]>
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> | |
23 | #include <linux/module.h> | |
24 | #include <linux/vmalloc.h> | |
25 | #include <linux/fs.h> | |
26 | #include <linux/mman.h> | |
27 | #include <linux/sched.h> | |
86ce8535 | 28 | #include <linux/kvm.h> |
749cf76c | 29 | #include <trace/events/kvm.h> |
b02386eb | 30 | #include <kvm/arm_pmu.h> |
749cf76c CD |
31 | |
32 | #define CREATE_TRACE_POINTS | |
33 | #include "trace.h" | |
34 | ||
749cf76c CD |
35 | #include <asm/uaccess.h> |
36 | #include <asm/ptrace.h> | |
37 | #include <asm/mman.h> | |
342cd0ab | 38 | #include <asm/tlbflush.h> |
5b3e5e5b | 39 | #include <asm/cacheflush.h> |
342cd0ab CD |
40 | #include <asm/virt.h> |
41 | #include <asm/kvm_arm.h> | |
42 | #include <asm/kvm_asm.h> | |
43 | #include <asm/kvm_mmu.h> | |
f7ed45be | 44 | #include <asm/kvm_emulate.h> |
5b3e5e5b | 45 | #include <asm/kvm_coproc.h> |
aa024c2f | 46 | #include <asm/kvm_psci.h> |
910917bb | 47 | #include <asm/sections.h> |
749cf76c CD |
48 | |
49 | #ifdef REQUIRES_VIRT | |
50 | __asm__(".arch_extension virt"); | |
51 | #endif | |
52 | ||
342cd0ab | 53 | static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); |
3de50da6 | 54 | static kvm_cpu_context_t __percpu *kvm_host_cpu_state; |
342cd0ab CD |
55 | static unsigned long hyp_default_vectors; |
56 | ||
1638a12d MZ |
57 | /* Per-CPU variable containing the currently running vcpu. */ |
58 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu); | |
59 | ||
f7ed45be CD |
60 | /* The VMID used in the VTTBR */ |
61 | static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1); | |
20475f78 VM |
62 | static u32 kvm_next_vmid; |
63 | static unsigned int kvm_vmid_bits __read_mostly; | |
f7ed45be | 64 | static DEFINE_SPINLOCK(kvm_vmid_lock); |
342cd0ab | 65 | |
c7da6fa4 PF |
66 | static bool vgic_present; |
67 | ||
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 | { | |
d5d8184d CD |
116 | int ret = 0; |
117 | ||
749cf76c CD |
118 | if (type) |
119 | return -EINVAL; | |
120 | ||
d5d8184d CD |
121 | ret = kvm_alloc_stage2_pgd(kvm); |
122 | if (ret) | |
123 | goto out_fail_alloc; | |
124 | ||
125 | ret = create_hyp_mappings(kvm, kvm + 1); | |
126 | if (ret) | |
127 | goto out_free_stage2_pgd; | |
128 | ||
6c3d63c9 | 129 | kvm_vgic_early_init(kvm); |
a1a64387 CD |
130 | kvm_timer_init(kvm); |
131 | ||
d5d8184d CD |
132 | /* Mark the initial VMID generation invalid */ |
133 | kvm->arch.vmid_gen = 0; | |
134 | ||
3caa2d8c | 135 | /* The maximum number of VCPUs is limited by the host's GIC model */ |
c7da6fa4 PF |
136 | kvm->arch.max_vcpus = vgic_present ? |
137 | kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; | |
3caa2d8c | 138 | |
d5d8184d CD |
139 | return ret; |
140 | out_free_stage2_pgd: | |
141 | kvm_free_stage2_pgd(kvm); | |
142 | out_fail_alloc: | |
143 | return ret; | |
749cf76c CD |
144 | } |
145 | ||
146 | int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) | |
147 | { | |
148 | return VM_FAULT_SIGBUS; | |
149 | } | |
150 | ||
749cf76c | 151 | |
d5d8184d CD |
152 | /** |
153 | * kvm_arch_destroy_vm - destroy the VM data structure | |
154 | * @kvm: pointer to the KVM struct | |
155 | */ | |
749cf76c CD |
156 | void kvm_arch_destroy_vm(struct kvm *kvm) |
157 | { | |
158 | int i; | |
159 | ||
d5d8184d CD |
160 | kvm_free_stage2_pgd(kvm); |
161 | ||
749cf76c CD |
162 | for (i = 0; i < KVM_MAX_VCPUS; ++i) { |
163 | if (kvm->vcpus[i]) { | |
164 | kvm_arch_vcpu_free(kvm->vcpus[i]); | |
165 | kvm->vcpus[i] = NULL; | |
166 | } | |
167 | } | |
c1bfb577 MZ |
168 | |
169 | kvm_vgic_destroy(kvm); | |
749cf76c CD |
170 | } |
171 | ||
784aa3d7 | 172 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
749cf76c CD |
173 | { |
174 | int r; | |
175 | switch (ext) { | |
1a89dd91 | 176 | case KVM_CAP_IRQCHIP: |
c7da6fa4 PF |
177 | r = vgic_present; |
178 | break; | |
d44758c0 | 179 | case KVM_CAP_IOEVENTFD: |
7330672b | 180 | case KVM_CAP_DEVICE_CTRL: |
749cf76c CD |
181 | case KVM_CAP_USER_MEMORY: |
182 | case KVM_CAP_SYNC_MMU: | |
183 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
184 | case KVM_CAP_ONE_REG: | |
aa024c2f | 185 | case KVM_CAP_ARM_PSCI: |
4447a208 | 186 | case KVM_CAP_ARM_PSCI_0_2: |
98047888 | 187 | case KVM_CAP_READONLY_MEM: |
ecccf0cc | 188 | case KVM_CAP_MP_STATE: |
749cf76c CD |
189 | r = 1; |
190 | break; | |
191 | case KVM_CAP_COALESCED_MMIO: | |
192 | r = KVM_COALESCED_MMIO_PAGE_OFFSET; | |
193 | break; | |
3401d546 CD |
194 | case KVM_CAP_ARM_SET_DEVICE_ADDR: |
195 | r = 1; | |
ca46e10f | 196 | break; |
749cf76c CD |
197 | case KVM_CAP_NR_VCPUS: |
198 | r = num_online_cpus(); | |
199 | break; | |
200 | case KVM_CAP_MAX_VCPUS: | |
201 | r = KVM_MAX_VCPUS; | |
202 | break; | |
203 | default: | |
17b1e31f | 204 | r = kvm_arch_dev_ioctl_check_extension(ext); |
749cf76c CD |
205 | break; |
206 | } | |
207 | return r; | |
208 | } | |
209 | ||
210 | long kvm_arch_dev_ioctl(struct file *filp, | |
211 | unsigned int ioctl, unsigned long arg) | |
212 | { | |
213 | return -EINVAL; | |
214 | } | |
215 | ||
749cf76c CD |
216 | |
217 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) | |
218 | { | |
219 | int err; | |
220 | struct kvm_vcpu *vcpu; | |
221 | ||
716139df CD |
222 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) { |
223 | err = -EBUSY; | |
224 | goto out; | |
225 | } | |
226 | ||
3caa2d8c AP |
227 | if (id >= kvm->arch.max_vcpus) { |
228 | err = -EINVAL; | |
229 | goto out; | |
230 | } | |
231 | ||
749cf76c CD |
232 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
233 | if (!vcpu) { | |
234 | err = -ENOMEM; | |
235 | goto out; | |
236 | } | |
237 | ||
238 | err = kvm_vcpu_init(vcpu, kvm, id); | |
239 | if (err) | |
240 | goto free_vcpu; | |
241 | ||
d5d8184d CD |
242 | err = create_hyp_mappings(vcpu, vcpu + 1); |
243 | if (err) | |
244 | goto vcpu_uninit; | |
245 | ||
749cf76c | 246 | return vcpu; |
d5d8184d CD |
247 | vcpu_uninit: |
248 | kvm_vcpu_uninit(vcpu); | |
749cf76c CD |
249 | free_vcpu: |
250 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
251 | out: | |
252 | return ERR_PTR(err); | |
253 | } | |
254 | ||
31928aa5 | 255 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
749cf76c | 256 | { |
6c3d63c9 | 257 | kvm_vgic_vcpu_early_init(vcpu); |
749cf76c CD |
258 | } |
259 | ||
260 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
261 | { | |
d5d8184d | 262 | kvm_mmu_free_memory_caches(vcpu); |
967f8427 | 263 | kvm_timer_vcpu_terminate(vcpu); |
c1bfb577 | 264 | kvm_vgic_vcpu_destroy(vcpu); |
5f0a714a | 265 | kvm_pmu_vcpu_destroy(vcpu); |
591d215a | 266 | kvm_vcpu_uninit(vcpu); |
d5d8184d | 267 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
749cf76c CD |
268 | } |
269 | ||
270 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) | |
271 | { | |
272 | kvm_arch_vcpu_free(vcpu); | |
273 | } | |
274 | ||
275 | int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) | |
276 | { | |
1a748478 | 277 | return kvm_timer_should_fire(vcpu); |
749cf76c CD |
278 | } |
279 | ||
d35268da CD |
280 | void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) |
281 | { | |
282 | kvm_timer_schedule(vcpu); | |
283 | } | |
284 | ||
285 | void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
286 | { | |
287 | kvm_timer_unschedule(vcpu); | |
288 | } | |
289 | ||
749cf76c CD |
290 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
291 | { | |
f7ed45be CD |
292 | /* Force users to call KVM_ARM_VCPU_INIT */ |
293 | vcpu->arch.target = -1; | |
f7fa034d | 294 | bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); |
1a89dd91 | 295 | |
967f8427 MZ |
296 | /* Set up the timer */ |
297 | kvm_timer_vcpu_init(vcpu); | |
298 | ||
84e690bf AB |
299 | kvm_arm_reset_debug_ptr(vcpu); |
300 | ||
749cf76c CD |
301 | return 0; |
302 | } | |
303 | ||
749cf76c CD |
304 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
305 | { | |
86ce8535 | 306 | vcpu->cpu = cpu; |
3de50da6 | 307 | vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state); |
5b3e5e5b | 308 | |
1638a12d | 309 | kvm_arm_set_running_vcpu(vcpu); |
749cf76c CD |
310 | } |
311 | ||
312 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
313 | { | |
e9b152cb CD |
314 | /* |
315 | * The arch-generic KVM code expects the cpu field of a vcpu to be -1 | |
316 | * if the vcpu is no longer assigned to a cpu. This is used for the | |
317 | * optimized make_all_cpus_request path. | |
318 | */ | |
319 | vcpu->cpu = -1; | |
320 | ||
1638a12d | 321 | kvm_arm_set_running_vcpu(NULL); |
9b4a3004 | 322 | kvm_timer_vcpu_put(vcpu); |
749cf76c CD |
323 | } |
324 | ||
749cf76c CD |
325 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
326 | struct kvm_mp_state *mp_state) | |
327 | { | |
3781528e | 328 | if (vcpu->arch.power_off) |
ecccf0cc AB |
329 | mp_state->mp_state = KVM_MP_STATE_STOPPED; |
330 | else | |
331 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
332 | ||
333 | return 0; | |
749cf76c CD |
334 | } |
335 | ||
336 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
337 | struct kvm_mp_state *mp_state) | |
338 | { | |
ecccf0cc AB |
339 | switch (mp_state->mp_state) { |
340 | case KVM_MP_STATE_RUNNABLE: | |
3781528e | 341 | vcpu->arch.power_off = false; |
ecccf0cc AB |
342 | break; |
343 | case KVM_MP_STATE_STOPPED: | |
3781528e | 344 | vcpu->arch.power_off = true; |
ecccf0cc AB |
345 | break; |
346 | default: | |
347 | return -EINVAL; | |
348 | } | |
349 | ||
350 | return 0; | |
749cf76c CD |
351 | } |
352 | ||
5b3e5e5b CD |
353 | /** |
354 | * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled | |
355 | * @v: The VCPU pointer | |
356 | * | |
357 | * If the guest CPU is not waiting for interrupts or an interrupt line is | |
358 | * asserted, the CPU is by definition runnable. | |
359 | */ | |
749cf76c CD |
360 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) |
361 | { | |
4f5f1dc0 | 362 | return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v)) |
3b92830a | 363 | && !v->arch.power_off && !v->arch.pause); |
749cf76c CD |
364 | } |
365 | ||
f7ed45be CD |
366 | /* Just ensure a guest exit from a particular CPU */ |
367 | static void exit_vm_noop(void *info) | |
368 | { | |
369 | } | |
370 | ||
371 | void force_vm_exit(const cpumask_t *mask) | |
372 | { | |
898f949f | 373 | preempt_disable(); |
f7ed45be | 374 | smp_call_function_many(mask, exit_vm_noop, NULL, true); |
898f949f | 375 | preempt_enable(); |
f7ed45be CD |
376 | } |
377 | ||
378 | /** | |
379 | * need_new_vmid_gen - check that the VMID is still valid | |
380 | * @kvm: The VM's VMID to checkt | |
381 | * | |
382 | * return true if there is a new generation of VMIDs being used | |
383 | * | |
384 | * The hardware supports only 256 values with the value zero reserved for the | |
385 | * host, so we check if an assigned value belongs to a previous generation, | |
386 | * which which requires us to assign a new value. If we're the first to use a | |
387 | * VMID for the new generation, we must flush necessary caches and TLBs on all | |
388 | * CPUs. | |
389 | */ | |
390 | static bool need_new_vmid_gen(struct kvm *kvm) | |
391 | { | |
392 | return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen)); | |
393 | } | |
394 | ||
395 | /** | |
396 | * update_vttbr - Update the VTTBR with a valid VMID before the guest runs | |
397 | * @kvm The guest that we are about to run | |
398 | * | |
399 | * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the | |
400 | * VM has a valid VMID, otherwise assigns a new one and flushes corresponding | |
401 | * caches and TLBs. | |
402 | */ | |
403 | static void update_vttbr(struct kvm *kvm) | |
404 | { | |
405 | phys_addr_t pgd_phys; | |
406 | u64 vmid; | |
407 | ||
408 | if (!need_new_vmid_gen(kvm)) | |
409 | return; | |
410 | ||
411 | spin_lock(&kvm_vmid_lock); | |
412 | ||
413 | /* | |
414 | * We need to re-check the vmid_gen here to ensure that if another vcpu | |
415 | * already allocated a valid vmid for this vm, then this vcpu should | |
416 | * use the same vmid. | |
417 | */ | |
418 | if (!need_new_vmid_gen(kvm)) { | |
419 | spin_unlock(&kvm_vmid_lock); | |
420 | return; | |
421 | } | |
422 | ||
423 | /* First user of a new VMID generation? */ | |
424 | if (unlikely(kvm_next_vmid == 0)) { | |
425 | atomic64_inc(&kvm_vmid_gen); | |
426 | kvm_next_vmid = 1; | |
427 | ||
428 | /* | |
429 | * On SMP we know no other CPUs can use this CPU's or each | |
430 | * other's VMID after force_vm_exit returns since the | |
431 | * kvm_vmid_lock blocks them from reentry to the guest. | |
432 | */ | |
433 | force_vm_exit(cpu_all_mask); | |
434 | /* | |
435 | * Now broadcast TLB + ICACHE invalidation over the inner | |
436 | * shareable domain to make sure all data structures are | |
437 | * clean. | |
438 | */ | |
439 | kvm_call_hyp(__kvm_flush_vm_context); | |
440 | } | |
441 | ||
442 | kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen); | |
443 | kvm->arch.vmid = kvm_next_vmid; | |
444 | kvm_next_vmid++; | |
20475f78 | 445 | kvm_next_vmid &= (1 << kvm_vmid_bits) - 1; |
f7ed45be CD |
446 | |
447 | /* update vttbr to be used with the new vmid */ | |
9163ee23 | 448 | pgd_phys = virt_to_phys(kvm->arch.pgd); |
dbff124e | 449 | BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK); |
20475f78 | 450 | vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits); |
dbff124e | 451 | kvm->arch.vttbr = pgd_phys | vmid; |
f7ed45be CD |
452 | |
453 | spin_unlock(&kvm_vmid_lock); | |
454 | } | |
455 | ||
f7ed45be CD |
456 | static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) |
457 | { | |
05971120 | 458 | struct kvm *kvm = vcpu->kvm; |
41a54482 | 459 | int ret = 0; |
e1ba0207 | 460 | |
f7ed45be CD |
461 | if (likely(vcpu->arch.has_run_once)) |
462 | return 0; | |
463 | ||
464 | vcpu->arch.has_run_once = true; | |
aa024c2f | 465 | |
01ac5e34 | 466 | /* |
6d3cfbe2 PM |
467 | * Map the VGIC hardware resources before running a vcpu the first |
468 | * time on this VM. | |
01ac5e34 | 469 | */ |
c2f58514 | 470 | if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) { |
05971120 | 471 | ret = kvm_vgic_map_resources(kvm); |
01ac5e34 MZ |
472 | if (ret) |
473 | return ret; | |
474 | } | |
475 | ||
05971120 CD |
476 | /* |
477 | * Enable the arch timers only if we have an in-kernel VGIC | |
478 | * and it has been properly initialized, since we cannot handle | |
479 | * interrupts from the virtual timer with a userspace gic. | |
480 | */ | |
481 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) | |
41a54482 | 482 | ret = kvm_timer_enable(vcpu); |
05971120 | 483 | |
41a54482 | 484 | return ret; |
f7ed45be CD |
485 | } |
486 | ||
c1426e4c EA |
487 | bool kvm_arch_intc_initialized(struct kvm *kvm) |
488 | { | |
489 | return vgic_initialized(kvm); | |
490 | } | |
491 | ||
b13216cf | 492 | void kvm_arm_halt_guest(struct kvm *kvm) |
3b92830a EA |
493 | { |
494 | int i; | |
495 | struct kvm_vcpu *vcpu; | |
496 | ||
497 | kvm_for_each_vcpu(i, vcpu, kvm) | |
498 | vcpu->arch.pause = true; | |
b13216cf | 499 | kvm_make_all_cpus_request(kvm, KVM_REQ_VCPU_EXIT); |
3b92830a EA |
500 | } |
501 | ||
35a2d585 CD |
502 | void kvm_arm_halt_vcpu(struct kvm_vcpu *vcpu) |
503 | { | |
504 | vcpu->arch.pause = true; | |
505 | kvm_vcpu_kick(vcpu); | |
506 | } | |
507 | ||
508 | void kvm_arm_resume_vcpu(struct kvm_vcpu *vcpu) | |
b13216cf CD |
509 | { |
510 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); | |
511 | ||
512 | vcpu->arch.pause = false; | |
513 | swake_up(wq); | |
514 | } | |
515 | ||
516 | void kvm_arm_resume_guest(struct kvm *kvm) | |
3b92830a EA |
517 | { |
518 | int i; | |
519 | struct kvm_vcpu *vcpu; | |
520 | ||
b13216cf CD |
521 | kvm_for_each_vcpu(i, vcpu, kvm) |
522 | kvm_arm_resume_vcpu(vcpu); | |
3b92830a EA |
523 | } |
524 | ||
3781528e | 525 | static void vcpu_sleep(struct kvm_vcpu *vcpu) |
aa024c2f | 526 | { |
8577370f | 527 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); |
aa024c2f | 528 | |
8577370f | 529 | swait_event_interruptible(*wq, ((!vcpu->arch.power_off) && |
3b92830a | 530 | (!vcpu->arch.pause))); |
aa024c2f MZ |
531 | } |
532 | ||
e8180dca AP |
533 | static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) |
534 | { | |
535 | return vcpu->arch.target >= 0; | |
536 | } | |
537 | ||
f7ed45be CD |
538 | /** |
539 | * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code | |
540 | * @vcpu: The VCPU pointer | |
541 | * @run: The kvm_run structure pointer used for userspace state exchange | |
542 | * | |
543 | * This function is called through the VCPU_RUN ioctl called from user space. It | |
544 | * will execute VM code in a loop until the time slice for the process is used | |
545 | * or some emulation is needed from user space in which case the function will | |
546 | * return with return value 0 and with the kvm_run structure filled in with the | |
547 | * required data for the requested emulation. | |
548 | */ | |
749cf76c CD |
549 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) |
550 | { | |
f7ed45be CD |
551 | int ret; |
552 | sigset_t sigsaved; | |
553 | ||
e8180dca | 554 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
f7ed45be CD |
555 | return -ENOEXEC; |
556 | ||
557 | ret = kvm_vcpu_first_run_init(vcpu); | |
558 | if (ret) | |
559 | return ret; | |
560 | ||
45e96ea6 CD |
561 | if (run->exit_reason == KVM_EXIT_MMIO) { |
562 | ret = kvm_handle_mmio_return(vcpu, vcpu->run); | |
563 | if (ret) | |
564 | return ret; | |
565 | } | |
566 | ||
f7ed45be CD |
567 | if (vcpu->sigset_active) |
568 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
569 | ||
570 | ret = 1; | |
571 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
572 | while (ret > 0) { | |
573 | /* | |
574 | * Check conditions before entering the guest | |
575 | */ | |
576 | cond_resched(); | |
577 | ||
578 | update_vttbr(vcpu->kvm); | |
579 | ||
3b92830a | 580 | if (vcpu->arch.power_off || vcpu->arch.pause) |
3781528e | 581 | vcpu_sleep(vcpu); |
aa024c2f | 582 | |
abdf5843 MZ |
583 | /* |
584 | * Preparing the interrupts to be injected also | |
585 | * involves poking the GIC, which must be done in a | |
586 | * non-preemptible context. | |
587 | */ | |
1b3d546d | 588 | preempt_disable(); |
b02386eb | 589 | kvm_pmu_flush_hwstate(vcpu); |
7e16aa81 | 590 | kvm_timer_flush_hwstate(vcpu); |
abdf5843 MZ |
591 | kvm_vgic_flush_hwstate(vcpu); |
592 | ||
f7ed45be CD |
593 | local_irq_disable(); |
594 | ||
595 | /* | |
596 | * Re-check atomic conditions | |
597 | */ | |
598 | if (signal_pending(current)) { | |
599 | ret = -EINTR; | |
600 | run->exit_reason = KVM_EXIT_INTR; | |
601 | } | |
602 | ||
101d3da0 | 603 | if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) || |
3b92830a | 604 | vcpu->arch.power_off || vcpu->arch.pause) { |
f7ed45be | 605 | local_irq_enable(); |
b02386eb | 606 | kvm_pmu_sync_hwstate(vcpu); |
4b4b4512 | 607 | kvm_timer_sync_hwstate(vcpu); |
1a89dd91 | 608 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 | 609 | preempt_enable(); |
f7ed45be CD |
610 | continue; |
611 | } | |
612 | ||
56c7f5e7 AB |
613 | kvm_arm_setup_debug(vcpu); |
614 | ||
f7ed45be CD |
615 | /************************************************************** |
616 | * Enter the guest | |
617 | */ | |
618 | trace_kvm_entry(*vcpu_pc(vcpu)); | |
ccf73aaf | 619 | __kvm_guest_enter(); |
f7ed45be CD |
620 | vcpu->mode = IN_GUEST_MODE; |
621 | ||
622 | ret = kvm_call_hyp(__kvm_vcpu_run, vcpu); | |
623 | ||
624 | vcpu->mode = OUTSIDE_GUEST_MODE; | |
b19e6892 | 625 | vcpu->stat.exits++; |
1b3d546d CD |
626 | /* |
627 | * Back from guest | |
628 | *************************************************************/ | |
629 | ||
56c7f5e7 AB |
630 | kvm_arm_clear_debug(vcpu); |
631 | ||
f7ed45be CD |
632 | /* |
633 | * We may have taken a host interrupt in HYP mode (ie | |
634 | * while executing the guest). This interrupt is still | |
635 | * pending, as we haven't serviced it yet! | |
636 | * | |
637 | * We're now back in SVC mode, with interrupts | |
638 | * disabled. Enabling the interrupts now will have | |
639 | * the effect of taking the interrupt again, in SVC | |
640 | * mode this time. | |
641 | */ | |
642 | local_irq_enable(); | |
643 | ||
644 | /* | |
1b3d546d CD |
645 | * We do local_irq_enable() before calling kvm_guest_exit() so |
646 | * that if a timer interrupt hits while running the guest we | |
647 | * account that tick as being spent in the guest. We enable | |
648 | * preemption after calling kvm_guest_exit() so that if we get | |
649 | * preempted we make sure ticks after that is not counted as | |
650 | * guest time. | |
651 | */ | |
652 | kvm_guest_exit(); | |
b5905dc1 | 653 | trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); |
1b3d546d | 654 | |
4b4b4512 | 655 | /* |
b02386eb SZ |
656 | * We must sync the PMU and timer state before the vgic state so |
657 | * that the vgic can properly sample the updated state of the | |
4b4b4512 CD |
658 | * interrupt line. |
659 | */ | |
b02386eb | 660 | kvm_pmu_sync_hwstate(vcpu); |
4b4b4512 CD |
661 | kvm_timer_sync_hwstate(vcpu); |
662 | ||
1a89dd91 | 663 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 MZ |
664 | |
665 | preempt_enable(); | |
666 | ||
f7ed45be CD |
667 | ret = handle_exit(vcpu, run, ret); |
668 | } | |
669 | ||
670 | if (vcpu->sigset_active) | |
671 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
672 | return ret; | |
749cf76c CD |
673 | } |
674 | ||
86ce8535 CD |
675 | static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) |
676 | { | |
677 | int bit_index; | |
678 | bool set; | |
679 | unsigned long *ptr; | |
680 | ||
681 | if (number == KVM_ARM_IRQ_CPU_IRQ) | |
682 | bit_index = __ffs(HCR_VI); | |
683 | else /* KVM_ARM_IRQ_CPU_FIQ */ | |
684 | bit_index = __ffs(HCR_VF); | |
685 | ||
686 | ptr = (unsigned long *)&vcpu->arch.irq_lines; | |
687 | if (level) | |
688 | set = test_and_set_bit(bit_index, ptr); | |
689 | else | |
690 | set = test_and_clear_bit(bit_index, ptr); | |
691 | ||
692 | /* | |
693 | * If we didn't change anything, no need to wake up or kick other CPUs | |
694 | */ | |
695 | if (set == level) | |
696 | return 0; | |
697 | ||
698 | /* | |
699 | * The vcpu irq_lines field was updated, wake up sleeping VCPUs and | |
700 | * trigger a world-switch round on the running physical CPU to set the | |
701 | * virtual IRQ/FIQ fields in the HCR appropriately. | |
702 | */ | |
703 | kvm_vcpu_kick(vcpu); | |
704 | ||
705 | return 0; | |
706 | } | |
707 | ||
79558f11 AG |
708 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, |
709 | bool line_status) | |
86ce8535 CD |
710 | { |
711 | u32 irq = irq_level->irq; | |
712 | unsigned int irq_type, vcpu_idx, irq_num; | |
713 | int nrcpus = atomic_read(&kvm->online_vcpus); | |
714 | struct kvm_vcpu *vcpu = NULL; | |
715 | bool level = irq_level->level; | |
716 | ||
717 | irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; | |
718 | vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; | |
719 | irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; | |
720 | ||
721 | trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); | |
722 | ||
5863c2ce MZ |
723 | switch (irq_type) { |
724 | case KVM_ARM_IRQ_TYPE_CPU: | |
725 | if (irqchip_in_kernel(kvm)) | |
726 | return -ENXIO; | |
86ce8535 | 727 | |
5863c2ce MZ |
728 | if (vcpu_idx >= nrcpus) |
729 | return -EINVAL; | |
86ce8535 | 730 | |
5863c2ce MZ |
731 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); |
732 | if (!vcpu) | |
733 | return -EINVAL; | |
86ce8535 | 734 | |
5863c2ce MZ |
735 | if (irq_num > KVM_ARM_IRQ_CPU_FIQ) |
736 | return -EINVAL; | |
737 | ||
738 | return vcpu_interrupt_line(vcpu, irq_num, level); | |
739 | case KVM_ARM_IRQ_TYPE_PPI: | |
740 | if (!irqchip_in_kernel(kvm)) | |
741 | return -ENXIO; | |
742 | ||
743 | if (vcpu_idx >= nrcpus) | |
744 | return -EINVAL; | |
745 | ||
746 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); | |
747 | if (!vcpu) | |
748 | return -EINVAL; | |
749 | ||
750 | if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) | |
751 | return -EINVAL; | |
86ce8535 | 752 | |
5863c2ce MZ |
753 | return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level); |
754 | case KVM_ARM_IRQ_TYPE_SPI: | |
755 | if (!irqchip_in_kernel(kvm)) | |
756 | return -ENXIO; | |
757 | ||
fd1d0ddf | 758 | if (irq_num < VGIC_NR_PRIVATE_IRQS) |
5863c2ce MZ |
759 | return -EINVAL; |
760 | ||
761 | return kvm_vgic_inject_irq(kvm, 0, irq_num, level); | |
762 | } | |
763 | ||
764 | return -EINVAL; | |
86ce8535 CD |
765 | } |
766 | ||
f7fa034d CD |
767 | static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, |
768 | const struct kvm_vcpu_init *init) | |
769 | { | |
770 | unsigned int i; | |
771 | int phys_target = kvm_target_cpu(); | |
772 | ||
773 | if (init->target != phys_target) | |
774 | return -EINVAL; | |
775 | ||
776 | /* | |
777 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
778 | * use the same target. | |
779 | */ | |
780 | if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) | |
781 | return -EINVAL; | |
782 | ||
783 | /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ | |
784 | for (i = 0; i < sizeof(init->features) * 8; i++) { | |
785 | bool set = (init->features[i / 32] & (1 << (i % 32))); | |
786 | ||
787 | if (set && i >= KVM_VCPU_MAX_FEATURES) | |
788 | return -ENOENT; | |
789 | ||
790 | /* | |
791 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
792 | * use the same feature set. | |
793 | */ | |
794 | if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && | |
795 | test_bit(i, vcpu->arch.features) != set) | |
796 | return -EINVAL; | |
797 | ||
798 | if (set) | |
799 | set_bit(i, vcpu->arch.features); | |
800 | } | |
801 | ||
802 | vcpu->arch.target = phys_target; | |
803 | ||
804 | /* Now we know what it is, we can reset it. */ | |
805 | return kvm_reset_vcpu(vcpu); | |
806 | } | |
807 | ||
808 | ||
478a8237 CD |
809 | static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, |
810 | struct kvm_vcpu_init *init) | |
811 | { | |
812 | int ret; | |
813 | ||
814 | ret = kvm_vcpu_set_target(vcpu, init); | |
815 | if (ret) | |
816 | return ret; | |
817 | ||
957db105 CD |
818 | /* |
819 | * Ensure a rebooted VM will fault in RAM pages and detect if the | |
820 | * guest MMU is turned off and flush the caches as needed. | |
821 | */ | |
822 | if (vcpu->arch.has_run_once) | |
823 | stage2_unmap_vm(vcpu->kvm); | |
824 | ||
b856a591 CD |
825 | vcpu_reset_hcr(vcpu); |
826 | ||
478a8237 | 827 | /* |
3781528e | 828 | * Handle the "start in power-off" case. |
478a8237 | 829 | */ |
03f1d4c1 | 830 | if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) |
3781528e | 831 | vcpu->arch.power_off = true; |
3ad8b3de | 832 | else |
3781528e | 833 | vcpu->arch.power_off = false; |
478a8237 CD |
834 | |
835 | return 0; | |
836 | } | |
837 | ||
f577f6c2 SZ |
838 | static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, |
839 | struct kvm_device_attr *attr) | |
840 | { | |
841 | int ret = -ENXIO; | |
842 | ||
843 | switch (attr->group) { | |
844 | default: | |
bb0c70bc | 845 | ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); |
f577f6c2 SZ |
846 | break; |
847 | } | |
848 | ||
849 | return ret; | |
850 | } | |
851 | ||
852 | static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, | |
853 | struct kvm_device_attr *attr) | |
854 | { | |
855 | int ret = -ENXIO; | |
856 | ||
857 | switch (attr->group) { | |
858 | default: | |
bb0c70bc | 859 | ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); |
f577f6c2 SZ |
860 | break; |
861 | } | |
862 | ||
863 | return ret; | |
864 | } | |
865 | ||
866 | static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, | |
867 | struct kvm_device_attr *attr) | |
868 | { | |
869 | int ret = -ENXIO; | |
870 | ||
871 | switch (attr->group) { | |
872 | default: | |
bb0c70bc | 873 | ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); |
f577f6c2 SZ |
874 | break; |
875 | } | |
876 | ||
877 | return ret; | |
878 | } | |
879 | ||
749cf76c CD |
880 | long kvm_arch_vcpu_ioctl(struct file *filp, |
881 | unsigned int ioctl, unsigned long arg) | |
882 | { | |
883 | struct kvm_vcpu *vcpu = filp->private_data; | |
884 | void __user *argp = (void __user *)arg; | |
f577f6c2 | 885 | struct kvm_device_attr attr; |
749cf76c CD |
886 | |
887 | switch (ioctl) { | |
888 | case KVM_ARM_VCPU_INIT: { | |
889 | struct kvm_vcpu_init init; | |
890 | ||
891 | if (copy_from_user(&init, argp, sizeof(init))) | |
892 | return -EFAULT; | |
893 | ||
478a8237 | 894 | return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); |
749cf76c CD |
895 | } |
896 | case KVM_SET_ONE_REG: | |
897 | case KVM_GET_ONE_REG: { | |
898 | struct kvm_one_reg reg; | |
e8180dca AP |
899 | |
900 | if (unlikely(!kvm_vcpu_initialized(vcpu))) | |
901 | return -ENOEXEC; | |
902 | ||
749cf76c CD |
903 | if (copy_from_user(®, argp, sizeof(reg))) |
904 | return -EFAULT; | |
905 | if (ioctl == KVM_SET_ONE_REG) | |
906 | return kvm_arm_set_reg(vcpu, ®); | |
907 | else | |
908 | return kvm_arm_get_reg(vcpu, ®); | |
909 | } | |
910 | case KVM_GET_REG_LIST: { | |
911 | struct kvm_reg_list __user *user_list = argp; | |
912 | struct kvm_reg_list reg_list; | |
913 | unsigned n; | |
914 | ||
e8180dca AP |
915 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
916 | return -ENOEXEC; | |
917 | ||
749cf76c CD |
918 | if (copy_from_user(®_list, user_list, sizeof(reg_list))) |
919 | return -EFAULT; | |
920 | n = reg_list.n; | |
921 | reg_list.n = kvm_arm_num_regs(vcpu); | |
922 | if (copy_to_user(user_list, ®_list, sizeof(reg_list))) | |
923 | return -EFAULT; | |
924 | if (n < reg_list.n) | |
925 | return -E2BIG; | |
926 | return kvm_arm_copy_reg_indices(vcpu, user_list->reg); | |
927 | } | |
f577f6c2 SZ |
928 | case KVM_SET_DEVICE_ATTR: { |
929 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
930 | return -EFAULT; | |
931 | return kvm_arm_vcpu_set_attr(vcpu, &attr); | |
932 | } | |
933 | case KVM_GET_DEVICE_ATTR: { | |
934 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
935 | return -EFAULT; | |
936 | return kvm_arm_vcpu_get_attr(vcpu, &attr); | |
937 | } | |
938 | case KVM_HAS_DEVICE_ATTR: { | |
939 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
940 | return -EFAULT; | |
941 | return kvm_arm_vcpu_has_attr(vcpu, &attr); | |
942 | } | |
749cf76c CD |
943 | default: |
944 | return -EINVAL; | |
945 | } | |
946 | } | |
947 | ||
53c810c3 MS |
948 | /** |
949 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
950 | * @kvm: kvm instance | |
951 | * @log: slot id and address to which we copy the log | |
952 | * | |
953 | * Steps 1-4 below provide general overview of dirty page logging. See | |
954 | * kvm_get_dirty_log_protect() function description for additional details. | |
955 | * | |
956 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
957 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
958 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
959 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
960 | * writes will be marked dirty for next log read. | |
961 | * | |
962 | * 1. Take a snapshot of the bit and clear it if needed. | |
963 | * 2. Write protect the corresponding page. | |
964 | * 3. Copy the snapshot to the userspace. | |
965 | * 4. Flush TLB's if needed. | |
966 | */ | |
749cf76c CD |
967 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
968 | { | |
53c810c3 MS |
969 | bool is_dirty = false; |
970 | int r; | |
971 | ||
972 | mutex_lock(&kvm->slots_lock); | |
973 | ||
974 | r = kvm_get_dirty_log_protect(kvm, log, &is_dirty); | |
975 | ||
976 | if (is_dirty) | |
977 | kvm_flush_remote_tlbs(kvm); | |
978 | ||
979 | mutex_unlock(&kvm->slots_lock); | |
980 | return r; | |
749cf76c CD |
981 | } |
982 | ||
3401d546 CD |
983 | static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, |
984 | struct kvm_arm_device_addr *dev_addr) | |
985 | { | |
330690cd CD |
986 | unsigned long dev_id, type; |
987 | ||
988 | dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> | |
989 | KVM_ARM_DEVICE_ID_SHIFT; | |
990 | type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> | |
991 | KVM_ARM_DEVICE_TYPE_SHIFT; | |
992 | ||
993 | switch (dev_id) { | |
994 | case KVM_ARM_DEVICE_VGIC_V2: | |
c7da6fa4 PF |
995 | if (!vgic_present) |
996 | return -ENXIO; | |
ce01e4e8 | 997 | return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); |
330690cd CD |
998 | default: |
999 | return -ENODEV; | |
1000 | } | |
3401d546 CD |
1001 | } |
1002 | ||
749cf76c CD |
1003 | long kvm_arch_vm_ioctl(struct file *filp, |
1004 | unsigned int ioctl, unsigned long arg) | |
1005 | { | |
3401d546 CD |
1006 | struct kvm *kvm = filp->private_data; |
1007 | void __user *argp = (void __user *)arg; | |
1008 | ||
1009 | switch (ioctl) { | |
5863c2ce | 1010 | case KVM_CREATE_IRQCHIP: { |
c7da6fa4 PF |
1011 | if (!vgic_present) |
1012 | return -ENXIO; | |
69ff5c61 | 1013 | return kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); |
5863c2ce | 1014 | } |
3401d546 CD |
1015 | case KVM_ARM_SET_DEVICE_ADDR: { |
1016 | struct kvm_arm_device_addr dev_addr; | |
1017 | ||
1018 | if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) | |
1019 | return -EFAULT; | |
1020 | return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); | |
1021 | } | |
42c4e0c7 AP |
1022 | case KVM_ARM_PREFERRED_TARGET: { |
1023 | int err; | |
1024 | struct kvm_vcpu_init init; | |
1025 | ||
1026 | err = kvm_vcpu_preferred_target(&init); | |
1027 | if (err) | |
1028 | return err; | |
1029 | ||
1030 | if (copy_to_user(argp, &init, sizeof(init))) | |
1031 | return -EFAULT; | |
1032 | ||
1033 | return 0; | |
1034 | } | |
3401d546 CD |
1035 | default: |
1036 | return -EINVAL; | |
1037 | } | |
749cf76c CD |
1038 | } |
1039 | ||
d157f4a5 | 1040 | static void cpu_init_hyp_mode(void *dummy) |
342cd0ab | 1041 | { |
dac288f7 MZ |
1042 | phys_addr_t boot_pgd_ptr; |
1043 | phys_addr_t pgd_ptr; | |
342cd0ab CD |
1044 | unsigned long hyp_stack_ptr; |
1045 | unsigned long stack_page; | |
1046 | unsigned long vector_ptr; | |
1047 | ||
1048 | /* Switch from the HYP stub to our own HYP init vector */ | |
5a677ce0 | 1049 | __hyp_set_vectors(kvm_get_idmap_vector()); |
342cd0ab | 1050 | |
dac288f7 MZ |
1051 | boot_pgd_ptr = kvm_mmu_get_boot_httbr(); |
1052 | pgd_ptr = kvm_mmu_get_httbr(); | |
1436c1aa | 1053 | stack_page = __this_cpu_read(kvm_arm_hyp_stack_page); |
342cd0ab | 1054 | hyp_stack_ptr = stack_page + PAGE_SIZE; |
a0bf9776 | 1055 | vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector); |
342cd0ab | 1056 | |
5a677ce0 | 1057 | __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr); |
35a2491a | 1058 | __cpu_init_stage2(); |
56c7f5e7 AB |
1059 | |
1060 | kvm_arm_init_debug(); | |
342cd0ab CD |
1061 | } |
1062 | ||
5f5560b1 JM |
1063 | static void cpu_hyp_reinit(void) |
1064 | { | |
1065 | if (is_kernel_in_hyp_mode()) { | |
1066 | /* | |
67f69197 | 1067 | * __cpu_init_stage2() is safe to call even if the PM |
5f5560b1 JM |
1068 | * event was cancelled before the CPU was reset. |
1069 | */ | |
67f69197 | 1070 | __cpu_init_stage2(); |
5f5560b1 JM |
1071 | } else { |
1072 | if (__hyp_get_vectors() == hyp_default_vectors) | |
1073 | cpu_init_hyp_mode(NULL); | |
1074 | } | |
1075 | } | |
1076 | ||
67f69197 | 1077 | static void cpu_hyp_reset(void) |
d157f4a5 | 1078 | { |
67f69197 AT |
1079 | phys_addr_t boot_pgd_ptr; |
1080 | phys_addr_t phys_idmap_start; | |
1081 | ||
1082 | if (!is_kernel_in_hyp_mode()) { | |
1083 | boot_pgd_ptr = kvm_mmu_get_boot_httbr(); | |
1084 | phys_idmap_start = kvm_get_idmap_start(); | |
1085 | ||
1086 | __cpu_reset_hyp_mode(boot_pgd_ptr, phys_idmap_start); | |
1087 | } | |
1088 | } | |
1089 | ||
1090 | static void _kvm_arch_hardware_enable(void *discard) | |
1091 | { | |
1092 | if (!__this_cpu_read(kvm_arm_hardware_enabled)) { | |
5f5560b1 | 1093 | cpu_hyp_reinit(); |
67f69197 | 1094 | __this_cpu_write(kvm_arm_hardware_enabled, 1); |
d157f4a5 | 1095 | } |
67f69197 | 1096 | } |
d157f4a5 | 1097 | |
67f69197 AT |
1098 | int kvm_arch_hardware_enable(void) |
1099 | { | |
1100 | _kvm_arch_hardware_enable(NULL); | |
1101 | return 0; | |
342cd0ab CD |
1102 | } |
1103 | ||
67f69197 AT |
1104 | static void _kvm_arch_hardware_disable(void *discard) |
1105 | { | |
1106 | if (__this_cpu_read(kvm_arm_hardware_enabled)) { | |
1107 | cpu_hyp_reset(); | |
1108 | __this_cpu_write(kvm_arm_hardware_enabled, 0); | |
1109 | } | |
1110 | } | |
1111 | ||
1112 | void kvm_arch_hardware_disable(void) | |
1113 | { | |
1114 | _kvm_arch_hardware_disable(NULL); | |
1115 | } | |
d157f4a5 | 1116 | |
1fcf7ce0 LP |
1117 | #ifdef CONFIG_CPU_PM |
1118 | static int hyp_init_cpu_pm_notifier(struct notifier_block *self, | |
1119 | unsigned long cmd, | |
1120 | void *v) | |
1121 | { | |
67f69197 AT |
1122 | /* |
1123 | * kvm_arm_hardware_enabled is left with its old value over | |
1124 | * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should | |
1125 | * re-enable hyp. | |
1126 | */ | |
1127 | switch (cmd) { | |
1128 | case CPU_PM_ENTER: | |
1129 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1130 | /* | |
1131 | * don't update kvm_arm_hardware_enabled here | |
1132 | * so that the hardware will be re-enabled | |
1133 | * when we resume. See below. | |
1134 | */ | |
1135 | cpu_hyp_reset(); | |
1136 | ||
1fcf7ce0 | 1137 | return NOTIFY_OK; |
67f69197 AT |
1138 | case CPU_PM_EXIT: |
1139 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1140 | /* The hardware was enabled before suspend. */ | |
1141 | cpu_hyp_reinit(); | |
1fcf7ce0 | 1142 | |
67f69197 AT |
1143 | return NOTIFY_OK; |
1144 | ||
1145 | default: | |
1146 | return NOTIFY_DONE; | |
1147 | } | |
1fcf7ce0 LP |
1148 | } |
1149 | ||
1150 | static struct notifier_block hyp_init_cpu_pm_nb = { | |
1151 | .notifier_call = hyp_init_cpu_pm_notifier, | |
1152 | }; | |
1153 | ||
1154 | static void __init hyp_cpu_pm_init(void) | |
1155 | { | |
1156 | cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); | |
1157 | } | |
06a71a24 SH |
1158 | static void __init hyp_cpu_pm_exit(void) |
1159 | { | |
1160 | cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); | |
1161 | } | |
1fcf7ce0 LP |
1162 | #else |
1163 | static inline void hyp_cpu_pm_init(void) | |
1164 | { | |
1165 | } | |
06a71a24 SH |
1166 | static inline void hyp_cpu_pm_exit(void) |
1167 | { | |
1168 | } | |
1fcf7ce0 LP |
1169 | #endif |
1170 | ||
1e947bad MZ |
1171 | static void teardown_common_resources(void) |
1172 | { | |
1173 | free_percpu(kvm_host_cpu_state); | |
1174 | } | |
1175 | ||
1176 | static int init_common_resources(void) | |
1177 | { | |
1178 | kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t); | |
1179 | if (!kvm_host_cpu_state) { | |
1180 | kvm_err("Cannot allocate host CPU state\n"); | |
1181 | return -ENOMEM; | |
1182 | } | |
1183 | ||
1184 | return 0; | |
1185 | } | |
1186 | ||
1187 | static int init_subsystems(void) | |
1188 | { | |
67f69197 | 1189 | int err = 0; |
1e947bad | 1190 | |
5f5560b1 | 1191 | /* |
67f69197 | 1192 | * Enable hardware so that subsystem initialisation can access EL2. |
5f5560b1 | 1193 | */ |
67f69197 | 1194 | on_each_cpu(_kvm_arch_hardware_enable, NULL, 1); |
5f5560b1 JM |
1195 | |
1196 | /* | |
1197 | * Register CPU lower-power notifier | |
1198 | */ | |
1199 | hyp_cpu_pm_init(); | |
1200 | ||
1e947bad MZ |
1201 | /* |
1202 | * Init HYP view of VGIC | |
1203 | */ | |
1204 | err = kvm_vgic_hyp_init(); | |
1205 | switch (err) { | |
1206 | case 0: | |
1207 | vgic_present = true; | |
1208 | break; | |
1209 | case -ENODEV: | |
1210 | case -ENXIO: | |
1211 | vgic_present = false; | |
67f69197 | 1212 | err = 0; |
1e947bad MZ |
1213 | break; |
1214 | default: | |
67f69197 | 1215 | goto out; |
1e947bad MZ |
1216 | } |
1217 | ||
1218 | /* | |
1219 | * Init HYP architected timer support | |
1220 | */ | |
1221 | err = kvm_timer_hyp_init(); | |
1222 | if (err) | |
67f69197 | 1223 | goto out; |
1e947bad MZ |
1224 | |
1225 | kvm_perf_init(); | |
1226 | kvm_coproc_table_init(); | |
1227 | ||
67f69197 AT |
1228 | out: |
1229 | on_each_cpu(_kvm_arch_hardware_disable, NULL, 1); | |
1230 | ||
1231 | return err; | |
1e947bad MZ |
1232 | } |
1233 | ||
1234 | static void teardown_hyp_mode(void) | |
1235 | { | |
1236 | int cpu; | |
1237 | ||
1238 | if (is_kernel_in_hyp_mode()) | |
1239 | return; | |
1240 | ||
1241 | free_hyp_pgds(); | |
1242 | for_each_possible_cpu(cpu) | |
1243 | free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); | |
06a71a24 | 1244 | hyp_cpu_pm_exit(); |
1e947bad MZ |
1245 | } |
1246 | ||
1247 | static int init_vhe_mode(void) | |
1248 | { | |
1e947bad MZ |
1249 | /* set size of VMID supported by CPU */ |
1250 | kvm_vmid_bits = kvm_get_vmid_bits(); | |
1251 | kvm_info("%d-bit VMID\n", kvm_vmid_bits); | |
1252 | ||
1253 | kvm_info("VHE mode initialized successfully\n"); | |
1254 | return 0; | |
1255 | } | |
1256 | ||
342cd0ab CD |
1257 | /** |
1258 | * Inits Hyp-mode on all online CPUs | |
1259 | */ | |
1260 | static int init_hyp_mode(void) | |
1261 | { | |
342cd0ab CD |
1262 | int cpu; |
1263 | int err = 0; | |
1264 | ||
1265 | /* | |
1266 | * Allocate Hyp PGD and setup Hyp identity mapping | |
1267 | */ | |
1268 | err = kvm_mmu_init(); | |
1269 | if (err) | |
1270 | goto out_err; | |
1271 | ||
1272 | /* | |
1273 | * It is probably enough to obtain the default on one | |
1274 | * CPU. It's unlikely to be different on the others. | |
1275 | */ | |
1276 | hyp_default_vectors = __hyp_get_vectors(); | |
1277 | ||
1278 | /* | |
1279 | * Allocate stack pages for Hypervisor-mode | |
1280 | */ | |
1281 | for_each_possible_cpu(cpu) { | |
1282 | unsigned long stack_page; | |
1283 | ||
1284 | stack_page = __get_free_page(GFP_KERNEL); | |
1285 | if (!stack_page) { | |
1286 | err = -ENOMEM; | |
1e947bad | 1287 | goto out_err; |
342cd0ab CD |
1288 | } |
1289 | ||
1290 | per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; | |
1291 | } | |
1292 | ||
342cd0ab CD |
1293 | /* |
1294 | * Map the Hyp-code called directly from the host | |
1295 | */ | |
588ab3f9 LT |
1296 | err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), |
1297 | kvm_ksym_ref(__hyp_text_end)); | |
342cd0ab CD |
1298 | if (err) { |
1299 | kvm_err("Cannot map world-switch code\n"); | |
1e947bad | 1300 | goto out_err; |
342cd0ab CD |
1301 | } |
1302 | ||
a0bf9776 AB |
1303 | err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), |
1304 | kvm_ksym_ref(__end_rodata)); | |
910917bb MZ |
1305 | if (err) { |
1306 | kvm_err("Cannot map rodata section\n"); | |
1e947bad | 1307 | goto out_err; |
910917bb MZ |
1308 | } |
1309 | ||
342cd0ab CD |
1310 | /* |
1311 | * Map the Hyp stack pages | |
1312 | */ | |
1313 | for_each_possible_cpu(cpu) { | |
1314 | char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); | |
1315 | err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE); | |
1316 | ||
1317 | if (err) { | |
1318 | kvm_err("Cannot map hyp stack\n"); | |
1e947bad | 1319 | goto out_err; |
342cd0ab CD |
1320 | } |
1321 | } | |
1322 | ||
342cd0ab | 1323 | for_each_possible_cpu(cpu) { |
3de50da6 | 1324 | kvm_cpu_context_t *cpu_ctxt; |
342cd0ab | 1325 | |
3de50da6 MZ |
1326 | cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu); |
1327 | err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1); | |
342cd0ab CD |
1328 | |
1329 | if (err) { | |
3de50da6 | 1330 | kvm_err("Cannot map host CPU state: %d\n", err); |
1e947bad | 1331 | goto out_err; |
342cd0ab CD |
1332 | } |
1333 | } | |
1334 | ||
d157f4a5 MZ |
1335 | #ifndef CONFIG_HOTPLUG_CPU |
1336 | free_boot_hyp_pgd(); | |
1337 | #endif | |
1338 | ||
20475f78 VM |
1339 | /* set size of VMID supported by CPU */ |
1340 | kvm_vmid_bits = kvm_get_vmid_bits(); | |
1341 | kvm_info("%d-bit VMID\n", kvm_vmid_bits); | |
1342 | ||
342cd0ab | 1343 | kvm_info("Hyp mode initialized successfully\n"); |
210552c1 | 1344 | |
342cd0ab | 1345 | return 0; |
1e947bad | 1346 | |
342cd0ab | 1347 | out_err: |
1e947bad | 1348 | teardown_hyp_mode(); |
342cd0ab CD |
1349 | kvm_err("error initializing Hyp mode: %d\n", err); |
1350 | return err; | |
1351 | } | |
1352 | ||
d4e071ce AP |
1353 | static void check_kvm_target_cpu(void *ret) |
1354 | { | |
1355 | *(int *)ret = kvm_target_cpu(); | |
1356 | } | |
1357 | ||
4429fc64 AP |
1358 | struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) |
1359 | { | |
1360 | struct kvm_vcpu *vcpu; | |
1361 | int i; | |
1362 | ||
1363 | mpidr &= MPIDR_HWID_BITMASK; | |
1364 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
1365 | if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) | |
1366 | return vcpu; | |
1367 | } | |
1368 | return NULL; | |
1369 | } | |
1370 | ||
342cd0ab CD |
1371 | /** |
1372 | * Initialize Hyp-mode and memory mappings on all CPUs. | |
1373 | */ | |
749cf76c CD |
1374 | int kvm_arch_init(void *opaque) |
1375 | { | |
342cd0ab | 1376 | int err; |
d4e071ce | 1377 | int ret, cpu; |
342cd0ab CD |
1378 | |
1379 | if (!is_hyp_mode_available()) { | |
1380 | kvm_err("HYP mode not available\n"); | |
1381 | return -ENODEV; | |
1382 | } | |
1383 | ||
d4e071ce AP |
1384 | for_each_online_cpu(cpu) { |
1385 | smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); | |
1386 | if (ret < 0) { | |
1387 | kvm_err("Error, CPU %d not supported!\n", cpu); | |
1388 | return -ENODEV; | |
1389 | } | |
342cd0ab CD |
1390 | } |
1391 | ||
1e947bad | 1392 | err = init_common_resources(); |
342cd0ab | 1393 | if (err) |
1e947bad | 1394 | return err; |
342cd0ab | 1395 | |
1e947bad MZ |
1396 | if (is_kernel_in_hyp_mode()) |
1397 | err = init_vhe_mode(); | |
1398 | else | |
1399 | err = init_hyp_mode(); | |
1400 | if (err) | |
d157f4a5 | 1401 | goto out_err; |
8146875d | 1402 | |
1e947bad MZ |
1403 | err = init_subsystems(); |
1404 | if (err) | |
1405 | goto out_hyp; | |
1fcf7ce0 | 1406 | |
749cf76c | 1407 | return 0; |
1e947bad MZ |
1408 | |
1409 | out_hyp: | |
1410 | teardown_hyp_mode(); | |
342cd0ab | 1411 | out_err: |
1e947bad | 1412 | teardown_common_resources(); |
342cd0ab | 1413 | return err; |
749cf76c CD |
1414 | } |
1415 | ||
1416 | /* NOP: Compiling as a module not supported */ | |
1417 | void kvm_arch_exit(void) | |
1418 | { | |
210552c1 | 1419 | kvm_perf_teardown(); |
749cf76c CD |
1420 | } |
1421 | ||
1422 | static int arm_init(void) | |
1423 | { | |
1424 | int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1425 | return rc; | |
1426 | } | |
1427 | ||
1428 | module_init(arm_init); |