]>
Commit | Line | Data |
---|---|---|
749cf76c CD |
1 | /* |
2 | * Copyright (C) 2012 - Virtual Open Systems and Columbia University | |
3 | * Author: Christoffer Dall <c.dall@virtualopensystems.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License, version 2, as | |
7 | * published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. | |
17 | */ | |
18 | ||
1fcf7ce0 | 19 | #include <linux/cpu_pm.h> |
749cf76c CD |
20 | #include <linux/errno.h> |
21 | #include <linux/err.h> | |
22 | #include <linux/kvm_host.h> | |
1085fdc6 | 23 | #include <linux/list.h> |
749cf76c CD |
24 | #include <linux/module.h> |
25 | #include <linux/vmalloc.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/mman.h> | |
28 | #include <linux/sched.h> | |
86ce8535 | 29 | #include <linux/kvm.h> |
749cf76c | 30 | #include <trace/events/kvm.h> |
b02386eb | 31 | #include <kvm/arm_pmu.h> |
749cf76c CD |
32 | |
33 | #define CREATE_TRACE_POINTS | |
34 | #include "trace.h" | |
35 | ||
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 | ||
67f69197 AT |
69 | static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled); |
70 | ||
1638a12d MZ |
71 | static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu) |
72 | { | |
73 | BUG_ON(preemptible()); | |
1436c1aa | 74 | __this_cpu_write(kvm_arm_running_vcpu, vcpu); |
1638a12d MZ |
75 | } |
76 | ||
77 | /** | |
78 | * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU. | |
79 | * Must be called from non-preemptible context | |
80 | */ | |
81 | struct kvm_vcpu *kvm_arm_get_running_vcpu(void) | |
82 | { | |
83 | BUG_ON(preemptible()); | |
1436c1aa | 84 | return __this_cpu_read(kvm_arm_running_vcpu); |
1638a12d MZ |
85 | } |
86 | ||
87 | /** | |
88 | * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
89 | */ | |
4000be42 | 90 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) |
1638a12d MZ |
91 | { |
92 | return &kvm_arm_running_vcpu; | |
93 | } | |
94 | ||
749cf76c CD |
95 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
96 | { | |
97 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; | |
98 | } | |
99 | ||
749cf76c CD |
100 | int kvm_arch_hardware_setup(void) |
101 | { | |
102 | return 0; | |
103 | } | |
104 | ||
749cf76c CD |
105 | void kvm_arch_check_processor_compat(void *rtn) |
106 | { | |
107 | *(int *)rtn = 0; | |
108 | } | |
109 | ||
749cf76c | 110 | |
d5d8184d CD |
111 | /** |
112 | * kvm_arch_init_vm - initializes a VM data structure | |
113 | * @kvm: pointer to the KVM struct | |
114 | */ | |
749cf76c CD |
115 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
116 | { | |
d5d8184d CD |
117 | int ret = 0; |
118 | ||
749cf76c CD |
119 | if (type) |
120 | return -EINVAL; | |
121 | ||
d5d8184d CD |
122 | ret = kvm_alloc_stage2_pgd(kvm); |
123 | if (ret) | |
124 | goto out_fail_alloc; | |
125 | ||
c8dddecd | 126 | ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP); |
d5d8184d CD |
127 | if (ret) |
128 | goto out_free_stage2_pgd; | |
129 | ||
6c3d63c9 | 130 | kvm_vgic_early_init(kvm); |
a1a64387 CD |
131 | kvm_timer_init(kvm); |
132 | ||
d5d8184d CD |
133 | /* Mark the initial VMID generation invalid */ |
134 | kvm->arch.vmid_gen = 0; | |
135 | ||
3caa2d8c | 136 | /* The maximum number of VCPUs is limited by the host's GIC model */ |
c7da6fa4 PF |
137 | kvm->arch.max_vcpus = vgic_present ? |
138 | kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; | |
3caa2d8c | 139 | |
d5d8184d CD |
140 | return ret; |
141 | out_free_stage2_pgd: | |
142 | kvm_free_stage2_pgd(kvm); | |
143 | out_fail_alloc: | |
144 | return ret; | |
749cf76c CD |
145 | } |
146 | ||
235539b4 LC |
147 | bool kvm_arch_has_vcpu_debugfs(void) |
148 | { | |
149 | return false; | |
150 | } | |
151 | ||
152 | int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu) | |
153 | { | |
154 | return 0; | |
155 | } | |
156 | ||
749cf76c CD |
157 | int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
158 | { | |
159 | return VM_FAULT_SIGBUS; | |
160 | } | |
161 | ||
749cf76c | 162 | |
d5d8184d CD |
163 | /** |
164 | * kvm_arch_destroy_vm - destroy the VM data structure | |
165 | * @kvm: pointer to the KVM struct | |
166 | */ | |
749cf76c CD |
167 | void kvm_arch_destroy_vm(struct kvm *kvm) |
168 | { | |
169 | int i; | |
170 | ||
171 | for (i = 0; i < KVM_MAX_VCPUS; ++i) { | |
172 | if (kvm->vcpus[i]) { | |
173 | kvm_arch_vcpu_free(kvm->vcpus[i]); | |
174 | kvm->vcpus[i] = NULL; | |
175 | } | |
176 | } | |
c1bfb577 MZ |
177 | |
178 | kvm_vgic_destroy(kvm); | |
749cf76c CD |
179 | } |
180 | ||
784aa3d7 | 181 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
749cf76c CD |
182 | { |
183 | int r; | |
184 | switch (ext) { | |
1a89dd91 | 185 | case KVM_CAP_IRQCHIP: |
c7da6fa4 PF |
186 | r = vgic_present; |
187 | break; | |
d44758c0 | 188 | case KVM_CAP_IOEVENTFD: |
7330672b | 189 | case KVM_CAP_DEVICE_CTRL: |
749cf76c CD |
190 | case KVM_CAP_USER_MEMORY: |
191 | case KVM_CAP_SYNC_MMU: | |
192 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
193 | case KVM_CAP_ONE_REG: | |
aa024c2f | 194 | case KVM_CAP_ARM_PSCI: |
4447a208 | 195 | case KVM_CAP_ARM_PSCI_0_2: |
98047888 | 196 | case KVM_CAP_READONLY_MEM: |
ecccf0cc | 197 | case KVM_CAP_MP_STATE: |
749cf76c CD |
198 | r = 1; |
199 | break; | |
200 | case KVM_CAP_COALESCED_MMIO: | |
201 | r = KVM_COALESCED_MMIO_PAGE_OFFSET; | |
202 | break; | |
3401d546 CD |
203 | case KVM_CAP_ARM_SET_DEVICE_ADDR: |
204 | r = 1; | |
ca46e10f | 205 | break; |
749cf76c CD |
206 | case KVM_CAP_NR_VCPUS: |
207 | r = num_online_cpus(); | |
208 | break; | |
209 | case KVM_CAP_MAX_VCPUS: | |
210 | r = KVM_MAX_VCPUS; | |
211 | break; | |
212 | default: | |
b46f01ce | 213 | r = kvm_arch_dev_ioctl_check_extension(kvm, ext); |
749cf76c CD |
214 | break; |
215 | } | |
216 | return r; | |
217 | } | |
218 | ||
219 | long kvm_arch_dev_ioctl(struct file *filp, | |
220 | unsigned int ioctl, unsigned long arg) | |
221 | { | |
222 | return -EINVAL; | |
223 | } | |
224 | ||
749cf76c CD |
225 | |
226 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) | |
227 | { | |
228 | int err; | |
229 | struct kvm_vcpu *vcpu; | |
230 | ||
716139df CD |
231 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) { |
232 | err = -EBUSY; | |
233 | goto out; | |
234 | } | |
235 | ||
3caa2d8c AP |
236 | if (id >= kvm->arch.max_vcpus) { |
237 | err = -EINVAL; | |
238 | goto out; | |
239 | } | |
240 | ||
749cf76c CD |
241 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
242 | if (!vcpu) { | |
243 | err = -ENOMEM; | |
244 | goto out; | |
245 | } | |
246 | ||
247 | err = kvm_vcpu_init(vcpu, kvm, id); | |
248 | if (err) | |
249 | goto free_vcpu; | |
250 | ||
c8dddecd | 251 | err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP); |
d5d8184d CD |
252 | if (err) |
253 | goto vcpu_uninit; | |
254 | ||
749cf76c | 255 | return vcpu; |
d5d8184d CD |
256 | vcpu_uninit: |
257 | kvm_vcpu_uninit(vcpu); | |
749cf76c CD |
258 | free_vcpu: |
259 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
260 | out: | |
261 | return ERR_PTR(err); | |
262 | } | |
263 | ||
31928aa5 | 264 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
749cf76c | 265 | { |
6c3d63c9 | 266 | kvm_vgic_vcpu_early_init(vcpu); |
749cf76c CD |
267 | } |
268 | ||
269 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
270 | { | |
d5d8184d | 271 | kvm_mmu_free_memory_caches(vcpu); |
967f8427 | 272 | kvm_timer_vcpu_terminate(vcpu); |
c1bfb577 | 273 | kvm_vgic_vcpu_destroy(vcpu); |
5f0a714a | 274 | kvm_pmu_vcpu_destroy(vcpu); |
591d215a | 275 | kvm_vcpu_uninit(vcpu); |
d5d8184d | 276 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
749cf76c CD |
277 | } |
278 | ||
279 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) | |
280 | { | |
281 | kvm_arch_vcpu_free(vcpu); | |
282 | } | |
283 | ||
284 | int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) | |
285 | { | |
1a748478 | 286 | return kvm_timer_should_fire(vcpu); |
749cf76c CD |
287 | } |
288 | ||
d35268da CD |
289 | void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) |
290 | { | |
291 | kvm_timer_schedule(vcpu); | |
292 | } | |
293 | ||
294 | void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) | |
295 | { | |
296 | kvm_timer_unschedule(vcpu); | |
297 | } | |
298 | ||
749cf76c CD |
299 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
300 | { | |
f7ed45be CD |
301 | /* Force users to call KVM_ARM_VCPU_INIT */ |
302 | vcpu->arch.target = -1; | |
f7fa034d | 303 | bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); |
1a89dd91 | 304 | |
967f8427 MZ |
305 | /* Set up the timer */ |
306 | kvm_timer_vcpu_init(vcpu); | |
307 | ||
84e690bf AB |
308 | kvm_arm_reset_debug_ptr(vcpu); |
309 | ||
749cf76c CD |
310 | return 0; |
311 | } | |
312 | ||
749cf76c CD |
313 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
314 | { | |
86ce8535 | 315 | vcpu->cpu = cpu; |
3de50da6 | 316 | vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state); |
5b3e5e5b | 317 | |
1638a12d | 318 | kvm_arm_set_running_vcpu(vcpu); |
749cf76c CD |
319 | } |
320 | ||
321 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
322 | { | |
e9b152cb CD |
323 | /* |
324 | * The arch-generic KVM code expects the cpu field of a vcpu to be -1 | |
325 | * if the vcpu is no longer assigned to a cpu. This is used for the | |
326 | * optimized make_all_cpus_request path. | |
327 | */ | |
328 | vcpu->cpu = -1; | |
329 | ||
1638a12d | 330 | kvm_arm_set_running_vcpu(NULL); |
9b4a3004 | 331 | kvm_timer_vcpu_put(vcpu); |
749cf76c CD |
332 | } |
333 | ||
749cf76c CD |
334 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
335 | struct kvm_mp_state *mp_state) | |
336 | { | |
3781528e | 337 | if (vcpu->arch.power_off) |
ecccf0cc AB |
338 | mp_state->mp_state = KVM_MP_STATE_STOPPED; |
339 | else | |
340 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
341 | ||
342 | return 0; | |
749cf76c CD |
343 | } |
344 | ||
345 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
346 | struct kvm_mp_state *mp_state) | |
347 | { | |
ecccf0cc AB |
348 | switch (mp_state->mp_state) { |
349 | case KVM_MP_STATE_RUNNABLE: | |
3781528e | 350 | vcpu->arch.power_off = false; |
ecccf0cc AB |
351 | break; |
352 | case KVM_MP_STATE_STOPPED: | |
3781528e | 353 | vcpu->arch.power_off = true; |
ecccf0cc AB |
354 | break; |
355 | default: | |
356 | return -EINVAL; | |
357 | } | |
358 | ||
359 | return 0; | |
749cf76c CD |
360 | } |
361 | ||
5b3e5e5b CD |
362 | /** |
363 | * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled | |
364 | * @v: The VCPU pointer | |
365 | * | |
366 | * If the guest CPU is not waiting for interrupts or an interrupt line is | |
367 | * asserted, the CPU is by definition runnable. | |
368 | */ | |
749cf76c CD |
369 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) |
370 | { | |
4f5f1dc0 | 371 | return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v)) |
3b92830a | 372 | && !v->arch.power_off && !v->arch.pause); |
749cf76c CD |
373 | } |
374 | ||
f7ed45be CD |
375 | /* Just ensure a guest exit from a particular CPU */ |
376 | static void exit_vm_noop(void *info) | |
377 | { | |
378 | } | |
379 | ||
380 | void force_vm_exit(const cpumask_t *mask) | |
381 | { | |
898f949f | 382 | preempt_disable(); |
f7ed45be | 383 | smp_call_function_many(mask, exit_vm_noop, NULL, true); |
898f949f | 384 | preempt_enable(); |
f7ed45be CD |
385 | } |
386 | ||
387 | /** | |
388 | * need_new_vmid_gen - check that the VMID is still valid | |
6a727b0b | 389 | * @kvm: The VM's VMID to check |
f7ed45be CD |
390 | * |
391 | * return true if there is a new generation of VMIDs being used | |
392 | * | |
393 | * The hardware supports only 256 values with the value zero reserved for the | |
394 | * host, so we check if an assigned value belongs to a previous generation, | |
395 | * which which requires us to assign a new value. If we're the first to use a | |
396 | * VMID for the new generation, we must flush necessary caches and TLBs on all | |
397 | * CPUs. | |
398 | */ | |
399 | static bool need_new_vmid_gen(struct kvm *kvm) | |
400 | { | |
401 | return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen)); | |
402 | } | |
403 | ||
404 | /** | |
405 | * update_vttbr - Update the VTTBR with a valid VMID before the guest runs | |
406 | * @kvm The guest that we are about to run | |
407 | * | |
408 | * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the | |
409 | * VM has a valid VMID, otherwise assigns a new one and flushes corresponding | |
410 | * caches and TLBs. | |
411 | */ | |
412 | static void update_vttbr(struct kvm *kvm) | |
413 | { | |
414 | phys_addr_t pgd_phys; | |
415 | u64 vmid; | |
416 | ||
417 | if (!need_new_vmid_gen(kvm)) | |
418 | return; | |
419 | ||
420 | spin_lock(&kvm_vmid_lock); | |
421 | ||
422 | /* | |
423 | * We need to re-check the vmid_gen here to ensure that if another vcpu | |
424 | * already allocated a valid vmid for this vm, then this vcpu should | |
425 | * use the same vmid. | |
426 | */ | |
427 | if (!need_new_vmid_gen(kvm)) { | |
428 | spin_unlock(&kvm_vmid_lock); | |
429 | return; | |
430 | } | |
431 | ||
432 | /* First user of a new VMID generation? */ | |
433 | if (unlikely(kvm_next_vmid == 0)) { | |
434 | atomic64_inc(&kvm_vmid_gen); | |
435 | kvm_next_vmid = 1; | |
436 | ||
437 | /* | |
438 | * On SMP we know no other CPUs can use this CPU's or each | |
439 | * other's VMID after force_vm_exit returns since the | |
440 | * kvm_vmid_lock blocks them from reentry to the guest. | |
441 | */ | |
442 | force_vm_exit(cpu_all_mask); | |
443 | /* | |
444 | * Now broadcast TLB + ICACHE invalidation over the inner | |
445 | * shareable domain to make sure all data structures are | |
446 | * clean. | |
447 | */ | |
448 | kvm_call_hyp(__kvm_flush_vm_context); | |
449 | } | |
450 | ||
451 | kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen); | |
452 | kvm->arch.vmid = kvm_next_vmid; | |
453 | kvm_next_vmid++; | |
20475f78 | 454 | kvm_next_vmid &= (1 << kvm_vmid_bits) - 1; |
f7ed45be CD |
455 | |
456 | /* update vttbr to be used with the new vmid */ | |
9163ee23 | 457 | pgd_phys = virt_to_phys(kvm->arch.pgd); |
dbff124e | 458 | BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK); |
20475f78 | 459 | vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits); |
dbff124e | 460 | kvm->arch.vttbr = pgd_phys | vmid; |
f7ed45be CD |
461 | |
462 | spin_unlock(&kvm_vmid_lock); | |
463 | } | |
464 | ||
f7ed45be CD |
465 | static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) |
466 | { | |
05971120 | 467 | struct kvm *kvm = vcpu->kvm; |
41a54482 | 468 | int ret = 0; |
e1ba0207 | 469 | |
f7ed45be CD |
470 | if (likely(vcpu->arch.has_run_once)) |
471 | return 0; | |
472 | ||
473 | vcpu->arch.has_run_once = true; | |
aa024c2f | 474 | |
01ac5e34 | 475 | /* |
6d3cfbe2 PM |
476 | * Map the VGIC hardware resources before running a vcpu the first |
477 | * time on this VM. | |
01ac5e34 | 478 | */ |
c2f58514 | 479 | if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) { |
05971120 | 480 | ret = kvm_vgic_map_resources(kvm); |
01ac5e34 MZ |
481 | if (ret) |
482 | return ret; | |
483 | } | |
484 | ||
05971120 CD |
485 | /* |
486 | * Enable the arch timers only if we have an in-kernel VGIC | |
487 | * and it has been properly initialized, since we cannot handle | |
488 | * interrupts from the virtual timer with a userspace gic. | |
489 | */ | |
490 | if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) | |
41a54482 | 491 | ret = kvm_timer_enable(vcpu); |
05971120 | 492 | |
41a54482 | 493 | return ret; |
f7ed45be CD |
494 | } |
495 | ||
c1426e4c EA |
496 | bool kvm_arch_intc_initialized(struct kvm *kvm) |
497 | { | |
498 | return vgic_initialized(kvm); | |
499 | } | |
500 | ||
b13216cf | 501 | void kvm_arm_halt_guest(struct kvm *kvm) |
3b92830a EA |
502 | { |
503 | int i; | |
504 | struct kvm_vcpu *vcpu; | |
505 | ||
506 | kvm_for_each_vcpu(i, vcpu, kvm) | |
507 | vcpu->arch.pause = true; | |
b13216cf | 508 | kvm_make_all_cpus_request(kvm, KVM_REQ_VCPU_EXIT); |
3b92830a EA |
509 | } |
510 | ||
35a2d585 CD |
511 | void kvm_arm_halt_vcpu(struct kvm_vcpu *vcpu) |
512 | { | |
513 | vcpu->arch.pause = true; | |
514 | kvm_vcpu_kick(vcpu); | |
515 | } | |
516 | ||
517 | void kvm_arm_resume_vcpu(struct kvm_vcpu *vcpu) | |
b13216cf CD |
518 | { |
519 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); | |
520 | ||
521 | vcpu->arch.pause = false; | |
522 | swake_up(wq); | |
523 | } | |
524 | ||
525 | void kvm_arm_resume_guest(struct kvm *kvm) | |
3b92830a EA |
526 | { |
527 | int i; | |
528 | struct kvm_vcpu *vcpu; | |
529 | ||
b13216cf CD |
530 | kvm_for_each_vcpu(i, vcpu, kvm) |
531 | kvm_arm_resume_vcpu(vcpu); | |
3b92830a EA |
532 | } |
533 | ||
3781528e | 534 | static void vcpu_sleep(struct kvm_vcpu *vcpu) |
aa024c2f | 535 | { |
8577370f | 536 | struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu); |
aa024c2f | 537 | |
8577370f | 538 | swait_event_interruptible(*wq, ((!vcpu->arch.power_off) && |
3b92830a | 539 | (!vcpu->arch.pause))); |
aa024c2f MZ |
540 | } |
541 | ||
e8180dca AP |
542 | static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) |
543 | { | |
544 | return vcpu->arch.target >= 0; | |
545 | } | |
546 | ||
f7ed45be CD |
547 | /** |
548 | * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code | |
549 | * @vcpu: The VCPU pointer | |
550 | * @run: The kvm_run structure pointer used for userspace state exchange | |
551 | * | |
552 | * This function is called through the VCPU_RUN ioctl called from user space. It | |
553 | * will execute VM code in a loop until the time slice for the process is used | |
554 | * or some emulation is needed from user space in which case the function will | |
555 | * return with return value 0 and with the kvm_run structure filled in with the | |
556 | * required data for the requested emulation. | |
557 | */ | |
749cf76c CD |
558 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) |
559 | { | |
f7ed45be CD |
560 | int ret; |
561 | sigset_t sigsaved; | |
562 | ||
e8180dca | 563 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
f7ed45be CD |
564 | return -ENOEXEC; |
565 | ||
566 | ret = kvm_vcpu_first_run_init(vcpu); | |
567 | if (ret) | |
568 | return ret; | |
569 | ||
45e96ea6 CD |
570 | if (run->exit_reason == KVM_EXIT_MMIO) { |
571 | ret = kvm_handle_mmio_return(vcpu, vcpu->run); | |
572 | if (ret) | |
573 | return ret; | |
574 | } | |
575 | ||
f7ed45be CD |
576 | if (vcpu->sigset_active) |
577 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
578 | ||
579 | ret = 1; | |
580 | run->exit_reason = KVM_EXIT_UNKNOWN; | |
581 | while (ret > 0) { | |
582 | /* | |
583 | * Check conditions before entering the guest | |
584 | */ | |
585 | cond_resched(); | |
586 | ||
587 | update_vttbr(vcpu->kvm); | |
588 | ||
3b92830a | 589 | if (vcpu->arch.power_off || vcpu->arch.pause) |
3781528e | 590 | vcpu_sleep(vcpu); |
aa024c2f | 591 | |
abdf5843 MZ |
592 | /* |
593 | * Preparing the interrupts to be injected also | |
594 | * involves poking the GIC, which must be done in a | |
595 | * non-preemptible context. | |
596 | */ | |
1b3d546d | 597 | preempt_disable(); |
b02386eb | 598 | kvm_pmu_flush_hwstate(vcpu); |
7e16aa81 | 599 | kvm_timer_flush_hwstate(vcpu); |
abdf5843 MZ |
600 | kvm_vgic_flush_hwstate(vcpu); |
601 | ||
f7ed45be CD |
602 | local_irq_disable(); |
603 | ||
604 | /* | |
605 | * Re-check atomic conditions | |
606 | */ | |
607 | if (signal_pending(current)) { | |
608 | ret = -EINTR; | |
609 | run->exit_reason = KVM_EXIT_INTR; | |
610 | } | |
611 | ||
101d3da0 | 612 | if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) || |
3b92830a | 613 | vcpu->arch.power_off || vcpu->arch.pause) { |
f7ed45be | 614 | local_irq_enable(); |
b02386eb | 615 | kvm_pmu_sync_hwstate(vcpu); |
4b4b4512 | 616 | kvm_timer_sync_hwstate(vcpu); |
1a89dd91 | 617 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 | 618 | preempt_enable(); |
f7ed45be CD |
619 | continue; |
620 | } | |
621 | ||
56c7f5e7 AB |
622 | kvm_arm_setup_debug(vcpu); |
623 | ||
f7ed45be CD |
624 | /************************************************************** |
625 | * Enter the guest | |
626 | */ | |
627 | trace_kvm_entry(*vcpu_pc(vcpu)); | |
6edaa530 | 628 | guest_enter_irqoff(); |
f7ed45be CD |
629 | vcpu->mode = IN_GUEST_MODE; |
630 | ||
631 | ret = kvm_call_hyp(__kvm_vcpu_run, vcpu); | |
632 | ||
633 | vcpu->mode = OUTSIDE_GUEST_MODE; | |
b19e6892 | 634 | vcpu->stat.exits++; |
1b3d546d CD |
635 | /* |
636 | * Back from guest | |
637 | *************************************************************/ | |
638 | ||
56c7f5e7 AB |
639 | kvm_arm_clear_debug(vcpu); |
640 | ||
f7ed45be CD |
641 | /* |
642 | * We may have taken a host interrupt in HYP mode (ie | |
643 | * while executing the guest). This interrupt is still | |
644 | * pending, as we haven't serviced it yet! | |
645 | * | |
646 | * We're now back in SVC mode, with interrupts | |
647 | * disabled. Enabling the interrupts now will have | |
648 | * the effect of taking the interrupt again, in SVC | |
649 | * mode this time. | |
650 | */ | |
651 | local_irq_enable(); | |
652 | ||
653 | /* | |
6edaa530 | 654 | * We do local_irq_enable() before calling guest_exit() so |
1b3d546d CD |
655 | * that if a timer interrupt hits while running the guest we |
656 | * account that tick as being spent in the guest. We enable | |
6edaa530 | 657 | * preemption after calling guest_exit() so that if we get |
1b3d546d CD |
658 | * preempted we make sure ticks after that is not counted as |
659 | * guest time. | |
660 | */ | |
6edaa530 | 661 | guest_exit(); |
b5905dc1 | 662 | trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); |
1b3d546d | 663 | |
4b4b4512 | 664 | /* |
b02386eb SZ |
665 | * We must sync the PMU and timer state before the vgic state so |
666 | * that the vgic can properly sample the updated state of the | |
4b4b4512 CD |
667 | * interrupt line. |
668 | */ | |
b02386eb | 669 | kvm_pmu_sync_hwstate(vcpu); |
4b4b4512 CD |
670 | kvm_timer_sync_hwstate(vcpu); |
671 | ||
1a89dd91 | 672 | kvm_vgic_sync_hwstate(vcpu); |
abdf5843 MZ |
673 | |
674 | preempt_enable(); | |
675 | ||
f7ed45be CD |
676 | ret = handle_exit(vcpu, run, ret); |
677 | } | |
678 | ||
679 | if (vcpu->sigset_active) | |
680 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
681 | return ret; | |
749cf76c CD |
682 | } |
683 | ||
86ce8535 CD |
684 | static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) |
685 | { | |
686 | int bit_index; | |
687 | bool set; | |
688 | unsigned long *ptr; | |
689 | ||
690 | if (number == KVM_ARM_IRQ_CPU_IRQ) | |
691 | bit_index = __ffs(HCR_VI); | |
692 | else /* KVM_ARM_IRQ_CPU_FIQ */ | |
693 | bit_index = __ffs(HCR_VF); | |
694 | ||
695 | ptr = (unsigned long *)&vcpu->arch.irq_lines; | |
696 | if (level) | |
697 | set = test_and_set_bit(bit_index, ptr); | |
698 | else | |
699 | set = test_and_clear_bit(bit_index, ptr); | |
700 | ||
701 | /* | |
702 | * If we didn't change anything, no need to wake up or kick other CPUs | |
703 | */ | |
704 | if (set == level) | |
705 | return 0; | |
706 | ||
707 | /* | |
708 | * The vcpu irq_lines field was updated, wake up sleeping VCPUs and | |
709 | * trigger a world-switch round on the running physical CPU to set the | |
710 | * virtual IRQ/FIQ fields in the HCR appropriately. | |
711 | */ | |
712 | kvm_vcpu_kick(vcpu); | |
713 | ||
714 | return 0; | |
715 | } | |
716 | ||
79558f11 AG |
717 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, |
718 | bool line_status) | |
86ce8535 CD |
719 | { |
720 | u32 irq = irq_level->irq; | |
721 | unsigned int irq_type, vcpu_idx, irq_num; | |
722 | int nrcpus = atomic_read(&kvm->online_vcpus); | |
723 | struct kvm_vcpu *vcpu = NULL; | |
724 | bool level = irq_level->level; | |
725 | ||
726 | irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; | |
727 | vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; | |
728 | irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; | |
729 | ||
730 | trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); | |
731 | ||
5863c2ce MZ |
732 | switch (irq_type) { |
733 | case KVM_ARM_IRQ_TYPE_CPU: | |
734 | if (irqchip_in_kernel(kvm)) | |
735 | return -ENXIO; | |
86ce8535 | 736 | |
5863c2ce MZ |
737 | if (vcpu_idx >= nrcpus) |
738 | return -EINVAL; | |
86ce8535 | 739 | |
5863c2ce MZ |
740 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); |
741 | if (!vcpu) | |
742 | return -EINVAL; | |
86ce8535 | 743 | |
5863c2ce MZ |
744 | if (irq_num > KVM_ARM_IRQ_CPU_FIQ) |
745 | return -EINVAL; | |
746 | ||
747 | return vcpu_interrupt_line(vcpu, irq_num, level); | |
748 | case KVM_ARM_IRQ_TYPE_PPI: | |
749 | if (!irqchip_in_kernel(kvm)) | |
750 | return -ENXIO; | |
751 | ||
752 | if (vcpu_idx >= nrcpus) | |
753 | return -EINVAL; | |
754 | ||
755 | vcpu = kvm_get_vcpu(kvm, vcpu_idx); | |
756 | if (!vcpu) | |
757 | return -EINVAL; | |
758 | ||
759 | if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) | |
760 | return -EINVAL; | |
86ce8535 | 761 | |
5863c2ce MZ |
762 | return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level); |
763 | case KVM_ARM_IRQ_TYPE_SPI: | |
764 | if (!irqchip_in_kernel(kvm)) | |
765 | return -ENXIO; | |
766 | ||
fd1d0ddf | 767 | if (irq_num < VGIC_NR_PRIVATE_IRQS) |
5863c2ce MZ |
768 | return -EINVAL; |
769 | ||
770 | return kvm_vgic_inject_irq(kvm, 0, irq_num, level); | |
771 | } | |
772 | ||
773 | return -EINVAL; | |
86ce8535 CD |
774 | } |
775 | ||
f7fa034d CD |
776 | static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, |
777 | const struct kvm_vcpu_init *init) | |
778 | { | |
779 | unsigned int i; | |
780 | int phys_target = kvm_target_cpu(); | |
781 | ||
782 | if (init->target != phys_target) | |
783 | return -EINVAL; | |
784 | ||
785 | /* | |
786 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
787 | * use the same target. | |
788 | */ | |
789 | if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) | |
790 | return -EINVAL; | |
791 | ||
792 | /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ | |
793 | for (i = 0; i < sizeof(init->features) * 8; i++) { | |
794 | bool set = (init->features[i / 32] & (1 << (i % 32))); | |
795 | ||
796 | if (set && i >= KVM_VCPU_MAX_FEATURES) | |
797 | return -ENOENT; | |
798 | ||
799 | /* | |
800 | * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must | |
801 | * use the same feature set. | |
802 | */ | |
803 | if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && | |
804 | test_bit(i, vcpu->arch.features) != set) | |
805 | return -EINVAL; | |
806 | ||
807 | if (set) | |
808 | set_bit(i, vcpu->arch.features); | |
809 | } | |
810 | ||
811 | vcpu->arch.target = phys_target; | |
812 | ||
813 | /* Now we know what it is, we can reset it. */ | |
814 | return kvm_reset_vcpu(vcpu); | |
815 | } | |
816 | ||
817 | ||
478a8237 CD |
818 | static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, |
819 | struct kvm_vcpu_init *init) | |
820 | { | |
821 | int ret; | |
822 | ||
823 | ret = kvm_vcpu_set_target(vcpu, init); | |
824 | if (ret) | |
825 | return ret; | |
826 | ||
957db105 CD |
827 | /* |
828 | * Ensure a rebooted VM will fault in RAM pages and detect if the | |
829 | * guest MMU is turned off and flush the caches as needed. | |
830 | */ | |
831 | if (vcpu->arch.has_run_once) | |
832 | stage2_unmap_vm(vcpu->kvm); | |
833 | ||
b856a591 CD |
834 | vcpu_reset_hcr(vcpu); |
835 | ||
478a8237 | 836 | /* |
3781528e | 837 | * Handle the "start in power-off" case. |
478a8237 | 838 | */ |
03f1d4c1 | 839 | if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) |
3781528e | 840 | vcpu->arch.power_off = true; |
3ad8b3de | 841 | else |
3781528e | 842 | vcpu->arch.power_off = false; |
478a8237 CD |
843 | |
844 | return 0; | |
845 | } | |
846 | ||
f577f6c2 SZ |
847 | static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, |
848 | struct kvm_device_attr *attr) | |
849 | { | |
850 | int ret = -ENXIO; | |
851 | ||
852 | switch (attr->group) { | |
853 | default: | |
bb0c70bc | 854 | ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); |
f577f6c2 SZ |
855 | break; |
856 | } | |
857 | ||
858 | return ret; | |
859 | } | |
860 | ||
861 | static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, | |
862 | struct kvm_device_attr *attr) | |
863 | { | |
864 | int ret = -ENXIO; | |
865 | ||
866 | switch (attr->group) { | |
867 | default: | |
bb0c70bc | 868 | ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); |
f577f6c2 SZ |
869 | break; |
870 | } | |
871 | ||
872 | return ret; | |
873 | } | |
874 | ||
875 | static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, | |
876 | struct kvm_device_attr *attr) | |
877 | { | |
878 | int ret = -ENXIO; | |
879 | ||
880 | switch (attr->group) { | |
881 | default: | |
bb0c70bc | 882 | ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); |
f577f6c2 SZ |
883 | break; |
884 | } | |
885 | ||
886 | return ret; | |
887 | } | |
888 | ||
749cf76c CD |
889 | long kvm_arch_vcpu_ioctl(struct file *filp, |
890 | unsigned int ioctl, unsigned long arg) | |
891 | { | |
892 | struct kvm_vcpu *vcpu = filp->private_data; | |
893 | void __user *argp = (void __user *)arg; | |
f577f6c2 | 894 | struct kvm_device_attr attr; |
749cf76c CD |
895 | |
896 | switch (ioctl) { | |
897 | case KVM_ARM_VCPU_INIT: { | |
898 | struct kvm_vcpu_init init; | |
899 | ||
900 | if (copy_from_user(&init, argp, sizeof(init))) | |
901 | return -EFAULT; | |
902 | ||
478a8237 | 903 | return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); |
749cf76c CD |
904 | } |
905 | case KVM_SET_ONE_REG: | |
906 | case KVM_GET_ONE_REG: { | |
907 | struct kvm_one_reg reg; | |
e8180dca AP |
908 | |
909 | if (unlikely(!kvm_vcpu_initialized(vcpu))) | |
910 | return -ENOEXEC; | |
911 | ||
749cf76c CD |
912 | if (copy_from_user(®, argp, sizeof(reg))) |
913 | return -EFAULT; | |
914 | if (ioctl == KVM_SET_ONE_REG) | |
915 | return kvm_arm_set_reg(vcpu, ®); | |
916 | else | |
917 | return kvm_arm_get_reg(vcpu, ®); | |
918 | } | |
919 | case KVM_GET_REG_LIST: { | |
920 | struct kvm_reg_list __user *user_list = argp; | |
921 | struct kvm_reg_list reg_list; | |
922 | unsigned n; | |
923 | ||
e8180dca AP |
924 | if (unlikely(!kvm_vcpu_initialized(vcpu))) |
925 | return -ENOEXEC; | |
926 | ||
749cf76c CD |
927 | if (copy_from_user(®_list, user_list, sizeof(reg_list))) |
928 | return -EFAULT; | |
929 | n = reg_list.n; | |
930 | reg_list.n = kvm_arm_num_regs(vcpu); | |
931 | if (copy_to_user(user_list, ®_list, sizeof(reg_list))) | |
932 | return -EFAULT; | |
933 | if (n < reg_list.n) | |
934 | return -E2BIG; | |
935 | return kvm_arm_copy_reg_indices(vcpu, user_list->reg); | |
936 | } | |
f577f6c2 SZ |
937 | case KVM_SET_DEVICE_ATTR: { |
938 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
939 | return -EFAULT; | |
940 | return kvm_arm_vcpu_set_attr(vcpu, &attr); | |
941 | } | |
942 | case KVM_GET_DEVICE_ATTR: { | |
943 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
944 | return -EFAULT; | |
945 | return kvm_arm_vcpu_get_attr(vcpu, &attr); | |
946 | } | |
947 | case KVM_HAS_DEVICE_ATTR: { | |
948 | if (copy_from_user(&attr, argp, sizeof(attr))) | |
949 | return -EFAULT; | |
950 | return kvm_arm_vcpu_has_attr(vcpu, &attr); | |
951 | } | |
749cf76c CD |
952 | default: |
953 | return -EINVAL; | |
954 | } | |
955 | } | |
956 | ||
53c810c3 MS |
957 | /** |
958 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
959 | * @kvm: kvm instance | |
960 | * @log: slot id and address to which we copy the log | |
961 | * | |
962 | * Steps 1-4 below provide general overview of dirty page logging. See | |
963 | * kvm_get_dirty_log_protect() function description for additional details. | |
964 | * | |
965 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
966 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
967 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
968 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
969 | * writes will be marked dirty for next log read. | |
970 | * | |
971 | * 1. Take a snapshot of the bit and clear it if needed. | |
972 | * 2. Write protect the corresponding page. | |
973 | * 3. Copy the snapshot to the userspace. | |
974 | * 4. Flush TLB's if needed. | |
975 | */ | |
749cf76c CD |
976 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
977 | { | |
53c810c3 MS |
978 | bool is_dirty = false; |
979 | int r; | |
980 | ||
981 | mutex_lock(&kvm->slots_lock); | |
982 | ||
983 | r = kvm_get_dirty_log_protect(kvm, log, &is_dirty); | |
984 | ||
985 | if (is_dirty) | |
986 | kvm_flush_remote_tlbs(kvm); | |
987 | ||
988 | mutex_unlock(&kvm->slots_lock); | |
989 | return r; | |
749cf76c CD |
990 | } |
991 | ||
3401d546 CD |
992 | static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, |
993 | struct kvm_arm_device_addr *dev_addr) | |
994 | { | |
330690cd CD |
995 | unsigned long dev_id, type; |
996 | ||
997 | dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> | |
998 | KVM_ARM_DEVICE_ID_SHIFT; | |
999 | type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> | |
1000 | KVM_ARM_DEVICE_TYPE_SHIFT; | |
1001 | ||
1002 | switch (dev_id) { | |
1003 | case KVM_ARM_DEVICE_VGIC_V2: | |
c7da6fa4 PF |
1004 | if (!vgic_present) |
1005 | return -ENXIO; | |
ce01e4e8 | 1006 | return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); |
330690cd CD |
1007 | default: |
1008 | return -ENODEV; | |
1009 | } | |
3401d546 CD |
1010 | } |
1011 | ||
749cf76c CD |
1012 | long kvm_arch_vm_ioctl(struct file *filp, |
1013 | unsigned int ioctl, unsigned long arg) | |
1014 | { | |
3401d546 CD |
1015 | struct kvm *kvm = filp->private_data; |
1016 | void __user *argp = (void __user *)arg; | |
1017 | ||
1018 | switch (ioctl) { | |
5863c2ce | 1019 | case KVM_CREATE_IRQCHIP: { |
a28ebea2 | 1020 | int ret; |
c7da6fa4 PF |
1021 | if (!vgic_present) |
1022 | return -ENXIO; | |
a28ebea2 CD |
1023 | mutex_lock(&kvm->lock); |
1024 | ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); | |
1025 | mutex_unlock(&kvm->lock); | |
1026 | return ret; | |
5863c2ce | 1027 | } |
3401d546 CD |
1028 | case KVM_ARM_SET_DEVICE_ADDR: { |
1029 | struct kvm_arm_device_addr dev_addr; | |
1030 | ||
1031 | if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) | |
1032 | return -EFAULT; | |
1033 | return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); | |
1034 | } | |
42c4e0c7 AP |
1035 | case KVM_ARM_PREFERRED_TARGET: { |
1036 | int err; | |
1037 | struct kvm_vcpu_init init; | |
1038 | ||
1039 | err = kvm_vcpu_preferred_target(&init); | |
1040 | if (err) | |
1041 | return err; | |
1042 | ||
1043 | if (copy_to_user(argp, &init, sizeof(init))) | |
1044 | return -EFAULT; | |
1045 | ||
1046 | return 0; | |
1047 | } | |
3401d546 CD |
1048 | default: |
1049 | return -EINVAL; | |
1050 | } | |
749cf76c CD |
1051 | } |
1052 | ||
d157f4a5 | 1053 | static void cpu_init_hyp_mode(void *dummy) |
342cd0ab | 1054 | { |
dac288f7 | 1055 | phys_addr_t pgd_ptr; |
342cd0ab CD |
1056 | unsigned long hyp_stack_ptr; |
1057 | unsigned long stack_page; | |
1058 | unsigned long vector_ptr; | |
1059 | ||
1060 | /* Switch from the HYP stub to our own HYP init vector */ | |
5a677ce0 | 1061 | __hyp_set_vectors(kvm_get_idmap_vector()); |
342cd0ab | 1062 | |
dac288f7 | 1063 | pgd_ptr = kvm_mmu_get_httbr(); |
1436c1aa | 1064 | stack_page = __this_cpu_read(kvm_arm_hyp_stack_page); |
342cd0ab | 1065 | hyp_stack_ptr = stack_page + PAGE_SIZE; |
a0bf9776 | 1066 | vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector); |
342cd0ab | 1067 | |
12fda812 | 1068 | __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr); |
35a2491a | 1069 | __cpu_init_stage2(); |
56c7f5e7 AB |
1070 | |
1071 | kvm_arm_init_debug(); | |
342cd0ab CD |
1072 | } |
1073 | ||
5f5560b1 JM |
1074 | static void cpu_hyp_reinit(void) |
1075 | { | |
1076 | if (is_kernel_in_hyp_mode()) { | |
1077 | /* | |
67f69197 | 1078 | * __cpu_init_stage2() is safe to call even if the PM |
5f5560b1 JM |
1079 | * event was cancelled before the CPU was reset. |
1080 | */ | |
67f69197 | 1081 | __cpu_init_stage2(); |
5f5560b1 JM |
1082 | } else { |
1083 | if (__hyp_get_vectors() == hyp_default_vectors) | |
1084 | cpu_init_hyp_mode(NULL); | |
1085 | } | |
1086 | } | |
1087 | ||
67f69197 | 1088 | static void cpu_hyp_reset(void) |
d157f4a5 | 1089 | { |
12fda812 | 1090 | if (!is_kernel_in_hyp_mode()) |
e537ecd7 MZ |
1091 | __cpu_reset_hyp_mode(hyp_default_vectors, |
1092 | kvm_get_idmap_start()); | |
67f69197 AT |
1093 | } |
1094 | ||
1095 | static void _kvm_arch_hardware_enable(void *discard) | |
1096 | { | |
1097 | if (!__this_cpu_read(kvm_arm_hardware_enabled)) { | |
5f5560b1 | 1098 | cpu_hyp_reinit(); |
67f69197 | 1099 | __this_cpu_write(kvm_arm_hardware_enabled, 1); |
d157f4a5 | 1100 | } |
67f69197 | 1101 | } |
d157f4a5 | 1102 | |
67f69197 AT |
1103 | int kvm_arch_hardware_enable(void) |
1104 | { | |
1105 | _kvm_arch_hardware_enable(NULL); | |
1106 | return 0; | |
342cd0ab CD |
1107 | } |
1108 | ||
67f69197 AT |
1109 | static void _kvm_arch_hardware_disable(void *discard) |
1110 | { | |
1111 | if (__this_cpu_read(kvm_arm_hardware_enabled)) { | |
1112 | cpu_hyp_reset(); | |
1113 | __this_cpu_write(kvm_arm_hardware_enabled, 0); | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | void kvm_arch_hardware_disable(void) | |
1118 | { | |
1119 | _kvm_arch_hardware_disable(NULL); | |
1120 | } | |
d157f4a5 | 1121 | |
1fcf7ce0 LP |
1122 | #ifdef CONFIG_CPU_PM |
1123 | static int hyp_init_cpu_pm_notifier(struct notifier_block *self, | |
1124 | unsigned long cmd, | |
1125 | void *v) | |
1126 | { | |
67f69197 AT |
1127 | /* |
1128 | * kvm_arm_hardware_enabled is left with its old value over | |
1129 | * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should | |
1130 | * re-enable hyp. | |
1131 | */ | |
1132 | switch (cmd) { | |
1133 | case CPU_PM_ENTER: | |
1134 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1135 | /* | |
1136 | * don't update kvm_arm_hardware_enabled here | |
1137 | * so that the hardware will be re-enabled | |
1138 | * when we resume. See below. | |
1139 | */ | |
1140 | cpu_hyp_reset(); | |
1141 | ||
1fcf7ce0 | 1142 | return NOTIFY_OK; |
67f69197 AT |
1143 | case CPU_PM_EXIT: |
1144 | if (__this_cpu_read(kvm_arm_hardware_enabled)) | |
1145 | /* The hardware was enabled before suspend. */ | |
1146 | cpu_hyp_reinit(); | |
1fcf7ce0 | 1147 | |
67f69197 AT |
1148 | return NOTIFY_OK; |
1149 | ||
1150 | default: | |
1151 | return NOTIFY_DONE; | |
1152 | } | |
1fcf7ce0 LP |
1153 | } |
1154 | ||
1155 | static struct notifier_block hyp_init_cpu_pm_nb = { | |
1156 | .notifier_call = hyp_init_cpu_pm_notifier, | |
1157 | }; | |
1158 | ||
1159 | static void __init hyp_cpu_pm_init(void) | |
1160 | { | |
1161 | cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); | |
1162 | } | |
06a71a24 SH |
1163 | static void __init hyp_cpu_pm_exit(void) |
1164 | { | |
1165 | cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); | |
1166 | } | |
1fcf7ce0 LP |
1167 | #else |
1168 | static inline void hyp_cpu_pm_init(void) | |
1169 | { | |
1170 | } | |
06a71a24 SH |
1171 | static inline void hyp_cpu_pm_exit(void) |
1172 | { | |
1173 | } | |
1fcf7ce0 LP |
1174 | #endif |
1175 | ||
1e947bad MZ |
1176 | static void teardown_common_resources(void) |
1177 | { | |
1178 | free_percpu(kvm_host_cpu_state); | |
1179 | } | |
1180 | ||
1181 | static int init_common_resources(void) | |
1182 | { | |
1183 | kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t); | |
1184 | if (!kvm_host_cpu_state) { | |
1185 | kvm_err("Cannot allocate host CPU state\n"); | |
1186 | return -ENOMEM; | |
1187 | } | |
1188 | ||
61349937 VM |
1189 | /* set size of VMID supported by CPU */ |
1190 | kvm_vmid_bits = kvm_get_vmid_bits(); | |
1191 | kvm_info("%d-bit VMID\n", kvm_vmid_bits); | |
1192 | ||
1e947bad MZ |
1193 | return 0; |
1194 | } | |
1195 | ||
1196 | static int init_subsystems(void) | |
1197 | { | |
67f69197 | 1198 | int err = 0; |
1e947bad | 1199 | |
5f5560b1 | 1200 | /* |
67f69197 | 1201 | * Enable hardware so that subsystem initialisation can access EL2. |
5f5560b1 | 1202 | */ |
67f69197 | 1203 | on_each_cpu(_kvm_arch_hardware_enable, NULL, 1); |
5f5560b1 JM |
1204 | |
1205 | /* | |
1206 | * Register CPU lower-power notifier | |
1207 | */ | |
1208 | hyp_cpu_pm_init(); | |
1209 | ||
1e947bad MZ |
1210 | /* |
1211 | * Init HYP view of VGIC | |
1212 | */ | |
1213 | err = kvm_vgic_hyp_init(); | |
1214 | switch (err) { | |
1215 | case 0: | |
1216 | vgic_present = true; | |
1217 | break; | |
1218 | case -ENODEV: | |
1219 | case -ENXIO: | |
1220 | vgic_present = false; | |
67f69197 | 1221 | err = 0; |
1e947bad MZ |
1222 | break; |
1223 | default: | |
67f69197 | 1224 | goto out; |
1e947bad MZ |
1225 | } |
1226 | ||
1227 | /* | |
1228 | * Init HYP architected timer support | |
1229 | */ | |
1230 | err = kvm_timer_hyp_init(); | |
1231 | if (err) | |
67f69197 | 1232 | goto out; |
1e947bad MZ |
1233 | |
1234 | kvm_perf_init(); | |
1235 | kvm_coproc_table_init(); | |
1236 | ||
67f69197 AT |
1237 | out: |
1238 | on_each_cpu(_kvm_arch_hardware_disable, NULL, 1); | |
1239 | ||
1240 | return err; | |
1e947bad MZ |
1241 | } |
1242 | ||
1243 | static void teardown_hyp_mode(void) | |
1244 | { | |
1245 | int cpu; | |
1246 | ||
1247 | if (is_kernel_in_hyp_mode()) | |
1248 | return; | |
1249 | ||
1250 | free_hyp_pgds(); | |
1251 | for_each_possible_cpu(cpu) | |
1252 | free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); | |
06a71a24 | 1253 | hyp_cpu_pm_exit(); |
1e947bad MZ |
1254 | } |
1255 | ||
1256 | static int init_vhe_mode(void) | |
1257 | { | |
1e947bad MZ |
1258 | kvm_info("VHE mode initialized successfully\n"); |
1259 | return 0; | |
1260 | } | |
1261 | ||
342cd0ab CD |
1262 | /** |
1263 | * Inits Hyp-mode on all online CPUs | |
1264 | */ | |
1265 | static int init_hyp_mode(void) | |
1266 | { | |
342cd0ab CD |
1267 | int cpu; |
1268 | int err = 0; | |
1269 | ||
1270 | /* | |
1271 | * Allocate Hyp PGD and setup Hyp identity mapping | |
1272 | */ | |
1273 | err = kvm_mmu_init(); | |
1274 | if (err) | |
1275 | goto out_err; | |
1276 | ||
1277 | /* | |
1278 | * It is probably enough to obtain the default on one | |
1279 | * CPU. It's unlikely to be different on the others. | |
1280 | */ | |
1281 | hyp_default_vectors = __hyp_get_vectors(); | |
1282 | ||
1283 | /* | |
1284 | * Allocate stack pages for Hypervisor-mode | |
1285 | */ | |
1286 | for_each_possible_cpu(cpu) { | |
1287 | unsigned long stack_page; | |
1288 | ||
1289 | stack_page = __get_free_page(GFP_KERNEL); | |
1290 | if (!stack_page) { | |
1291 | err = -ENOMEM; | |
1e947bad | 1292 | goto out_err; |
342cd0ab CD |
1293 | } |
1294 | ||
1295 | per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; | |
1296 | } | |
1297 | ||
342cd0ab CD |
1298 | /* |
1299 | * Map the Hyp-code called directly from the host | |
1300 | */ | |
588ab3f9 | 1301 | err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), |
59002705 | 1302 | kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC); |
342cd0ab CD |
1303 | if (err) { |
1304 | kvm_err("Cannot map world-switch code\n"); | |
1e947bad | 1305 | goto out_err; |
342cd0ab CD |
1306 | } |
1307 | ||
a0bf9776 | 1308 | err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), |
74a6b888 | 1309 | kvm_ksym_ref(__end_rodata), PAGE_HYP_RO); |
910917bb MZ |
1310 | if (err) { |
1311 | kvm_err("Cannot map rodata section\n"); | |
1e947bad | 1312 | goto out_err; |
910917bb MZ |
1313 | } |
1314 | ||
342cd0ab CD |
1315 | /* |
1316 | * Map the Hyp stack pages | |
1317 | */ | |
1318 | for_each_possible_cpu(cpu) { | |
1319 | char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); | |
c8dddecd MZ |
1320 | err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE, |
1321 | PAGE_HYP); | |
342cd0ab CD |
1322 | |
1323 | if (err) { | |
1324 | kvm_err("Cannot map hyp stack\n"); | |
1e947bad | 1325 | goto out_err; |
342cd0ab CD |
1326 | } |
1327 | } | |
1328 | ||
342cd0ab | 1329 | for_each_possible_cpu(cpu) { |
3de50da6 | 1330 | kvm_cpu_context_t *cpu_ctxt; |
342cd0ab | 1331 | |
3de50da6 | 1332 | cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu); |
c8dddecd | 1333 | err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP); |
342cd0ab CD |
1334 | |
1335 | if (err) { | |
3de50da6 | 1336 | kvm_err("Cannot map host CPU state: %d\n", err); |
1e947bad | 1337 | goto out_err; |
342cd0ab CD |
1338 | } |
1339 | } | |
1340 | ||
1341 | kvm_info("Hyp mode initialized successfully\n"); | |
210552c1 | 1342 | |
342cd0ab | 1343 | return 0; |
1e947bad | 1344 | |
342cd0ab | 1345 | out_err: |
1e947bad | 1346 | teardown_hyp_mode(); |
342cd0ab CD |
1347 | kvm_err("error initializing Hyp mode: %d\n", err); |
1348 | return err; | |
1349 | } | |
1350 | ||
d4e071ce AP |
1351 | static void check_kvm_target_cpu(void *ret) |
1352 | { | |
1353 | *(int *)ret = kvm_target_cpu(); | |
1354 | } | |
1355 | ||
4429fc64 AP |
1356 | struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) |
1357 | { | |
1358 | struct kvm_vcpu *vcpu; | |
1359 | int i; | |
1360 | ||
1361 | mpidr &= MPIDR_HWID_BITMASK; | |
1362 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
1363 | if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) | |
1364 | return vcpu; | |
1365 | } | |
1366 | return NULL; | |
1367 | } | |
1368 | ||
342cd0ab CD |
1369 | /** |
1370 | * Initialize Hyp-mode and memory mappings on all CPUs. | |
1371 | */ | |
749cf76c CD |
1372 | int kvm_arch_init(void *opaque) |
1373 | { | |
342cd0ab | 1374 | int err; |
d4e071ce | 1375 | int ret, cpu; |
342cd0ab CD |
1376 | |
1377 | if (!is_hyp_mode_available()) { | |
1378 | kvm_err("HYP mode not available\n"); | |
1379 | return -ENODEV; | |
1380 | } | |
1381 | ||
d4e071ce AP |
1382 | for_each_online_cpu(cpu) { |
1383 | smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); | |
1384 | if (ret < 0) { | |
1385 | kvm_err("Error, CPU %d not supported!\n", cpu); | |
1386 | return -ENODEV; | |
1387 | } | |
342cd0ab CD |
1388 | } |
1389 | ||
1e947bad | 1390 | err = init_common_resources(); |
342cd0ab | 1391 | if (err) |
1e947bad | 1392 | return err; |
342cd0ab | 1393 | |
1e947bad MZ |
1394 | if (is_kernel_in_hyp_mode()) |
1395 | err = init_vhe_mode(); | |
1396 | else | |
1397 | err = init_hyp_mode(); | |
1398 | if (err) | |
d157f4a5 | 1399 | goto out_err; |
8146875d | 1400 | |
1e947bad MZ |
1401 | err = init_subsystems(); |
1402 | if (err) | |
1403 | goto out_hyp; | |
1fcf7ce0 | 1404 | |
749cf76c | 1405 | return 0; |
1e947bad MZ |
1406 | |
1407 | out_hyp: | |
1408 | teardown_hyp_mode(); | |
342cd0ab | 1409 | out_err: |
1e947bad | 1410 | teardown_common_resources(); |
342cd0ab | 1411 | return err; |
749cf76c CD |
1412 | } |
1413 | ||
1414 | /* NOP: Compiling as a module not supported */ | |
1415 | void kvm_arch_exit(void) | |
1416 | { | |
210552c1 | 1417 | kvm_perf_teardown(); |
749cf76c CD |
1418 | } |
1419 | ||
1420 | static int arm_init(void) | |
1421 | { | |
1422 | int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); | |
1423 | return rc; | |
1424 | } | |
1425 | ||
1426 | module_init(arm_init); |