]>
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 | 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 | ||
67f69197 AT |
68 | static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled); |
69 | ||
1638a12d MZ |
70 | static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu) |
71 | { | |
72 | BUG_ON(preemptible()); | |
1436c1aa | 73 | __this_cpu_write(kvm_arm_running_vcpu, vcpu); |
1638a12d MZ |
74 | } |
75 | ||
76 | /** | |
77 | * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU. | |
78 | * Must be called from non-preemptible context | |
79 | */ | |
80 | struct kvm_vcpu *kvm_arm_get_running_vcpu(void) | |
81 | { | |
82 | BUG_ON(preemptible()); | |
1436c1aa | 83 | return __this_cpu_read(kvm_arm_running_vcpu); |
1638a12d MZ |
84 | } |
85 | ||
86 | /** | |
87 | * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
88 | */ | |
4000be42 | 89 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) |
1638a12d MZ |
90 | { |
91 | return &kvm_arm_running_vcpu; | |
92 | } | |
93 | ||
749cf76c CD |
94 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
95 | { | |
96 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; | |
97 | } | |
98 | ||
749cf76c CD |
99 | int kvm_arch_hardware_setup(void) |
100 | { | |
101 | return 0; | |
102 | } | |
103 | ||
749cf76c CD |
104 | void kvm_arch_check_processor_compat(void *rtn) |
105 | { | |
106 | *(int *)rtn = 0; | |
107 | } | |
108 | ||
749cf76c | 109 | |
d5d8184d CD |
110 | /** |
111 | * kvm_arch_init_vm - initializes a VM data structure | |
112 | * @kvm: pointer to the KVM struct | |
113 | */ | |
749cf76c CD |
114 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
115 | { | |
94d0e598 | 116 | int ret, cpu; |
d5d8184d | 117 | |
749cf76c CD |
118 | if (type) |
119 | return -EINVAL; | |
120 | ||
94d0e598 MZ |
121 | kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran)); |
122 | if (!kvm->arch.last_vcpu_ran) | |
123 | return -ENOMEM; | |
124 | ||
125 | for_each_possible_cpu(cpu) | |
126 | *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1; | |
127 | ||
d5d8184d CD |
128 | ret = kvm_alloc_stage2_pgd(kvm); |
129 | if (ret) | |
130 | goto out_fail_alloc; | |
131 | ||
c8dddecd | 132 | ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP); |
d5d8184d CD |
133 | if (ret) |
134 | goto out_free_stage2_pgd; | |
135 | ||
6c3d63c9 | 136 | kvm_vgic_early_init(kvm); |
a1a64387 | 137 | |
d5d8184d CD |
138 | /* Mark the initial VMID generation invalid */ |
139 | kvm->arch.vmid_gen = 0; | |
140 | ||
3caa2d8c | 141 | /* The maximum number of VCPUs is limited by the host's GIC model */ |
c7da6fa4 PF |
142 | kvm->arch.max_vcpus = vgic_present ? |
143 | kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; | |
3caa2d8c | 144 | |
d5d8184d CD |
145 | return ret; |
146 | out_free_stage2_pgd: | |
147 | kvm_free_stage2_pgd(kvm); | |
148 | out_fail_alloc: | |
94d0e598 MZ |
149 | free_percpu(kvm->arch.last_vcpu_ran); |
150 | kvm->arch.last_vcpu_ran = NULL; | |
d5d8184d | 151 | return ret; |
749cf76c CD |
152 | } |
153 | ||
235539b4 LC |
154 | bool kvm_arch_has_vcpu_debugfs(void) |
155 | { | |
156 | return false; | |
157 | } | |
158 | ||
159 | int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu) | |
160 | { | |
161 | return 0; | |
162 | } | |
163 | ||
749cf76c CD |
164 | int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
165 | { | |
166 | return VM_FAULT_SIGBUS; | |
167 | } | |
168 | ||
749cf76c | 169 | |
d5d8184d CD |
170 | /** |
171 | * kvm_arch_destroy_vm - destroy the VM data structure | |
172 | * @kvm: pointer to the KVM struct | |
173 | */ | |
749cf76c CD |
174 | void kvm_arch_destroy_vm(struct kvm *kvm) |
175 | { | |
176 | int i; | |
177 | ||
94d0e598 MZ |
178 | free_percpu(kvm->arch.last_vcpu_ran); |
179 | kvm->arch.last_vcpu_ran = NULL; | |
180 | ||
749cf76c CD |
181 | for (i = 0; i < KVM_MAX_VCPUS; ++i) { |
182 | if (kvm->vcpus[i]) { | |
183 | kvm_arch_vcpu_free(kvm->vcpus[i]); | |
184 | kvm->vcpus[i] = NULL; | |
185 | } | |
186 | } | |
c1bfb577 MZ |
187 | |
188 | kvm_vgic_destroy(kvm); | |
749cf76c CD |
189 | } |
190 | ||
784aa3d7 | 191 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
749cf76c CD |
192 | { |
193 | int r; | |
194 | switch (ext) { | |
1a89dd91 | 195 | case KVM_CAP_IRQCHIP: |
c7da6fa4 PF |
196 | r = vgic_present; |
197 | break; | |
d44758c0 | 198 | case KVM_CAP_IOEVENTFD: |
7330672b | 199 | case KVM_CAP_DEVICE_CTRL: |
749cf76c CD |
200 | case KVM_CAP_USER_MEMORY: |
201 | case KVM_CAP_SYNC_MMU: | |
202 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
203 | case KVM_CAP_ONE_REG: | |
aa024c2f | 204 | case KVM_CAP_ARM_PSCI: |
4447a208 | 205 | case KVM_CAP_ARM_PSCI_0_2: |
98047888 | 206 | case KVM_CAP_READONLY_MEM: |
ecccf0cc | 207 | case KVM_CAP_MP_STATE: |
460df4c1 | 208 | case KVM_CAP_IMMEDIATE_EXIT: |
749cf76c CD |
209 | r = 1; |
210 | break; | |
3401d546 CD |
211 | case KVM_CAP_ARM_SET_DEVICE_ADDR: |
212 | r = 1; | |
ca46e10f | 213 | break; |
749cf76c CD |
214 | case KVM_CAP_NR_VCPUS: |
215 | r = num_online_cpus(); | |
216 | break; | |
217 | case KVM_CAP_MAX_VCPUS: | |
218 | r = KVM_MAX_VCPUS; | |
219 | break; | |
7af4df85 LC |
220 | case KVM_CAP_NR_MEMSLOTS: |
221 | r = KVM_USER_MEM_SLOTS; | |
222 | break; | |
2988509d VM |
223 | case KVM_CAP_MSI_DEVID: |
224 | if (!kvm) | |
225 | r = -EINVAL; | |
226 | else | |
227 | r = kvm->arch.vgic.msis_require_devid; | |
228 | break; | |
f7214e60 CD |
229 | case KVM_CAP_ARM_USER_IRQ: |
230 | /* | |
231 | * 1: EL1_VTIMER, EL1_PTIMER, and PMU. | |
232 | * (bump this number if adding more devices) | |
233 | */ | |
234 | r = 1; | |
235 | break; | |
749cf76c | 236 | default: |
b46f01ce | 237 | r = kvm_arch_dev_ioctl_check_extension(kvm, ext); |
749cf76c CD |
238 | break; |
239 | } | |
240 | return r; | |
241 | } | |
242 | ||
243 | long kvm_arch_dev_ioctl(struct file *filp, | |
244 | unsigned int ioctl, unsigned long arg) | |
245 | { | |
246 | return -EINVAL; | |
247 | } | |
248 | ||
749cf76c CD |
249 | |
250 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) | |
251 | { | |
252 | int err; | |
253 | struct kvm_vcpu *vcpu; | |
254 | ||
716139df CD |
255 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) { |
256 | err = -EBUSY; | |
257 | goto out; | |
258 | } | |
259 | ||
3caa2d8c AP |
260 | if (id >= kvm->arch.max_vcpus) { |
261 | err = -EINVAL; | |
262 | goto out; | |
263 | } | |
264 | ||
749cf76c CD |
265 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
266 | if (!vcpu) { | |
267 | err = -ENOMEM; | |
268 | goto out; | |
269 | } | |
270 | ||
271 | err = kvm_vcpu_init(vcpu, kvm, id); | |
272 | if (err) | |
273 | goto free_vcpu; | |
274 | ||
c8dddecd | 275 | err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP); |
d5d8184d CD |
276 | if (err) |
277 | goto vcpu_uninit; | |
278 | ||
749cf76c | 279 | return vcpu; |
d5d8184d CD |
280 | vcpu_uninit: |
281 | kvm_vcpu_uninit(vcpu); | |
749cf76c CD |
282 | free_vcpu: |
283 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
284 | out: | |
285 | return ERR_PTR(err); | |
286 | } | |
287 | ||
31928aa5 | 288 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
749cf76c | 289 | { |
6c3d63c9 | 290 | kvm_vgic_vcpu_early_init(vcpu); |
749cf76c CD |
291 | } |
292 | ||
293 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
294 | { | |
d5d8184d | 295 | kvm_mmu_free_memory_caches(vcpu); |
967f8427 | 296 | kvm_timer_vcpu_terminate(vcpu); |
c1bfb577 | 297 | kvm_vgic_vcpu_destroy(vcpu); |
5f0a714a | 298 | kvm_pmu_vcpu_destroy(vcpu); |
591d215a | 299 | kvm_vcpu_uninit(vcpu); |
d5d8184d | 300 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
749cf76c CD |
301 | } |
302 | ||
303 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) | |
304 | { | |
305 | kvm_arch_vcpu_free(vcpu); | |
306 | } | |
307 | ||
308 | int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) | |
309 | { | |
fb280e97 JL |
310 | return kvm_timer_should_fire(vcpu_vtimer(vcpu)) || |
311 | kvm_timer_should_fire(vcpu_ptimer(vcpu)); | |
749cf76c CD |
312 | } |
313 | ||
d35268da CD |
314 | void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) |
315 | { | |
316 | kvm_timer_schedule(vcpu); | |
317 | } | |
318 | ||
319 | void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
320 | { | |
321 | kvm_timer_unschedule(vcpu); | |
322 | } | |
323 | ||
749cf76c CD |
324 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
325 | { | |
f7ed45be CD |
326 | /* Force users to call KVM_ARM_VCPU_INIT */ |
327 | vcpu->arch.target = -1; | |
f7fa034d | 328 | bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); |
1a89dd91 | 329 | |
967f8427 MZ |
330 | /* Set up the timer */ |
331 | kvm_timer_vcpu_init(vcpu); | |
332 | ||
84e690bf AB |
333 | kvm_arm_reset_debug_ptr(vcpu); |
334 | ||
1aab6f46 | 335 | return kvm_vgic_vcpu_init(vcpu); |
749cf76c CD |
336 | } |
337 | ||
749cf76c CD |
338 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
339 | { | |
94d0e598 MZ |
340 | int *last_ran; |
341 | ||
342 | last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran); | |
343 | ||
344 | /* | |
345 | * We might get preempted before the vCPU actually runs, but | |
346 | * over-invalidation doesn't affect correctness. | |
347 | */ | |
348 | if (*last_ran != vcpu->vcpu_id) { | |
349 | kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu); | |
350 | *last_ran = vcpu->vcpu_id; | |
351 | } | |
352 | ||
86ce8535 | 353 | vcpu->cpu = cpu; |
3de50da6 | 354 | vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state); |
5b3e5e5b | 355 | |
1638a12d | 356 | kvm_arm_set_running_vcpu(vcpu); |
328e5664 CD |
357 | |
358 | kvm_vgic_load(vcpu); | |
749cf76c CD |
359 | } |
360 | ||
361 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
362 | { | |
328e5664 CD |
363 | kvm_vgic_put(vcpu); |
364 | ||
e9b152cb CD |
365 | vcpu->cpu = -1; |
366 | ||
1638a12d | 367 | kvm_arm_set_running_vcpu(NULL); |
9b4a3004 | 368 | kvm_timer_vcpu_put(vcpu); |
749cf76c CD |
369 | } |
370 | ||
424c989b AJ |
371 | static void vcpu_power_off(struct kvm_vcpu *vcpu) |
372 | { | |
373 | vcpu->arch.power_off = true; | |
7b244e2b | 374 | kvm_make_request(KVM_REQ_SLEEP, vcpu); |
424c989b AJ |
375 | kvm_vcpu_kick(vcpu); |
376 | } | |
377 | ||
749cf76c CD |
378 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
379 | struct kvm_mp_state *mp_state) | |
380 | { | |
3781528e | 381 | if (vcpu->arch.power_off) |
ecccf0cc AB |
382 | mp_state->mp_state = KVM_MP_STATE_STOPPED; |
383 | else | |
384 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
385 | ||
386 | return 0; | |
749cf76c CD |
387 | } |
388 | ||
389 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
390 | struct kvm_mp_state *mp_state) | |
391 | { | |
ecccf0cc AB |
392 | switch (mp_state->mp_state) { |
393 | case KVM_MP_STATE_RUNNABLE: | |
3781528e | 394 | vcpu->arch.power_off = false; |
ecccf0cc AB |
395 | break; |
396 | case KVM_MP_STATE_STOPPED: | |
424c989b | 397 | vcpu_power_off(vcpu); |
ecccf0cc AB |
398 | break; |
399 | default: | |
400 | return -EINVAL; | |
401 | } | |
402 | ||
403 | return 0; | |
749cf76c CD |
404 | } |
405 | ||
5b3e5e5b CD |
406 | /** |
407 | * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled | |
408 | * @v: The VCPU pointer | |
409 | * | |
410 | * If the guest CPU is not waiting for interrupts or an interrupt line is | |
411 | * asserted, the CPU is by definition runnable. | |
412 | */ | |
749cf76c CD |
413 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) |
414 | { | |
4f5f1dc0 | 415 | return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v)) |
3b92830a | 416 | && !v->arch.power_off && !v->arch.pause); |
749cf76c CD |
417 | } |
418 | ||
f7ed45be CD |
419 | /* Just ensure a guest exit from a particular CPU */ |
420 | static void exit_vm_noop(void *info) | |
421 | { | |
422 | } | |
423 | ||
424 | void force_vm_exit(const cpumask_t *mask) | |
425 | { | |
898f949f | 426 | preempt_disable(); |
f7ed45be | 427 | smp_call_function_many(mask, exit_vm_noop, NULL, true); |
898f949f | 428 | preempt_enable(); |
f7ed45be CD |
429 | } |
430 | ||
431 | /** | |
432 | * need_new_vmid_gen - check that the VMID is still valid | |
6a727b0b | 433 | * @kvm: The VM's VMID to check |
f7ed45be CD |
434 | * |
435 | * return true if there is a new generation of VMIDs being used | |
436 | * | |
437 | * The hardware supports only 256 values with the value zero reserved for the | |
438 | * host, so we check if an assigned value belongs to a previous generation, | |
439 | * which which requires us to assign a new value. If we're the first to use a | |
440 | * VMID for the new generation, we must flush necessary caches and TLBs on all | |
441 | * CPUs. | |
442 | */ | |
443 | static bool need_new_vmid_gen(struct kvm *kvm) | |
444 | { | |
445 | return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen)); | |
446 | } | |
447 | ||
448 | /** | |
449 | * update_vttbr - Update the VTTBR with a valid VMID before the guest runs | |
450 | * @kvm The guest that we are about to run | |
451 | * | |
452 | * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the | |
453 | * VM has a valid VMID, otherwise assigns a new one and flushes corresponding | |
454 | * caches and TLBs. | |
455 | */ | |
456 | static void update_vttbr(struct kvm *kvm) | |
457 | { | |
458 | phys_addr_t pgd_phys; | |
459 | u64 vmid; | |
460 | ||
461 | if (!need_new_vmid_gen(kvm)) | |
462 | return; | |
463 | ||
464 | spin_lock(&kvm_vmid_lock); | |
465 | ||
466 | /* | |
467 | * We need to re-check the vmid_gen here to ensure that if another vcpu | |
468 | * already allocated a valid vmid for this vm, then this vcpu should | |
469 | * use the same vmid. | |
470 | */ | |
471 | if (!need_new_vmid_gen(kvm)) { | |
472 | spin_unlock(&kvm_vmid_lock); | |
473 | return; | |
474 | } | |
475 | ||
476 | /* First user of a new VMID generation? */ | |
477 | if (unlikely(kvm_next_vmid == 0)) { | |
478 | atomic64_inc(&kvm_vmid_gen); | |
479 | kvm_next_vmid = 1; | |
480 | ||
481 | /* | |
482 | * On SMP we know no other CPUs can use this CPU's or each | |
483 | * other's VMID after force_vm_exit returns since the | |
484 | * kvm_vmid_lock blocks them from reentry to the guest. | |
485 | */ | |
486 | force_vm_exit(cpu_all_mask); | |
487 | /* | |
488 | * Now broadcast TLB + ICACHE invalidation over the inner | |
489 | * shareable domain to make sure all data structures are | |
490 | * clean. | |
491 | */ | |
492 | kvm_call_hyp(__kvm_flush_vm_context); | |
493 | } | |
494 | ||
495 | kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen); | |
496 | kvm->arch.vmid = kvm_next_vmid; | |
497 | kvm_next_vmid++; | |
20475f78 | 498 | kvm_next_vmid &= (1 << kvm_vmid_bits) - 1; |
f7ed45be CD |
499 | |
500 | /* update vttbr to be used with the new vmid */ | |
9163ee23 | 501 | pgd_phys = virt_to_phys(kvm->arch.pgd); |
dbff124e | 502 | BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK); |
20475f78 | 503 | vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits); |
dbff124e | 504 | kvm->arch.vttbr = pgd_phys | vmid; |
f7ed45be CD |
505 | |
506 | spin_unlock(&kvm_vmid_lock); | |
507 | } | |
508 | ||
f7ed45be CD |
509 | static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) |
510 | { | |
05971120 | 511 | struct kvm *kvm = vcpu->kvm; |
41a54482 | 512 | int ret = 0; |
e1ba0207 | 513 | |
f7ed45be CD |
514 | if (likely(vcpu->arch.has_run_once)) |
515 | return 0; | |
516 | ||
517 | vcpu->arch.has_run_once = true; | |
aa024c2f | 518 | |
01ac5e34 | 519 | /* |
6d3cfbe2 PM |
520 | * Map the VGIC hardware resources before running a vcpu the first |
521 | * time on this VM. | |
01ac5e34 | 522 | */ |
c2f58514 | 523 | if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) { |
05971120 | 524 | ret = kvm_vgic_map_resources(kvm); |
01ac5e34 MZ |
525 | if (ret) |
526 | return ret; | |
527 | } | |
528 | ||
d9e13977 | 529 | ret = kvm_timer_enable(vcpu); |
a2befacf CD |
530 | if (ret) |
531 | return ret; | |
532 | ||
533 | ret = kvm_arm_pmu_v3_enable(vcpu); | |
05971120 | 534 | |
41a54482 | 535 | return ret; |
f7ed45be CD |
536 | } |
537 | ||
c1426e4c EA |
538 | bool kvm_arch_intc_initialized(struct kvm *kvm) |
539 | { | |
540 | return vgic_initialized(kvm); | |
541 | } | |
542 | ||
b13216cf | 543 | void kvm_arm_halt_guest(struct kvm *kvm) |
3b92830a EA |
544 | { |
545 | int i; | |
546 | struct kvm_vcpu *vcpu; | |
547 | ||
548 | kvm_for_each_vcpu(i, vcpu, kvm) | |
549 | vcpu->arch.pause = true; | |
7b244e2b | 550 | kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP); |
3b92830a EA |
551 | } |
552 | ||
b13216cf | 553 | void kvm_arm_resume_guest(struct kvm *kvm) |
3b92830a EA |
554 | { |
555 | int i; | |
556 | struct kvm_vcpu *vcpu; | |
557 | ||
abd72296 CD |
558 | kvm_for_each_vcpu(i, vcpu, kvm) { |
559 | vcpu->arch.pause = false; | |
560 | swake_up(kvm_arch_vcpu_wq(vcpu)); | |
561 | } | |
3b92830a EA |
562 | } |
563 | ||
7b244e2b | 564 | static void vcpu_req_sleep(struct kvm_vcpu *vcpu) |
aa024c2f | 565 | { |
8577370f | 566 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); |
aa024c2f | 567 | |
8577370f | 568 | swait_event_interruptible(*wq, ((!vcpu->arch.power_off) && |
3b92830a | 569 | (!vcpu->arch.pause))); |
0592c005 | 570 | |
424c989b | 571 | if (vcpu->arch.power_off || vcpu->arch.pause) { |
0592c005 | 572 | /* Awaken to handle a signal, request we sleep again later. */ |
7b244e2b | 573 | kvm_make_request(KVM_REQ_SLEEP, vcpu); |
0592c005 | 574 | } |
aa024c2f MZ |
575 | } |
576 | ||
e8180dca AP |
577 | static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) |
578 | { | |
579 | return vcpu->arch.target >= 0; | |
580 | } | |
581 | ||
0592c005 AJ |
582 | static void check_vcpu_requests(struct kvm_vcpu *vcpu) |
583 | { | |
584 | if (kvm_request_pending(vcpu)) { | |
7b244e2b AJ |
585 | if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) |
586 | vcpu_req_sleep(vcpu); | |
325f9c64 AJ |
587 | |
588 | /* | |
589 | * Clear IRQ_PENDING requests that were made to guarantee | |
590 | * that a VCPU sees new virtual interrupts. | |
591 | */ | |
592 | kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu); | |
0592c005 AJ |
593 | } |
594 | } | |
595 | ||
f7ed45be CD |
596 | /** |
597 | * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code | |
598 | * @vcpu: The VCPU pointer | |
599 | * @run: The kvm_run structure pointer used for userspace state exchange | |
600 | * | |
601 | * This function is called through the VCPU_RUN ioctl called from user space. It | |
602 | * will execute VM code in a loop until the time slice for the process is used | |
603 | * or some emulation is needed from user space in which case the function will | |
604 | * return with return value 0 and with the kvm_run structure filled in with the | |
605 | * required data for the requested emulation. | |
606 | */ | |
749cf76c CD |
607 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) |
608 | { | |
f7ed45be CD |
609 | int ret; |
610 | sigset_t sigsaved; | |
611 | ||
e8180dca | 612 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
f7ed45be CD |
613 | return -ENOEXEC; |
614 | ||
615 | ret = kvm_vcpu_first_run_init(vcpu); | |
616 | if (ret) | |
617 | return ret; | |
618 | ||
45e96ea6 CD |
619 | if (run->exit_reason == KVM_EXIT_MMIO) { |
620 | ret = kvm_handle_mmio_return(vcpu, vcpu->run); | |
621 | if (ret) | |
622 | return ret; | |
623 | } | |
624 | ||
460df4c1 PB |
625 | if (run->immediate_exit) |
626 | return -EINTR; | |
627 | ||
f7ed45be CD |
628 | if (vcpu->sigset_active) |
629 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
630 | ||
631 | ret = 1; | |
632 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
633 | while (ret > 0) { | |
634 | /* | |
635 | * Check conditions before entering the guest | |
636 | */ | |
637 | cond_resched(); | |
638 | ||
639 | update_vttbr(vcpu->kvm); | |
640 | ||
0592c005 AJ |
641 | check_vcpu_requests(vcpu); |
642 | ||
abdf5843 MZ |
643 | /* |
644 | * Preparing the interrupts to be injected also | |
645 | * involves poking the GIC, which must be done in a | |
646 | * non-preemptible context. | |
647 | */ | |
1b3d546d | 648 | preempt_disable(); |
328e5664 | 649 | |
b02386eb | 650 | kvm_pmu_flush_hwstate(vcpu); |
328e5664 | 651 | |
7e16aa81 | 652 | kvm_timer_flush_hwstate(vcpu); |
abdf5843 MZ |
653 | kvm_vgic_flush_hwstate(vcpu); |
654 | ||
f7ed45be CD |
655 | local_irq_disable(); |
656 | ||
657 | /* | |
d9e13977 | 658 | * If we have a singal pending, or need to notify a userspace |
3dbbdf78 CD |
659 | * irqchip about timer or PMU level changes, then we exit (and |
660 | * update the timer level state in kvm_timer_update_run | |
661 | * below). | |
f7ed45be | 662 | */ |
d9e13977 | 663 | if (signal_pending(current) || |
3dbbdf78 CD |
664 | kvm_timer_should_notify_user(vcpu) || |
665 | kvm_pmu_should_notify_user(vcpu)) { | |
f7ed45be CD |
666 | ret = -EINTR; |
667 | run->exit_reason = KVM_EXIT_INTR; | |
668 | } | |
669 | ||
6a6d73be AJ |
670 | /* |
671 | * Ensure we set mode to IN_GUEST_MODE after we disable | |
672 | * interrupts and before the final VCPU requests check. | |
673 | * See the comment in kvm_vcpu_exiting_guest_mode() and | |
674 | * Documentation/virtual/kvm/vcpu-requests.rst | |
675 | */ | |
676 | smp_store_mb(vcpu->mode, IN_GUEST_MODE); | |
677 | ||
101d3da0 | 678 | if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) || |
424c989b | 679 | kvm_request_pending(vcpu)) { |
6a6d73be | 680 | vcpu->mode = OUTSIDE_GUEST_MODE; |
f7ed45be | 681 | local_irq_enable(); |
b02386eb | 682 | kvm_pmu_sync_hwstate(vcpu); |
4b4b4512 | 683 | kvm_timer_sync_hwstate(vcpu); |
1a89dd91 | 684 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 | 685 | preempt_enable(); |
f7ed45be CD |
686 | continue; |
687 | } | |
688 | ||
56c7f5e7 AB |
689 | kvm_arm_setup_debug(vcpu); |
690 | ||
f7ed45be CD |
691 | /************************************************************** |
692 | * Enter the guest | |
693 | */ | |
694 | trace_kvm_entry(*vcpu_pc(vcpu)); | |
6edaa530 | 695 | guest_enter_irqoff(); |
f7ed45be CD |
696 | |
697 | ret = kvm_call_hyp(__kvm_vcpu_run, vcpu); | |
698 | ||
699 | vcpu->mode = OUTSIDE_GUEST_MODE; | |
b19e6892 | 700 | vcpu->stat.exits++; |
1b3d546d CD |
701 | /* |
702 | * Back from guest | |
703 | *************************************************************/ | |
704 | ||
56c7f5e7 AB |
705 | kvm_arm_clear_debug(vcpu); |
706 | ||
f7ed45be CD |
707 | /* |
708 | * We may have taken a host interrupt in HYP mode (ie | |
709 | * while executing the guest). This interrupt is still | |
710 | * pending, as we haven't serviced it yet! | |
711 | * | |
712 | * We're now back in SVC mode, with interrupts | |
713 | * disabled. Enabling the interrupts now will have | |
714 | * the effect of taking the interrupt again, in SVC | |
715 | * mode this time. | |
716 | */ | |
717 | local_irq_enable(); | |
718 | ||
719 | /* | |
6edaa530 | 720 | * We do local_irq_enable() before calling guest_exit() so |
1b3d546d CD |
721 | * that if a timer interrupt hits while running the guest we |
722 | * account that tick as being spent in the guest. We enable | |
6edaa530 | 723 | * preemption after calling guest_exit() so that if we get |
1b3d546d CD |
724 | * preempted we make sure ticks after that is not counted as |
725 | * guest time. | |
726 | */ | |
6edaa530 | 727 | guest_exit(); |
b5905dc1 | 728 | trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); |
1b3d546d | 729 | |
4b4b4512 | 730 | /* |
b02386eb SZ |
731 | * We must sync the PMU and timer state before the vgic state so |
732 | * that the vgic can properly sample the updated state of the | |
4b4b4512 CD |
733 | * interrupt line. |
734 | */ | |
b02386eb | 735 | kvm_pmu_sync_hwstate(vcpu); |
4b4b4512 CD |
736 | kvm_timer_sync_hwstate(vcpu); |
737 | ||
1a89dd91 | 738 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 MZ |
739 | |
740 | preempt_enable(); | |
741 | ||
f7ed45be CD |
742 | ret = handle_exit(vcpu, run, ret); |
743 | } | |
744 | ||
d9e13977 | 745 | /* Tell userspace about in-kernel device output levels */ |
3dbbdf78 CD |
746 | if (unlikely(!irqchip_in_kernel(vcpu->kvm))) { |
747 | kvm_timer_update_run(vcpu); | |
748 | kvm_pmu_update_run(vcpu); | |
749 | } | |
d9e13977 | 750 | |
f7ed45be CD |
751 | if (vcpu->sigset_active) |
752 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
753 | return ret; | |
749cf76c CD |
754 | } |
755 | ||
86ce8535 CD |
756 | static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) |
757 | { | |
758 | int bit_index; | |
759 | bool set; | |
760 | unsigned long *ptr; | |
761 | ||
762 | if (number == KVM_ARM_IRQ_CPU_IRQ) | |
763 | bit_index = __ffs(HCR_VI); | |
764 | else /* KVM_ARM_IRQ_CPU_FIQ */ | |
765 | bit_index = __ffs(HCR_VF); | |
766 | ||
767 | ptr = (unsigned long *)&vcpu->arch.irq_lines; | |
768 | if (level) | |
769 | set = test_and_set_bit(bit_index, ptr); | |
770 | else | |
771 | set = test_and_clear_bit(bit_index, ptr); | |
772 | ||
773 | /* | |
774 | * If we didn't change anything, no need to wake up or kick other CPUs | |
775 | */ | |
776 | if (set == level) | |
777 | return 0; | |
778 | ||
779 | /* | |
780 | * The vcpu irq_lines field was updated, wake up sleeping VCPUs and | |
781 | * trigger a world-switch round on the running physical CPU to set the | |
782 | * virtual IRQ/FIQ fields in the HCR appropriately. | |
783 | */ | |
325f9c64 | 784 | kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu); |
86ce8535 CD |
785 | kvm_vcpu_kick(vcpu); |
786 | ||
787 | return 0; | |
788 | } | |
789 | ||
79558f11 AG |
790 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, |
791 | bool line_status) | |
86ce8535 CD |
792 | { |
793 | u32 irq = irq_level->irq; | |
794 | unsigned int irq_type, vcpu_idx, irq_num; | |
795 | int nrcpus = atomic_read(&kvm->online_vcpus); | |
796 | struct kvm_vcpu *vcpu = NULL; | |
797 | bool level = irq_level->level; | |
798 | ||
799 | irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; | |
800 | vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; | |
801 | irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; | |
802 | ||
803 | trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); | |
804 | ||
5863c2ce MZ |
805 | switch (irq_type) { |
806 | case KVM_ARM_IRQ_TYPE_CPU: | |
807 | if (irqchip_in_kernel(kvm)) | |
808 | return -ENXIO; | |
86ce8535 | 809 | |
5863c2ce MZ |
810 | if (vcpu_idx >= nrcpus) |
811 | return -EINVAL; | |
86ce8535 | 812 | |
5863c2ce MZ |
813 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); |
814 | if (!vcpu) | |
815 | return -EINVAL; | |
86ce8535 | 816 | |
5863c2ce MZ |
817 | if (irq_num > KVM_ARM_IRQ_CPU_FIQ) |
818 | return -EINVAL; | |
819 | ||
820 | return vcpu_interrupt_line(vcpu, irq_num, level); | |
821 | case KVM_ARM_IRQ_TYPE_PPI: | |
822 | if (!irqchip_in_kernel(kvm)) | |
823 | return -ENXIO; | |
824 | ||
825 | if (vcpu_idx >= nrcpus) | |
826 | return -EINVAL; | |
827 | ||
828 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); | |
829 | if (!vcpu) | |
830 | return -EINVAL; | |
831 | ||
832 | if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) | |
833 | return -EINVAL; | |
86ce8535 | 834 | |
cb3f0ad8 | 835 | return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL); |
5863c2ce MZ |
836 | case KVM_ARM_IRQ_TYPE_SPI: |
837 | if (!irqchip_in_kernel(kvm)) | |
838 | return -ENXIO; | |
839 | ||
fd1d0ddf | 840 | if (irq_num < VGIC_NR_PRIVATE_IRQS) |
5863c2ce MZ |
841 | return -EINVAL; |
842 | ||
cb3f0ad8 | 843 | return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL); |
5863c2ce MZ |
844 | } |
845 | ||
846 | return -EINVAL; | |
86ce8535 CD |
847 | } |
848 | ||
f7fa034d CD |
849 | static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, |
850 | const struct kvm_vcpu_init *init) | |
851 | { | |
852 | unsigned int i; | |
853 | int phys_target = kvm_target_cpu(); | |
854 | ||
855 | if (init->target != phys_target) | |
856 | return -EINVAL; | |
857 | ||
858 | /* | |
859 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
860 | * use the same target. | |
861 | */ | |
862 | if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) | |
863 | return -EINVAL; | |
864 | ||
865 | /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ | |
866 | for (i = 0; i < sizeof(init->features) * 8; i++) { | |
867 | bool set = (init->features[i / 32] & (1 << (i % 32))); | |
868 | ||
869 | if (set && i >= KVM_VCPU_MAX_FEATURES) | |
870 | return -ENOENT; | |
871 | ||
872 | /* | |
873 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
874 | * use the same feature set. | |
875 | */ | |
876 | if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && | |
877 | test_bit(i, vcpu->arch.features) != set) | |
878 | return -EINVAL; | |
879 | ||
880 | if (set) | |
881 | set_bit(i, vcpu->arch.features); | |
882 | } | |
883 | ||
884 | vcpu->arch.target = phys_target; | |
885 | ||
886 | /* Now we know what it is, we can reset it. */ | |
887 | return kvm_reset_vcpu(vcpu); | |
888 | } | |
889 | ||
890 | ||
478a8237 CD |
891 | static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, |
892 | struct kvm_vcpu_init *init) | |
893 | { | |
894 | int ret; | |
895 | ||
896 | ret = kvm_vcpu_set_target(vcpu, init); | |
897 | if (ret) | |
898 | return ret; | |
899 | ||
957db105 CD |
900 | /* |
901 | * Ensure a rebooted VM will fault in RAM pages and detect if the | |
902 | * guest MMU is turned off and flush the caches as needed. | |
903 | */ | |
904 | if (vcpu->arch.has_run_once) | |
905 | stage2_unmap_vm(vcpu->kvm); | |
906 | ||
b856a591 CD |
907 | vcpu_reset_hcr(vcpu); |
908 | ||
478a8237 | 909 | /* |
3781528e | 910 | * Handle the "start in power-off" case. |
478a8237 | 911 | */ |
03f1d4c1 | 912 | if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) |
424c989b | 913 | vcpu_power_off(vcpu); |
3ad8b3de | 914 | else |
3781528e | 915 | vcpu->arch.power_off = false; |
478a8237 CD |
916 | |
917 | return 0; | |
918 | } | |
919 | ||
f577f6c2 SZ |
920 | static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, |
921 | struct kvm_device_attr *attr) | |
922 | { | |
923 | int ret = -ENXIO; | |
924 | ||
925 | switch (attr->group) { | |
926 | default: | |
bb0c70bc | 927 | ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); |
f577f6c2 SZ |
928 | break; |
929 | } | |
930 | ||
931 | return ret; | |
932 | } | |
933 | ||
934 | static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, | |
935 | struct kvm_device_attr *attr) | |
936 | { | |
937 | int ret = -ENXIO; | |
938 | ||
939 | switch (attr->group) { | |
940 | default: | |
bb0c70bc | 941 | ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); |
f577f6c2 SZ |
942 | break; |
943 | } | |
944 | ||
945 | return ret; | |
946 | } | |
947 | ||
948 | static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, | |
949 | struct kvm_device_attr *attr) | |
950 | { | |
951 | int ret = -ENXIO; | |
952 | ||
953 | switch (attr->group) { | |
954 | default: | |
bb0c70bc | 955 | ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); |
f577f6c2 SZ |
956 | break; |
957 | } | |
958 | ||
959 | return ret; | |
960 | } | |
961 | ||
749cf76c CD |
962 | long kvm_arch_vcpu_ioctl(struct file *filp, |
963 | unsigned int ioctl, unsigned long arg) | |
964 | { | |
965 | struct kvm_vcpu *vcpu = filp->private_data; | |
966 | void __user *argp = (void __user *)arg; | |
f577f6c2 | 967 | struct kvm_device_attr attr; |
749cf76c CD |
968 | |
969 | switch (ioctl) { | |
970 | case KVM_ARM_VCPU_INIT: { | |
971 | struct kvm_vcpu_init init; | |
972 | ||
973 | if (copy_from_user(&init, argp, sizeof(init))) | |
974 | return -EFAULT; | |
975 | ||
478a8237 | 976 | return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); |
749cf76c CD |
977 | } |
978 | case KVM_SET_ONE_REG: | |
979 | case KVM_GET_ONE_REG: { | |
980 | struct kvm_one_reg reg; | |
e8180dca AP |
981 | |
982 | if (unlikely(!kvm_vcpu_initialized(vcpu))) | |
983 | return -ENOEXEC; | |
984 | ||
749cf76c CD |
985 | if (copy_from_user(®, argp, sizeof(reg))) |
986 | return -EFAULT; | |
987 | if (ioctl == KVM_SET_ONE_REG) | |
988 | return kvm_arm_set_reg(vcpu, ®); | |
989 | else | |
990 | return kvm_arm_get_reg(vcpu, ®); | |
991 | } | |
992 | case KVM_GET_REG_LIST: { | |
993 | struct kvm_reg_list __user *user_list = argp; | |
994 | struct kvm_reg_list reg_list; | |
995 | unsigned n; | |
996 | ||
e8180dca AP |
997 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
998 | return -ENOEXEC; | |
999 | ||
749cf76c CD |
1000 | if (copy_from_user(®_list, user_list, sizeof(reg_list))) |
1001 | return -EFAULT; | |
1002 | n = reg_list.n; | |
1003 | reg_list.n = kvm_arm_num_regs(vcpu); | |
1004 | if (copy_to_user(user_list, ®_list, sizeof(reg_list))) | |
1005 | return -EFAULT; | |
1006 | if (n < reg_list.n) | |
1007 | return -E2BIG; | |
1008 | return kvm_arm_copy_reg_indices(vcpu, user_list->reg); | |
1009 | } | |
f577f6c2 SZ |
1010 | case KVM_SET_DEVICE_ATTR: { |
1011 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
1012 | return -EFAULT; | |
1013 | return kvm_arm_vcpu_set_attr(vcpu, &attr); | |
1014 | } | |
1015 | case KVM_GET_DEVICE_ATTR: { | |
1016 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
1017 | return -EFAULT; | |
1018 | return kvm_arm_vcpu_get_attr(vcpu, &attr); | |
1019 | } | |
1020 | case KVM_HAS_DEVICE_ATTR: { | |
1021 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
1022 | return -EFAULT; | |
1023 | return kvm_arm_vcpu_has_attr(vcpu, &attr); | |
1024 | } | |
749cf76c CD |
1025 | default: |
1026 | return -EINVAL; | |
1027 | } | |
1028 | } | |
1029 | ||
53c810c3 MS |
1030 | /** |
1031 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
1032 | * @kvm: kvm instance | |
1033 | * @log: slot id and address to which we copy the log | |
1034 | * | |
1035 | * Steps 1-4 below provide general overview of dirty page logging. See | |
1036 | * kvm_get_dirty_log_protect() function description for additional details. | |
1037 | * | |
1038 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
1039 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
1040 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
1041 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
1042 | * writes will be marked dirty for next log read. | |
1043 | * | |
1044 | * 1. Take a snapshot of the bit and clear it if needed. | |
1045 | * 2. Write protect the corresponding page. | |
1046 | * 3. Copy the snapshot to the userspace. | |
1047 | * 4. Flush TLB's if needed. | |
1048 | */ | |
749cf76c CD |
1049 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
1050 | { | |
53c810c3 MS |
1051 | bool is_dirty = false; |
1052 | int r; | |
1053 | ||
1054 | mutex_lock(&kvm->slots_lock); | |
1055 | ||
1056 | r = kvm_get_dirty_log_protect(kvm, log, &is_dirty); | |
1057 | ||
1058 | if (is_dirty) | |
1059 | kvm_flush_remote_tlbs(kvm); | |
1060 | ||
1061 | mutex_unlock(&kvm->slots_lock); | |
1062 | return r; | |
749cf76c CD |
1063 | } |
1064 | ||
3401d546 CD |
1065 | static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, |
1066 | struct kvm_arm_device_addr *dev_addr) | |
1067 | { | |
330690cd CD |
1068 | unsigned long dev_id, type; |
1069 | ||
1070 | dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> | |
1071 | KVM_ARM_DEVICE_ID_SHIFT; | |
1072 | type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> | |
1073 | KVM_ARM_DEVICE_TYPE_SHIFT; | |
1074 | ||
1075 | switch (dev_id) { | |
1076 | case KVM_ARM_DEVICE_VGIC_V2: | |
c7da6fa4 PF |
1077 | if (!vgic_present) |
1078 | return -ENXIO; | |
ce01e4e8 | 1079 | return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); |
330690cd CD |
1080 | default: |
1081 | return -ENODEV; | |
1082 | } | |
3401d546 CD |
1083 | } |
1084 | ||
749cf76c CD |
1085 | long kvm_arch_vm_ioctl(struct file *filp, |
1086 | unsigned int ioctl, unsigned long arg) | |
1087 | { | |
3401d546 CD |
1088 | struct kvm *kvm = filp->private_data; |
1089 | void __user *argp = (void __user *)arg; | |
1090 | ||
1091 | switch (ioctl) { | |
5863c2ce | 1092 | case KVM_CREATE_IRQCHIP: { |
a28ebea2 | 1093 | int ret; |
c7da6fa4 PF |
1094 | if (!vgic_present) |
1095 | return -ENXIO; | |
a28ebea2 CD |
1096 | mutex_lock(&kvm->lock); |
1097 | ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); | |
1098 | mutex_unlock(&kvm->lock); | |
1099 | return ret; | |
5863c2ce | 1100 | } |
3401d546 CD |
1101 | case KVM_ARM_SET_DEVICE_ADDR: { |
1102 | struct kvm_arm_device_addr dev_addr; | |
1103 | ||
1104 | if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) | |
1105 | return -EFAULT; | |
1106 | return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); | |
1107 | } | |
42c4e0c7 AP |
1108 | case KVM_ARM_PREFERRED_TARGET: { |
1109 | int err; | |
1110 | struct kvm_vcpu_init init; | |
1111 | ||
1112 | err = kvm_vcpu_preferred_target(&init); | |
1113 | if (err) | |
1114 | return err; | |
1115 | ||
1116 | if (copy_to_user(argp, &init, sizeof(init))) | |
1117 | return -EFAULT; | |
1118 | ||
1119 | return 0; | |
1120 | } | |
3401d546 CD |
1121 | default: |
1122 | return -EINVAL; | |
1123 | } | |
749cf76c CD |
1124 | } |
1125 | ||
d157f4a5 | 1126 | static void cpu_init_hyp_mode(void *dummy) |
342cd0ab | 1127 | { |
dac288f7 | 1128 | phys_addr_t pgd_ptr; |
342cd0ab CD |
1129 | unsigned long hyp_stack_ptr; |
1130 | unsigned long stack_page; | |
1131 | unsigned long vector_ptr; | |
1132 | ||
1133 | /* Switch from the HYP stub to our own HYP init vector */ | |
5a677ce0 | 1134 | __hyp_set_vectors(kvm_get_idmap_vector()); |
342cd0ab | 1135 | |
dac288f7 | 1136 | pgd_ptr = kvm_mmu_get_httbr(); |
1436c1aa | 1137 | stack_page = __this_cpu_read(kvm_arm_hyp_stack_page); |
342cd0ab | 1138 | hyp_stack_ptr = stack_page + PAGE_SIZE; |
757a12f7 | 1139 | vector_ptr = (unsigned long)kvm_get_hyp_vector(); |
342cd0ab | 1140 | |
12fda812 | 1141 | __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr); |
35a2491a | 1142 | __cpu_init_stage2(); |
56c7f5e7 AB |
1143 | |
1144 | kvm_arm_init_debug(); | |
342cd0ab CD |
1145 | } |
1146 | ||
47eb3cba MZ |
1147 | static void cpu_hyp_reset(void) |
1148 | { | |
1149 | if (!is_kernel_in_hyp_mode()) | |
1150 | __hyp_reset_vectors(); | |
1151 | } | |
1152 | ||
5f5560b1 JM |
1153 | static void cpu_hyp_reinit(void) |
1154 | { | |
47eb3cba MZ |
1155 | cpu_hyp_reset(); |
1156 | ||
5f5560b1 JM |
1157 | if (is_kernel_in_hyp_mode()) { |
1158 | /* | |
67f69197 | 1159 | * __cpu_init_stage2() is safe to call even if the PM |
5f5560b1 JM |
1160 | * event was cancelled before the CPU was reset. |
1161 | */ | |
67f69197 | 1162 | __cpu_init_stage2(); |
02d50cda | 1163 | kvm_timer_init_vhe(); |
5f5560b1 | 1164 | } else { |
47eb3cba | 1165 | cpu_init_hyp_mode(NULL); |
5f5560b1 | 1166 | } |
5b0d2cc2 CD |
1167 | |
1168 | if (vgic_present) | |
1169 | kvm_vgic_init_cpu_hardware(); | |
5f5560b1 JM |
1170 | } |
1171 | ||
67f69197 AT |
1172 | static void _kvm_arch_hardware_enable(void *discard) |
1173 | { | |
1174 | if (!__this_cpu_read(kvm_arm_hardware_enabled)) { | |
5f5560b1 | 1175 | cpu_hyp_reinit(); |
67f69197 | 1176 | __this_cpu_write(kvm_arm_hardware_enabled, 1); |
d157f4a5 | 1177 | } |
67f69197 | 1178 | } |
d157f4a5 | 1179 | |
67f69197 AT |
1180 | int kvm_arch_hardware_enable(void) |
1181 | { | |
1182 | _kvm_arch_hardware_enable(NULL); | |
1183 | return 0; | |
342cd0ab CD |
1184 | } |
1185 | ||
67f69197 AT |
1186 | static void _kvm_arch_hardware_disable(void *discard) |
1187 | { | |
1188 | if (__this_cpu_read(kvm_arm_hardware_enabled)) { | |
1189 | cpu_hyp_reset(); | |
1190 | __this_cpu_write(kvm_arm_hardware_enabled, 0); | |
1191 | } | |
1192 | } | |
1193 | ||
1194 | void kvm_arch_hardware_disable(void) | |
1195 | { | |
1196 | _kvm_arch_hardware_disable(NULL); | |
1197 | } | |
d157f4a5 | 1198 | |
1fcf7ce0 LP |
1199 | #ifdef CONFIG_CPU_PM |
1200 | static int hyp_init_cpu_pm_notifier(struct notifier_block *self, | |
1201 | unsigned long cmd, | |
1202 | void *v) | |
1203 | { | |
67f69197 AT |
1204 | /* |
1205 | * kvm_arm_hardware_enabled is left with its old value over | |
1206 | * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should | |
1207 | * re-enable hyp. | |
1208 | */ | |
1209 | switch (cmd) { | |
1210 | case CPU_PM_ENTER: | |
1211 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1212 | /* | |
1213 | * don't update kvm_arm_hardware_enabled here | |
1214 | * so that the hardware will be re-enabled | |
1215 | * when we resume. See below. | |
1216 | */ | |
1217 | cpu_hyp_reset(); | |
1218 | ||
1fcf7ce0 | 1219 | return NOTIFY_OK; |
67f69197 AT |
1220 | case CPU_PM_EXIT: |
1221 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1222 | /* The hardware was enabled before suspend. */ | |
1223 | cpu_hyp_reinit(); | |
1fcf7ce0 | 1224 | |
67f69197 AT |
1225 | return NOTIFY_OK; |
1226 | ||
1227 | default: | |
1228 | return NOTIFY_DONE; | |
1229 | } | |
1fcf7ce0 LP |
1230 | } |
1231 | ||
1232 | static struct notifier_block hyp_init_cpu_pm_nb = { | |
1233 | .notifier_call = hyp_init_cpu_pm_notifier, | |
1234 | }; | |
1235 | ||
1236 | static void __init hyp_cpu_pm_init(void) | |
1237 | { | |
1238 | cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); | |
1239 | } | |
06a71a24 SH |
1240 | static void __init hyp_cpu_pm_exit(void) |
1241 | { | |
1242 | cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); | |
1243 | } | |
1fcf7ce0 LP |
1244 | #else |
1245 | static inline void hyp_cpu_pm_init(void) | |
1246 | { | |
1247 | } | |
06a71a24 SH |
1248 | static inline void hyp_cpu_pm_exit(void) |
1249 | { | |
1250 | } | |
1fcf7ce0 LP |
1251 | #endif |
1252 | ||
1e947bad MZ |
1253 | static void teardown_common_resources(void) |
1254 | { | |
1255 | free_percpu(kvm_host_cpu_state); | |
1256 | } | |
1257 | ||
1258 | static int init_common_resources(void) | |
1259 | { | |
1260 | kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t); | |
1261 | if (!kvm_host_cpu_state) { | |
1262 | kvm_err("Cannot allocate host CPU state\n"); | |
1263 | return -ENOMEM; | |
1264 | } | |
1265 | ||
61349937 VM |
1266 | /* set size of VMID supported by CPU */ |
1267 | kvm_vmid_bits = kvm_get_vmid_bits(); | |
1268 | kvm_info("%d-bit VMID\n", kvm_vmid_bits); | |
1269 | ||
1e947bad MZ |
1270 | return 0; |
1271 | } | |
1272 | ||
1273 | static int init_subsystems(void) | |
1274 | { | |
67f69197 | 1275 | int err = 0; |
1e947bad | 1276 | |
5f5560b1 | 1277 | /* |
67f69197 | 1278 | * Enable hardware so that subsystem initialisation can access EL2. |
5f5560b1 | 1279 | */ |
67f69197 | 1280 | on_each_cpu(_kvm_arch_hardware_enable, NULL, 1); |
5f5560b1 JM |
1281 | |
1282 | /* | |
1283 | * Register CPU lower-power notifier | |
1284 | */ | |
1285 | hyp_cpu_pm_init(); | |
1286 | ||
1e947bad MZ |
1287 | /* |
1288 | * Init HYP view of VGIC | |
1289 | */ | |
1290 | err = kvm_vgic_hyp_init(); | |
1291 | switch (err) { | |
1292 | case 0: | |
1293 | vgic_present = true; | |
1294 | break; | |
1295 | case -ENODEV: | |
1296 | case -ENXIO: | |
1297 | vgic_present = false; | |
67f69197 | 1298 | err = 0; |
1e947bad MZ |
1299 | break; |
1300 | default: | |
67f69197 | 1301 | goto out; |
1e947bad MZ |
1302 | } |
1303 | ||
1304 | /* | |
1305 | * Init HYP architected timer support | |
1306 | */ | |
1307 | err = kvm_timer_hyp_init(); | |
1308 | if (err) | |
67f69197 | 1309 | goto out; |
1e947bad MZ |
1310 | |
1311 | kvm_perf_init(); | |
1312 | kvm_coproc_table_init(); | |
1313 | ||
67f69197 AT |
1314 | out: |
1315 | on_each_cpu(_kvm_arch_hardware_disable, NULL, 1); | |
1316 | ||
1317 | return err; | |
1e947bad MZ |
1318 | } |
1319 | ||
1320 | static void teardown_hyp_mode(void) | |
1321 | { | |
1322 | int cpu; | |
1323 | ||
1324 | if (is_kernel_in_hyp_mode()) | |
1325 | return; | |
1326 | ||
1327 | free_hyp_pgds(); | |
1328 | for_each_possible_cpu(cpu) | |
1329 | free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); | |
06a71a24 | 1330 | hyp_cpu_pm_exit(); |
1e947bad MZ |
1331 | } |
1332 | ||
1333 | static int init_vhe_mode(void) | |
1334 | { | |
1e947bad MZ |
1335 | kvm_info("VHE mode initialized successfully\n"); |
1336 | return 0; | |
1337 | } | |
1338 | ||
342cd0ab CD |
1339 | /** |
1340 | * Inits Hyp-mode on all online CPUs | |
1341 | */ | |
1342 | static int init_hyp_mode(void) | |
1343 | { | |
342cd0ab CD |
1344 | int cpu; |
1345 | int err = 0; | |
1346 | ||
1347 | /* | |
1348 | * Allocate Hyp PGD and setup Hyp identity mapping | |
1349 | */ | |
1350 | err = kvm_mmu_init(); | |
1351 | if (err) | |
1352 | goto out_err; | |
1353 | ||
342cd0ab CD |
1354 | /* |
1355 | * Allocate stack pages for Hypervisor-mode | |
1356 | */ | |
1357 | for_each_possible_cpu(cpu) { | |
1358 | unsigned long stack_page; | |
1359 | ||
1360 | stack_page = __get_free_page(GFP_KERNEL); | |
1361 | if (!stack_page) { | |
1362 | err = -ENOMEM; | |
1e947bad | 1363 | goto out_err; |
342cd0ab CD |
1364 | } |
1365 | ||
1366 | per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; | |
1367 | } | |
1368 | ||
342cd0ab CD |
1369 | /* |
1370 | * Map the Hyp-code called directly from the host | |
1371 | */ | |
588ab3f9 | 1372 | err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), |
59002705 | 1373 | kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC); |
342cd0ab CD |
1374 | if (err) { |
1375 | kvm_err("Cannot map world-switch code\n"); | |
1e947bad | 1376 | goto out_err; |
342cd0ab CD |
1377 | } |
1378 | ||
a0bf9776 | 1379 | err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), |
74a6b888 | 1380 | kvm_ksym_ref(__end_rodata), PAGE_HYP_RO); |
910917bb MZ |
1381 | if (err) { |
1382 | kvm_err("Cannot map rodata section\n"); | |
c8ea0395 MZ |
1383 | goto out_err; |
1384 | } | |
1385 | ||
1386 | err = create_hyp_mappings(kvm_ksym_ref(__bss_start), | |
1387 | kvm_ksym_ref(__bss_stop), PAGE_HYP_RO); | |
1388 | if (err) { | |
1389 | kvm_err("Cannot map bss section\n"); | |
1e947bad | 1390 | goto out_err; |
910917bb MZ |
1391 | } |
1392 | ||
757a12f7 MZ |
1393 | err = kvm_map_vectors(); |
1394 | if (err) { | |
1395 | kvm_err("Cannot map vectors\n"); | |
1396 | goto out_err; | |
1397 | } | |
1398 | ||
342cd0ab CD |
1399 | /* |
1400 | * Map the Hyp stack pages | |
1401 | */ | |
1402 | for_each_possible_cpu(cpu) { | |
1403 | char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); | |
c8dddecd MZ |
1404 | err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE, |
1405 | PAGE_HYP); | |
342cd0ab CD |
1406 | |
1407 | if (err) { | |
1408 | kvm_err("Cannot map hyp stack\n"); | |
1e947bad | 1409 | goto out_err; |
342cd0ab CD |
1410 | } |
1411 | } | |
1412 | ||
342cd0ab | 1413 | for_each_possible_cpu(cpu) { |
3de50da6 | 1414 | kvm_cpu_context_t *cpu_ctxt; |
342cd0ab | 1415 | |
3de50da6 | 1416 | cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu); |
c8dddecd | 1417 | err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP); |
342cd0ab CD |
1418 | |
1419 | if (err) { | |
3de50da6 | 1420 | kvm_err("Cannot map host CPU state: %d\n", err); |
1e947bad | 1421 | goto out_err; |
342cd0ab CD |
1422 | } |
1423 | } | |
1424 | ||
1425 | kvm_info("Hyp mode initialized successfully\n"); | |
210552c1 | 1426 | |
342cd0ab | 1427 | return 0; |
1e947bad | 1428 | |
342cd0ab | 1429 | out_err: |
1e947bad | 1430 | teardown_hyp_mode(); |
342cd0ab CD |
1431 | kvm_err("error initializing Hyp mode: %d\n", err); |
1432 | return err; | |
1433 | } | |
1434 | ||
d4e071ce AP |
1435 | static void check_kvm_target_cpu(void *ret) |
1436 | { | |
1437 | *(int *)ret = kvm_target_cpu(); | |
1438 | } | |
1439 | ||
4429fc64 AP |
1440 | struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) |
1441 | { | |
1442 | struct kvm_vcpu *vcpu; | |
1443 | int i; | |
1444 | ||
1445 | mpidr &= MPIDR_HWID_BITMASK; | |
1446 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
1447 | if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) | |
1448 | return vcpu; | |
1449 | } | |
1450 | return NULL; | |
1451 | } | |
1452 | ||
342cd0ab CD |
1453 | /** |
1454 | * Initialize Hyp-mode and memory mappings on all CPUs. | |
1455 | */ | |
749cf76c CD |
1456 | int kvm_arch_init(void *opaque) |
1457 | { | |
342cd0ab | 1458 | int err; |
d4e071ce | 1459 | int ret, cpu; |
342cd0ab CD |
1460 | |
1461 | if (!is_hyp_mode_available()) { | |
1462 | kvm_err("HYP mode not available\n"); | |
1463 | return -ENODEV; | |
1464 | } | |
1465 | ||
d4e071ce AP |
1466 | for_each_online_cpu(cpu) { |
1467 | smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); | |
1468 | if (ret < 0) { | |
1469 | kvm_err("Error, CPU %d not supported!\n", cpu); | |
1470 | return -ENODEV; | |
1471 | } | |
342cd0ab CD |
1472 | } |
1473 | ||
1e947bad | 1474 | err = init_common_resources(); |
342cd0ab | 1475 | if (err) |
1e947bad | 1476 | return err; |
342cd0ab | 1477 | |
1e947bad MZ |
1478 | if (is_kernel_in_hyp_mode()) |
1479 | err = init_vhe_mode(); | |
1480 | else | |
1481 | err = init_hyp_mode(); | |
1482 | if (err) | |
d157f4a5 | 1483 | goto out_err; |
8146875d | 1484 | |
1e947bad MZ |
1485 | err = init_subsystems(); |
1486 | if (err) | |
1487 | goto out_hyp; | |
1fcf7ce0 | 1488 | |
749cf76c | 1489 | return 0; |
1e947bad MZ |
1490 | |
1491 | out_hyp: | |
1492 | teardown_hyp_mode(); | |
342cd0ab | 1493 | out_err: |
1e947bad | 1494 | teardown_common_resources(); |
342cd0ab | 1495 | return err; |
749cf76c CD |
1496 | } |
1497 | ||
1498 | /* NOP: Compiling as a module not supported */ | |
1499 | void kvm_arch_exit(void) | |
1500 | { | |
210552c1 | 1501 | kvm_perf_teardown(); |
749cf76c CD |
1502 | } |
1503 | ||
1504 | static int arm_init(void) | |
1505 | { | |
1506 | int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1507 | return rc; | |
1508 | } | |
1509 | ||
1510 | module_init(arm_init); |