
749cf76c CD |
20 | #include <linux/errno.h> |
21 | #include <linux/err.h> | |
22 | #include <linux/kvm_host.h> | |
1085fdc6 | 23 | #include <linux/list.h> |
749cf76c CD |
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 | 30 | #include <trace/events/kvm.h> |
b02386eb | 31 | #include <kvm/arm_pmu.h> |
749cf76c CD |
32 | |
33 | #define CREATE_TRACE_POINTS | |
34 | #include "trace.h" | |
35 | ||
7c0f6ba6 | 36 | #include <linux/uaccess.h> |
749cf76c CD |
37 | #include <asm/ptrace.h> |
38 | #include <asm/mman.h> | |
342cd0ab | 39 | #include <asm/tlbflush.h> |
5b3e5e5b | 40 | #include <asm/cacheflush.h> |
342cd0ab CD |
41 | #include <asm/virt.h> |
42 | #include <asm/kvm_arm.h> | |
43 | #include <asm/kvm_asm.h> | |
44 | #include <asm/kvm_mmu.h> | |
f7ed45be | 45 | #include <asm/kvm_emulate.h> |
5b3e5e5b | 46 | #include <asm/kvm_coproc.h> |
aa024c2f | 47 | #include <asm/kvm_psci.h> |
910917bb | 48 | #include <asm/sections.h> |
749cf76c CD |
49 | |
50 | #ifdef REQUIRES_VIRT | |
51 | __asm__(".arch_extension virt"); | |
52 | #endif | |
53 | ||
342cd0ab | 54 | static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); |
3de50da6 | 55 | static kvm_cpu_context_t __percpu *kvm_host_cpu_state; |
342cd0ab CD |
56 | static unsigned long hyp_default_vectors; |
57 | ||
1638a12d MZ |
58 | /* Per-CPU variable containing the currently running vcpu. */ |
59 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu); | |
60 | ||
f7ed45be CD |
61 | /* The VMID used in the VTTBR */ |
62 | static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1); | |
20475f78 VM |
63 | static u32 kvm_next_vmid; |
64 | static unsigned int kvm_vmid_bits __read_mostly; | |
f7ed45be | 65 | static DEFINE_SPINLOCK(kvm_vmid_lock); |
342cd0ab | 66 | |
c7da6fa4 PF |
67 | static bool vgic_present; |
68 | ||
67f69197 AT |
69 | static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled); |
70 | ||
1638a12d MZ |
71 | static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu) |
72 | { | |
73 | BUG_ON(preemptible()); | |
1436c1aa | 74 | __this_cpu_write(kvm_arm_running_vcpu, vcpu); |
1638a12d MZ |
75 | } |
76 | ||
77 | /** | |
78 | * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU. | |
79 | * Must be called from non-preemptible context | |
80 | */ | |
81 | struct kvm_vcpu *kvm_arm_get_running_vcpu(void) | |
82 | { | |
83 | BUG_ON(preemptible()); | |
1436c1aa | 84 | return __this_cpu_read(kvm_arm_running_vcpu); |
1638a12d MZ |
85 | } |
86 | ||
87 | /** | |
88 | * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
89 | */ | |
4000be42 | 90 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) |
1638a12d MZ |
91 | { |
92 | return &kvm_arm_running_vcpu; | |
93 | } | |
94 | ||
749cf76c CD |
95 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
96 | { | |
97 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; | |
98 | } | |
99 | ||
749cf76c CD |
100 | int kvm_arch_hardware_setup(void) |
101 | { | |
102 | return 0; | |
103 | } | |
104 | ||
749cf76c CD |
105 | void kvm_arch_check_processor_compat(void *rtn) |
106 | { | |
107 | *(int *)rtn = 0; | |
108 | } | |
109 | ||
749cf76c | 110 | |
d5d8184d CD |
111 | /** |
112 | * kvm_arch_init_vm - initializes a VM data structure | |
113 | * @kvm: pointer to the KVM struct | |
114 | */ | |
749cf76c CD |
115 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
116 | { | |
94d0e598 | 117 | int ret, cpu; |
d5d8184d | 118 | |
749cf76c CD |
119 | if (type) |
120 | return -EINVAL; | |
121 | ||
94d0e598 MZ |
122 | kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran)); |
123 | if (!kvm->arch.last_vcpu_ran) | |
124 | return -ENOMEM; | |
125 | ||
126 | for_each_possible_cpu(cpu) | |
127 | *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1; | |
128 | ||
d5d8184d CD |
129 | ret = kvm_alloc_stage2_pgd(kvm); |
130 | if (ret) | |
131 | goto out_fail_alloc; | |
132 | ||
c8dddecd | 133 | ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP); |
d5d8184d CD |
134 | if (ret) |
135 | goto out_free_stage2_pgd; | |
136 | ||
6c3d63c9 | 137 | kvm_vgic_early_init(kvm); |
a1a64387 | 138 | |
d5d8184d CD |
139 | /* Mark the initial VMID generation invalid */ |
140 | kvm->arch.vmid_gen = 0; | |
141 | ||
3caa2d8c | 142 | /* The maximum number of VCPUs is limited by the host's GIC model */ |
c7da6fa4 PF |
143 | kvm->arch.max_vcpus = vgic_present ? |
144 | kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; | |
3caa2d8c | 145 | |
d5d8184d CD |
146 | return ret; |
147 | out_free_stage2_pgd: | |
148 | kvm_free_stage2_pgd(kvm); | |
149 | out_fail_alloc: | |
94d0e598 MZ |
150 | free_percpu(kvm->arch.last_vcpu_ran); |
151 | kvm->arch.last_vcpu_ran = NULL; | |
d5d8184d | 152 | return ret; |
749cf76c CD |
153 | } |
154 | ||
235539b4 LC |
155 | bool kvm_arch_has_vcpu_debugfs(void) |
156 | { | |
157 | return false; | |
158 | } | |
159 | ||
160 | int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu) | |
161 | { | |
162 | return 0; | |
163 | } | |
164 | ||
749cf76c CD |
165 | int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
166 | { | |
167 | return VM_FAULT_SIGBUS; | |
168 | } | |
169 | ||
749cf76c | 170 | |
d5d8184d CD |
171 | /** |
172 | * kvm_arch_destroy_vm - destroy the VM data structure | |
173 | * @kvm: pointer to the KVM struct | |
174 | */ | |
749cf76c CD |
175 | void kvm_arch_destroy_vm(struct kvm *kvm) |
176 | { | |
177 | int i; | |
178 | ||
94d0e598 MZ |
179 | free_percpu(kvm->arch.last_vcpu_ran); |
180 | kvm->arch.last_vcpu_ran = NULL; | |
181 | ||
749cf76c CD |
182 | for (i = 0; i < KVM_MAX_VCPUS; ++i) { |
183 | if (kvm->vcpus[i]) { | |
184 | kvm_arch_vcpu_free(kvm->vcpus[i]); | |
185 | kvm->vcpus[i] = NULL; | |
186 | } | |
187 | } | |
c1bfb577 MZ |
188 | |
189 | kvm_vgic_destroy(kvm); | |
749cf76c CD |
190 | } |
191 | ||
784aa3d7 | 192 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
749cf76c CD |
193 | { |
194 | int r; | |
195 | switch (ext) { | |
1a89dd91 | 196 | case KVM_CAP_IRQCHIP: |
c7da6fa4 PF |
197 | r = vgic_present; |
198 | break; | |
d44758c0 | 199 | case KVM_CAP_IOEVENTFD: |
7330672b | 200 | case KVM_CAP_DEVICE_CTRL: |
749cf76c CD |
201 | case KVM_CAP_USER_MEMORY: |
202 | case KVM_CAP_SYNC_MMU: | |
203 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
204 | case KVM_CAP_ONE_REG: | |
aa024c2f | 205 | case KVM_CAP_ARM_PSCI: |
4447a208 | 206 | case KVM_CAP_ARM_PSCI_0_2: |
98047888 | 207 | case KVM_CAP_READONLY_MEM: |
ecccf0cc | 208 | case KVM_CAP_MP_STATE: |
460df4c1 | 209 | case KVM_CAP_IMMEDIATE_EXIT: |
749cf76c CD |
210 | r = 1; |
211 | break; | |
212 | case KVM_CAP_COALESCED_MMIO: | |
213 | r = KVM_COALESCED_MMIO_PAGE_OFFSET; | |
214 | break; | |
3401d546 CD |
215 | case KVM_CAP_ARM_SET_DEVICE_ADDR: |
216 | r = 1; | |
ca46e10f | 217 | break; |
749cf76c CD |
218 | case KVM_CAP_NR_VCPUS: |
219 | r = num_online_cpus(); | |
220 | break; | |
221 | case KVM_CAP_MAX_VCPUS: | |
222 | r = KVM_MAX_VCPUS; | |
223 | break; | |
7af4df85 LC |
224 | case KVM_CAP_NR_MEMSLOTS: |
225 | r = KVM_USER_MEM_SLOTS; | |
226 | break; | |
2988509d VM |
227 | case KVM_CAP_MSI_DEVID: |
228 | if (!kvm) | |
229 | r = -EINVAL; | |
230 | else | |
231 | r = kvm->arch.vgic.msis_require_devid; | |
232 | break; | |
749cf76c | 233 | default: |
b46f01ce | 234 | r = kvm_arch_dev_ioctl_check_extension(kvm, ext); |
749cf76c CD |
235 | break; |
236 | } | |
237 | return r; | |
238 | } | |
239 | ||
240 | long kvm_arch_dev_ioctl(struct file *filp, | |
241 | unsigned int ioctl, unsigned long arg) | |
242 | { | |
243 | return -EINVAL; | |
244 | } | |
245 | ||
749cf76c CD |
246 | |
247 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) | |
248 | { | |
249 | int err; | |
250 | struct kvm_vcpu *vcpu; | |
251 | ||
716139df CD |
252 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) { |
253 | err = -EBUSY; | |
254 | goto out; | |
255 | } | |
256 | ||
3caa2d8c AP |
257 | if (id >= kvm->arch.max_vcpus) { |
258 | err = -EINVAL; | |
259 | goto out; | |
260 | } | |
261 | ||
749cf76c CD |
262 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
263 | if (!vcpu) { | |
264 | err = -ENOMEM; | |
265 | goto out; | |
266 | } | |
267 | ||
268 | err = kvm_vcpu_init(vcpu, kvm, id); | |
269 | if (err) | |
270 | goto free_vcpu; | |
271 | ||
c8dddecd | 272 | err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP); |
d5d8184d CD |
273 | if (err) |
274 | goto vcpu_uninit; | |
275 | ||
749cf76c | 276 | return vcpu; |
d5d8184d CD |
277 | vcpu_uninit: |
278 | kvm_vcpu_uninit(vcpu); | |
749cf76c CD |
279 | free_vcpu: |
280 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
281 | out: | |
282 | return ERR_PTR(err); | |
283 | } | |
284 | ||
31928aa5 | 285 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
749cf76c | 286 | { |
6c3d63c9 | 287 | kvm_vgic_vcpu_early_init(vcpu); |
749cf76c CD |
288 | } |
289 | ||
290 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
291 | { | |
d5d8184d | 292 | kvm_mmu_free_memory_caches(vcpu); |
967f8427 | 293 | kvm_timer_vcpu_terminate(vcpu); |
c1bfb577 | 294 | kvm_vgic_vcpu_destroy(vcpu); |
5f0a714a | 295 | kvm_pmu_vcpu_destroy(vcpu); |
591d215a | 296 | kvm_vcpu_uninit(vcpu); |
d5d8184d | 297 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
749cf76c CD |
298 | } |
299 | ||
300 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) | |
301 | { | |
302 | kvm_arch_vcpu_free(vcpu); | |
303 | } | |
304 | ||
305 | int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) | |
306 | { | |
fb280e97 JL |
307 | return kvm_timer_should_fire(vcpu_vtimer(vcpu)) || |
308 | kvm_timer_should_fire(vcpu_ptimer(vcpu)); | |
749cf76c CD |
309 | } |
310 | ||
d35268da CD |
311 | void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) |
312 | { | |
313 | kvm_timer_schedule(vcpu); | |
314 | } | |
315 | ||
316 | void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
317 | { | |
318 | kvm_timer_unschedule(vcpu); | |
319 | } | |
320 | ||
749cf76c CD |
321 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
322 | { | |
f7ed45be CD |
323 | /* Force users to call KVM_ARM_VCPU_INIT */ |
324 | vcpu->arch.target = -1; | |
f7fa034d | 325 | bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); |
1a89dd91 | 326 | |
967f8427 MZ |
327 | /* Set up the timer */ |
328 | kvm_timer_vcpu_init(vcpu); | |
329 | ||
84e690bf AB |
330 | kvm_arm_reset_debug_ptr(vcpu); |
331 | ||
749cf76c CD |
332 | return 0; |
333 | } | |
334 | ||
749cf76c CD |
335 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
336 | { | |
94d0e598 MZ |
337 | int *last_ran; |
338 | ||
339 | last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran); | |
340 | ||
341 | /* | |
342 | * We might get preempted before the vCPU actually runs, but | |
343 | * over-invalidation doesn't affect correctness. | |
344 | */ | |
345 | if (*last_ran != vcpu->vcpu_id) { | |
346 | kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu); | |
347 | *last_ran = vcpu->vcpu_id; | |
348 | } | |
349 | ||
86ce8535 | 350 | vcpu->cpu = cpu; |
3de50da6 | 351 | vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state); |
5b3e5e5b | 352 | |
1638a12d | 353 | kvm_arm_set_running_vcpu(vcpu); |
328e5664 CD |
354 | |
355 | kvm_vgic_load(vcpu); | |
749cf76c CD |
356 | } |
357 | ||
358 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
359 | { | |
328e5664 CD |
360 | kvm_vgic_put(vcpu); |
361 | ||
e9b152cb CD |
362 | vcpu->cpu = -1; |
363 | ||
1638a12d | 364 | kvm_arm_set_running_vcpu(NULL); |
9b4a3004 | 365 | kvm_timer_vcpu_put(vcpu); |
749cf76c CD |
366 | } |
367 | ||
749cf76c CD |
368 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
369 | struct kvm_mp_state *mp_state) | |
370 | { | |
3781528e | 371 | if (vcpu->arch.power_off) |
ecccf0cc AB |
372 | mp_state->mp_state = KVM_MP_STATE_STOPPED; |
373 | else | |
374 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
375 | ||
376 | return 0; | |
749cf76c CD |
377 | } |
378 | ||
379 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
380 | struct kvm_mp_state *mp_state) | |
381 | { | |
ecccf0cc AB |
382 | switch (mp_state->mp_state) { |
383 | case KVM_MP_STATE_RUNNABLE: | |
3781528e | 384 | vcpu->arch.power_off = false; |
ecccf0cc AB |
385 | break; |
386 | case KVM_MP_STATE_STOPPED: | |
3781528e | 387 | vcpu->arch.power_off = true; |
ecccf0cc AB |
388 | break; |
389 | default: | |
390 | return -EINVAL; | |
391 | } | |
392 | ||
393 | return 0; | |
749cf76c CD |
394 | } |
395 | ||
5b3e5e5b CD |
396 | /** |
397 | * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled | |
398 | * @v: The VCPU pointer | |
399 | * | |
400 | * If the guest CPU is not waiting for interrupts or an interrupt line is | |
401 | * asserted, the CPU is by definition runnable. | |
402 | */ | |
749cf76c CD |
403 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) |
404 | { | |
4f5f1dc0 | 405 | return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v)) |
3b92830a | 406 | && !v->arch.power_off && !v->arch.pause); |
749cf76c CD |
407 | } |
408 | ||
f7ed45be CD |
409 | /* Just ensure a guest exit from a particular CPU */ |
410 | static void exit_vm_noop(void *info) | |
411 | { | |
412 | } | |
413 | ||
414 | void force_vm_exit(const cpumask_t *mask) | |
415 | { | |
898f949f | 416 | preempt_disable(); |
f7ed45be | 417 | smp_call_function_many(mask, exit_vm_noop, NULL, true); |
898f949f | 418 | preempt_enable(); |
f7ed45be CD |
419 | } |
420 | ||
421 | /** | |
422 | * need_new_vmid_gen - check that the VMID is still valid | |
6a727b0b | 423 | * @kvm: The VM's VMID to check |
f7ed45be CD |
424 | * |
425 | * return true if there is a new generation of VMIDs being used | |
426 | * | |
427 | * The hardware supports only 256 values with the value zero reserved for the | |
428 | * host, so we check if an assigned value belongs to a previous generation, | |
429 | * which which requires us to assign a new value. If we're the first to use a | |
430 | * VMID for the new generation, we must flush necessary caches and TLBs on all | |
431 | * CPUs. | |
432 | */ | |
433 | static bool need_new_vmid_gen(struct kvm *kvm) | |
434 | { | |
435 | return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen)); | |
436 | } | |
437 | ||
438 | /** | |
439 | * update_vttbr - Update the VTTBR with a valid VMID before the guest runs | |
440 | * @kvm The guest that we are about to run | |
441 | * | |
442 | * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the | |
443 | * VM has a valid VMID, otherwise assigns a new one and flushes corresponding | |
444 | * caches and TLBs. | |
445 | */ | |
446 | static void update_vttbr(struct kvm *kvm) | |
447 | { | |
448 | phys_addr_t pgd_phys; | |
449 | u64 vmid; | |
450 | ||
451 | if (!need_new_vmid_gen(kvm)) | |
452 | return; | |
453 | ||
454 | spin_lock(&kvm_vmid_lock); | |
455 | ||
456 | /* | |
457 | * We need to re-check the vmid_gen here to ensure that if another vcpu | |
458 | * already allocated a valid vmid for this vm, then this vcpu should | |
459 | * use the same vmid. | |
460 | */ | |
461 | if (!need_new_vmid_gen(kvm)) { | |
462 | spin_unlock(&kvm_vmid_lock); | |
463 | return; | |
464 | } | |
465 | ||
466 | /* First user of a new VMID generation? */ | |
467 | if (unlikely(kvm_next_vmid == 0)) { | |
468 | atomic64_inc(&kvm_vmid_gen); | |
469 | kvm_next_vmid = 1; | |
470 | ||
471 | /* | |
472 | * On SMP we know no other CPUs can use this CPU's or each | |
473 | * other's VMID after force_vm_exit returns since the | |
474 | * kvm_vmid_lock blocks them from reentry to the guest. | |
475 | */ | |
476 | force_vm_exit(cpu_all_mask); | |
477 | /* | |
478 | * Now broadcast TLB + ICACHE invalidation over the inner | |
479 | * shareable domain to make sure all data structures are | |
480 | * clean. | |
481 | */ | |
482 | kvm_call_hyp(__kvm_flush_vm_context); | |
483 | } | |
484 | ||
485 | kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen); | |
486 | kvm->arch.vmid = kvm_next_vmid; | |
487 | kvm_next_vmid++; | |
20475f78 | 488 | kvm_next_vmid &= (1 << kvm_vmid_bits) - 1; |
f7ed45be CD |
489 | |
490 | /* update vttbr to be used with the new vmid */ | |
9163ee23 | 491 | pgd_phys = virt_to_phys(kvm->arch.pgd); |
dbff124e | 492 | BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK); |
20475f78 | 493 | vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits); |
dbff124e | 494 | kvm->arch.vttbr = pgd_phys | vmid; |
f7ed45be CD |
495 | |
496 | spin_unlock(&kvm_vmid_lock); | |
497 | } | |
498 | ||
f7ed45be CD |
499 | static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) |
500 | { | |
05971120 | 501 | struct kvm *kvm = vcpu->kvm; |
41a54482 | 502 | int ret = 0; |
e1ba0207 | 503 | |
f7ed45be CD |
504 | if (likely(vcpu->arch.has_run_once)) |
505 | return 0; | |
506 | ||
507 | vcpu->arch.has_run_once = true; | |
aa024c2f | 508 | |
01ac5e34 | 509 | /* |
6d3cfbe2 PM |
510 | * Map the VGIC hardware resources before running a vcpu the first |
511 | * time on this VM. | |
01ac5e34 | 512 | */ |
c2f58514 | 513 | if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) { |
05971120 | 514 | ret = kvm_vgic_map_resources(kvm); |
01ac5e34 MZ |
515 | if (ret) |
516 | return ret; | |
517 | } | |
518 | ||
05971120 CD |
519 | /* |
520 | * Enable the arch timers only if we have an in-kernel VGIC | |
521 | * and it has been properly initialized, since we cannot handle | |
522 | * interrupts from the virtual timer with a userspace gic. | |
523 | */ | |
524 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) | |
41a54482 | 525 | ret = kvm_timer_enable(vcpu); |
05971120 | 526 | |
41a54482 | 527 | return ret; |
f7ed45be CD |
528 | } |
529 | ||
c1426e4c EA |
530 | bool kvm_arch_intc_initialized(struct kvm *kvm) |
531 | { | |
532 | return vgic_initialized(kvm); | |
533 | } | |
534 | ||
b13216cf | 535 | void kvm_arm_halt_guest(struct kvm *kvm) |
3b92830a EA |
536 | { |
537 | int i; | |
538 | struct kvm_vcpu *vcpu; | |
539 | ||
540 | kvm_for_each_vcpu(i, vcpu, kvm) | |
541 | vcpu->arch.pause = true; | |
b13216cf | 542 | kvm_make_all_cpus_request(kvm, KVM_REQ_VCPU_EXIT); |
3b92830a EA |
543 | } |
544 | ||
35a2d585 CD |
545 | void kvm_arm_halt_vcpu(struct kvm_vcpu *vcpu) |
546 | { | |
547 | vcpu->arch.pause = true; | |
548 | kvm_vcpu_kick(vcpu); | |
549 | } | |
550 | ||
551 | void kvm_arm_resume_vcpu(struct kvm_vcpu *vcpu) | |
b13216cf CD |
552 | { |
553 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); | |
554 | ||
555 | vcpu->arch.pause = false; | |
556 | swake_up(wq); | |
557 | } | |
558 | ||
559 | void kvm_arm_resume_guest(struct kvm *kvm) | |
3b92830a EA |
560 | { |
561 | int i; | |
562 | struct kvm_vcpu *vcpu; | |
563 | ||
b13216cf CD |
564 | kvm_for_each_vcpu(i, vcpu, kvm) |
565 | kvm_arm_resume_vcpu(vcpu); | |
3b92830a EA |
566 | } |
567 | ||
3781528e | 568 | static void vcpu_sleep(struct kvm_vcpu *vcpu) |
aa024c2f | 569 | { |
8577370f | 570 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); |
aa024c2f | 571 | |
8577370f | 572 | swait_event_interruptible(*wq, ((!vcpu->arch.power_off) && |
3b92830a | 573 | (!vcpu->arch.pause))); |
aa024c2f MZ |
574 | } |
575 | ||
e8180dca AP |
576 | static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) |
577 | { | |
578 | return vcpu->arch.target >= 0; | |
579 | } | |
580 | ||
f7ed45be CD |
581 | /** |
582 | * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code | |
583 | * @vcpu: The VCPU pointer | |
584 | * @run: The kvm_run structure pointer used for userspace state exchange | |
585 | * | |
586 | * This function is called through the VCPU_RUN ioctl called from user space. It | |
587 | * will execute VM code in a loop until the time slice for the process is used | |
588 | * or some emulation is needed from user space in which case the function will | |
589 | * return with return value 0 and with the kvm_run structure filled in with the | |
590 | * required data for the requested emulation. | |
591 | */ | |
749cf76c CD |
592 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) |
593 | { | |
f7ed45be CD |
594 | int ret; |
595 | sigset_t sigsaved; | |
596 | ||
e8180dca | 597 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
f7ed45be CD |
598 | return -ENOEXEC; |
599 | ||
600 | ret = kvm_vcpu_first_run_init(vcpu); | |
601 | if (ret) | |
602 | return ret; | |
603 | ||
45e96ea6 CD |
604 | if (run->exit_reason == KVM_EXIT_MMIO) { |
605 | ret = kvm_handle_mmio_return(vcpu, vcpu->run); | |
606 | if (ret) | |
607 | return ret; | |
608 | } | |
609 | ||
460df4c1 PB |
610 | if (run->immediate_exit) |
611 | return -EINTR; | |
612 | ||
f7ed45be CD |
613 | if (vcpu->sigset_active) |
614 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
615 | ||
616 | ret = 1; | |
617 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
618 | while (ret > 0) { | |
619 | /* | |
620 | * Check conditions before entering the guest | |
621 | */ | |
622 | cond_resched(); | |
623 | ||
624 | update_vttbr(vcpu->kvm); | |
625 | ||
3b92830a | 626 | if (vcpu->arch.power_off || vcpu->arch.pause) |
3781528e | 627 | vcpu_sleep(vcpu); |
aa024c2f | 628 | |
abdf5843 MZ |
629 | /* |
630 | * Preparing the interrupts to be injected also | |
631 | * involves poking the GIC, which must be done in a | |
632 | * non-preemptible context. | |
633 | */ | |
1b3d546d | 634 | preempt_disable(); |
328e5664 | 635 | |
b02386eb | 636 | kvm_pmu_flush_hwstate(vcpu); |
328e5664 | 637 | |
7e16aa81 | 638 | kvm_timer_flush_hwstate(vcpu); |
abdf5843 MZ |
639 | kvm_vgic_flush_hwstate(vcpu); |
640 | ||
f7ed45be CD |
641 | local_irq_disable(); |
642 | ||
643 | /* | |
644 | * Re-check atomic conditions | |
645 | */ | |
646 | if (signal_pending(current)) { | |
647 | ret = -EINTR; | |
648 | run->exit_reason = KVM_EXIT_INTR; | |
649 | } | |
650 | ||
101d3da0 | 651 | if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) || |
3b92830a | 652 | vcpu->arch.power_off || vcpu->arch.pause) { |
f7ed45be | 653 | local_irq_enable(); |
b02386eb | 654 | kvm_pmu_sync_hwstate(vcpu); |
4b4b4512 | 655 | kvm_timer_sync_hwstate(vcpu); |
1a89dd91 | 656 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 | 657 | preempt_enable(); |
f7ed45be CD |
658 | continue; |
659 | } | |
660 | ||
56c7f5e7 AB |
661 | kvm_arm_setup_debug(vcpu); |
662 | ||
f7ed45be CD |
663 | /************************************************************** |
664 | * Enter the guest | |
665 | */ | |
666 | trace_kvm_entry(*vcpu_pc(vcpu)); | |
6edaa530 | 667 | guest_enter_irqoff(); |
f7ed45be CD |
668 | vcpu->mode = IN_GUEST_MODE; |
669 | ||
670 | ret = kvm_call_hyp(__kvm_vcpu_run, vcpu); | |
671 | ||
672 | vcpu->mode = OUTSIDE_GUEST_MODE; | |
b19e6892 | 673 | vcpu->stat.exits++; |
1b3d546d CD |
674 | /* |
675 | * Back from guest | |
676 | *************************************************************/ | |
677 | ||
56c7f5e7 AB |
678 | kvm_arm_clear_debug(vcpu); |
679 | ||
f7ed45be CD |
680 | /* |
681 | * We may have taken a host interrupt in HYP mode (ie | |
682 | * while executing the guest). This interrupt is still | |
683 | * pending, as we haven't serviced it yet! | |
684 | * | |
685 | * We're now back in SVC mode, with interrupts | |
686 | * disabled. Enabling the interrupts now will have | |
687 | * the effect of taking the interrupt again, in SVC | |
688 | * mode this time. | |
689 | */ | |
690 | local_irq_enable(); | |
691 | ||
692 | /* | |
6edaa530 | 693 | * We do local_irq_enable() before calling guest_exit() so |
1b3d546d CD |
694 | * that if a timer interrupt hits while running the guest we |
695 | * account that tick as being spent in the guest. We enable | |
6edaa530 | 696 | * preemption after calling guest_exit() so that if we get |
1b3d546d CD |
697 | * preempted we make sure ticks after that is not counted as |
698 | * guest time. | |
699 | */ | |
6edaa530 | 700 | guest_exit(); |
b5905dc1 | 701 | trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); |
1b3d546d | 702 | |
4b4b4512 | 703 | /* |
b02386eb SZ |
704 | * We must sync the PMU and timer state before the vgic state so |
705 | * that the vgic can properly sample the updated state of the | |
4b4b4512 CD |
706 | * interrupt line. |
707 | */ | |
b02386eb | 708 | kvm_pmu_sync_hwstate(vcpu); |
4b4b4512 CD |
709 | kvm_timer_sync_hwstate(vcpu); |
710 | ||
1a89dd91 | 711 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 MZ |
712 | |
713 | preempt_enable(); | |
714 | ||
f7ed45be CD |
715 | ret = handle_exit(vcpu, run, ret); |
716 | } | |
717 | ||
718 | if (vcpu->sigset_active) | |
719 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
720 | return ret; | |
749cf76c CD |
721 | } |
722 | ||
86ce8535 CD |
723 | static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) |
724 | { | |
725 | int bit_index; | |
726 | bool set; | |
727 | unsigned long *ptr; | |
728 | ||
729 | if (number == KVM_ARM_IRQ_CPU_IRQ) | |
730 | bit_index = __ffs(HCR_VI); | |
731 | else /* KVM_ARM_IRQ_CPU_FIQ */ | |
732 | bit_index = __ffs(HCR_VF); | |
733 | ||
734 | ptr = (unsigned long *)&vcpu->arch.irq_lines; | |
735 | if (level) | |
736 | set = test_and_set_bit(bit_index, ptr); | |
737 | else | |
738 | set = test_and_clear_bit(bit_index, ptr); | |
739 | ||
740 | /* | |
741 | * If we didn't change anything, no need to wake up or kick other CPUs | |
742 | */ | |
743 | if (set == level) | |
744 | return 0; | |
745 | ||
746 | /* | |
747 | * The vcpu irq_lines field was updated, wake up sleeping VCPUs and | |
748 | * trigger a world-switch round on the running physical CPU to set the | |
749 | * virtual IRQ/FIQ fields in the HCR appropriately. | |
750 | */ | |
751 | kvm_vcpu_kick(vcpu); | |
752 | ||
753 | return 0; | |
754 | } | |
755 | ||
79558f11 AG |
756 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, |
757 | bool line_status) | |
86ce8535 CD |
758 | { |
759 | u32 irq = irq_level->irq; | |
760 | unsigned int irq_type, vcpu_idx, irq_num; | |
761 | int nrcpus = atomic_read(&kvm->online_vcpus); | |
762 | struct kvm_vcpu *vcpu = NULL; | |
763 | bool level = irq_level->level; | |
764 | ||
765 | irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; | |
766 | vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; | |
767 | irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; | |
768 | ||
769 | trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); | |
770 | ||
5863c2ce MZ |
771 | switch (irq_type) { |
772 | case KVM_ARM_IRQ_TYPE_CPU: | |
773 | if (irqchip_in_kernel(kvm)) | |
774 | return -ENXIO; | |
86ce8535 | 775 | |
5863c2ce MZ |
776 | if (vcpu_idx >= nrcpus) |
777 | return -EINVAL; | |
86ce8535 | 778 | |
5863c2ce MZ |
779 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); |
780 | if (!vcpu) | |
781 | return -EINVAL; | |
86ce8535 | 782 | |
5863c2ce MZ |
783 | if (irq_num > KVM_ARM_IRQ_CPU_FIQ) |
784 | return -EINVAL; | |
785 | ||
786 | return vcpu_interrupt_line(vcpu, irq_num, level); | |
787 | case KVM_ARM_IRQ_TYPE_PPI: | |
788 | if (!irqchip_in_kernel(kvm)) | |
789 | return -ENXIO; | |
790 | ||
791 | if (vcpu_idx >= nrcpus) | |
792 | return -EINVAL; | |
793 | ||
794 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); | |
795 | if (!vcpu) | |
796 | return -EINVAL; | |
797 | ||
798 | if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) | |
799 | return -EINVAL; | |
86ce8535 | 800 | |
5863c2ce MZ |
801 | return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level); |
802 | case KVM_ARM_IRQ_TYPE_SPI: | |
803 | if (!irqchip_in_kernel(kvm)) | |
804 | return -ENXIO; | |
805 | ||
fd1d0ddf | 806 | if (irq_num < VGIC_NR_PRIVATE_IRQS) |
5863c2ce MZ |
807 | return -EINVAL; |
808 | ||
809 | return kvm_vgic_inject_irq(kvm, 0, irq_num, level); | |
810 | } | |
811 | ||
812 | return -EINVAL; | |
86ce8535 CD |
813 | } |
814 | ||
f7fa034d CD |
815 | static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, |
816 | const struct kvm_vcpu_init *init) | |
817 | { | |
818 | unsigned int i; | |
819 | int phys_target = kvm_target_cpu(); | |
820 | ||
821 | if (init->target != phys_target) | |
822 | return -EINVAL; | |
823 | ||
824 | /* | |
825 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
826 | * use the same target. | |
827 | */ | |
828 | if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) | |
829 | return -EINVAL; | |
830 | ||
831 | /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ | |
832 | for (i = 0; i < sizeof(init->features) * 8; i++) { | |
833 | bool set = (init->features[i / 32] & (1 << (i % 32))); | |
834 | ||
835 | if (set && i >= KVM_VCPU_MAX_FEATURES) | |
836 | return -ENOENT; | |
837 | ||
838 | /* | |
839 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
840 | * use the same feature set. | |
841 | */ | |
842 | if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && | |
843 | test_bit(i, vcpu->arch.features) != set) | |
844 | return -EINVAL; | |
845 | ||
846 | if (set) | |
847 | set_bit(i, vcpu->arch.features); | |
848 | } | |
849 | ||
850 | vcpu->arch.target = phys_target; | |
851 | ||
852 | /* Now we know what it is, we can reset it. */ | |
853 | return kvm_reset_vcpu(vcpu); | |
854 | } | |
855 | ||
856 | ||
478a8237 CD |
857 | static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, |
858 | struct kvm_vcpu_init *init) | |
859 | { | |
860 | int ret; | |
861 | ||
862 | ret = kvm_vcpu_set_target(vcpu, init); | |
863 | if (ret) | |
864 | return ret; | |
865 | ||
957db105 CD |
866 | /* |
867 | * Ensure a rebooted VM will fault in RAM pages and detect if the | |
868 | * guest MMU is turned off and flush the caches as needed. | |
869 | */ | |
870 | if (vcpu->arch.has_run_once) | |
871 | stage2_unmap_vm(vcpu->kvm); | |
872 | ||
b856a591 CD |
873 | vcpu_reset_hcr(vcpu); |
874 | ||
478a8237 | 875 | /* |
3781528e | 876 | * Handle the "start in power-off" case. |
478a8237 | 877 | */ |
03f1d4c1 | 878 | if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) |
3781528e | 879 | vcpu->arch.power_off = true; |
3ad8b3de | 880 | else |
3781528e | 881 | vcpu->arch.power_off = false; |
478a8237 CD |
882 | |
883 | return 0; | |
884 | } | |
885 | ||
f577f6c2 SZ |
886 | static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, |
887 | struct kvm_device_attr *attr) | |
888 | { | |
889 | int ret = -ENXIO; | |
890 | ||
891 | switch (attr->group) { | |
892 | default: | |
bb0c70bc | 893 | ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); |
f577f6c2 SZ |
894 | break; |
895 | } | |
896 | ||
897 | return ret; | |
898 | } | |
899 | ||
900 | static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, | |
901 | struct kvm_device_attr *attr) | |
902 | { | |
903 | int ret = -ENXIO; | |
904 | ||
905 | switch (attr->group) { | |
906 | default: | |
bb0c70bc | 907 | ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); |
f577f6c2 SZ |
908 | break; |
909 | } | |
910 | ||
911 | return ret; | |
912 | } | |
913 | ||
914 | static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, | |
915 | struct kvm_device_attr *attr) | |
916 | { | |
917 | int ret = -ENXIO; | |
918 | ||
919 | switch (attr->group) { | |
920 | default: | |
bb0c70bc | 921 | ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); |
f577f6c2 SZ |
922 | break; |
923 | } | |
924 | ||
925 | return ret; | |
926 | } | |
927 | ||
749cf76c CD |
928 | long kvm_arch_vcpu_ioctl(struct file *filp, |
929 | unsigned int ioctl, unsigned long arg) | |
930 | { | |
931 | struct kvm_vcpu *vcpu = filp->private_data; | |
932 | void __user *argp = (void __user *)arg; | |
f577f6c2 | 933 | struct kvm_device_attr attr; |
749cf76c CD |
934 | |
935 | switch (ioctl) { | |
936 | case KVM_ARM_VCPU_INIT: { | |
937 | struct kvm_vcpu_init init; | |
938 | ||
939 | if (copy_from_user(&init, argp, sizeof(init))) | |
940 | return -EFAULT; | |
941 | ||
478a8237 | 942 | return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); |
749cf76c CD |
943 | } |
944 | case KVM_SET_ONE_REG: | |
945 | case KVM_GET_ONE_REG: { | |
946 | struct kvm_one_reg reg; | |
e8180dca AP |
947 | |
948 | if (unlikely(!kvm_vcpu_initialized(vcpu))) | |
949 | return -ENOEXEC; | |
950 | ||
749cf76c CD |
951 | if (copy_from_user(®, argp, sizeof(reg))) |
952 | return -EFAULT; | |
953 | if (ioctl == KVM_SET_ONE_REG) | |
954 | return kvm_arm_set_reg(vcpu, ®); | |
955 | else | |
956 | return kvm_arm_get_reg(vcpu, ®); | |
957 | } | |
958 | case KVM_GET_REG_LIST: { | |
959 | struct kvm_reg_list __user *user_list = argp; | |
960 | struct kvm_reg_list reg_list; | |
961 | unsigned n; | |
962 | ||
e8180dca AP |
963 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
964 | return -ENOEXEC; | |
965 | ||
749cf76c CD |
966 | if (copy_from_user(®_list, user_list, sizeof(reg_list))) |
967 | return -EFAULT; | |
968 | n = reg_list.n; | |
969 | reg_list.n = kvm_arm_num_regs(vcpu); | |
970 | if (copy_to_user(user_list, ®_list, sizeof(reg_list))) | |
971 | return -EFAULT; | |
972 | if (n < reg_list.n) | |
973 | return -E2BIG; | |
974 | return kvm_arm_copy_reg_indices(vcpu, user_list->reg); | |
975 | } | |
f577f6c2 SZ |
976 | case KVM_SET_DEVICE_ATTR: { |
977 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
978 | return -EFAULT; | |
979 | return kvm_arm_vcpu_set_attr(vcpu, &attr); | |
980 | } | |
981 | case KVM_GET_DEVICE_ATTR: { | |
982 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
983 | return -EFAULT; | |
984 | return kvm_arm_vcpu_get_attr(vcpu, &attr); | |
985 | } | |
986 | case KVM_HAS_DEVICE_ATTR: { | |
987 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
988 | return -EFAULT; | |
989 | return kvm_arm_vcpu_has_attr(vcpu, &attr); | |
990 | } | |
749cf76c CD |
991 | default: |
992 | return -EINVAL; | |
993 | } | |
994 | } | |
995 | ||
53c810c3 MS |
996 | /** |
997 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
998 | * @kvm: kvm instance | |
999 | * @log: slot id and address to which we copy the log | |
1000 | * | |
1001 | * Steps 1-4 below provide general overview of dirty page logging. See | |
1002 | * kvm_get_dirty_log_protect() function description for additional details. | |
1003 | * | |
1004 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
1005 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
1006 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
1007 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
1008 | * writes will be marked dirty for next log read. | |
1009 | * | |
1010 | * 1. Take a snapshot of the bit and clear it if needed. | |
1011 | * 2. Write protect the corresponding page. | |
1012 | * 3. Copy the snapshot to the userspace. | |
1013 | * 4. Flush TLB's if needed. | |
1014 | */ | |
749cf76c CD |
1015 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
1016 | { | |
53c810c3 MS |
1017 | bool is_dirty = false; |
1018 | int r; | |
1019 | ||
1020 | mutex_lock(&kvm->slots_lock); | |
1021 | ||
1022 | r = kvm_get_dirty_log_protect(kvm, log, &is_dirty); | |
1023 | ||
1024 | if (is_dirty) | |
1025 | kvm_flush_remote_tlbs(kvm); | |
1026 | ||
1027 | mutex_unlock(&kvm->slots_lock); | |
1028 | return r; | |
749cf76c CD |
1029 | } |
1030 | ||
3401d546 CD |
1031 | static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, |
1032 | struct kvm_arm_device_addr *dev_addr) | |
1033 | { | |
330690cd CD |
1034 | unsigned long dev_id, type; |
1035 | ||
1036 | dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> | |
1037 | KVM_ARM_DEVICE_ID_SHIFT; | |
1038 | type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> | |
1039 | KVM_ARM_DEVICE_TYPE_SHIFT; | |
1040 | ||
1041 | switch (dev_id) { | |
1042 | case KVM_ARM_DEVICE_VGIC_V2: | |
c7da6fa4 PF |
1043 | if (!vgic_present) |
1044 | return -ENXIO; | |
ce01e4e8 | 1045 | return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); |
330690cd CD |
1046 | default: |
1047 | return -ENODEV; | |
1048 | } | |
3401d546 CD |
1049 | } |
1050 | ||
749cf76c CD |
1051 | long kvm_arch_vm_ioctl(struct file *filp, |
1052 | unsigned int ioctl, unsigned long arg) | |
1053 | { | |
3401d546 CD |
1054 | struct kvm *kvm = filp->private_data; |
1055 | void __user *argp = (void __user *)arg; | |
1056 | ||
1057 | switch (ioctl) { | |
5863c2ce | 1058 | case KVM_CREATE_IRQCHIP: { |
a28ebea2 | 1059 | int ret; |
c7da6fa4 PF |
1060 | if (!vgic_present) |
1061 | return -ENXIO; | |
a28ebea2 CD |
1062 | mutex_lock(&kvm->lock); |
1063 | ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); | |
1064 | mutex_unlock(&kvm->lock); | |
1065 | return ret; | |
5863c2ce | 1066 | } |
3401d546 CD |
1067 | case KVM_ARM_SET_DEVICE_ADDR: { |
1068 | struct kvm_arm_device_addr dev_addr; | |
1069 | ||
1070 | if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) | |
1071 | return -EFAULT; | |
1072 | return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); | |
1073 | } | |
42c4e0c7 AP |
1074 | case KVM_ARM_PREFERRED_TARGET: { |
1075 | int err; | |
1076 | struct kvm_vcpu_init init; | |
1077 | ||
1078 | err = kvm_vcpu_preferred_target(&init); | |
1079 | if (err) | |
1080 | return err; | |
1081 | ||
1082 | if (copy_to_user(argp, &init, sizeof(init))) | |
1083 | return -EFAULT; | |
1084 | ||
1085 | return 0; | |
1086 | } | |
3401d546 CD |
1087 | default: |
1088 | return -EINVAL; | |
1089 | } | |
749cf76c CD |
1090 | } |
1091 | ||
d157f4a5 | 1092 | static void cpu_init_hyp_mode(void *dummy) |
342cd0ab | 1093 | { |
dac288f7 | 1094 | phys_addr_t pgd_ptr; |
342cd0ab CD |
1095 | unsigned long hyp_stack_ptr; |
1096 | unsigned long stack_page; | |
1097 | unsigned long vector_ptr; | |
1098 | ||
1099 | /* Switch from the HYP stub to our own HYP init vector */ | |
5a677ce0 | 1100 | __hyp_set_vectors(kvm_get_idmap_vector()); |
342cd0ab | 1101 | |
dac288f7 | 1102 | pgd_ptr = kvm_mmu_get_httbr(); |
1436c1aa | 1103 | stack_page = __this_cpu_read(kvm_arm_hyp_stack_page); |
342cd0ab | 1104 | hyp_stack_ptr = stack_page + PAGE_SIZE; |
a0bf9776 | 1105 | vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector); |
342cd0ab | 1106 | |
12fda812 | 1107 | __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr); |
35a2491a | 1108 | __cpu_init_stage2(); |
56c7f5e7 | 1109 | |
488f94d7 JL |
1110 | if (is_kernel_in_hyp_mode()) |
1111 | kvm_timer_init_vhe(); | |
1112 | ||
56c7f5e7 | 1113 | kvm_arm_init_debug(); |
342cd0ab CD |
1114 | } |
1115 | ||
5f5560b1 JM |
1116 | static void cpu_hyp_reinit(void) |
1117 | { | |
1118 | if (is_kernel_in_hyp_mode()) { | |
1119 | /* | |
67f69197 | 1120 | * __cpu_init_stage2() is safe to call even if the PM |
5f5560b1 JM |
1121 | * event was cancelled before the CPU was reset. |
1122 | */ | |
67f69197 | 1123 | __cpu_init_stage2(); |
5f5560b1 JM |
1124 | } else { |
1125 | if (__hyp_get_vectors() == hyp_default_vectors) | |
1126 | cpu_init_hyp_mode(NULL); | |
1127 | } | |
1128 | } | |
1129 | ||
67f69197 | 1130 | static void cpu_hyp_reset(void) |
d157f4a5 | 1131 | { |
12fda812 | 1132 | if (!is_kernel_in_hyp_mode()) |
0fb26593 | 1133 | __hyp_reset_vectors(); |
67f69197 AT |
1134 | } |
1135 | ||
1136 | static void _kvm_arch_hardware_enable(void *discard) | |
1137 | { | |
1138 | if (!__this_cpu_read(kvm_arm_hardware_enabled)) { | |
5f5560b1 | 1139 | cpu_hyp_reinit(); |
67f69197 | 1140 | __this_cpu_write(kvm_arm_hardware_enabled, 1); |
d157f4a5 | 1141 | } |
67f69197 | 1142 | } |
d157f4a5 | 1143 | |
67f69197 AT |
1144 | int kvm_arch_hardware_enable(void) |
1145 | { | |
1146 | _kvm_arch_hardware_enable(NULL); | |
1147 | return 0; | |
342cd0ab CD |
1148 | } |
1149 | ||
67f69197 AT |
1150 | static void _kvm_arch_hardware_disable(void *discard) |
1151 | { | |
1152 | if (__this_cpu_read(kvm_arm_hardware_enabled)) { | |
1153 | cpu_hyp_reset(); | |
1154 | __this_cpu_write(kvm_arm_hardware_enabled, 0); | |
1155 | } | |
1156 | } | |
1157 | ||
1158 | void kvm_arch_hardware_disable(void) | |
1159 | { | |
1160 | _kvm_arch_hardware_disable(NULL); | |
1161 | } | |
d157f4a5 | 1162 | |
1fcf7ce0 LP |
1163 | #ifdef CONFIG_CPU_PM |
1164 | static int hyp_init_cpu_pm_notifier(struct notifier_block *self, | |
1165 | unsigned long cmd, | |
1166 | void *v) | |
1167 | { | |
67f69197 AT |
1168 | /* |
1169 | * kvm_arm_hardware_enabled is left with its old value over | |
1170 | * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should | |
1171 | * re-enable hyp. | |
1172 | */ | |
1173 | switch (cmd) { | |
1174 | case CPU_PM_ENTER: | |
1175 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1176 | /* | |
1177 | * don't update kvm_arm_hardware_enabled here | |
1178 | * so that the hardware will be re-enabled | |
1179 | * when we resume. See below. | |
1180 | */ | |
1181 | cpu_hyp_reset(); | |
1182 | ||
1fcf7ce0 | 1183 | return NOTIFY_OK; |
67f69197 AT |
1184 | case CPU_PM_EXIT: |
1185 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1186 | /* The hardware was enabled before suspend. */ | |
1187 | cpu_hyp_reinit(); | |
1fcf7ce0 | 1188 | |
67f69197 AT |
1189 | return NOTIFY_OK; |
1190 | ||
1191 | default: | |
1192 | return NOTIFY_DONE; | |
1193 | } | |
1fcf7ce0 LP |
1194 | } |
1195 | ||
1196 | static struct notifier_block hyp_init_cpu_pm_nb = { | |
1197 | .notifier_call = hyp_init_cpu_pm_notifier, | |
1198 | }; | |
1199 | ||
1200 | static void __init hyp_cpu_pm_init(void) | |
1201 | { | |
1202 | cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); | |
1203 | } | |
06a71a24 SH |
1204 | static void __init hyp_cpu_pm_exit(void) |
1205 | { | |
1206 | cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); | |
1207 | } | |
1fcf7ce0 LP |
1208 | #else |
1209 | static inline void hyp_cpu_pm_init(void) | |
1210 | { | |
1211 | } | |
06a71a24 SH |
1212 | static inline void hyp_cpu_pm_exit(void) |
1213 | { | |
1214 | } | |
1fcf7ce0 LP |
1215 | #endif |
1216 | ||
1e947bad MZ |
1217 | static void teardown_common_resources(void) |
1218 | { | |
1219 | free_percpu(kvm_host_cpu_state); | |
1220 | } | |
1221 | ||
1222 | static int init_common_resources(void) | |
1223 | { | |
1224 | kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t); | |
1225 | if (!kvm_host_cpu_state) { | |
1226 | kvm_err("Cannot allocate host CPU state\n"); | |
1227 | return -ENOMEM; | |
1228 | } | |
1229 | ||
61349937 VM |
1230 | /* set size of VMID supported by CPU */ |
1231 | kvm_vmid_bits = kvm_get_vmid_bits(); | |
1232 | kvm_info("%d-bit VMID\n", kvm_vmid_bits); | |
1233 | ||
1e947bad MZ |
1234 | return 0; |
1235 | } | |
1236 | ||
1237 | static int init_subsystems(void) | |
1238 | { | |
67f69197 | 1239 | int err = 0; |
1e947bad | 1240 | |
5f5560b1 | 1241 | /* |
67f69197 | 1242 | * Enable hardware so that subsystem initialisation can access EL2. |
5f5560b1 | 1243 | */ |
67f69197 | 1244 | on_each_cpu(_kvm_arch_hardware_enable, NULL, 1); |
5f5560b1 JM |
1245 | |
1246 | /* | |
1247 | * Register CPU lower-power notifier | |
1248 | */ | |
1249 | hyp_cpu_pm_init(); | |
1250 | ||
1e947bad MZ |
1251 | /* |
1252 | * Init HYP view of VGIC | |
1253 | */ | |
1254 | err = kvm_vgic_hyp_init(); | |
1255 | switch (err) { | |
1256 | case 0: | |
1257 | vgic_present = true; | |
1258 | break; | |
1259 | case -ENODEV: | |
1260 | case -ENXIO: | |
1261 | vgic_present = false; | |
67f69197 | 1262 | err = 0; |
1e947bad MZ |
1263 | break; |
1264 | default: | |
67f69197 | 1265 | goto out; |
1e947bad MZ |
1266 | } |
1267 | ||
1268 | /* | |
1269 | * Init HYP architected timer support | |
1270 | */ | |
1271 | err = kvm_timer_hyp_init(); | |
1272 | if (err) | |
67f69197 | 1273 | goto out; |
1e947bad MZ |
1274 | |
1275 | kvm_perf_init(); | |
1276 | kvm_coproc_table_init(); | |
1277 | ||
67f69197 AT |
1278 | out: |
1279 | on_each_cpu(_kvm_arch_hardware_disable, NULL, 1); | |
1280 | ||
1281 | return err; | |
1e947bad MZ |
1282 | } |
1283 | ||
1284 | static void teardown_hyp_mode(void) | |
1285 | { | |
1286 | int cpu; | |
1287 | ||
1288 | if (is_kernel_in_hyp_mode()) | |
1289 | return; | |
1290 | ||
1291 | free_hyp_pgds(); | |
1292 | for_each_possible_cpu(cpu) | |
1293 | free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); | |
06a71a24 | 1294 | hyp_cpu_pm_exit(); |
1e947bad MZ |
1295 | } |
1296 | ||
1297 | static int init_vhe_mode(void) | |
1298 | { | |
1e947bad MZ |
1299 | kvm_info("VHE mode initialized successfully\n"); |
1300 | return 0; | |
1301 | } | |
1302 | ||
342cd0ab CD |
1303 | /** |
1304 | * Inits Hyp-mode on all online CPUs | |
1305 | */ | |
1306 | static int init_hyp_mode(void) | |
1307 | { | |
342cd0ab CD |
1308 | int cpu; |
1309 | int err = 0; | |
1310 | ||
1311 | /* | |
1312 | * Allocate Hyp PGD and setup Hyp identity mapping | |
1313 | */ | |
1314 | err = kvm_mmu_init(); | |
1315 | if (err) | |
1316 | goto out_err; | |
1317 | ||
1318 | /* | |
1319 | * It is probably enough to obtain the default on one | |
1320 | * CPU. It's unlikely to be different on the others. | |
1321 | */ | |
1322 | hyp_default_vectors = __hyp_get_vectors(); | |
1323 | ||
1324 | /* | |
1325 | * Allocate stack pages for Hypervisor-mode | |
1326 | */ | |
1327 | for_each_possible_cpu(cpu) { | |
1328 | unsigned long stack_page; | |
1329 | ||
1330 | stack_page = __get_free_page(GFP_KERNEL); | |
1331 | if (!stack_page) { | |
1332 | err = -ENOMEM; | |
1e947bad | 1333 | goto out_err; |
342cd0ab CD |
1334 | } |
1335 | ||
1336 | per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; | |
1337 | } | |
1338 | ||
342cd0ab CD |
1339 | /* |
1340 | * Map the Hyp-code called directly from the host | |
1341 | */ | |
588ab3f9 | 1342 | err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), |
59002705 | 1343 | kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC); |
342cd0ab CD |
1344 | if (err) { |
1345 | kvm_err("Cannot map world-switch code\n"); | |
1e947bad | 1346 | goto out_err; |
342cd0ab CD |
1347 | } |
1348 | ||
a0bf9776 | 1349 | err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), |
74a6b888 | 1350 | kvm_ksym_ref(__end_rodata), PAGE_HYP_RO); |
910917bb MZ |
1351 | if (err) { |
1352 | kvm_err("Cannot map rodata section\n"); | |
c8ea0395 MZ |
1353 | goto out_err; |
1354 | } | |
1355 | ||
1356 | err = create_hyp_mappings(kvm_ksym_ref(__bss_start), | |
1357 | kvm_ksym_ref(__bss_stop), PAGE_HYP_RO); | |
1358 | if (err) { | |
1359 | kvm_err("Cannot map bss section\n"); | |
1e947bad | 1360 | goto out_err; |
910917bb MZ |
1361 | } |
1362 | ||
342cd0ab CD |
1363 | /* |
1364 | * Map the Hyp stack pages | |
1365 | */ | |
1366 | for_each_possible_cpu(cpu) { | |
1367 | char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); | |
c8dddecd MZ |
1368 | err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE, |
1369 | PAGE_HYP); | |
342cd0ab CD |
1370 | |
1371 | if (err) { | |
1372 | kvm_err("Cannot map hyp stack\n"); | |
1e947bad | 1373 | goto out_err; |
342cd0ab CD |
1374 | } |
1375 | } | |
1376 | ||
342cd0ab | 1377 | for_each_possible_cpu(cpu) { |
3de50da6 | 1378 | kvm_cpu_context_t *cpu_ctxt; |
342cd0ab | 1379 | |
3de50da6 | 1380 | cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu); |
c8dddecd | 1381 | err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP); |
342cd0ab CD |
1382 | |
1383 | if (err) { | |
3de50da6 | 1384 | kvm_err("Cannot map host CPU state: %d\n", err); |
1e947bad | 1385 | goto out_err; |
342cd0ab CD |
1386 | } |
1387 | } | |
1388 | ||
1389 | kvm_info("Hyp mode initialized successfully\n"); | |
210552c1 | 1390 | |
342cd0ab | 1391 | return 0; |
1e947bad | 1392 | |
342cd0ab | 1393 | out_err: |
1e947bad | 1394 | teardown_hyp_mode(); |
342cd0ab CD |
1395 | kvm_err("error initializing Hyp mode: %d\n", err); |
1396 | return err; | |
1397 | } | |
1398 | ||
d4e071ce AP |
1399 | static void check_kvm_target_cpu(void *ret) |
1400 | { | |
1401 | *(int *)ret = kvm_target_cpu(); | |
1402 | } | |
1403 | ||
4429fc64 AP |
1404 | struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) |
1405 | { | |
1406 | struct kvm_vcpu *vcpu; | |
1407 | int i; | |
1408 | ||
1409 | mpidr &= MPIDR_HWID_BITMASK; | |
1410 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
1411 | if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) | |
1412 | return vcpu; | |
1413 | } | |
1414 | return NULL; | |
1415 | } | |
1416 | ||
342cd0ab CD |
1417 | /** |
1418 | * Initialize Hyp-mode and memory mappings on all CPUs. | |
1419 | */ | |
749cf76c CD |
1420 | int kvm_arch_init(void *opaque) |
1421 | { | |
342cd0ab | 1422 | int err; |
d4e071ce | 1423 | int ret, cpu; |
342cd0ab CD |
1424 | |
1425 | if (!is_hyp_mode_available()) { | |
1426 | kvm_err("HYP mode not available\n"); | |
1427 | return -ENODEV; | |
1428 | } | |
1429 | ||
d4e071ce AP |
1430 | for_each_online_cpu(cpu) { |
1431 | smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); | |
1432 | if (ret < 0) { | |
1433 | kvm_err("Error, CPU %d not supported!\n", cpu); | |
1434 | return -ENODEV; | |
1435 | } | |
342cd0ab CD |
1436 | } |
1437 | ||
1e947bad | 1438 | err = init_common_resources(); |
342cd0ab | 1439 | if (err) |
1e947bad | 1440 | return err; |
342cd0ab | 1441 | |
1e947bad MZ |
1442 | if (is_kernel_in_hyp_mode()) |
1443 | err = init_vhe_mode(); | |
1444 | else | |
1445 | err = init_hyp_mode(); | |
1446 | if (err) | |
d157f4a5 | 1447 | goto out_err; |
8146875d | 1448 | |
1e947bad MZ |
1449 | err = init_subsystems(); |
1450 | if (err) | |
1451 | goto out_hyp; | |
1fcf7ce0 | 1452 | |
749cf76c | 1453 | return 0; |
1e947bad MZ |
1454 | |
1455 | out_hyp: | |
1456 | teardown_hyp_mode(); | |
342cd0ab | 1457 | out_err: |
1e947bad | 1458 | teardown_common_resources(); |
342cd0ab | 1459 | return err; |
749cf76c CD |
1460 | } |
1461 | ||
1462 | /* NOP: Compiling as a module not supported */ | |
1463 | void kvm_arch_exit(void) | |
1464 | { | |
210552c1 | 1465 | kvm_perf_teardown(); |
749cf76c CD |
1466 | } |
1467 | ||
1468 | static int arm_init(void) | |
1469 | { | |
1470 | int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1471 | return rc; | |
1472 | } | |
1473 | ||
1474 | module_init(arm_init); |