Commit | Line | Data |
---|---|---|
749cf76c CD |
1 | /* |
2 | * Copyright (C) 2012 - Virtual Open Systems and Columbia University | |
3 | * Author: Christoffer Dall <c.dall@virtualopensystems.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License, version 2, as | |
7 | * published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. | |
17 | */ | |
18 | ||
d157f4a5 | 19 | #include <linux/cpu.h> |
1fcf7ce0 | 20 | #include <linux/cpu_pm.h> |
749cf76c CD |
21 | #include <linux/errno.h> |
22 | #include <linux/err.h> | |
23 | #include <linux/kvm_host.h> | |
24 | #include <linux/module.h> | |
25 | #include <linux/vmalloc.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/mman.h> | |
28 | #include <linux/sched.h> | |
86ce8535 | 29 | #include <linux/kvm.h> |
749cf76c | 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 | ||
749cf76c CD |
36 | #include <asm/uaccess.h> |
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 | ||
1638a12d MZ |
69 | static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu) |
70 | { | |
71 | BUG_ON(preemptible()); | |
1436c1aa | 72 | __this_cpu_write(kvm_arm_running_vcpu, vcpu); |
1638a12d MZ |
73 | } |
74 | ||
75 | /** | |
76 | * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU. | |
77 | * Must be called from non-preemptible context | |
78 | */ | |
79 | struct kvm_vcpu *kvm_arm_get_running_vcpu(void) | |
80 | { | |
81 | BUG_ON(preemptible()); | |
1436c1aa | 82 | return __this_cpu_read(kvm_arm_running_vcpu); |
1638a12d MZ |
83 | } |
84 | ||
85 | /** | |
86 | * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
87 | */ | |
4000be42 | 88 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) |
1638a12d MZ |
89 | { |
90 | return &kvm_arm_running_vcpu; | |
91 | } | |
92 | ||
13a34e06 | 93 | int kvm_arch_hardware_enable(void) |
749cf76c CD |
94 | { |
95 | return 0; | |
96 | } | |
97 | ||
98 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) | |
99 | { | |
100 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; | |
101 | } | |
102 | ||
749cf76c CD |
103 | int kvm_arch_hardware_setup(void) |
104 | { | |
105 | return 0; | |
106 | } | |
107 | ||
749cf76c CD |
108 | void kvm_arch_check_processor_compat(void *rtn) |
109 | { | |
110 | *(int *)rtn = 0; | |
111 | } | |
112 | ||
749cf76c | 113 | |
d5d8184d CD |
114 | /** |
115 | * kvm_arch_init_vm - initializes a VM data structure | |
116 | * @kvm: pointer to the KVM struct | |
117 | */ | |
749cf76c CD |
118 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
119 | { | |
d5d8184d CD |
120 | int ret = 0; |
121 | ||
749cf76c CD |
122 | if (type) |
123 | return -EINVAL; | |
124 | ||
d5d8184d CD |
125 | ret = kvm_alloc_stage2_pgd(kvm); |
126 | if (ret) | |
127 | goto out_fail_alloc; | |
128 | ||
129 | ret = create_hyp_mappings(kvm, kvm + 1); | |
130 | if (ret) | |
131 | goto out_free_stage2_pgd; | |
132 | ||
6c3d63c9 | 133 | kvm_vgic_early_init(kvm); |
a1a64387 CD |
134 | kvm_timer_init(kvm); |
135 | ||
d5d8184d CD |
136 | /* Mark the initial VMID generation invalid */ |
137 | kvm->arch.vmid_gen = 0; | |
138 | ||
3caa2d8c | 139 | /* The maximum number of VCPUs is limited by the host's GIC model */ |
c7da6fa4 PF |
140 | kvm->arch.max_vcpus = vgic_present ? |
141 | kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; | |
3caa2d8c | 142 | |
d5d8184d CD |
143 | return ret; |
144 | out_free_stage2_pgd: | |
145 | kvm_free_stage2_pgd(kvm); | |
146 | out_fail_alloc: | |
147 | return ret; | |
749cf76c CD |
148 | } |
149 | ||
150 | int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) | |
151 | { | |
152 | return VM_FAULT_SIGBUS; | |
153 | } | |
154 | ||
749cf76c | 155 | |
d5d8184d CD |
156 | /** |
157 | * kvm_arch_destroy_vm - destroy the VM data structure | |
158 | * @kvm: pointer to the KVM struct | |
159 | */ | |
749cf76c CD |
160 | void kvm_arch_destroy_vm(struct kvm *kvm) |
161 | { | |
162 | int i; | |
163 | ||
d5d8184d CD |
164 | kvm_free_stage2_pgd(kvm); |
165 | ||
749cf76c CD |
166 | for (i = 0; i < KVM_MAX_VCPUS; ++i) { |
167 | if (kvm->vcpus[i]) { | |
168 | kvm_arch_vcpu_free(kvm->vcpus[i]); | |
169 | kvm->vcpus[i] = NULL; | |
170 | } | |
171 | } | |
c1bfb577 MZ |
172 | |
173 | kvm_vgic_destroy(kvm); | |
749cf76c CD |
174 | } |
175 | ||
784aa3d7 | 176 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
749cf76c CD |
177 | { |
178 | int r; | |
179 | switch (ext) { | |
1a89dd91 | 180 | case KVM_CAP_IRQCHIP: |
c7da6fa4 PF |
181 | r = vgic_present; |
182 | break; | |
d44758c0 | 183 | case KVM_CAP_IOEVENTFD: |
7330672b | 184 | case KVM_CAP_DEVICE_CTRL: |
749cf76c CD |
185 | case KVM_CAP_USER_MEMORY: |
186 | case KVM_CAP_SYNC_MMU: | |
187 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
188 | case KVM_CAP_ONE_REG: | |
aa024c2f | 189 | case KVM_CAP_ARM_PSCI: |
4447a208 | 190 | case KVM_CAP_ARM_PSCI_0_2: |
98047888 | 191 | case KVM_CAP_READONLY_MEM: |
ecccf0cc | 192 | case KVM_CAP_MP_STATE: |
749cf76c CD |
193 | r = 1; |
194 | break; | |
195 | case KVM_CAP_COALESCED_MMIO: | |
196 | r = KVM_COALESCED_MMIO_PAGE_OFFSET; | |
197 | break; | |
3401d546 CD |
198 | case KVM_CAP_ARM_SET_DEVICE_ADDR: |
199 | r = 1; | |
ca46e10f | 200 | break; |
749cf76c CD |
201 | case KVM_CAP_NR_VCPUS: |
202 | r = num_online_cpus(); | |
203 | break; | |
204 | case KVM_CAP_MAX_VCPUS: | |
205 | r = KVM_MAX_VCPUS; | |
206 | break; | |
207 | default: | |
17b1e31f | 208 | r = kvm_arch_dev_ioctl_check_extension(ext); |
749cf76c CD |
209 | break; |
210 | } | |
211 | return r; | |
212 | } | |
213 | ||
214 | long kvm_arch_dev_ioctl(struct file *filp, | |
215 | unsigned int ioctl, unsigned long arg) | |
216 | { | |
217 | return -EINVAL; | |
218 | } | |
219 | ||
749cf76c CD |
220 | |
221 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) | |
222 | { | |
223 | int err; | |
224 | struct kvm_vcpu *vcpu; | |
225 | ||
716139df CD |
226 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) { |
227 | err = -EBUSY; | |
228 | goto out; | |
229 | } | |
230 | ||
3caa2d8c AP |
231 | if (id >= kvm->arch.max_vcpus) { |
232 | err = -EINVAL; | |
233 | goto out; | |
234 | } | |
235 | ||
749cf76c CD |
236 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
237 | if (!vcpu) { | |
238 | err = -ENOMEM; | |
239 | goto out; | |
240 | } | |
241 | ||
242 | err = kvm_vcpu_init(vcpu, kvm, id); | |
243 | if (err) | |
244 | goto free_vcpu; | |
245 | ||
d5d8184d CD |
246 | err = create_hyp_mappings(vcpu, vcpu + 1); |
247 | if (err) | |
248 | goto vcpu_uninit; | |
249 | ||
749cf76c | 250 | return vcpu; |
d5d8184d CD |
251 | vcpu_uninit: |
252 | kvm_vcpu_uninit(vcpu); | |
749cf76c CD |
253 | free_vcpu: |
254 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
255 | out: | |
256 | return ERR_PTR(err); | |
257 | } | |
258 | ||
31928aa5 | 259 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
749cf76c | 260 | { |
6c3d63c9 | 261 | kvm_vgic_vcpu_early_init(vcpu); |
749cf76c CD |
262 | } |
263 | ||
264 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
265 | { | |
d5d8184d | 266 | kvm_mmu_free_memory_caches(vcpu); |
967f8427 | 267 | kvm_timer_vcpu_terminate(vcpu); |
c1bfb577 | 268 | kvm_vgic_vcpu_destroy(vcpu); |
5f0a714a | 269 | kvm_pmu_vcpu_destroy(vcpu); |
d5d8184d | 270 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
749cf76c CD |
271 | } |
272 | ||
273 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) | |
274 | { | |
275 | kvm_arch_vcpu_free(vcpu); | |
276 | } | |
277 | ||
278 | int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) | |
279 | { | |
1a748478 | 280 | return kvm_timer_should_fire(vcpu); |
749cf76c CD |
281 | } |
282 | ||
d35268da CD |
283 | void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) |
284 | { | |
285 | kvm_timer_schedule(vcpu); | |
286 | } | |
287 | ||
288 | void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
289 | { | |
290 | kvm_timer_unschedule(vcpu); | |
291 | } | |
292 | ||
749cf76c CD |
293 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
294 | { | |
f7ed45be CD |
295 | /* Force users to call KVM_ARM_VCPU_INIT */ |
296 | vcpu->arch.target = -1; | |
f7fa034d | 297 | bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); |
1a89dd91 | 298 | |
967f8427 MZ |
299 | /* Set up the timer */ |
300 | kvm_timer_vcpu_init(vcpu); | |
301 | ||
84e690bf AB |
302 | kvm_arm_reset_debug_ptr(vcpu); |
303 | ||
749cf76c CD |
304 | return 0; |
305 | } | |
306 | ||
749cf76c CD |
307 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
308 | { | |
86ce8535 | 309 | vcpu->cpu = cpu; |
3de50da6 | 310 | vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state); |
5b3e5e5b | 311 | |
1638a12d | 312 | kvm_arm_set_running_vcpu(vcpu); |
749cf76c CD |
313 | } |
314 | ||
315 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
316 | { | |
e9b152cb CD |
317 | /* |
318 | * The arch-generic KVM code expects the cpu field of a vcpu to be -1 | |
319 | * if the vcpu is no longer assigned to a cpu. This is used for the | |
320 | * optimized make_all_cpus_request path. | |
321 | */ | |
322 | vcpu->cpu = -1; | |
323 | ||
1638a12d | 324 | kvm_arm_set_running_vcpu(NULL); |
9b4a3004 | 325 | kvm_timer_vcpu_put(vcpu); |
749cf76c CD |
326 | } |
327 | ||
749cf76c CD |
328 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
329 | struct kvm_mp_state *mp_state) | |
330 | { | |
3781528e | 331 | if (vcpu->arch.power_off) |
ecccf0cc AB |
332 | mp_state->mp_state = KVM_MP_STATE_STOPPED; |
333 | else | |
334 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
335 | ||
336 | return 0; | |
749cf76c CD |
337 | } |
338 | ||
339 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
340 | struct kvm_mp_state *mp_state) | |
341 | { | |
ecccf0cc AB |
342 | switch (mp_state->mp_state) { |
343 | case KVM_MP_STATE_RUNNABLE: | |
3781528e | 344 | vcpu->arch.power_off = false; |
ecccf0cc AB |
345 | break; |
346 | case KVM_MP_STATE_STOPPED: | |
3781528e | 347 | vcpu->arch.power_off = true; |
ecccf0cc AB |
348 | break; |
349 | default: | |
350 | return -EINVAL; | |
351 | } | |
352 | ||
353 | return 0; | |
749cf76c CD |
354 | } |
355 | ||
5b3e5e5b CD |
356 | /** |
357 | * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled | |
358 | * @v: The VCPU pointer | |
359 | * | |
360 | * If the guest CPU is not waiting for interrupts or an interrupt line is | |
361 | * asserted, the CPU is by definition runnable. | |
362 | */ | |
749cf76c CD |
363 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) |
364 | { | |
4f5f1dc0 | 365 | return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v)) |
3b92830a | 366 | && !v->arch.power_off && !v->arch.pause); |
749cf76c CD |
367 | } |
368 | ||
f7ed45be CD |
369 | /* Just ensure a guest exit from a particular CPU */ |
370 | static void exit_vm_noop(void *info) | |
371 | { | |
372 | } | |
373 | ||
374 | void force_vm_exit(const cpumask_t *mask) | |
375 | { | |
898f949f | 376 | preempt_disable(); |
f7ed45be | 377 | smp_call_function_many(mask, exit_vm_noop, NULL, true); |
898f949f | 378 | preempt_enable(); |
f7ed45be CD |
379 | } |
380 | ||
381 | /** | |
382 | * need_new_vmid_gen - check that the VMID is still valid | |
383 | * @kvm: The VM's VMID to checkt | |
384 | * | |
385 | * return true if there is a new generation of VMIDs being used | |
386 | * | |
387 | * The hardware supports only 256 values with the value zero reserved for the | |
388 | * host, so we check if an assigned value belongs to a previous generation, | |
389 | * which which requires us to assign a new value. If we're the first to use a | |
390 | * VMID for the new generation, we must flush necessary caches and TLBs on all | |
391 | * CPUs. | |
392 | */ | |
393 | static bool need_new_vmid_gen(struct kvm *kvm) | |
394 | { | |
395 | return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen)); | |
396 | } | |
397 | ||
398 | /** | |
399 | * update_vttbr - Update the VTTBR with a valid VMID before the guest runs | |
400 | * @kvm The guest that we are about to run | |
401 | * | |
402 | * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the | |
403 | * VM has a valid VMID, otherwise assigns a new one and flushes corresponding | |
404 | * caches and TLBs. | |
405 | */ | |
406 | static void update_vttbr(struct kvm *kvm) | |
407 | { | |
408 | phys_addr_t pgd_phys; | |
409 | u64 vmid; | |
410 | ||
411 | if (!need_new_vmid_gen(kvm)) | |
412 | return; | |
413 | ||
414 | spin_lock(&kvm_vmid_lock); | |
415 | ||
416 | /* | |
417 | * We need to re-check the vmid_gen here to ensure that if another vcpu | |
418 | * already allocated a valid vmid for this vm, then this vcpu should | |
419 | * use the same vmid. | |
420 | */ | |
421 | if (!need_new_vmid_gen(kvm)) { | |
422 | spin_unlock(&kvm_vmid_lock); | |
423 | return; | |
424 | } | |
425 | ||
426 | /* First user of a new VMID generation? */ | |
427 | if (unlikely(kvm_next_vmid == 0)) { | |
428 | atomic64_inc(&kvm_vmid_gen); | |
429 | kvm_next_vmid = 1; | |
430 | ||
431 | /* | |
432 | * On SMP we know no other CPUs can use this CPU's or each | |
433 | * other's VMID after force_vm_exit returns since the | |
434 | * kvm_vmid_lock blocks them from reentry to the guest. | |
435 | */ | |
436 | force_vm_exit(cpu_all_mask); | |
437 | /* | |
438 | * Now broadcast TLB + ICACHE invalidation over the inner | |
439 | * shareable domain to make sure all data structures are | |
440 | * clean. | |
441 | */ | |
442 | kvm_call_hyp(__kvm_flush_vm_context); | |
443 | } | |
444 | ||
445 | kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen); | |
446 | kvm->arch.vmid = kvm_next_vmid; | |
447 | kvm_next_vmid++; | |
20475f78 | 448 | kvm_next_vmid &= (1 << kvm_vmid_bits) - 1; |
f7ed45be CD |
449 | |
450 | /* update vttbr to be used with the new vmid */ | |
38f791a4 | 451 | pgd_phys = virt_to_phys(kvm_get_hwpgd(kvm)); |
dbff124e | 452 | BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK); |
20475f78 | 453 | vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits); |
dbff124e | 454 | kvm->arch.vttbr = pgd_phys | vmid; |
f7ed45be CD |
455 | |
456 | spin_unlock(&kvm_vmid_lock); | |
457 | } | |
458 | ||
f7ed45be CD |
459 | static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) |
460 | { | |
05971120 | 461 | struct kvm *kvm = vcpu->kvm; |
e1ba0207 CD |
462 | int ret; |
463 | ||
f7ed45be CD |
464 | if (likely(vcpu->arch.has_run_once)) |
465 | return 0; | |
466 | ||
467 | vcpu->arch.has_run_once = true; | |
aa024c2f | 468 | |
01ac5e34 | 469 | /* |
6d3cfbe2 PM |
470 | * Map the VGIC hardware resources before running a vcpu the first |
471 | * time on this VM. | |
01ac5e34 | 472 | */ |
c2f58514 | 473 | if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) { |
05971120 | 474 | ret = kvm_vgic_map_resources(kvm); |
01ac5e34 MZ |
475 | if (ret) |
476 | return ret; | |
477 | } | |
478 | ||
05971120 CD |
479 | /* |
480 | * Enable the arch timers only if we have an in-kernel VGIC | |
481 | * and it has been properly initialized, since we cannot handle | |
482 | * interrupts from the virtual timer with a userspace gic. | |
483 | */ | |
484 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) | |
485 | kvm_timer_enable(kvm); | |
486 | ||
f7ed45be CD |
487 | return 0; |
488 | } | |
489 | ||
c1426e4c EA |
490 | bool kvm_arch_intc_initialized(struct kvm *kvm) |
491 | { | |
492 | return vgic_initialized(kvm); | |
493 | } | |
494 | ||
3b92830a EA |
495 | static void kvm_arm_halt_guest(struct kvm *kvm) __maybe_unused; |
496 | static void kvm_arm_resume_guest(struct kvm *kvm) __maybe_unused; | |
497 | ||
498 | static void kvm_arm_halt_guest(struct kvm *kvm) | |
499 | { | |
500 | int i; | |
501 | struct kvm_vcpu *vcpu; | |
502 | ||
503 | kvm_for_each_vcpu(i, vcpu, kvm) | |
504 | vcpu->arch.pause = true; | |
505 | force_vm_exit(cpu_all_mask); | |
506 | } | |
507 | ||
508 | static void kvm_arm_resume_guest(struct kvm *kvm) | |
509 | { | |
510 | int i; | |
511 | struct kvm_vcpu *vcpu; | |
512 | ||
513 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
8577370f | 514 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); |
3b92830a EA |
515 | |
516 | vcpu->arch.pause = false; | |
8577370f | 517 | swake_up(wq); |
3b92830a EA |
518 | } |
519 | } | |
520 | ||
3781528e | 521 | static void vcpu_sleep(struct kvm_vcpu *vcpu) |
aa024c2f | 522 | { |
8577370f | 523 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); |
aa024c2f | 524 | |
8577370f | 525 | swait_event_interruptible(*wq, ((!vcpu->arch.power_off) && |
3b92830a | 526 | (!vcpu->arch.pause))); |
aa024c2f MZ |
527 | } |
528 | ||
e8180dca AP |
529 | static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) |
530 | { | |
531 | return vcpu->arch.target >= 0; | |
532 | } | |
533 | ||
f7ed45be CD |
534 | /** |
535 | * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code | |
536 | * @vcpu: The VCPU pointer | |
537 | * @run: The kvm_run structure pointer used for userspace state exchange | |
538 | * | |
539 | * This function is called through the VCPU_RUN ioctl called from user space. It | |
540 | * will execute VM code in a loop until the time slice for the process is used | |
541 | * or some emulation is needed from user space in which case the function will | |
542 | * return with return value 0 and with the kvm_run structure filled in with the | |
543 | * required data for the requested emulation. | |
544 | */ | |
749cf76c CD |
545 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) |
546 | { | |
f7ed45be CD |
547 | int ret; |
548 | sigset_t sigsaved; | |
549 | ||
e8180dca | 550 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
f7ed45be CD |
551 | return -ENOEXEC; |
552 | ||
553 | ret = kvm_vcpu_first_run_init(vcpu); | |
554 | if (ret) | |
555 | return ret; | |
556 | ||
45e96ea6 CD |
557 | if (run->exit_reason == KVM_EXIT_MMIO) { |
558 | ret = kvm_handle_mmio_return(vcpu, vcpu->run); | |
559 | if (ret) | |
560 | return ret; | |
561 | } | |
562 | ||
f7ed45be CD |
563 | if (vcpu->sigset_active) |
564 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
565 | ||
566 | ret = 1; | |
567 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
568 | while (ret > 0) { | |
569 | /* | |
570 | * Check conditions before entering the guest | |
571 | */ | |
572 | cond_resched(); | |
573 | ||
574 | update_vttbr(vcpu->kvm); | |
575 | ||
3b92830a | 576 | if (vcpu->arch.power_off || vcpu->arch.pause) |
3781528e | 577 | vcpu_sleep(vcpu); |
aa024c2f | 578 | |
abdf5843 MZ |
579 | /* |
580 | * Preparing the interrupts to be injected also | |
581 | * involves poking the GIC, which must be done in a | |
582 | * non-preemptible context. | |
583 | */ | |
1b3d546d | 584 | preempt_disable(); |
b02386eb | 585 | kvm_pmu_flush_hwstate(vcpu); |
7e16aa81 | 586 | kvm_timer_flush_hwstate(vcpu); |
abdf5843 MZ |
587 | kvm_vgic_flush_hwstate(vcpu); |
588 | ||
f7ed45be CD |
589 | local_irq_disable(); |
590 | ||
591 | /* | |
592 | * Re-check atomic conditions | |
593 | */ | |
594 | if (signal_pending(current)) { | |
595 | ret = -EINTR; | |
596 | run->exit_reason = KVM_EXIT_INTR; | |
597 | } | |
598 | ||
101d3da0 | 599 | if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) || |
3b92830a | 600 | vcpu->arch.power_off || vcpu->arch.pause) { |
f7ed45be | 601 | local_irq_enable(); |
b02386eb | 602 | kvm_pmu_sync_hwstate(vcpu); |
4b4b4512 | 603 | kvm_timer_sync_hwstate(vcpu); |
1a89dd91 | 604 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 | 605 | preempt_enable(); |
f7ed45be CD |
606 | continue; |
607 | } | |
608 | ||
56c7f5e7 AB |
609 | kvm_arm_setup_debug(vcpu); |
610 | ||
f7ed45be CD |
611 | /************************************************************** |
612 | * Enter the guest | |
613 | */ | |
614 | trace_kvm_entry(*vcpu_pc(vcpu)); | |
ccf73aaf | 615 | __kvm_guest_enter(); |
f7ed45be CD |
616 | vcpu->mode = IN_GUEST_MODE; |
617 | ||
618 | ret = kvm_call_hyp(__kvm_vcpu_run, vcpu); | |
619 | ||
620 | vcpu->mode = OUTSIDE_GUEST_MODE; | |
b19e6892 | 621 | vcpu->stat.exits++; |
1b3d546d CD |
622 | /* |
623 | * Back from guest | |
624 | *************************************************************/ | |
625 | ||
56c7f5e7 AB |
626 | kvm_arm_clear_debug(vcpu); |
627 | ||
f7ed45be CD |
628 | /* |
629 | * We may have taken a host interrupt in HYP mode (ie | |
630 | * while executing the guest). This interrupt is still | |
631 | * pending, as we haven't serviced it yet! | |
632 | * | |
633 | * We're now back in SVC mode, with interrupts | |
634 | * disabled. Enabling the interrupts now will have | |
635 | * the effect of taking the interrupt again, in SVC | |
636 | * mode this time. | |
637 | */ | |
638 | local_irq_enable(); | |
639 | ||
640 | /* | |
1b3d546d CD |
641 | * We do local_irq_enable() before calling kvm_guest_exit() so |
642 | * that if a timer interrupt hits while running the guest we | |
643 | * account that tick as being spent in the guest. We enable | |
644 | * preemption after calling kvm_guest_exit() so that if we get | |
645 | * preempted we make sure ticks after that is not counted as | |
646 | * guest time. | |
647 | */ | |
648 | kvm_guest_exit(); | |
b5905dc1 | 649 | trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); |
1b3d546d | 650 | |
4b4b4512 | 651 | /* |
b02386eb SZ |
652 | * We must sync the PMU and timer state before the vgic state so |
653 | * that the vgic can properly sample the updated state of the | |
4b4b4512 CD |
654 | * interrupt line. |
655 | */ | |
b02386eb | 656 | kvm_pmu_sync_hwstate(vcpu); |
4b4b4512 CD |
657 | kvm_timer_sync_hwstate(vcpu); |
658 | ||
1a89dd91 | 659 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 MZ |
660 | |
661 | preempt_enable(); | |
662 | ||
f7ed45be CD |
663 | ret = handle_exit(vcpu, run, ret); |
664 | } | |
665 | ||
666 | if (vcpu->sigset_active) | |
667 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
668 | return ret; | |
749cf76c CD |
669 | } |
670 | ||
86ce8535 CD |
671 | static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) |
672 | { | |
673 | int bit_index; | |
674 | bool set; | |
675 | unsigned long *ptr; | |
676 | ||
677 | if (number == KVM_ARM_IRQ_CPU_IRQ) | |
678 | bit_index = __ffs(HCR_VI); | |
679 | else /* KVM_ARM_IRQ_CPU_FIQ */ | |
680 | bit_index = __ffs(HCR_VF); | |
681 | ||
682 | ptr = (unsigned long *)&vcpu->arch.irq_lines; | |
683 | if (level) | |
684 | set = test_and_set_bit(bit_index, ptr); | |
685 | else | |
686 | set = test_and_clear_bit(bit_index, ptr); | |
687 | ||
688 | /* | |
689 | * If we didn't change anything, no need to wake up or kick other CPUs | |
690 | */ | |
691 | if (set == level) | |
692 | return 0; | |
693 | ||
694 | /* | |
695 | * The vcpu irq_lines field was updated, wake up sleeping VCPUs and | |
696 | * trigger a world-switch round on the running physical CPU to set the | |
697 | * virtual IRQ/FIQ fields in the HCR appropriately. | |
698 | */ | |
699 | kvm_vcpu_kick(vcpu); | |
700 | ||
701 | return 0; | |
702 | } | |
703 | ||
79558f11 AG |
704 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, |
705 | bool line_status) | |
86ce8535 CD |
706 | { |
707 | u32 irq = irq_level->irq; | |
708 | unsigned int irq_type, vcpu_idx, irq_num; | |
709 | int nrcpus = atomic_read(&kvm->online_vcpus); | |
710 | struct kvm_vcpu *vcpu = NULL; | |
711 | bool level = irq_level->level; | |
712 | ||
713 | irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; | |
714 | vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; | |
715 | irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; | |
716 | ||
717 | trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); | |
718 | ||
5863c2ce MZ |
719 | switch (irq_type) { |
720 | case KVM_ARM_IRQ_TYPE_CPU: | |
721 | if (irqchip_in_kernel(kvm)) | |
722 | return -ENXIO; | |
86ce8535 | 723 | |
5863c2ce MZ |
724 | if (vcpu_idx >= nrcpus) |
725 | return -EINVAL; | |
86ce8535 | 726 | |
5863c2ce MZ |
727 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); |
728 | if (!vcpu) | |
729 | return -EINVAL; | |
86ce8535 | 730 | |
5863c2ce MZ |
731 | if (irq_num > KVM_ARM_IRQ_CPU_FIQ) |
732 | return -EINVAL; | |
733 | ||
734 | return vcpu_interrupt_line(vcpu, irq_num, level); | |
735 | case KVM_ARM_IRQ_TYPE_PPI: | |
736 | if (!irqchip_in_kernel(kvm)) | |
737 | return -ENXIO; | |
738 | ||
739 | if (vcpu_idx >= nrcpus) | |
740 | return -EINVAL; | |
741 | ||
742 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); | |
743 | if (!vcpu) | |
744 | return -EINVAL; | |
745 | ||
746 | if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) | |
747 | return -EINVAL; | |
86ce8535 | 748 | |
5863c2ce MZ |
749 | return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level); |
750 | case KVM_ARM_IRQ_TYPE_SPI: | |
751 | if (!irqchip_in_kernel(kvm)) | |
752 | return -ENXIO; | |
753 | ||
fd1d0ddf | 754 | if (irq_num < VGIC_NR_PRIVATE_IRQS) |
5863c2ce MZ |
755 | return -EINVAL; |
756 | ||
757 | return kvm_vgic_inject_irq(kvm, 0, irq_num, level); | |
758 | } | |
759 | ||
760 | return -EINVAL; | |
86ce8535 CD |
761 | } |
762 | ||
f7fa034d CD |
763 | static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, |
764 | const struct kvm_vcpu_init *init) | |
765 | { | |
766 | unsigned int i; | |
767 | int phys_target = kvm_target_cpu(); | |
768 | ||
769 | if (init->target != phys_target) | |
770 | return -EINVAL; | |
771 | ||
772 | /* | |
773 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
774 | * use the same target. | |
775 | */ | |
776 | if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) | |
777 | return -EINVAL; | |
778 | ||
779 | /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ | |
780 | for (i = 0; i < sizeof(init->features) * 8; i++) { | |
781 | bool set = (init->features[i / 32] & (1 << (i % 32))); | |
782 | ||
783 | if (set && i >= KVM_VCPU_MAX_FEATURES) | |
784 | return -ENOENT; | |
785 | ||
786 | /* | |
787 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
788 | * use the same feature set. | |
789 | */ | |
790 | if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && | |
791 | test_bit(i, vcpu->arch.features) != set) | |
792 | return -EINVAL; | |
793 | ||
794 | if (set) | |
795 | set_bit(i, vcpu->arch.features); | |
796 | } | |
797 | ||
798 | vcpu->arch.target = phys_target; | |
799 | ||
800 | /* Now we know what it is, we can reset it. */ | |
801 | return kvm_reset_vcpu(vcpu); | |
802 | } | |
803 | ||
804 | ||
478a8237 CD |
805 | static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, |
806 | struct kvm_vcpu_init *init) | |
807 | { | |
808 | int ret; | |
809 | ||
810 | ret = kvm_vcpu_set_target(vcpu, init); | |
811 | if (ret) | |
812 | return ret; | |
813 | ||
957db105 CD |
814 | /* |
815 | * Ensure a rebooted VM will fault in RAM pages and detect if the | |
816 | * guest MMU is turned off and flush the caches as needed. | |
817 | */ | |
818 | if (vcpu->arch.has_run_once) | |
819 | stage2_unmap_vm(vcpu->kvm); | |
820 | ||
b856a591 CD |
821 | vcpu_reset_hcr(vcpu); |
822 | ||
478a8237 | 823 | /* |
3781528e | 824 | * Handle the "start in power-off" case. |
478a8237 | 825 | */ |
03f1d4c1 | 826 | if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) |
3781528e | 827 | vcpu->arch.power_off = true; |
3ad8b3de | 828 | else |
3781528e | 829 | vcpu->arch.power_off = false; |
478a8237 CD |
830 | |
831 | return 0; | |
832 | } | |
833 | ||
f577f6c2 SZ |
834 | static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, |
835 | struct kvm_device_attr *attr) | |
836 | { | |
837 | int ret = -ENXIO; | |
838 | ||
839 | switch (attr->group) { | |
840 | default: | |
bb0c70bc | 841 | ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); |
f577f6c2 SZ |
842 | break; |
843 | } | |
844 | ||
845 | return ret; | |
846 | } | |
847 | ||
848 | static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, | |
849 | struct kvm_device_attr *attr) | |
850 | { | |
851 | int ret = -ENXIO; | |
852 | ||
853 | switch (attr->group) { | |
854 | default: | |
bb0c70bc | 855 | ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); |
f577f6c2 SZ |
856 | break; |
857 | } | |
858 | ||
859 | return ret; | |
860 | } | |
861 | ||
862 | static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, | |
863 | struct kvm_device_attr *attr) | |
864 | { | |
865 | int ret = -ENXIO; | |
866 | ||
867 | switch (attr->group) { | |
868 | default: | |
bb0c70bc | 869 | ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); |
f577f6c2 SZ |
870 | break; |
871 | } | |
872 | ||
873 | return ret; | |
874 | } | |
875 | ||
749cf76c CD |
876 | long kvm_arch_vcpu_ioctl(struct file *filp, |
877 | unsigned int ioctl, unsigned long arg) | |
878 | { | |
879 | struct kvm_vcpu *vcpu = filp->private_data; | |
880 | void __user *argp = (void __user *)arg; | |
f577f6c2 | 881 | struct kvm_device_attr attr; |
749cf76c CD |
882 | |
883 | switch (ioctl) { | |
884 | case KVM_ARM_VCPU_INIT: { | |
885 | struct kvm_vcpu_init init; | |
886 | ||
887 | if (copy_from_user(&init, argp, sizeof(init))) | |
888 | return -EFAULT; | |
889 | ||
478a8237 | 890 | return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); |
749cf76c CD |
891 | } |
892 | case KVM_SET_ONE_REG: | |
893 | case KVM_GET_ONE_REG: { | |
894 | struct kvm_one_reg reg; | |
e8180dca AP |
895 | |
896 | if (unlikely(!kvm_vcpu_initialized(vcpu))) | |
897 | return -ENOEXEC; | |
898 | ||
749cf76c CD |
899 | if (copy_from_user(®, argp, sizeof(reg))) |
900 | return -EFAULT; | |
901 | if (ioctl == KVM_SET_ONE_REG) | |
902 | return kvm_arm_set_reg(vcpu, ®); | |
903 | else | |
904 | return kvm_arm_get_reg(vcpu, ®); | |
905 | } | |
906 | case KVM_GET_REG_LIST: { | |
907 | struct kvm_reg_list __user *user_list = argp; | |
908 | struct kvm_reg_list reg_list; | |
909 | unsigned n; | |
910 | ||
e8180dca AP |
911 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
912 | return -ENOEXEC; | |
913 | ||
749cf76c CD |
914 | if (copy_from_user(®_list, user_list, sizeof(reg_list))) |
915 | return -EFAULT; | |
916 | n = reg_list.n; | |
917 | reg_list.n = kvm_arm_num_regs(vcpu); | |
918 | if (copy_to_user(user_list, ®_list, sizeof(reg_list))) | |
919 | return -EFAULT; | |
920 | if (n < reg_list.n) | |
921 | return -E2BIG; | |
922 | return kvm_arm_copy_reg_indices(vcpu, user_list->reg); | |
923 | } | |
f577f6c2 SZ |
924 | case KVM_SET_DEVICE_ATTR: { |
925 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
926 | return -EFAULT; | |
927 | return kvm_arm_vcpu_set_attr(vcpu, &attr); | |
928 | } | |
929 | case KVM_GET_DEVICE_ATTR: { | |
930 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
931 | return -EFAULT; | |
932 | return kvm_arm_vcpu_get_attr(vcpu, &attr); | |
933 | } | |
934 | case KVM_HAS_DEVICE_ATTR: { | |
935 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
936 | return -EFAULT; | |
937 | return kvm_arm_vcpu_has_attr(vcpu, &attr); | |
938 | } | |
749cf76c CD |
939 | default: |
940 | return -EINVAL; | |
941 | } | |
942 | } | |
943 | ||
53c810c3 MS |
944 | /** |
945 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
946 | * @kvm: kvm instance | |
947 | * @log: slot id and address to which we copy the log | |
948 | * | |
949 | * Steps 1-4 below provide general overview of dirty page logging. See | |
950 | * kvm_get_dirty_log_protect() function description for additional details. | |
951 | * | |
952 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
953 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
954 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
955 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
956 | * writes will be marked dirty for next log read. | |
957 | * | |
958 | * 1. Take a snapshot of the bit and clear it if needed. | |
959 | * 2. Write protect the corresponding page. | |
960 | * 3. Copy the snapshot to the userspace. | |
961 | * 4. Flush TLB's if needed. | |
962 | */ | |
749cf76c CD |
963 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
964 | { | |
53c810c3 MS |
965 | bool is_dirty = false; |
966 | int r; | |
967 | ||
968 | mutex_lock(&kvm->slots_lock); | |
969 | ||
970 | r = kvm_get_dirty_log_protect(kvm, log, &is_dirty); | |
971 | ||
972 | if (is_dirty) | |
973 | kvm_flush_remote_tlbs(kvm); | |
974 | ||
975 | mutex_unlock(&kvm->slots_lock); | |
976 | return r; | |
749cf76c CD |
977 | } |
978 | ||
3401d546 CD |
979 | static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, |
980 | struct kvm_arm_device_addr *dev_addr) | |
981 | { | |
330690cd CD |
982 | unsigned long dev_id, type; |
983 | ||
984 | dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> | |
985 | KVM_ARM_DEVICE_ID_SHIFT; | |
986 | type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> | |
987 | KVM_ARM_DEVICE_TYPE_SHIFT; | |
988 | ||
989 | switch (dev_id) { | |
990 | case KVM_ARM_DEVICE_VGIC_V2: | |
c7da6fa4 PF |
991 | if (!vgic_present) |
992 | return -ENXIO; | |
ce01e4e8 | 993 | return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); |
330690cd CD |
994 | default: |
995 | return -ENODEV; | |
996 | } | |
3401d546 CD |
997 | } |
998 | ||
749cf76c CD |
999 | long kvm_arch_vm_ioctl(struct file *filp, |
1000 | unsigned int ioctl, unsigned long arg) | |
1001 | { | |
3401d546 CD |
1002 | struct kvm *kvm = filp->private_data; |
1003 | void __user *argp = (void __user *)arg; | |
1004 | ||
1005 | switch (ioctl) { | |
5863c2ce | 1006 | case KVM_CREATE_IRQCHIP: { |
c7da6fa4 PF |
1007 | if (!vgic_present) |
1008 | return -ENXIO; | |
69ff5c61 | 1009 | return kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); |
5863c2ce | 1010 | } |
3401d546 CD |
1011 | case KVM_ARM_SET_DEVICE_ADDR: { |
1012 | struct kvm_arm_device_addr dev_addr; | |
1013 | ||
1014 | if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) | |
1015 | return -EFAULT; | |
1016 | return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); | |
1017 | } | |
42c4e0c7 AP |
1018 | case KVM_ARM_PREFERRED_TARGET: { |
1019 | int err; | |
1020 | struct kvm_vcpu_init init; | |
1021 | ||
1022 | err = kvm_vcpu_preferred_target(&init); | |
1023 | if (err) | |
1024 | return err; | |
1025 | ||
1026 | if (copy_to_user(argp, &init, sizeof(init))) | |
1027 | return -EFAULT; | |
1028 | ||
1029 | return 0; | |
1030 | } | |
3401d546 CD |
1031 | default: |
1032 | return -EINVAL; | |
1033 | } | |
749cf76c CD |
1034 | } |
1035 | ||
1e947bad MZ |
1036 | static void cpu_init_stage2(void *dummy) |
1037 | { | |
1038 | __cpu_init_stage2(); | |
1039 | } | |
1040 | ||
d157f4a5 | 1041 | static void cpu_init_hyp_mode(void *dummy) |
342cd0ab | 1042 | { |
dac288f7 MZ |
1043 | phys_addr_t boot_pgd_ptr; |
1044 | phys_addr_t pgd_ptr; | |
342cd0ab CD |
1045 | unsigned long hyp_stack_ptr; |
1046 | unsigned long stack_page; | |
1047 | unsigned long vector_ptr; | |
1048 | ||
1049 | /* Switch from the HYP stub to our own HYP init vector */ | |
5a677ce0 | 1050 | __hyp_set_vectors(kvm_get_idmap_vector()); |
342cd0ab | 1051 | |
dac288f7 MZ |
1052 | boot_pgd_ptr = kvm_mmu_get_boot_httbr(); |
1053 | pgd_ptr = kvm_mmu_get_httbr(); | |
1436c1aa | 1054 | stack_page = __this_cpu_read(kvm_arm_hyp_stack_page); |
342cd0ab | 1055 | hyp_stack_ptr = stack_page + PAGE_SIZE; |
a0bf9776 | 1056 | vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector); |
342cd0ab | 1057 | |
5a677ce0 | 1058 | __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr); |
35a2491a | 1059 | __cpu_init_stage2(); |
56c7f5e7 AB |
1060 | |
1061 | kvm_arm_init_debug(); | |
342cd0ab CD |
1062 | } |
1063 | ||
5f5560b1 JM |
1064 | static void cpu_hyp_reinit(void) |
1065 | { | |
1066 | if (is_kernel_in_hyp_mode()) { | |
1067 | /* | |
1068 | * cpu_init_stage2() is safe to call even if the PM | |
1069 | * event was cancelled before the CPU was reset. | |
1070 | */ | |
1071 | cpu_init_stage2(NULL); | |
1072 | } else { | |
1073 | if (__hyp_get_vectors() == hyp_default_vectors) | |
1074 | cpu_init_hyp_mode(NULL); | |
1075 | } | |
1076 | } | |
1077 | ||
d157f4a5 MZ |
1078 | static int hyp_init_cpu_notify(struct notifier_block *self, |
1079 | unsigned long action, void *cpu) | |
1080 | { | |
1081 | switch (action) { | |
1082 | case CPU_STARTING: | |
1083 | case CPU_STARTING_FROZEN: | |
5f5560b1 | 1084 | cpu_hyp_reinit(); |
d157f4a5 MZ |
1085 | } |
1086 | ||
1087 | return NOTIFY_OK; | |
342cd0ab CD |
1088 | } |
1089 | ||
d157f4a5 MZ |
1090 | static struct notifier_block hyp_init_cpu_nb = { |
1091 | .notifier_call = hyp_init_cpu_notify, | |
1092 | }; | |
1093 | ||
1fcf7ce0 LP |
1094 | #ifdef CONFIG_CPU_PM |
1095 | static int hyp_init_cpu_pm_notifier(struct notifier_block *self, | |
1096 | unsigned long cmd, | |
1097 | void *v) | |
1098 | { | |
5f5560b1 JM |
1099 | if (cmd == CPU_PM_EXIT) { |
1100 | cpu_hyp_reinit(); | |
1fcf7ce0 LP |
1101 | return NOTIFY_OK; |
1102 | } | |
1103 | ||
1104 | return NOTIFY_DONE; | |
1105 | } | |
1106 | ||
1107 | static struct notifier_block hyp_init_cpu_pm_nb = { | |
1108 | .notifier_call = hyp_init_cpu_pm_notifier, | |
1109 | }; | |
1110 | ||
1111 | static void __init hyp_cpu_pm_init(void) | |
1112 | { | |
1113 | cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); | |
1114 | } | |
06a71a24 SH |
1115 | static void __init hyp_cpu_pm_exit(void) |
1116 | { | |
1117 | cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); | |
1118 | } | |
1fcf7ce0 LP |
1119 | #else |
1120 | static inline void hyp_cpu_pm_init(void) | |
1121 | { | |
1122 | } | |
06a71a24 SH |
1123 | static inline void hyp_cpu_pm_exit(void) |
1124 | { | |
1125 | } | |
1fcf7ce0 LP |
1126 | #endif |
1127 | ||
1e947bad MZ |
1128 | static void teardown_common_resources(void) |
1129 | { | |
1130 | free_percpu(kvm_host_cpu_state); | |
1131 | } | |
1132 | ||
1133 | static int init_common_resources(void) | |
1134 | { | |
1135 | kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t); | |
1136 | if (!kvm_host_cpu_state) { | |
1137 | kvm_err("Cannot allocate host CPU state\n"); | |
1138 | return -ENOMEM; | |
1139 | } | |
1140 | ||
1141 | return 0; | |
1142 | } | |
1143 | ||
1144 | static int init_subsystems(void) | |
1145 | { | |
1146 | int err; | |
1147 | ||
5f5560b1 JM |
1148 | /* |
1149 | * Register CPU Hotplug notifier | |
1150 | */ | |
06a71a24 | 1151 | err = register_cpu_notifier(&hyp_init_cpu_nb); |
5f5560b1 JM |
1152 | if (err) { |
1153 | kvm_err("Cannot register KVM init CPU notifier (%d)\n", err); | |
1154 | return err; | |
1155 | } | |
1156 | ||
1157 | /* | |
1158 | * Register CPU lower-power notifier | |
1159 | */ | |
1160 | hyp_cpu_pm_init(); | |
1161 | ||
1e947bad MZ |
1162 | /* |
1163 | * Init HYP view of VGIC | |
1164 | */ | |
1165 | err = kvm_vgic_hyp_init(); | |
1166 | switch (err) { | |
1167 | case 0: | |
1168 | vgic_present = true; | |
1169 | break; | |
1170 | case -ENODEV: | |
1171 | case -ENXIO: | |
1172 | vgic_present = false; | |
1173 | break; | |
1174 | default: | |
1175 | return err; | |
1176 | } | |
1177 | ||
1178 | /* | |
1179 | * Init HYP architected timer support | |
1180 | */ | |
1181 | err = kvm_timer_hyp_init(); | |
1182 | if (err) | |
1183 | return err; | |
1184 | ||
1185 | kvm_perf_init(); | |
1186 | kvm_coproc_table_init(); | |
1187 | ||
1188 | return 0; | |
1189 | } | |
1190 | ||
1191 | static void teardown_hyp_mode(void) | |
1192 | { | |
1193 | int cpu; | |
1194 | ||
1195 | if (is_kernel_in_hyp_mode()) | |
1196 | return; | |
1197 | ||
1198 | free_hyp_pgds(); | |
1199 | for_each_possible_cpu(cpu) | |
1200 | free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); | |
06a71a24 SH |
1201 | unregister_cpu_notifier(&hyp_init_cpu_nb); |
1202 | hyp_cpu_pm_exit(); | |
1e947bad MZ |
1203 | } |
1204 | ||
1205 | static int init_vhe_mode(void) | |
1206 | { | |
1207 | /* | |
1208 | * Execute the init code on each CPU. | |
1209 | */ | |
1210 | on_each_cpu(cpu_init_stage2, NULL, 1); | |
1211 | ||
1212 | /* set size of VMID supported by CPU */ | |
1213 | kvm_vmid_bits = kvm_get_vmid_bits(); | |
1214 | kvm_info("%d-bit VMID\n", kvm_vmid_bits); | |
1215 | ||
1216 | kvm_info("VHE mode initialized successfully\n"); | |
1217 | return 0; | |
1218 | } | |
1219 | ||
342cd0ab CD |
1220 | /** |
1221 | * Inits Hyp-mode on all online CPUs | |
1222 | */ | |
1223 | static int init_hyp_mode(void) | |
1224 | { | |
342cd0ab CD |
1225 | int cpu; |
1226 | int err = 0; | |
1227 | ||
1228 | /* | |
1229 | * Allocate Hyp PGD and setup Hyp identity mapping | |
1230 | */ | |
1231 | err = kvm_mmu_init(); | |
1232 | if (err) | |
1233 | goto out_err; | |
1234 | ||
1235 | /* | |
1236 | * It is probably enough to obtain the default on one | |
1237 | * CPU. It's unlikely to be different on the others. | |
1238 | */ | |
1239 | hyp_default_vectors = __hyp_get_vectors(); | |
1240 | ||
1241 | /* | |
1242 | * Allocate stack pages for Hypervisor-mode | |
1243 | */ | |
1244 | for_each_possible_cpu(cpu) { | |
1245 | unsigned long stack_page; | |
1246 | ||
1247 | stack_page = __get_free_page(GFP_KERNEL); | |
1248 | if (!stack_page) { | |
1249 | err = -ENOMEM; | |
1e947bad | 1250 | goto out_err; |
342cd0ab CD |
1251 | } |
1252 | ||
1253 | per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; | |
1254 | } | |
1255 | ||
342cd0ab CD |
1256 | /* |
1257 | * Map the Hyp-code called directly from the host | |
1258 | */ | |
588ab3f9 LT |
1259 | err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), |
1260 | kvm_ksym_ref(__hyp_text_end)); | |
342cd0ab CD |
1261 | if (err) { |
1262 | kvm_err("Cannot map world-switch code\n"); | |
1e947bad | 1263 | goto out_err; |
342cd0ab CD |
1264 | } |
1265 | ||
a0bf9776 AB |
1266 | err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), |
1267 | kvm_ksym_ref(__end_rodata)); | |
910917bb MZ |
1268 | if (err) { |
1269 | kvm_err("Cannot map rodata section\n"); | |
1e947bad | 1270 | goto out_err; |
910917bb MZ |
1271 | } |
1272 | ||
342cd0ab CD |
1273 | /* |
1274 | * Map the Hyp stack pages | |
1275 | */ | |
1276 | for_each_possible_cpu(cpu) { | |
1277 | char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); | |
1278 | err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE); | |
1279 | ||
1280 | if (err) { | |
1281 | kvm_err("Cannot map hyp stack\n"); | |
1e947bad | 1282 | goto out_err; |
342cd0ab CD |
1283 | } |
1284 | } | |
1285 | ||
342cd0ab | 1286 | for_each_possible_cpu(cpu) { |
3de50da6 | 1287 | kvm_cpu_context_t *cpu_ctxt; |
342cd0ab | 1288 | |
3de50da6 MZ |
1289 | cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu); |
1290 | err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1); | |
342cd0ab CD |
1291 | |
1292 | if (err) { | |
3de50da6 | 1293 | kvm_err("Cannot map host CPU state: %d\n", err); |
1e947bad | 1294 | goto out_err; |
342cd0ab CD |
1295 | } |
1296 | } | |
1297 | ||
d157f4a5 MZ |
1298 | /* |
1299 | * Execute the init code on each CPU. | |
1300 | */ | |
1301 | on_each_cpu(cpu_init_hyp_mode, NULL, 1); | |
1302 | ||
d157f4a5 MZ |
1303 | #ifndef CONFIG_HOTPLUG_CPU |
1304 | free_boot_hyp_pgd(); | |
1305 | #endif | |
1306 | ||
20475f78 VM |
1307 | /* set size of VMID supported by CPU */ |
1308 | kvm_vmid_bits = kvm_get_vmid_bits(); | |
1309 | kvm_info("%d-bit VMID\n", kvm_vmid_bits); | |
1310 | ||
342cd0ab | 1311 | kvm_info("Hyp mode initialized successfully\n"); |
210552c1 | 1312 | |
342cd0ab | 1313 | return 0; |
1e947bad | 1314 | |
342cd0ab | 1315 | out_err: |
1e947bad | 1316 | teardown_hyp_mode(); |
342cd0ab CD |
1317 | kvm_err("error initializing Hyp mode: %d\n", err); |
1318 | return err; | |
1319 | } | |
1320 | ||
d4e071ce AP |
1321 | static void check_kvm_target_cpu(void *ret) |
1322 | { | |
1323 | *(int *)ret = kvm_target_cpu(); | |
1324 | } | |
1325 | ||
4429fc64 AP |
1326 | struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) |
1327 | { | |
1328 | struct kvm_vcpu *vcpu; | |
1329 | int i; | |
1330 | ||
1331 | mpidr &= MPIDR_HWID_BITMASK; | |
1332 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
1333 | if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) | |
1334 | return vcpu; | |
1335 | } | |
1336 | return NULL; | |
1337 | } | |
1338 | ||
342cd0ab CD |
1339 | /** |
1340 | * Initialize Hyp-mode and memory mappings on all CPUs. | |
1341 | */ | |
749cf76c CD |
1342 | int kvm_arch_init(void *opaque) |
1343 | { | |
342cd0ab | 1344 | int err; |
d4e071ce | 1345 | int ret, cpu; |
342cd0ab CD |
1346 | |
1347 | if (!is_hyp_mode_available()) { | |
1348 | kvm_err("HYP mode not available\n"); | |
1349 | return -ENODEV; | |
1350 | } | |
1351 | ||
d4e071ce AP |
1352 | for_each_online_cpu(cpu) { |
1353 | smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); | |
1354 | if (ret < 0) { | |
1355 | kvm_err("Error, CPU %d not supported!\n", cpu); | |
1356 | return -ENODEV; | |
1357 | } | |
342cd0ab CD |
1358 | } |
1359 | ||
1e947bad | 1360 | err = init_common_resources(); |
342cd0ab | 1361 | if (err) |
1e947bad | 1362 | return err; |
342cd0ab | 1363 | |
1e947bad MZ |
1364 | if (is_kernel_in_hyp_mode()) |
1365 | err = init_vhe_mode(); | |
1366 | else | |
1367 | err = init_hyp_mode(); | |
1368 | if (err) | |
d157f4a5 | 1369 | goto out_err; |
8146875d | 1370 | |
1e947bad MZ |
1371 | err = init_subsystems(); |
1372 | if (err) | |
1373 | goto out_hyp; | |
1fcf7ce0 | 1374 | |
749cf76c | 1375 | return 0; |
1e947bad MZ |
1376 | |
1377 | out_hyp: | |
1378 | teardown_hyp_mode(); | |
342cd0ab | 1379 | out_err: |
1e947bad | 1380 | teardown_common_resources(); |
342cd0ab | 1381 | return err; |
749cf76c CD |
1382 | } |
1383 | ||
1384 | /* NOP: Compiling as a module not supported */ | |
1385 | void kvm_arch_exit(void) | |
1386 | { | |
210552c1 | 1387 | kvm_perf_teardown(); |
749cf76c CD |
1388 | } |
1389 | ||
1390 | static int arm_init(void) | |
1391 | { | |
1392 | int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1393 | return rc; | |
1394 | } | |
1395 | ||
1396 | module_init(arm_init); |