]>
Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
6aa8b732 AK |
2 | /* |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * This module enables machines with Intel VT-x extensions to run virtual | |
6 | * machines without emulation or binary translation. | |
7 | * | |
8 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 9 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
10 | * |
11 | * Authors: | |
12 | * Avi Kivity <avi@qumranet.com> | |
13 | * Yaniv Kamay <yaniv@qumranet.com> | |
6aa8b732 AK |
14 | */ |
15 | ||
af669ac6 | 16 | #include <kvm/iodev.h> |
6aa8b732 | 17 | |
edf88417 | 18 | #include <linux/kvm_host.h> |
6aa8b732 AK |
19 | #include <linux/kvm.h> |
20 | #include <linux/module.h> | |
21 | #include <linux/errno.h> | |
6aa8b732 | 22 | #include <linux/percpu.h> |
6aa8b732 AK |
23 | #include <linux/mm.h> |
24 | #include <linux/miscdevice.h> | |
25 | #include <linux/vmalloc.h> | |
6aa8b732 | 26 | #include <linux/reboot.h> |
6aa8b732 AK |
27 | #include <linux/debugfs.h> |
28 | #include <linux/highmem.h> | |
29 | #include <linux/file.h> | |
fb3600cc | 30 | #include <linux/syscore_ops.h> |
774c47f1 | 31 | #include <linux/cpu.h> |
174cd4b1 | 32 | #include <linux/sched/signal.h> |
6e84f315 | 33 | #include <linux/sched/mm.h> |
03441a34 | 34 | #include <linux/sched/stat.h> |
d9e368d6 AK |
35 | #include <linux/cpumask.h> |
36 | #include <linux/smp.h> | |
d6d28168 | 37 | #include <linux/anon_inodes.h> |
04d2cc77 | 38 | #include <linux/profile.h> |
7aa81cc0 | 39 | #include <linux/kvm_para.h> |
6fc138d2 | 40 | #include <linux/pagemap.h> |
8d4e1288 | 41 | #include <linux/mman.h> |
35149e21 | 42 | #include <linux/swap.h> |
e56d532f | 43 | #include <linux/bitops.h> |
547de29e | 44 | #include <linux/spinlock.h> |
6ff5894c | 45 | #include <linux/compat.h> |
bc6678a3 | 46 | #include <linux/srcu.h> |
8f0b1ab6 | 47 | #include <linux/hugetlb.h> |
5a0e3ad6 | 48 | #include <linux/slab.h> |
743eeb0b SL |
49 | #include <linux/sort.h> |
50 | #include <linux/bsearch.h> | |
c011d23b | 51 | #include <linux/io.h> |
2eb06c30 | 52 | #include <linux/lockdep.h> |
c57c8046 | 53 | #include <linux/kthread.h> |
2fdef3a2 | 54 | #include <linux/suspend.h> |
6aa8b732 | 55 | |
e495606d | 56 | #include <asm/processor.h> |
2ea75be3 | 57 | #include <asm/ioctl.h> |
7c0f6ba6 | 58 | #include <linux/uaccess.h> |
6aa8b732 | 59 | |
5f94c174 | 60 | #include "coalesced_mmio.h" |
af585b92 | 61 | #include "async_pf.h" |
531810ca | 62 | #include "mmu_lock.h" |
3c3c29fd | 63 | #include "vfio.h" |
5f94c174 | 64 | |
229456fc MT |
65 | #define CREATE_TRACE_POINTS |
66 | #include <trace/events/kvm.h> | |
67 | ||
fb04a1ed PX |
68 | #include <linux/kvm_dirty_ring.h> |
69 | ||
536a6f88 JF |
70 | /* Worst case buffer size needed for holding an integer. */ |
71 | #define ITOA_MAX_LEN 12 | |
72 | ||
6aa8b732 AK |
73 | MODULE_AUTHOR("Qumranet"); |
74 | MODULE_LICENSE("GPL"); | |
75 | ||
920552b2 | 76 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 77 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
039c5d1b | 78 | module_param(halt_poll_ns, uint, 0644); |
ec76d819 | 79 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 80 | |
aca6ff29 | 81 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 82 | unsigned int halt_poll_ns_grow = 2; |
039c5d1b | 83 | module_param(halt_poll_ns_grow, uint, 0644); |
ec76d819 | 84 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 | 85 | |
49113d36 NW |
86 | /* The start value to grow halt_poll_ns from */ |
87 | unsigned int halt_poll_ns_grow_start = 10000; /* 10us */ | |
88 | module_param(halt_poll_ns_grow_start, uint, 0644); | |
89 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start); | |
90 | ||
aca6ff29 | 91 | /* Default resets per-vcpu halt_poll_ns . */ |
ec76d819 | 92 | unsigned int halt_poll_ns_shrink; |
039c5d1b | 93 | module_param(halt_poll_ns_shrink, uint, 0644); |
ec76d819 | 94 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 95 | |
fa40a821 MT |
96 | /* |
97 | * Ordering of locks: | |
98 | * | |
b7d409de | 99 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
100 | */ |
101 | ||
0d9ce162 | 102 | DEFINE_MUTEX(kvm_lock); |
4a937f96 | 103 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 104 | LIST_HEAD(vm_list); |
133de902 | 105 | |
7f59f492 | 106 | static cpumask_var_t cpus_hardware_enabled; |
f4fee932 | 107 | static int kvm_usage_count; |
10474ae8 | 108 | static atomic_t hardware_enable_failed; |
1b6c0168 | 109 | |
aaba298c | 110 | static struct kmem_cache *kvm_vcpu_cache; |
1165f5fe | 111 | |
15ad7146 | 112 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
7495e22b | 113 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu); |
15ad7146 | 114 | |
76f7c879 | 115 | struct dentry *kvm_debugfs_dir; |
e23a808b | 116 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 117 | |
09cbcef6 | 118 | static const struct file_operations stat_fops_per_vm; |
536a6f88 | 119 | |
bccf2150 AK |
120 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
121 | unsigned long arg); | |
de8e5d74 | 122 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
123 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
124 | unsigned long arg); | |
7ddfd3e0 MZ |
125 | #define KVM_COMPAT(c) .compat_ioctl = (c) |
126 | #else | |
9cb09e7c MZ |
127 | /* |
128 | * For architectures that don't implement a compat infrastructure, | |
129 | * adopt a double line of defense: | |
130 | * - Prevent a compat task from opening /dev/kvm | |
131 | * - If the open has been done by a 64bit task, and the KVM fd | |
132 | * passed to a compat task, let the ioctls fail. | |
133 | */ | |
7ddfd3e0 MZ |
134 | static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl, |
135 | unsigned long arg) { return -EINVAL; } | |
b9876e6d MZ |
136 | |
137 | static int kvm_no_compat_open(struct inode *inode, struct file *file) | |
138 | { | |
139 | return is_compat_task() ? -ENODEV : 0; | |
140 | } | |
141 | #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \ | |
142 | .open = kvm_no_compat_open | |
1dda606c | 143 | #endif |
10474ae8 AG |
144 | static int hardware_enable_all(void); |
145 | static void hardware_disable_all(void); | |
bccf2150 | 146 | |
e93f8a0f | 147 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 148 | |
52480137 | 149 | __visible bool kvm_rebooting; |
b7c4145b | 150 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 151 | |
286de8f6 CI |
152 | #define KVM_EVENT_CREATE_VM 0 |
153 | #define KVM_EVENT_DESTROY_VM 1 | |
154 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); | |
155 | static unsigned long long kvm_createvm_count; | |
156 | static unsigned long long kvm_active_vms; | |
157 | ||
baff59cc VK |
158 | static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask); |
159 | ||
e649b3f0 ET |
160 | __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
161 | unsigned long start, unsigned long end) | |
b1394e74 RK |
162 | { |
163 | } | |
164 | ||
a78986aa SC |
165 | bool kvm_is_zone_device_pfn(kvm_pfn_t pfn) |
166 | { | |
167 | /* | |
168 | * The metadata used by is_zone_device_page() to determine whether or | |
169 | * not a page is ZONE_DEVICE is guaranteed to be valid if and only if | |
170 | * the device has been pinned, e.g. by get_user_pages(). WARN if the | |
171 | * page_count() is zero to help detect bad usage of this helper. | |
172 | */ | |
173 | if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn)))) | |
174 | return false; | |
175 | ||
176 | return is_zone_device_page(pfn_to_page(pfn)); | |
177 | } | |
178 | ||
ba049e93 | 179 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
cbff90a7 | 180 | { |
a78986aa SC |
181 | /* |
182 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting | |
183 | * perspective they are "normal" pages, albeit with slightly different | |
184 | * usage rules. | |
185 | */ | |
11feeb49 | 186 | if (pfn_valid(pfn)) |
a78986aa | 187 | return PageReserved(pfn_to_page(pfn)) && |
7df003c8 | 188 | !is_zero_pfn(pfn) && |
a78986aa | 189 | !kvm_is_zone_device_pfn(pfn); |
cbff90a7 BAY |
190 | |
191 | return true; | |
192 | } | |
193 | ||
bccf2150 AK |
194 | /* |
195 | * Switches to specified vcpu, until a matching vcpu_put() | |
196 | */ | |
ec7660cc | 197 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 198 | { |
ec7660cc | 199 | int cpu = get_cpu(); |
7495e22b PB |
200 | |
201 | __this_cpu_write(kvm_running_vcpu, vcpu); | |
15ad7146 | 202 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 203 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 204 | put_cpu(); |
6aa8b732 | 205 | } |
2f1fe811 | 206 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 207 | |
313a3dc7 | 208 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 209 | { |
15ad7146 | 210 | preempt_disable(); |
313a3dc7 | 211 | kvm_arch_vcpu_put(vcpu); |
15ad7146 | 212 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
7495e22b | 213 | __this_cpu_write(kvm_running_vcpu, NULL); |
15ad7146 | 214 | preempt_enable(); |
6aa8b732 | 215 | } |
2f1fe811 | 216 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 217 | |
7a97cec2 PB |
218 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
219 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
220 | { | |
221 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
222 | ||
223 | /* | |
224 | * We need to wait for the VCPU to reenable interrupts and get out of | |
225 | * READING_SHADOW_PAGE_TABLES mode. | |
226 | */ | |
227 | if (req & KVM_REQUEST_WAIT) | |
228 | return mode != OUTSIDE_GUEST_MODE; | |
229 | ||
230 | /* | |
231 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
232 | */ | |
233 | return mode == IN_GUEST_MODE; | |
234 | } | |
235 | ||
d9e368d6 AK |
236 | static void ack_flush(void *_completed) |
237 | { | |
d9e368d6 AK |
238 | } |
239 | ||
620b2438 | 240 | static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait) |
b49defe8 | 241 | { |
b49defe8 PB |
242 | if (cpumask_empty(cpus)) |
243 | return false; | |
244 | ||
245 | smp_call_function_many(cpus, ack_flush, NULL, wait); | |
246 | return true; | |
247 | } | |
248 | ||
ae0946cd | 249 | static void kvm_make_vcpu_request(struct kvm *kvm, struct kvm_vcpu *vcpu, |
620b2438 | 250 | unsigned int req, struct cpumask *tmp, |
ae0946cd VK |
251 | int current_cpu) |
252 | { | |
253 | int cpu; | |
254 | ||
255 | kvm_make_request(req, vcpu); | |
256 | ||
257 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) | |
258 | return; | |
259 | ||
ae0946cd VK |
260 | /* |
261 | * Note, the vCPU could get migrated to a different pCPU at any point | |
262 | * after kvm_request_needs_ipi(), which could result in sending an IPI | |
263 | * to the previous pCPU. But, that's OK because the purpose of the IPI | |
264 | * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is | |
265 | * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES | |
266 | * after this point is also OK, as the requirement is only that KVM wait | |
267 | * for vCPUs that were reading SPTEs _before_ any changes were | |
268 | * finalized. See kvm_vcpu_kick() for more details on handling requests. | |
269 | */ | |
270 | if (kvm_request_needs_ipi(vcpu, req)) { | |
271 | cpu = READ_ONCE(vcpu->cpu); | |
272 | if (cpu != -1 && cpu != current_cpu) | |
273 | __cpumask_set_cpu(cpu, tmp); | |
274 | } | |
275 | } | |
276 | ||
7053df4e | 277 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
620b2438 | 278 | unsigned long *vcpu_bitmap) |
d9e368d6 | 279 | { |
d9e368d6 | 280 | struct kvm_vcpu *vcpu; |
620b2438 | 281 | struct cpumask *cpus; |
ae0946cd | 282 | int i, me; |
7053df4e | 283 | bool called; |
6ef7a1bc | 284 | |
3cba4130 | 285 | me = get_cpu(); |
7053df4e | 286 | |
620b2438 VK |
287 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
288 | cpumask_clear(cpus); | |
289 | ||
ae0946cd VK |
290 | for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) { |
291 | vcpu = kvm_get_vcpu(kvm, i); | |
381cecc5 | 292 | if (!vcpu) |
7053df4e | 293 | continue; |
620b2438 | 294 | kvm_make_vcpu_request(kvm, vcpu, req, cpus, me); |
49846896 | 295 | } |
7053df4e | 296 | |
620b2438 | 297 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); |
3cba4130 | 298 | put_cpu(); |
7053df4e VK |
299 | |
300 | return called; | |
301 | } | |
302 | ||
54163a34 SS |
303 | bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req, |
304 | struct kvm_vcpu *except) | |
7053df4e | 305 | { |
ae0946cd | 306 | struct kvm_vcpu *vcpu; |
baff59cc | 307 | struct cpumask *cpus; |
7053df4e | 308 | bool called; |
ae0946cd | 309 | int i, me; |
7053df4e | 310 | |
ae0946cd VK |
311 | me = get_cpu(); |
312 | ||
baff59cc VK |
313 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
314 | cpumask_clear(cpus); | |
315 | ||
ae0946cd VK |
316 | kvm_for_each_vcpu(i, vcpu, kvm) { |
317 | if (vcpu == except) | |
318 | continue; | |
319 | kvm_make_vcpu_request(kvm, vcpu, req, cpus, me); | |
320 | } | |
321 | ||
322 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); | |
323 | put_cpu(); | |
7053df4e | 324 | |
49846896 | 325 | return called; |
d9e368d6 AK |
326 | } |
327 | ||
54163a34 SS |
328 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
329 | { | |
330 | return kvm_make_all_cpus_request_except(kvm, req, NULL); | |
331 | } | |
a2486020 | 332 | EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request); |
54163a34 | 333 | |
a6d51016 | 334 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 335 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 336 | { |
3cc4e148 | 337 | ++kvm->stat.generic.remote_tlb_flush_requests; |
6bc6db00 | 338 | |
4ae3cb3a LT |
339 | /* |
340 | * We want to publish modifications to the page tables before reading | |
341 | * mode. Pairs with a memory barrier in arch-specific code. | |
342 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
343 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
344 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
345 | * | |
346 | * There is already an smp_mb__after_atomic() before | |
347 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
348 | * barrier here. | |
349 | */ | |
b08660e5 TL |
350 | if (!kvm_arch_flush_remote_tlb(kvm) |
351 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
0193cc90 | 352 | ++kvm->stat.generic.remote_tlb_flush; |
2e53d63a | 353 | } |
2ba9f0d8 | 354 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 355 | #endif |
2e53d63a | 356 | |
49846896 RR |
357 | void kvm_reload_remote_mmus(struct kvm *kvm) |
358 | { | |
445b8236 | 359 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 360 | } |
2e53d63a | 361 | |
6926f95a SC |
362 | #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE |
363 | static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc, | |
364 | gfp_t gfp_flags) | |
365 | { | |
366 | gfp_flags |= mc->gfp_zero; | |
367 | ||
368 | if (mc->kmem_cache) | |
369 | return kmem_cache_alloc(mc->kmem_cache, gfp_flags); | |
370 | else | |
371 | return (void *)__get_free_page(gfp_flags); | |
372 | } | |
373 | ||
374 | int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min) | |
375 | { | |
376 | void *obj; | |
377 | ||
378 | if (mc->nobjs >= min) | |
379 | return 0; | |
380 | while (mc->nobjs < ARRAY_SIZE(mc->objects)) { | |
381 | obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT); | |
382 | if (!obj) | |
383 | return mc->nobjs >= min ? 0 : -ENOMEM; | |
384 | mc->objects[mc->nobjs++] = obj; | |
385 | } | |
386 | return 0; | |
387 | } | |
388 | ||
389 | int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc) | |
390 | { | |
391 | return mc->nobjs; | |
392 | } | |
393 | ||
394 | void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
395 | { | |
396 | while (mc->nobjs) { | |
397 | if (mc->kmem_cache) | |
398 | kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]); | |
399 | else | |
400 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
401 | } | |
402 | } | |
403 | ||
404 | void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
405 | { | |
406 | void *p; | |
407 | ||
408 | if (WARN_ON(!mc->nobjs)) | |
409 | p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT); | |
410 | else | |
411 | p = mc->objects[--mc->nobjs]; | |
412 | BUG_ON(!p); | |
413 | return p; | |
414 | } | |
415 | #endif | |
416 | ||
8bd826d6 | 417 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 418 | { |
fb3f0f51 RR |
419 | mutex_init(&vcpu->mutex); |
420 | vcpu->cpu = -1; | |
fb3f0f51 RR |
421 | vcpu->kvm = kvm; |
422 | vcpu->vcpu_id = id; | |
34bb10b7 | 423 | vcpu->pid = NULL; |
da4ad88c | 424 | rcuwait_init(&vcpu->wait); |
af585b92 | 425 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 426 | |
bf9f6ac8 FW |
427 | vcpu->pre_pcpu = -1; |
428 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
429 | ||
4c088493 R |
430 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
431 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 432 | vcpu->preempted = false; |
d73eb57b | 433 | vcpu->ready = false; |
d5c48deb | 434 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
fe22ed82 | 435 | vcpu->last_used_slot = 0; |
fb3f0f51 | 436 | } |
fb3f0f51 | 437 | |
27592ae8 | 438 | static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
4543bdc0 | 439 | { |
fb04a1ed | 440 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
4543bdc0 | 441 | kvm_arch_vcpu_destroy(vcpu); |
e529ef66 | 442 | |
9941d224 SC |
443 | /* |
444 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
445 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
446 | * are already gone. | |
447 | */ | |
448 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
449 | ||
8bd826d6 | 450 | free_page((unsigned long)vcpu->run); |
e529ef66 | 451 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 | 452 | } |
27592ae8 MZ |
453 | |
454 | void kvm_destroy_vcpus(struct kvm *kvm) | |
455 | { | |
456 | unsigned int i; | |
457 | struct kvm_vcpu *vcpu; | |
458 | ||
459 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
460 | kvm_vcpu_destroy(vcpu); | |
c5b07754 | 461 | xa_erase(&kvm->vcpu_array, i); |
27592ae8 MZ |
462 | } |
463 | ||
464 | atomic_set(&kvm->online_vcpus, 0); | |
465 | } | |
466 | EXPORT_SYMBOL_GPL(kvm_destroy_vcpus); | |
4543bdc0 | 467 | |
e930bffe AA |
468 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
469 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
470 | { | |
471 | return container_of(mn, struct kvm, mmu_notifier); | |
472 | } | |
473 | ||
e649b3f0 ET |
474 | static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn, |
475 | struct mm_struct *mm, | |
476 | unsigned long start, unsigned long end) | |
477 | { | |
478 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
479 | int idx; | |
480 | ||
481 | idx = srcu_read_lock(&kvm->srcu); | |
482 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
483 | srcu_read_unlock(&kvm->srcu, idx); | |
484 | } | |
485 | ||
3039bcc7 SC |
486 | typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range); |
487 | ||
f922bd9b SC |
488 | typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start, |
489 | unsigned long end); | |
490 | ||
3039bcc7 SC |
491 | struct kvm_hva_range { |
492 | unsigned long start; | |
493 | unsigned long end; | |
494 | pte_t pte; | |
495 | hva_handler_t handler; | |
f922bd9b | 496 | on_lock_fn_t on_lock; |
3039bcc7 SC |
497 | bool flush_on_ret; |
498 | bool may_block; | |
499 | }; | |
500 | ||
f922bd9b SC |
501 | /* |
502 | * Use a dedicated stub instead of NULL to indicate that there is no callback | |
503 | * function/handler. The compiler technically can't guarantee that a real | |
504 | * function will have a non-zero address, and so it will generate code to | |
505 | * check for !NULL, whereas comparing against a stub will be elided at compile | |
506 | * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9). | |
507 | */ | |
508 | static void kvm_null_fn(void) | |
509 | { | |
510 | ||
511 | } | |
512 | #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn) | |
513 | ||
3039bcc7 SC |
514 | static __always_inline int __kvm_handle_hva_range(struct kvm *kvm, |
515 | const struct kvm_hva_range *range) | |
516 | { | |
8931a454 | 517 | bool ret = false, locked = false; |
f922bd9b | 518 | struct kvm_gfn_range gfn_range; |
3039bcc7 SC |
519 | struct kvm_memory_slot *slot; |
520 | struct kvm_memslots *slots; | |
3039bcc7 SC |
521 | int i, idx; |
522 | ||
f922bd9b SC |
523 | /* A null handler is allowed if and only if on_lock() is provided. */ |
524 | if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) && | |
525 | IS_KVM_NULL_FN(range->handler))) | |
526 | return 0; | |
527 | ||
3039bcc7 SC |
528 | idx = srcu_read_lock(&kvm->srcu); |
529 | ||
530 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
531 | slots = __kvm_memslots(kvm, i); | |
532 | kvm_for_each_memslot(slot, slots) { | |
533 | unsigned long hva_start, hva_end; | |
534 | ||
535 | hva_start = max(range->start, slot->userspace_addr); | |
536 | hva_end = min(range->end, slot->userspace_addr + | |
537 | (slot->npages << PAGE_SHIFT)); | |
538 | if (hva_start >= hva_end) | |
539 | continue; | |
540 | ||
541 | /* | |
542 | * To optimize for the likely case where the address | |
543 | * range is covered by zero or one memslots, don't | |
544 | * bother making these conditional (to avoid writes on | |
545 | * the second or later invocation of the handler). | |
546 | */ | |
547 | gfn_range.pte = range->pte; | |
548 | gfn_range.may_block = range->may_block; | |
549 | ||
550 | /* | |
551 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
552 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
553 | */ | |
554 | gfn_range.start = hva_to_gfn_memslot(hva_start, slot); | |
555 | gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot); | |
556 | gfn_range.slot = slot; | |
557 | ||
8931a454 SC |
558 | if (!locked) { |
559 | locked = true; | |
560 | KVM_MMU_LOCK(kvm); | |
071064f1 PB |
561 | if (!IS_KVM_NULL_FN(range->on_lock)) |
562 | range->on_lock(kvm, range->start, range->end); | |
563 | if (IS_KVM_NULL_FN(range->handler)) | |
564 | break; | |
8931a454 | 565 | } |
3039bcc7 SC |
566 | ret |= range->handler(kvm, &gfn_range); |
567 | } | |
568 | } | |
569 | ||
6bc6db00 | 570 | if (range->flush_on_ret && ret) |
3039bcc7 SC |
571 | kvm_flush_remote_tlbs(kvm); |
572 | ||
8931a454 SC |
573 | if (locked) |
574 | KVM_MMU_UNLOCK(kvm); | |
f922bd9b | 575 | |
3039bcc7 SC |
576 | srcu_read_unlock(&kvm->srcu, idx); |
577 | ||
578 | /* The notifiers are averse to booleans. :-( */ | |
579 | return (int)ret; | |
580 | } | |
581 | ||
582 | static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn, | |
583 | unsigned long start, | |
584 | unsigned long end, | |
585 | pte_t pte, | |
586 | hva_handler_t handler) | |
587 | { | |
588 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
589 | const struct kvm_hva_range range = { | |
590 | .start = start, | |
591 | .end = end, | |
592 | .pte = pte, | |
593 | .handler = handler, | |
f922bd9b | 594 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
595 | .flush_on_ret = true, |
596 | .may_block = false, | |
597 | }; | |
3039bcc7 | 598 | |
f922bd9b | 599 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 SC |
600 | } |
601 | ||
602 | static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn, | |
603 | unsigned long start, | |
604 | unsigned long end, | |
605 | hva_handler_t handler) | |
606 | { | |
607 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
608 | const struct kvm_hva_range range = { | |
609 | .start = start, | |
610 | .end = end, | |
611 | .pte = __pte(0), | |
612 | .handler = handler, | |
f922bd9b | 613 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
614 | .flush_on_ret = false, |
615 | .may_block = false, | |
616 | }; | |
3039bcc7 | 617 | |
f922bd9b | 618 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 | 619 | } |
3da0dd43 IE |
620 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
621 | struct mm_struct *mm, | |
622 | unsigned long address, | |
623 | pte_t pte) | |
624 | { | |
625 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
626 | ||
501b9185 SC |
627 | trace_kvm_set_spte_hva(address); |
628 | ||
c13fda23 | 629 | /* |
52ac8b35 | 630 | * .change_pte() must be surrounded by .invalidate_range_{start,end}(). |
071064f1 PB |
631 | * If mmu_notifier_count is zero, then no in-progress invalidations, |
632 | * including this one, found a relevant memslot at start(); rechecking | |
633 | * memslots here is unnecessary. Note, a false positive (count elevated | |
634 | * by a different invalidation) is sub-optimal but functionally ok. | |
c13fda23 | 635 | */ |
52ac8b35 | 636 | WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count)); |
071064f1 PB |
637 | if (!READ_ONCE(kvm->mmu_notifier_count)) |
638 | return; | |
c13fda23 | 639 | |
3039bcc7 | 640 | kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn); |
3da0dd43 IE |
641 | } |
642 | ||
edb298c6 | 643 | void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 644 | unsigned long end) |
e930bffe | 645 | { |
e930bffe AA |
646 | /* |
647 | * The count increase must become visible at unlock time as no | |
648 | * spte can be established without taking the mmu_lock and | |
649 | * count is also read inside the mmu_lock critical section. | |
650 | */ | |
651 | kvm->mmu_notifier_count++; | |
4a42d848 | 652 | if (likely(kvm->mmu_notifier_count == 1)) { |
f922bd9b SC |
653 | kvm->mmu_notifier_range_start = start; |
654 | kvm->mmu_notifier_range_end = end; | |
4a42d848 DS |
655 | } else { |
656 | /* | |
657 | * Fully tracking multiple concurrent ranges has dimishing | |
658 | * returns. Keep things simple and just find the minimal range | |
659 | * which includes the current and new ranges. As there won't be | |
660 | * enough information to subtract a range after its invalidate | |
661 | * completes, any ranges invalidated concurrently will | |
662 | * accumulate and persist until all outstanding invalidates | |
663 | * complete. | |
664 | */ | |
665 | kvm->mmu_notifier_range_start = | |
f922bd9b | 666 | min(kvm->mmu_notifier_range_start, start); |
4a42d848 | 667 | kvm->mmu_notifier_range_end = |
f922bd9b | 668 | max(kvm->mmu_notifier_range_end, end); |
4a42d848 | 669 | } |
f922bd9b | 670 | } |
3039bcc7 | 671 | |
f922bd9b SC |
672 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
673 | const struct mmu_notifier_range *range) | |
674 | { | |
675 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
676 | const struct kvm_hva_range hva_range = { | |
677 | .start = range->start, | |
678 | .end = range->end, | |
679 | .pte = __pte(0), | |
680 | .handler = kvm_unmap_gfn_range, | |
681 | .on_lock = kvm_inc_notifier_count, | |
682 | .flush_on_ret = true, | |
683 | .may_block = mmu_notifier_range_blockable(range), | |
684 | }; | |
565f3be2 | 685 | |
f922bd9b SC |
686 | trace_kvm_unmap_hva_range(range->start, range->end); |
687 | ||
52ac8b35 PB |
688 | /* |
689 | * Prevent memslot modification between range_start() and range_end() | |
690 | * so that conditionally locking provides the same result in both | |
691 | * functions. Without that guarantee, the mmu_notifier_count | |
692 | * adjustments will be imbalanced. | |
693 | * | |
694 | * Pairs with the decrement in range_end(). | |
695 | */ | |
696 | spin_lock(&kvm->mn_invalidate_lock); | |
697 | kvm->mn_active_invalidate_count++; | |
698 | spin_unlock(&kvm->mn_invalidate_lock); | |
699 | ||
f922bd9b | 700 | __kvm_handle_hva_range(kvm, &hva_range); |
93065ac7 | 701 | |
e649b3f0 | 702 | return 0; |
e930bffe AA |
703 | } |
704 | ||
edb298c6 | 705 | void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 706 | unsigned long end) |
e930bffe | 707 | { |
e930bffe AA |
708 | /* |
709 | * This sequence increase will notify the kvm page fault that | |
710 | * the page that is going to be mapped in the spte could have | |
711 | * been freed. | |
712 | */ | |
713 | kvm->mmu_notifier_seq++; | |
a355aa54 | 714 | smp_wmb(); |
e930bffe AA |
715 | /* |
716 | * The above sequence increase must be visible before the | |
a355aa54 PM |
717 | * below count decrease, which is ensured by the smp_wmb above |
718 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
719 | */ |
720 | kvm->mmu_notifier_count--; | |
f922bd9b SC |
721 | } |
722 | ||
723 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
724 | const struct mmu_notifier_range *range) | |
725 | { | |
726 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
727 | const struct kvm_hva_range hva_range = { | |
728 | .start = range->start, | |
729 | .end = range->end, | |
730 | .pte = __pte(0), | |
731 | .handler = (void *)kvm_null_fn, | |
732 | .on_lock = kvm_dec_notifier_count, | |
733 | .flush_on_ret = false, | |
734 | .may_block = mmu_notifier_range_blockable(range), | |
735 | }; | |
52ac8b35 | 736 | bool wake; |
f922bd9b SC |
737 | |
738 | __kvm_handle_hva_range(kvm, &hva_range); | |
e930bffe | 739 | |
52ac8b35 PB |
740 | /* Pairs with the increment in range_start(). */ |
741 | spin_lock(&kvm->mn_invalidate_lock); | |
742 | wake = (--kvm->mn_active_invalidate_count == 0); | |
743 | spin_unlock(&kvm->mn_invalidate_lock); | |
744 | ||
745 | /* | |
746 | * There can only be one waiter, since the wait happens under | |
747 | * slots_lock. | |
748 | */ | |
749 | if (wake) | |
750 | rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait); | |
751 | ||
e930bffe AA |
752 | BUG_ON(kvm->mmu_notifier_count < 0); |
753 | } | |
754 | ||
755 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
756 | struct mm_struct *mm, | |
57128468 ALC |
757 | unsigned long start, |
758 | unsigned long end) | |
e930bffe | 759 | { |
501b9185 SC |
760 | trace_kvm_age_hva(start, end); |
761 | ||
3039bcc7 | 762 | return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn); |
e930bffe AA |
763 | } |
764 | ||
1d7715c6 VD |
765 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
766 | struct mm_struct *mm, | |
767 | unsigned long start, | |
768 | unsigned long end) | |
769 | { | |
501b9185 SC |
770 | trace_kvm_age_hva(start, end); |
771 | ||
1d7715c6 VD |
772 | /* |
773 | * Even though we do not flush TLB, this will still adversely | |
774 | * affect performance on pre-Haswell Intel EPT, where there is | |
775 | * no EPT Access Bit to clear so that we have to tear down EPT | |
776 | * tables instead. If we find this unacceptable, we can always | |
777 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
778 | * do anything on clear_young. | |
779 | * | |
780 | * Also note that currently we never issue secondary TLB flushes | |
781 | * from clear_young, leaving this job up to the regular system | |
782 | * cadence. If we find this inaccurate, we might come up with a | |
783 | * more sophisticated heuristic later. | |
784 | */ | |
3039bcc7 | 785 | return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn); |
1d7715c6 VD |
786 | } |
787 | ||
8ee53820 AA |
788 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
789 | struct mm_struct *mm, | |
790 | unsigned long address) | |
791 | { | |
501b9185 SC |
792 | trace_kvm_test_age_hva(address); |
793 | ||
3039bcc7 SC |
794 | return kvm_handle_hva_range_no_flush(mn, address, address + 1, |
795 | kvm_test_age_gfn); | |
8ee53820 AA |
796 | } |
797 | ||
85db06e5 MT |
798 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
799 | struct mm_struct *mm) | |
800 | { | |
801 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
802 | int idx; |
803 | ||
804 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 805 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 806 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
807 | } |
808 | ||
e930bffe | 809 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e649b3f0 | 810 | .invalidate_range = kvm_mmu_notifier_invalidate_range, |
e930bffe AA |
811 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
812 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
813 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 814 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 815 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 816 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 817 | .release = kvm_mmu_notifier_release, |
e930bffe | 818 | }; |
4c07b0a4 AK |
819 | |
820 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
821 | { | |
822 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
823 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
824 | } | |
825 | ||
826 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
827 | ||
828 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
829 | { | |
830 | return 0; | |
831 | } | |
832 | ||
e930bffe AA |
833 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
834 | ||
2fdef3a2 SS |
835 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
836 | static int kvm_pm_notifier_call(struct notifier_block *bl, | |
837 | unsigned long state, | |
838 | void *unused) | |
839 | { | |
840 | struct kvm *kvm = container_of(bl, struct kvm, pm_notifier); | |
841 | ||
842 | return kvm_arch_pm_notifier(kvm, state); | |
843 | } | |
844 | ||
845 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
846 | { | |
847 | kvm->pm_notifier.notifier_call = kvm_pm_notifier_call; | |
848 | /* Suspend KVM before we suspend ftrace, RCU, etc. */ | |
849 | kvm->pm_notifier.priority = INT_MAX; | |
850 | register_pm_notifier(&kvm->pm_notifier); | |
851 | } | |
852 | ||
853 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
854 | { | |
855 | unregister_pm_notifier(&kvm->pm_notifier); | |
856 | } | |
857 | #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
858 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
859 | { | |
860 | } | |
861 | ||
862 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
863 | { | |
864 | } | |
865 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
866 | ||
a47d2b07 | 867 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
868 | { |
869 | int i; | |
a47d2b07 | 870 | struct kvm_memslots *slots; |
bf3e05bc | 871 | |
b12ce36a | 872 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
a47d2b07 PB |
873 | if (!slots) |
874 | return NULL; | |
875 | ||
bf3e05bc | 876 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
36947254 | 877 | slots->id_to_index[i] = -1; |
a47d2b07 PB |
878 | |
879 | return slots; | |
880 | } | |
881 | ||
882 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
883 | { | |
884 | if (!memslot->dirty_bitmap) | |
885 | return; | |
886 | ||
887 | kvfree(memslot->dirty_bitmap); | |
888 | memslot->dirty_bitmap = NULL; | |
889 | } | |
890 | ||
e96c81ee | 891 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 892 | { |
e96c81ee | 893 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 894 | |
e96c81ee | 895 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 896 | |
e96c81ee SC |
897 | slot->flags = 0; |
898 | slot->npages = 0; | |
a47d2b07 PB |
899 | } |
900 | ||
901 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
902 | { | |
903 | struct kvm_memory_slot *memslot; | |
904 | ||
905 | if (!slots) | |
906 | return; | |
907 | ||
908 | kvm_for_each_memslot(memslot, slots) | |
e96c81ee | 909 | kvm_free_memslot(kvm, memslot); |
a47d2b07 PB |
910 | |
911 | kvfree(slots); | |
bf3e05bc XG |
912 | } |
913 | ||
bc9e9e67 JZ |
914 | static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc) |
915 | { | |
916 | switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) { | |
917 | case KVM_STATS_TYPE_INSTANT: | |
918 | return 0444; | |
919 | case KVM_STATS_TYPE_CUMULATIVE: | |
920 | case KVM_STATS_TYPE_PEAK: | |
921 | default: | |
922 | return 0644; | |
923 | } | |
924 | } | |
925 | ||
926 | ||
536a6f88 JF |
927 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
928 | { | |
929 | int i; | |
bc9e9e67 JZ |
930 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
931 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
932 | |
933 | if (!kvm->debugfs_dentry) | |
934 | return; | |
935 | ||
936 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
937 | ||
9d5a1dce LC |
938 | if (kvm->debugfs_stat_data) { |
939 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
940 | kfree(kvm->debugfs_stat_data[i]); | |
941 | kfree(kvm->debugfs_stat_data); | |
942 | } | |
536a6f88 JF |
943 | } |
944 | ||
945 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
946 | { | |
85cd39af PB |
947 | static DEFINE_MUTEX(kvm_debugfs_lock); |
948 | struct dentry *dent; | |
536a6f88 JF |
949 | char dir_name[ITOA_MAX_LEN * 2]; |
950 | struct kvm_stat_data *stat_data; | |
bc9e9e67 | 951 | const struct _kvm_stats_desc *pdesc; |
3165af73 | 952 | int i, ret; |
bc9e9e67 JZ |
953 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
954 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
955 | |
956 | if (!debugfs_initialized()) | |
957 | return 0; | |
958 | ||
959 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
85cd39af PB |
960 | mutex_lock(&kvm_debugfs_lock); |
961 | dent = debugfs_lookup(dir_name, kvm_debugfs_dir); | |
962 | if (dent) { | |
963 | pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name); | |
964 | dput(dent); | |
965 | mutex_unlock(&kvm_debugfs_lock); | |
966 | return 0; | |
967 | } | |
968 | dent = debugfs_create_dir(dir_name, kvm_debugfs_dir); | |
969 | mutex_unlock(&kvm_debugfs_lock); | |
970 | if (IS_ERR(dent)) | |
971 | return 0; | |
536a6f88 | 972 | |
85cd39af | 973 | kvm->debugfs_dentry = dent; |
536a6f88 JF |
974 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
975 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 976 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
977 | if (!kvm->debugfs_stat_data) |
978 | return -ENOMEM; | |
979 | ||
bc9e9e67 JZ |
980 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
981 | pdesc = &kvm_vm_stats_desc[i]; | |
b12ce36a | 982 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
983 | if (!stat_data) |
984 | return -ENOMEM; | |
985 | ||
986 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
987 | stat_data->desc = pdesc; |
988 | stat_data->kind = KVM_STAT_VM; | |
989 | kvm->debugfs_stat_data[i] = stat_data; | |
990 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
991 | kvm->debugfs_dentry, stat_data, | |
992 | &stat_fops_per_vm); | |
993 | } | |
994 | ||
995 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
996 | pdesc = &kvm_vcpu_stats_desc[i]; | |
b12ce36a | 997 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
998 | if (!stat_data) |
999 | return -ENOMEM; | |
1000 | ||
1001 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1002 | stat_data->desc = pdesc; |
1003 | stat_data->kind = KVM_STAT_VCPU; | |
004d62eb | 1004 | kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data; |
bc9e9e67 | 1005 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), |
09cbcef6 MP |
1006 | kvm->debugfs_dentry, stat_data, |
1007 | &stat_fops_per_vm); | |
536a6f88 | 1008 | } |
3165af73 PX |
1009 | |
1010 | ret = kvm_arch_create_vm_debugfs(kvm); | |
1011 | if (ret) { | |
1012 | kvm_destroy_vm_debugfs(kvm); | |
1013 | return i; | |
1014 | } | |
1015 | ||
536a6f88 JF |
1016 | return 0; |
1017 | } | |
1018 | ||
1aa9b957 JS |
1019 | /* |
1020 | * Called after the VM is otherwise initialized, but just before adding it to | |
1021 | * the vm_list. | |
1022 | */ | |
1023 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
1024 | { | |
1025 | return 0; | |
1026 | } | |
1027 | ||
1028 | /* | |
1029 | * Called just after removing the VM from the vm_list, but before doing any | |
1030 | * other destruction. | |
1031 | */ | |
1032 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
1033 | { | |
1034 | } | |
1035 | ||
3165af73 PX |
1036 | /* |
1037 | * Called after per-vm debugfs created. When called kvm->debugfs_dentry should | |
1038 | * be setup already, so we can create arch-specific debugfs entries under it. | |
1039 | * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so | |
1040 | * a per-arch destroy interface is not needed. | |
1041 | */ | |
1042 | int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm) | |
1043 | { | |
1044 | return 0; | |
1045 | } | |
1046 | ||
e08b9637 | 1047 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 1048 | { |
d89f5eff | 1049 | struct kvm *kvm = kvm_arch_alloc_vm(); |
9121923c JM |
1050 | int r = -ENOMEM; |
1051 | int i; | |
6aa8b732 | 1052 | |
d89f5eff JK |
1053 | if (!kvm) |
1054 | return ERR_PTR(-ENOMEM); | |
1055 | ||
531810ca | 1056 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 1057 | mmgrab(current->mm); |
e9ad4ec8 PB |
1058 | kvm->mm = current->mm; |
1059 | kvm_eventfd_init(kvm); | |
1060 | mutex_init(&kvm->lock); | |
1061 | mutex_init(&kvm->irq_lock); | |
1062 | mutex_init(&kvm->slots_lock); | |
b10a038e | 1063 | mutex_init(&kvm->slots_arch_lock); |
52ac8b35 PB |
1064 | spin_lock_init(&kvm->mn_invalidate_lock); |
1065 | rcuwait_init(&kvm->mn_memslots_update_rcuwait); | |
c5b07754 | 1066 | xa_init(&kvm->vcpu_array); |
52ac8b35 | 1067 | |
e9ad4ec8 PB |
1068 | INIT_LIST_HEAD(&kvm->devices); |
1069 | ||
1e702d9a AW |
1070 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
1071 | ||
8a44119a PB |
1072 | if (init_srcu_struct(&kvm->srcu)) |
1073 | goto out_err_no_srcu; | |
1074 | if (init_srcu_struct(&kvm->irq_srcu)) | |
1075 | goto out_err_no_irq_srcu; | |
1076 | ||
e2d3fcaf | 1077 | refcount_set(&kvm->users_count, 1); |
f481b069 | 1078 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4bd518f1 | 1079 | struct kvm_memslots *slots = kvm_alloc_memslots(); |
9121923c | 1080 | |
4bd518f1 | 1081 | if (!slots) |
a97b0e77 | 1082 | goto out_err_no_arch_destroy_vm; |
0e32958e | 1083 | /* Generations must be different for each address space. */ |
164bf7e5 | 1084 | slots->generation = i; |
4bd518f1 | 1085 | rcu_assign_pointer(kvm->memslots[i], slots); |
f481b069 | 1086 | } |
00f034a1 | 1087 | |
e93f8a0f | 1088 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 1089 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 1090 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 1091 | if (!kvm->buses[i]) |
a97b0e77 | 1092 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 1093 | } |
e930bffe | 1094 | |
acd05785 DM |
1095 | kvm->max_halt_poll_ns = halt_poll_ns; |
1096 | ||
e08b9637 | 1097 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 1098 | if (r) |
a97b0e77 | 1099 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
1100 | |
1101 | r = hardware_enable_all(); | |
1102 | if (r) | |
719d93cd | 1103 | goto out_err_no_disable; |
10474ae8 | 1104 | |
c77dcacb | 1105 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 1106 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 1107 | #endif |
6aa8b732 | 1108 | |
74b5c5bf | 1109 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
1110 | if (r) |
1111 | goto out_err_no_mmu_notifier; | |
1112 | ||
1113 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
1114 | if (r) |
1115 | goto out_err; | |
1116 | ||
0d9ce162 | 1117 | mutex_lock(&kvm_lock); |
5e58cfe4 | 1118 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 1119 | mutex_unlock(&kvm_lock); |
d89f5eff | 1120 | |
2ecd9d29 | 1121 | preempt_notifier_inc(); |
2fdef3a2 | 1122 | kvm_init_pm_notifier(kvm); |
2ecd9d29 | 1123 | |
f17abe9a | 1124 | return kvm; |
10474ae8 AG |
1125 | |
1126 | out_err: | |
1aa9b957 JS |
1127 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1128 | if (kvm->mmu_notifier.ops) | |
1129 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
1130 | #endif | |
1131 | out_err_no_mmu_notifier: | |
10474ae8 | 1132 | hardware_disable_all(); |
719d93cd | 1133 | out_err_no_disable: |
a97b0e77 | 1134 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 1135 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 1136 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 1137 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 1138 | kfree(kvm_get_bus(kvm, i)); |
f481b069 | 1139 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 1140 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
8a44119a PB |
1141 | cleanup_srcu_struct(&kvm->irq_srcu); |
1142 | out_err_no_irq_srcu: | |
1143 | cleanup_srcu_struct(&kvm->srcu); | |
1144 | out_err_no_srcu: | |
d89f5eff | 1145 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 1146 | mmdrop(current->mm); |
10474ae8 | 1147 | return ERR_PTR(r); |
f17abe9a AK |
1148 | } |
1149 | ||
07f0a7bd SW |
1150 | static void kvm_destroy_devices(struct kvm *kvm) |
1151 | { | |
e6e3b5a6 | 1152 | struct kvm_device *dev, *tmp; |
07f0a7bd | 1153 | |
a28ebea2 CD |
1154 | /* |
1155 | * We do not need to take the kvm->lock here, because nobody else | |
1156 | * has a reference to the struct kvm at this point and therefore | |
1157 | * cannot access the devices list anyhow. | |
1158 | */ | |
e6e3b5a6 GT |
1159 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
1160 | list_del(&dev->vm_node); | |
07f0a7bd SW |
1161 | dev->ops->destroy(dev); |
1162 | } | |
1163 | } | |
1164 | ||
f17abe9a AK |
1165 | static void kvm_destroy_vm(struct kvm *kvm) |
1166 | { | |
e93f8a0f | 1167 | int i; |
6d4e4c4f AK |
1168 | struct mm_struct *mm = kvm->mm; |
1169 | ||
2fdef3a2 | 1170 | kvm_destroy_pm_notifier(kvm); |
286de8f6 | 1171 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 1172 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 1173 | kvm_arch_sync_events(kvm); |
0d9ce162 | 1174 | mutex_lock(&kvm_lock); |
133de902 | 1175 | list_del(&kvm->vm_list); |
0d9ce162 | 1176 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
1177 | kvm_arch_pre_destroy_vm(kvm); |
1178 | ||
399ec807 | 1179 | kvm_free_irq_routing(kvm); |
df630b8c | 1180 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 1181 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 1182 | |
4a12f951 CB |
1183 | if (bus) |
1184 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
1185 | kvm->buses[i] = NULL; |
1186 | } | |
980da6ce | 1187 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
1188 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1189 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
52ac8b35 PB |
1190 | /* |
1191 | * At this point, pending calls to invalidate_range_start() | |
1192 | * have completed but no more MMU notifiers will run, so | |
1193 | * mn_active_invalidate_count may remain unbalanced. | |
1194 | * No threads can be waiting in install_new_memslots as the | |
1195 | * last reference on KVM has been dropped, but freeing | |
1196 | * memslots would deadlock without this manual intervention. | |
1197 | */ | |
1198 | WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait)); | |
1199 | kvm->mn_active_invalidate_count = 0; | |
f00be0ca | 1200 | #else |
2df72e9b | 1201 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 1202 | #endif |
d19a9cd2 | 1203 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 1204 | kvm_destroy_devices(kvm); |
f481b069 | 1205 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 1206 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
820b3fcd | 1207 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
1208 | cleanup_srcu_struct(&kvm->srcu); |
1209 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 1210 | preempt_notifier_dec(); |
10474ae8 | 1211 | hardware_disable_all(); |
6d4e4c4f | 1212 | mmdrop(mm); |
f17abe9a AK |
1213 | } |
1214 | ||
d39f13b0 IE |
1215 | void kvm_get_kvm(struct kvm *kvm) |
1216 | { | |
e3736c3e | 1217 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
1218 | } |
1219 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
1220 | ||
605c7130 PX |
1221 | /* |
1222 | * Make sure the vm is not during destruction, which is a safe version of | |
1223 | * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise. | |
1224 | */ | |
1225 | bool kvm_get_kvm_safe(struct kvm *kvm) | |
1226 | { | |
1227 | return refcount_inc_not_zero(&kvm->users_count); | |
1228 | } | |
1229 | EXPORT_SYMBOL_GPL(kvm_get_kvm_safe); | |
1230 | ||
d39f13b0 IE |
1231 | void kvm_put_kvm(struct kvm *kvm) |
1232 | { | |
e3736c3e | 1233 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
1234 | kvm_destroy_vm(kvm); |
1235 | } | |
1236 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
1237 | ||
149487bd SC |
1238 | /* |
1239 | * Used to put a reference that was taken on behalf of an object associated | |
1240 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
1241 | * of the new file descriptor fails and the reference cannot be transferred to | |
1242 | * its final owner. In such cases, the caller is still actively using @kvm and | |
1243 | * will fail miserably if the refcount unexpectedly hits zero. | |
1244 | */ | |
1245 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
1246 | { | |
1247 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
1248 | } | |
1249 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 1250 | |
f17abe9a AK |
1251 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
1252 | { | |
1253 | struct kvm *kvm = filp->private_data; | |
1254 | ||
721eecbf GH |
1255 | kvm_irqfd_release(kvm); |
1256 | ||
d39f13b0 | 1257 | kvm_put_kvm(kvm); |
6aa8b732 AK |
1258 | return 0; |
1259 | } | |
1260 | ||
515a0127 TY |
1261 | /* |
1262 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 1263 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 1264 | */ |
3c9bd400 | 1265 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 1266 | { |
515a0127 | 1267 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 1268 | |
b12ce36a | 1269 | memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
1270 | if (!memslot->dirty_bitmap) |
1271 | return -ENOMEM; | |
1272 | ||
a36a57b1 TY |
1273 | return 0; |
1274 | } | |
1275 | ||
bf3e05bc | 1276 | /* |
0577d1ab SC |
1277 | * Delete a memslot by decrementing the number of used slots and shifting all |
1278 | * other entries in the array forward one spot. | |
bf3e05bc | 1279 | */ |
0577d1ab SC |
1280 | static inline void kvm_memslot_delete(struct kvm_memslots *slots, |
1281 | struct kvm_memory_slot *memslot) | |
bf3e05bc | 1282 | { |
063584d4 | 1283 | struct kvm_memory_slot *mslots = slots->memslots; |
0577d1ab | 1284 | int i; |
f85e2cb5 | 1285 | |
0577d1ab SC |
1286 | if (WARN_ON(slots->id_to_index[memslot->id] == -1)) |
1287 | return; | |
0e60b079 | 1288 | |
0577d1ab SC |
1289 | slots->used_slots--; |
1290 | ||
87689270 DM |
1291 | if (atomic_read(&slots->last_used_slot) >= slots->used_slots) |
1292 | atomic_set(&slots->last_used_slot, 0); | |
0774a964 | 1293 | |
0577d1ab | 1294 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) { |
7f379cff IM |
1295 | mslots[i] = mslots[i + 1]; |
1296 | slots->id_to_index[mslots[i].id] = i; | |
7f379cff | 1297 | } |
0577d1ab SC |
1298 | mslots[i] = *memslot; |
1299 | slots->id_to_index[memslot->id] = -1; | |
1300 | } | |
1301 | ||
1302 | /* | |
1303 | * "Insert" a new memslot by incrementing the number of used slots. Returns | |
1304 | * the new slot's initial index into the memslots array. | |
1305 | */ | |
1306 | static inline int kvm_memslot_insert_back(struct kvm_memslots *slots) | |
1307 | { | |
1308 | return slots->used_slots++; | |
1309 | } | |
1310 | ||
1311 | /* | |
1312 | * Move a changed memslot backwards in the array by shifting existing slots | |
1313 | * with a higher GFN toward the front of the array. Note, the changed memslot | |
1314 | * itself is not preserved in the array, i.e. not swapped at this time, only | |
1315 | * its new index into the array is tracked. Returns the changed memslot's | |
1316 | * current index into the memslots array. | |
1317 | */ | |
1318 | static inline int kvm_memslot_move_backward(struct kvm_memslots *slots, | |
1319 | struct kvm_memory_slot *memslot) | |
1320 | { | |
1321 | struct kvm_memory_slot *mslots = slots->memslots; | |
1322 | int i; | |
1323 | ||
1324 | if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) || | |
1325 | WARN_ON_ONCE(!slots->used_slots)) | |
1326 | return -1; | |
efbeec70 PB |
1327 | |
1328 | /* | |
0577d1ab SC |
1329 | * Move the target memslot backward in the array by shifting existing |
1330 | * memslots with a higher GFN (than the target memslot) towards the | |
1331 | * front of the array. | |
efbeec70 | 1332 | */ |
0577d1ab SC |
1333 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) { |
1334 | if (memslot->base_gfn > mslots[i + 1].base_gfn) | |
1335 | break; | |
1336 | ||
1337 | WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn); | |
f85e2cb5 | 1338 | |
0577d1ab SC |
1339 | /* Shift the next memslot forward one and update its index. */ |
1340 | mslots[i] = mslots[i + 1]; | |
1341 | slots->id_to_index[mslots[i].id] = i; | |
1342 | } | |
1343 | return i; | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * Move a changed memslot forwards in the array by shifting existing slots with | |
1348 | * a lower GFN toward the back of the array. Note, the changed memslot itself | |
1349 | * is not preserved in the array, i.e. not swapped at this time, only its new | |
1350 | * index into the array is tracked. Returns the changed memslot's final index | |
1351 | * into the memslots array. | |
1352 | */ | |
1353 | static inline int kvm_memslot_move_forward(struct kvm_memslots *slots, | |
1354 | struct kvm_memory_slot *memslot, | |
1355 | int start) | |
1356 | { | |
1357 | struct kvm_memory_slot *mslots = slots->memslots; | |
1358 | int i; | |
1359 | ||
1360 | for (i = start; i > 0; i--) { | |
1361 | if (memslot->base_gfn < mslots[i - 1].base_gfn) | |
1362 | break; | |
1363 | ||
1364 | WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn); | |
1365 | ||
1366 | /* Shift the next memslot back one and update its index. */ | |
1367 | mslots[i] = mslots[i - 1]; | |
1368 | slots->id_to_index[mslots[i].id] = i; | |
1369 | } | |
1370 | return i; | |
1371 | } | |
1372 | ||
1373 | /* | |
1374 | * Re-sort memslots based on their GFN to account for an added, deleted, or | |
1375 | * moved memslot. Sorting memslots by GFN allows using a binary search during | |
1376 | * memslot lookup. | |
1377 | * | |
1378 | * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry | |
1379 | * at memslots[0] has the highest GFN. | |
1380 | * | |
1381 | * The sorting algorithm takes advantage of having initially sorted memslots | |
1382 | * and knowing the position of the changed memslot. Sorting is also optimized | |
1383 | * by not swapping the updated memslot and instead only shifting other memslots | |
1384 | * and tracking the new index for the update memslot. Only once its final | |
1385 | * index is known is the updated memslot copied into its position in the array. | |
1386 | * | |
1387 | * - When deleting a memslot, the deleted memslot simply needs to be moved to | |
1388 | * the end of the array. | |
1389 | * | |
1390 | * - When creating a memslot, the algorithm "inserts" the new memslot at the | |
1391 | * end of the array and then it forward to its correct location. | |
1392 | * | |
1393 | * - When moving a memslot, the algorithm first moves the updated memslot | |
1394 | * backward to handle the scenario where the memslot's GFN was changed to a | |
1395 | * lower value. update_memslots() then falls through and runs the same flow | |
1396 | * as creating a memslot to move the memslot forward to handle the scenario | |
1397 | * where its GFN was changed to a higher value. | |
1398 | * | |
1399 | * Note, slots are sorted from highest->lowest instead of lowest->highest for | |
1400 | * historical reasons. Originally, invalid memslots where denoted by having | |
1401 | * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots | |
1402 | * to the end of the array. The current algorithm uses dedicated logic to | |
1403 | * delete a memslot and thus does not rely on invalid memslots having GFN=0. | |
1404 | * | |
1405 | * The other historical motiviation for highest->lowest was to improve the | |
1406 | * performance of memslot lookup. KVM originally used a linear search starting | |
1407 | * at memslots[0]. On x86, the largest memslot usually has one of the highest, | |
1408 | * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a | |
1409 | * single memslot above the 4gb boundary. As the largest memslot is also the | |
1410 | * most likely to be referenced, sorting it to the front of the array was | |
1411 | * advantageous. The current binary search starts from the middle of the array | |
1412 | * and uses an LRU pointer to improve performance for all memslots and GFNs. | |
1413 | */ | |
1414 | static void update_memslots(struct kvm_memslots *slots, | |
1415 | struct kvm_memory_slot *memslot, | |
1416 | enum kvm_mr_change change) | |
1417 | { | |
1418 | int i; | |
1419 | ||
1420 | if (change == KVM_MR_DELETE) { | |
1421 | kvm_memslot_delete(slots, memslot); | |
1422 | } else { | |
1423 | if (change == KVM_MR_CREATE) | |
1424 | i = kvm_memslot_insert_back(slots); | |
1425 | else | |
1426 | i = kvm_memslot_move_backward(slots, memslot); | |
1427 | i = kvm_memslot_move_forward(slots, memslot, i); | |
1428 | ||
1429 | /* | |
1430 | * Copy the memslot to its new position in memslots and update | |
1431 | * its index accordingly. | |
1432 | */ | |
1433 | slots->memslots[i] = *memslot; | |
1434 | slots->id_to_index[memslot->id] = i; | |
1435 | } | |
bf3e05bc XG |
1436 | } |
1437 | ||
09170a49 | 1438 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 1439 | { |
4d8b81ab XG |
1440 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1441 | ||
0f8a4de3 | 1442 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
1443 | valid_flags |= KVM_MEM_READONLY; |
1444 | #endif | |
1445 | ||
1446 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1447 | return -EINVAL; |
1448 | ||
1449 | return 0; | |
1450 | } | |
1451 | ||
7ec4fb44 | 1452 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 1453 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 1454 | { |
f481b069 | 1455 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
361209e0 | 1456 | u64 gen = old_memslots->generation; |
7ec4fb44 | 1457 | |
361209e0 SC |
1458 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1459 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1460 | |
52ac8b35 PB |
1461 | /* |
1462 | * Do not store the new memslots while there are invalidations in | |
071064f1 PB |
1463 | * progress, otherwise the locking in invalidate_range_start and |
1464 | * invalidate_range_end will be unbalanced. | |
52ac8b35 PB |
1465 | */ |
1466 | spin_lock(&kvm->mn_invalidate_lock); | |
1467 | prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
1468 | while (kvm->mn_active_invalidate_count) { | |
1469 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1470 | spin_unlock(&kvm->mn_invalidate_lock); | |
1471 | schedule(); | |
1472 | spin_lock(&kvm->mn_invalidate_lock); | |
1473 | } | |
1474 | finish_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
f481b069 | 1475 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
52ac8b35 | 1476 | spin_unlock(&kvm->mn_invalidate_lock); |
b10a038e BG |
1477 | |
1478 | /* | |
1479 | * Acquired in kvm_set_memslot. Must be released before synchronize | |
1480 | * SRCU below in order to avoid deadlock with another thread | |
1481 | * acquiring the slots_arch_lock in an srcu critical section. | |
1482 | */ | |
1483 | mutex_unlock(&kvm->slots_arch_lock); | |
1484 | ||
7ec4fb44 | 1485 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1486 | |
ee3d1570 | 1487 | /* |
361209e0 | 1488 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1489 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1490 | * the number of address spaces. This provides a unique and easily |
1491 | * identifiable generation number while the memslots are in flux. | |
1492 | */ | |
1493 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1494 | ||
1495 | /* | |
4bd518f1 PB |
1496 | * Generations must be unique even across address spaces. We do not need |
1497 | * a global counter for that, instead the generation space is evenly split | |
1498 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1499 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1500 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1501 | */ |
164bf7e5 | 1502 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 1503 | |
15248258 | 1504 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1505 | |
15248258 | 1506 | slots->generation = gen; |
e59dbe09 TY |
1507 | |
1508 | return old_memslots; | |
7ec4fb44 GN |
1509 | } |
1510 | ||
ddc12f2a BG |
1511 | static size_t kvm_memslots_size(int slots) |
1512 | { | |
1513 | return sizeof(struct kvm_memslots) + | |
1514 | (sizeof(struct kvm_memory_slot) * slots); | |
1515 | } | |
1516 | ||
1517 | static void kvm_copy_memslots(struct kvm_memslots *to, | |
1518 | struct kvm_memslots *from) | |
1519 | { | |
1520 | memcpy(to, from, kvm_memslots_size(from->used_slots)); | |
1521 | } | |
1522 | ||
36947254 SC |
1523 | /* |
1524 | * Note, at a minimum, the current number of used slots must be allocated, even | |
1525 | * when deleting a memslot, as we need a complete duplicate of the memslots for | |
1526 | * use when invalidating a memslot prior to deleting/moving the memslot. | |
1527 | */ | |
1528 | static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old, | |
1529 | enum kvm_mr_change change) | |
1530 | { | |
1531 | struct kvm_memslots *slots; | |
ddc12f2a | 1532 | size_t new_size; |
36947254 SC |
1533 | |
1534 | if (change == KVM_MR_CREATE) | |
ddc12f2a | 1535 | new_size = kvm_memslots_size(old->used_slots + 1); |
36947254 | 1536 | else |
ddc12f2a | 1537 | new_size = kvm_memslots_size(old->used_slots); |
36947254 SC |
1538 | |
1539 | slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT); | |
1540 | if (likely(slots)) | |
ddc12f2a | 1541 | kvm_copy_memslots(slots, old); |
36947254 SC |
1542 | |
1543 | return slots; | |
1544 | } | |
1545 | ||
cf47f50b SC |
1546 | static int kvm_set_memslot(struct kvm *kvm, |
1547 | const struct kvm_userspace_memory_region *mem, | |
cf47f50b SC |
1548 | struct kvm_memory_slot *new, int as_id, |
1549 | enum kvm_mr_change change) | |
1550 | { | |
bda44d84 | 1551 | struct kvm_memory_slot *slot, old; |
cf47f50b SC |
1552 | struct kvm_memslots *slots; |
1553 | int r; | |
1554 | ||
b10a038e BG |
1555 | /* |
1556 | * Released in install_new_memslots. | |
1557 | * | |
1558 | * Must be held from before the current memslots are copied until | |
1559 | * after the new memslots are installed with rcu_assign_pointer, | |
1560 | * then released before the synchronize srcu in install_new_memslots. | |
1561 | * | |
1562 | * When modifying memslots outside of the slots_lock, must be held | |
1563 | * before reading the pointer to the current memslots until after all | |
1564 | * changes to those memslots are complete. | |
1565 | * | |
1566 | * These rules ensure that installing new memslots does not lose | |
1567 | * changes made to the previous memslots. | |
1568 | */ | |
1569 | mutex_lock(&kvm->slots_arch_lock); | |
1570 | ||
36947254 | 1571 | slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change); |
b10a038e BG |
1572 | if (!slots) { |
1573 | mutex_unlock(&kvm->slots_arch_lock); | |
cf47f50b | 1574 | return -ENOMEM; |
b10a038e | 1575 | } |
cf47f50b SC |
1576 | |
1577 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { | |
1578 | /* | |
1579 | * Note, the INVALID flag needs to be in the appropriate entry | |
1580 | * in the freshly allocated memslots, not in @old or @new. | |
1581 | */ | |
bda44d84 | 1582 | slot = id_to_memslot(slots, new->id); |
cf47f50b SC |
1583 | slot->flags |= KVM_MEMSLOT_INVALID; |
1584 | ||
1585 | /* | |
b10a038e BG |
1586 | * We can re-use the memory from the old memslots. |
1587 | * It will be overwritten with a copy of the new memslots | |
1588 | * after reacquiring the slots_arch_lock below. | |
cf47f50b SC |
1589 | */ |
1590 | slots = install_new_memslots(kvm, as_id, slots); | |
1591 | ||
1592 | /* From this point no new shadow pages pointing to a deleted, | |
1593 | * or moved, memslot will be created. | |
1594 | * | |
1595 | * validation of sp->gfn happens in: | |
1596 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1597 | * - kvm_is_visible_gfn (mmu_check_root) | |
1598 | */ | |
1599 | kvm_arch_flush_shadow_memslot(kvm, slot); | |
b10a038e BG |
1600 | |
1601 | /* Released in install_new_memslots. */ | |
1602 | mutex_lock(&kvm->slots_arch_lock); | |
1603 | ||
1604 | /* | |
1605 | * The arch-specific fields of the memslots could have changed | |
1606 | * between releasing the slots_arch_lock in | |
1607 | * install_new_memslots and here, so get a fresh copy of the | |
1608 | * slots. | |
1609 | */ | |
1610 | kvm_copy_memslots(slots, __kvm_memslots(kvm, as_id)); | |
cf47f50b SC |
1611 | } |
1612 | ||
bda44d84 SC |
1613 | /* |
1614 | * Make a full copy of the old memslot, the pointer will become stale | |
1615 | * when the memslots are re-sorted by update_memslots(), and the old | |
1616 | * memslot needs to be referenced after calling update_memslots(), e.g. | |
1617 | * to free its resources and for arch specific behavior. This needs to | |
1618 | * happen *after* (re)acquiring slots_arch_lock. | |
1619 | */ | |
1620 | slot = id_to_memslot(slots, new->id); | |
1621 | if (slot) { | |
1622 | old = *slot; | |
1623 | } else { | |
1624 | WARN_ON_ONCE(change != KVM_MR_CREATE); | |
1625 | memset(&old, 0, sizeof(old)); | |
1626 | old.id = new->id; | |
1627 | old.as_id = as_id; | |
1628 | } | |
1629 | ||
1630 | /* Copy the arch-specific data, again after (re)acquiring slots_arch_lock. */ | |
1631 | memcpy(&new->arch, &old.arch, sizeof(old.arch)); | |
1632 | ||
cf47f50b SC |
1633 | r = kvm_arch_prepare_memory_region(kvm, new, mem, change); |
1634 | if (r) | |
1635 | goto out_slots; | |
1636 | ||
1637 | update_memslots(slots, new, change); | |
1638 | slots = install_new_memslots(kvm, as_id, slots); | |
1639 | ||
bda44d84 SC |
1640 | kvm_arch_commit_memory_region(kvm, mem, &old, new, change); |
1641 | ||
1642 | /* Free the old memslot's metadata. Note, this is the full copy!!! */ | |
1643 | if (change == KVM_MR_DELETE) | |
1644 | kvm_free_memslot(kvm, &old); | |
cf47f50b SC |
1645 | |
1646 | kvfree(slots); | |
1647 | return 0; | |
1648 | ||
1649 | out_slots: | |
b10a038e | 1650 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
bda44d84 | 1651 | slot = id_to_memslot(slots, new->id); |
b10a038e | 1652 | slot->flags &= ~KVM_MEMSLOT_INVALID; |
cf47f50b | 1653 | slots = install_new_memslots(kvm, as_id, slots); |
b10a038e BG |
1654 | } else { |
1655 | mutex_unlock(&kvm->slots_arch_lock); | |
1656 | } | |
cf47f50b SC |
1657 | kvfree(slots); |
1658 | return r; | |
1659 | } | |
1660 | ||
5c0b4f3d SC |
1661 | static int kvm_delete_memslot(struct kvm *kvm, |
1662 | const struct kvm_userspace_memory_region *mem, | |
1663 | struct kvm_memory_slot *old, int as_id) | |
1664 | { | |
1665 | struct kvm_memory_slot new; | |
5c0b4f3d SC |
1666 | |
1667 | if (!old->npages) | |
1668 | return -EINVAL; | |
1669 | ||
1670 | memset(&new, 0, sizeof(new)); | |
1671 | new.id = old->id; | |
9e9eb226 PX |
1672 | /* |
1673 | * This is only for debugging purpose; it should never be referenced | |
1674 | * for a removed memslot. | |
1675 | */ | |
1676 | new.as_id = as_id; | |
5c0b4f3d | 1677 | |
bda44d84 | 1678 | return kvm_set_memslot(kvm, mem, &new, as_id, KVM_MR_DELETE); |
5c0b4f3d SC |
1679 | } |
1680 | ||
6aa8b732 AK |
1681 | /* |
1682 | * Allocate some memory and give it an address in the guest physical address | |
1683 | * space. | |
1684 | * | |
1685 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1686 | * |
02d5d55b | 1687 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1688 | */ |
f78e0e2e | 1689 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 1690 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 1691 | { |
6aa8b732 | 1692 | struct kvm_memory_slot old, new; |
163da372 | 1693 | struct kvm_memory_slot *tmp; |
f64c0398 | 1694 | enum kvm_mr_change change; |
163da372 SC |
1695 | int as_id, id; |
1696 | int r; | |
6aa8b732 | 1697 | |
a50d64d6 XG |
1698 | r = check_memory_region_flags(mem); |
1699 | if (r) | |
71a4c30b | 1700 | return r; |
a50d64d6 | 1701 | |
f481b069 PB |
1702 | as_id = mem->slot >> 16; |
1703 | id = (u16)mem->slot; | |
1704 | ||
6aa8b732 | 1705 | /* General sanity checks */ |
6b285a55 SC |
1706 | if ((mem->memory_size & (PAGE_SIZE - 1)) || |
1707 | (mem->memory_size != (unsigned long)mem->memory_size)) | |
71a4c30b | 1708 | return -EINVAL; |
6aa8b732 | 1709 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 1710 | return -EINVAL; |
fa3d315a | 1711 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 1712 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 1713 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 1714 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 1715 | mem->memory_size)) |
71a4c30b | 1716 | return -EINVAL; |
f481b069 | 1717 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 1718 | return -EINVAL; |
6aa8b732 | 1719 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 1720 | return -EINVAL; |
6aa8b732 | 1721 | |
5c0b4f3d SC |
1722 | /* |
1723 | * Make a full copy of the old memslot, the pointer will become stale | |
1724 | * when the memslots are re-sorted by update_memslots(), and the old | |
1725 | * memslot needs to be referenced after calling update_memslots(), e.g. | |
0dff0846 | 1726 | * to free its resources and for arch specific behavior. |
5c0b4f3d | 1727 | */ |
0577d1ab SC |
1728 | tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
1729 | if (tmp) { | |
1730 | old = *tmp; | |
1731 | tmp = NULL; | |
1732 | } else { | |
1733 | memset(&old, 0, sizeof(old)); | |
1734 | old.id = id; | |
1735 | } | |
163da372 | 1736 | |
5c0b4f3d SC |
1737 | if (!mem->memory_size) |
1738 | return kvm_delete_memslot(kvm, mem, &old, as_id); | |
1739 | ||
9e9eb226 | 1740 | new.as_id = as_id; |
f481b069 | 1741 | new.id = id; |
163da372 SC |
1742 | new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
1743 | new.npages = mem->memory_size >> PAGE_SHIFT; | |
6aa8b732 | 1744 | new.flags = mem->flags; |
414de7ab | 1745 | new.userspace_addr = mem->userspace_addr; |
6aa8b732 | 1746 | |
163da372 SC |
1747 | if (new.npages > KVM_MEM_MAX_NR_PAGES) |
1748 | return -EINVAL; | |
1749 | ||
5c0b4f3d SC |
1750 | if (!old.npages) { |
1751 | change = KVM_MR_CREATE; | |
163da372 | 1752 | new.dirty_bitmap = NULL; |
5c0b4f3d SC |
1753 | } else { /* Modify an existing slot. */ |
1754 | if ((new.userspace_addr != old.userspace_addr) || | |
163da372 | 1755 | (new.npages != old.npages) || |
5c0b4f3d | 1756 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) |
71a4c30b | 1757 | return -EINVAL; |
09170a49 | 1758 | |
163da372 | 1759 | if (new.base_gfn != old.base_gfn) |
5c0b4f3d SC |
1760 | change = KVM_MR_MOVE; |
1761 | else if (new.flags != old.flags) | |
1762 | change = KVM_MR_FLAGS_ONLY; | |
1763 | else /* Nothing to change. */ | |
1764 | return 0; | |
163da372 | 1765 | |
bda44d84 | 1766 | /* Copy dirty_bitmap from the current memslot. */ |
163da372 | 1767 | new.dirty_bitmap = old.dirty_bitmap; |
09170a49 | 1768 | } |
6aa8b732 | 1769 | |
f64c0398 | 1770 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee | 1771 | /* Check for overlaps */ |
163da372 SC |
1772 | kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) { |
1773 | if (tmp->id == id) | |
0a706bee | 1774 | continue; |
163da372 SC |
1775 | if (!((new.base_gfn + new.npages <= tmp->base_gfn) || |
1776 | (new.base_gfn >= tmp->base_gfn + tmp->npages))) | |
71a4c30b | 1777 | return -EEXIST; |
0a706bee | 1778 | } |
6aa8b732 | 1779 | } |
6aa8b732 | 1780 | |
414de7ab SC |
1781 | /* Allocate/free page dirty bitmap as needed */ |
1782 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1783 | new.dirty_bitmap = NULL; | |
044c59c4 | 1784 | else if (!new.dirty_bitmap && !kvm->dirty_ring_size) { |
3c9bd400 | 1785 | r = kvm_alloc_dirty_bitmap(&new); |
71a4c30b SC |
1786 | if (r) |
1787 | return r; | |
3c9bd400 JZ |
1788 | |
1789 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1790 | bitmap_set(new.dirty_bitmap, 0, new.npages); | |
6aa8b732 AK |
1791 | } |
1792 | ||
bda44d84 | 1793 | r = kvm_set_memslot(kvm, mem, &new, as_id, change); |
cf47f50b SC |
1794 | if (r) |
1795 | goto out_bitmap; | |
82ce2c96 | 1796 | |
5c0b4f3d SC |
1797 | if (old.dirty_bitmap && !new.dirty_bitmap) |
1798 | kvm_destroy_dirty_bitmap(&old); | |
6aa8b732 AK |
1799 | return 0; |
1800 | ||
bd0e96fd SC |
1801 | out_bitmap: |
1802 | if (new.dirty_bitmap && !old.dirty_bitmap) | |
1803 | kvm_destroy_dirty_bitmap(&new); | |
6aa8b732 | 1804 | return r; |
210c7c4d | 1805 | } |
f78e0e2e SY |
1806 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1807 | ||
1808 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1809 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1810 | { |
1811 | int r; | |
1812 | ||
79fac95e | 1813 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1814 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1815 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1816 | return r; |
1817 | } | |
210c7c4d IE |
1818 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1819 | ||
7940876e SH |
1820 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1821 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1822 | { |
f481b069 | 1823 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1824 | return -EINVAL; |
09170a49 | 1825 | |
47ae31e2 | 1826 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1827 | } |
1828 | ||
0dff0846 | 1829 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
1830 | /** |
1831 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
1832 | * @kvm: pointer to kvm instance | |
1833 | * @log: slot id and address to which we copy the log | |
1834 | * @is_dirty: set to '1' if any dirty pages were found | |
1835 | * @memslot: set to the associated memslot, always valid on success | |
1836 | */ | |
1837 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
1838 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 1839 | { |
9f6b8029 | 1840 | struct kvm_memslots *slots; |
843574a3 | 1841 | int i, as_id, id; |
87bf6e7d | 1842 | unsigned long n; |
6aa8b732 AK |
1843 | unsigned long any = 0; |
1844 | ||
b2cc64c4 PX |
1845 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1846 | if (kvm->dirty_ring_size) | |
1847 | return -ENXIO; | |
1848 | ||
2a49f61d SC |
1849 | *memslot = NULL; |
1850 | *is_dirty = 0; | |
1851 | ||
f481b069 PB |
1852 | as_id = log->slot >> 16; |
1853 | id = (u16)log->slot; | |
1854 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 1855 | return -EINVAL; |
6aa8b732 | 1856 | |
f481b069 | 1857 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 1858 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 1859 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 1860 | return -ENOENT; |
6aa8b732 | 1861 | |
2a49f61d SC |
1862 | kvm_arch_sync_dirty_log(kvm, *memslot); |
1863 | ||
1864 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 1865 | |
cd1a4a98 | 1866 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 1867 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 1868 | |
2a49f61d | 1869 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 1870 | return -EFAULT; |
6aa8b732 | 1871 | |
5bb064dc ZX |
1872 | if (any) |
1873 | *is_dirty = 1; | |
843574a3 | 1874 | return 0; |
6aa8b732 | 1875 | } |
2ba9f0d8 | 1876 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1877 | |
0dff0846 | 1878 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 1879 | /** |
b8b00220 | 1880 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 1881 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
1882 | * @kvm: pointer to kvm instance |
1883 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
1884 | * |
1885 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1886 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1887 | * following order: | |
1888 | * | |
1889 | * 1. Take a snapshot of the bit and clear it if needed. | |
1890 | * 2. Write protect the corresponding page. | |
1891 | * 3. Copy the snapshot to the userspace. | |
1892 | * 4. Upon return caller flushes TLB's if needed. | |
1893 | * | |
1894 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1895 | * entry. This is not a problem because the page is reported dirty using | |
1896 | * the snapshot taken before and step 4 ensures that writes done after | |
1897 | * exiting to userspace will be logged for the next call. | |
1898 | * | |
1899 | */ | |
0dff0846 | 1900 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 1901 | { |
9f6b8029 | 1902 | struct kvm_memslots *slots; |
ba0513b5 | 1903 | struct kvm_memory_slot *memslot; |
58d6db34 | 1904 | int i, as_id, id; |
ba0513b5 MS |
1905 | unsigned long n; |
1906 | unsigned long *dirty_bitmap; | |
1907 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 1908 | bool flush; |
ba0513b5 | 1909 | |
b2cc64c4 PX |
1910 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1911 | if (kvm->dirty_ring_size) | |
1912 | return -ENXIO; | |
1913 | ||
f481b069 PB |
1914 | as_id = log->slot >> 16; |
1915 | id = (u16)log->slot; | |
1916 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 1917 | return -EINVAL; |
ba0513b5 | 1918 | |
f481b069 PB |
1919 | slots = __kvm_memslots(kvm, as_id); |
1920 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
1921 | if (!memslot || !memslot->dirty_bitmap) |
1922 | return -ENOENT; | |
ba0513b5 MS |
1923 | |
1924 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 1925 | |
0dff0846 SC |
1926 | kvm_arch_sync_dirty_log(kvm, memslot); |
1927 | ||
ba0513b5 | 1928 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 1929 | flush = false; |
2a31b9db PB |
1930 | if (kvm->manual_dirty_log_protect) { |
1931 | /* | |
1932 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
1933 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
1934 | * is some code duplication between this function and | |
1935 | * kvm_get_dirty_log, but hopefully all architecture | |
1936 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
1937 | * can be eliminated. | |
1938 | */ | |
1939 | dirty_bitmap_buffer = dirty_bitmap; | |
1940 | } else { | |
1941 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
1942 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 1943 | |
531810ca | 1944 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
1945 | for (i = 0; i < n / sizeof(long); i++) { |
1946 | unsigned long mask; | |
1947 | gfn_t offset; | |
ba0513b5 | 1948 | |
2a31b9db PB |
1949 | if (!dirty_bitmap[i]) |
1950 | continue; | |
1951 | ||
0dff0846 | 1952 | flush = true; |
2a31b9db PB |
1953 | mask = xchg(&dirty_bitmap[i], 0); |
1954 | dirty_bitmap_buffer[i] = mask; | |
1955 | ||
a67794ca LT |
1956 | offset = i * BITS_PER_LONG; |
1957 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1958 | offset, mask); | |
2a31b9db | 1959 | } |
531810ca | 1960 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
1961 | } |
1962 | ||
0dff0846 SC |
1963 | if (flush) |
1964 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
1965 | ||
2a31b9db PB |
1966 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
1967 | return -EFAULT; | |
1968 | return 0; | |
1969 | } | |
0dff0846 SC |
1970 | |
1971 | ||
1972 | /** | |
1973 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
1974 | * @kvm: kvm instance | |
1975 | * @log: slot id and address to which we copy the log | |
1976 | * | |
1977 | * Steps 1-4 below provide general overview of dirty page logging. See | |
1978 | * kvm_get_dirty_log_protect() function description for additional details. | |
1979 | * | |
1980 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
1981 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
1982 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
1983 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
1984 | * writes will be marked dirty for next log read. | |
1985 | * | |
1986 | * 1. Take a snapshot of the bit and clear it if needed. | |
1987 | * 2. Write protect the corresponding page. | |
1988 | * 3. Copy the snapshot to the userspace. | |
1989 | * 4. Flush TLB's if needed. | |
1990 | */ | |
1991 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
1992 | struct kvm_dirty_log *log) | |
1993 | { | |
1994 | int r; | |
1995 | ||
1996 | mutex_lock(&kvm->slots_lock); | |
1997 | ||
1998 | r = kvm_get_dirty_log_protect(kvm, log); | |
1999 | ||
2000 | mutex_unlock(&kvm->slots_lock); | |
2001 | return r; | |
2002 | } | |
2a31b9db PB |
2003 | |
2004 | /** | |
2005 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
2006 | * and reenable dirty page tracking for the corresponding pages. | |
2007 | * @kvm: pointer to kvm instance | |
2008 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
2009 | */ | |
0dff0846 SC |
2010 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
2011 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
2012 | { |
2013 | struct kvm_memslots *slots; | |
2014 | struct kvm_memory_slot *memslot; | |
98938aa8 | 2015 | int as_id, id; |
2a31b9db | 2016 | gfn_t offset; |
98938aa8 | 2017 | unsigned long i, n; |
2a31b9db PB |
2018 | unsigned long *dirty_bitmap; |
2019 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2020 | bool flush; |
2a31b9db | 2021 | |
b2cc64c4 PX |
2022 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
2023 | if (kvm->dirty_ring_size) | |
2024 | return -ENXIO; | |
2025 | ||
2a31b9db PB |
2026 | as_id = log->slot >> 16; |
2027 | id = (u16)log->slot; | |
2028 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
2029 | return -EINVAL; | |
2030 | ||
76d58e0f | 2031 | if (log->first_page & 63) |
2a31b9db PB |
2032 | return -EINVAL; |
2033 | ||
2034 | slots = __kvm_memslots(kvm, as_id); | |
2035 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2036 | if (!memslot || !memslot->dirty_bitmap) |
2037 | return -ENOENT; | |
2a31b9db PB |
2038 | |
2039 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 2040 | |
4ddc9204 | 2041 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
2042 | |
2043 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
2044 | log->num_pages > memslot->npages - log->first_page || |
2045 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
2046 | return -EINVAL; | |
98938aa8 | 2047 | |
0dff0846 SC |
2048 | kvm_arch_sync_dirty_log(kvm, memslot); |
2049 | ||
2050 | flush = false; | |
2a31b9db PB |
2051 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
2052 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
2053 | return -EFAULT; | |
ba0513b5 | 2054 | |
531810ca | 2055 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
2056 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
2057 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
2058 | i++, offset += BITS_PER_LONG) { |
2059 | unsigned long mask = *dirty_bitmap_buffer++; | |
2060 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
2061 | if (!mask) | |
ba0513b5 MS |
2062 | continue; |
2063 | ||
2a31b9db | 2064 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 2065 | |
2a31b9db PB |
2066 | /* |
2067 | * mask contains the bits that really have been cleared. This | |
2068 | * never includes any bits beyond the length of the memslot (if | |
2069 | * the length is not aligned to 64 pages), therefore it is not | |
2070 | * a problem if userspace sets them in log->dirty_bitmap. | |
2071 | */ | |
58d2930f | 2072 | if (mask) { |
0dff0846 | 2073 | flush = true; |
58d2930f TY |
2074 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
2075 | offset, mask); | |
2076 | } | |
ba0513b5 | 2077 | } |
531810ca | 2078 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 2079 | |
0dff0846 SC |
2080 | if (flush) |
2081 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2082 | ||
58d6db34 | 2083 | return 0; |
ba0513b5 | 2084 | } |
0dff0846 SC |
2085 | |
2086 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
2087 | struct kvm_clear_dirty_log *log) | |
2088 | { | |
2089 | int r; | |
2090 | ||
2091 | mutex_lock(&kvm->slots_lock); | |
2092 | ||
2093 | r = kvm_clear_dirty_log_protect(kvm, log); | |
2094 | ||
2095 | mutex_unlock(&kvm->slots_lock); | |
2096 | return r; | |
2097 | } | |
2098 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 2099 | |
49c7754c GN |
2100 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
2101 | { | |
2102 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
2103 | } | |
a1f4d395 | 2104 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 2105 | |
8e73485c PB |
2106 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
2107 | { | |
fe22ed82 DM |
2108 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
2109 | struct kvm_memory_slot *slot; | |
2110 | int slot_index; | |
2111 | ||
2112 | slot = try_get_memslot(slots, vcpu->last_used_slot, gfn); | |
2113 | if (slot) | |
2114 | return slot; | |
2115 | ||
2116 | /* | |
2117 | * Fall back to searching all memslots. We purposely use | |
2118 | * search_memslots() instead of __gfn_to_memslot() to avoid | |
2119 | * thrashing the VM-wide last_used_index in kvm_memslots. | |
2120 | */ | |
2121 | slot = search_memslots(slots, gfn, &slot_index); | |
2122 | if (slot) { | |
2123 | vcpu->last_used_slot = slot_index; | |
2124 | return slot; | |
2125 | } | |
2126 | ||
2127 | return NULL; | |
8e73485c | 2128 | } |
e72436bc | 2129 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot); |
8e73485c | 2130 | |
33e94154 | 2131 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 2132 | { |
bf3e05bc | 2133 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 2134 | |
c36b7150 | 2135 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
2136 | } |
2137 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
2138 | ||
995decb6 VK |
2139 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
2140 | { | |
2141 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2142 | ||
2143 | return kvm_is_visible_memslot(memslot); | |
2144 | } | |
2145 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
2146 | ||
f9b84e19 | 2147 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
2148 | { |
2149 | struct vm_area_struct *vma; | |
2150 | unsigned long addr, size; | |
2151 | ||
2152 | size = PAGE_SIZE; | |
2153 | ||
42cde48b | 2154 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
2155 | if (kvm_is_error_hva(addr)) |
2156 | return PAGE_SIZE; | |
2157 | ||
d8ed45c5 | 2158 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
2159 | vma = find_vma(current->mm, addr); |
2160 | if (!vma) | |
2161 | goto out; | |
2162 | ||
2163 | size = vma_kernel_pagesize(vma); | |
2164 | ||
2165 | out: | |
d8ed45c5 | 2166 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
2167 | |
2168 | return size; | |
2169 | } | |
2170 | ||
4d8b81ab XG |
2171 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
2172 | { | |
2173 | return slot->flags & KVM_MEM_READONLY; | |
2174 | } | |
2175 | ||
4d8b81ab XG |
2176 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2177 | gfn_t *nr_pages, bool write) | |
539cb660 | 2178 | { |
bc6678a3 | 2179 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 2180 | return KVM_HVA_ERR_BAD; |
48987781 | 2181 | |
4d8b81ab XG |
2182 | if (memslot_is_readonly(slot) && write) |
2183 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
2184 | |
2185 | if (nr_pages) | |
2186 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
2187 | ||
4d8b81ab | 2188 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 2189 | } |
48987781 | 2190 | |
4d8b81ab XG |
2191 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2192 | gfn_t *nr_pages) | |
2193 | { | |
2194 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 2195 | } |
48987781 | 2196 | |
4d8b81ab | 2197 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 2198 | gfn_t gfn) |
4d8b81ab XG |
2199 | { |
2200 | return gfn_to_hva_many(slot, gfn, NULL); | |
2201 | } | |
2202 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
2203 | ||
48987781 XG |
2204 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
2205 | { | |
49c7754c | 2206 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 2207 | } |
0d150298 | 2208 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 2209 | |
8e73485c PB |
2210 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
2211 | { | |
2212 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
2213 | } | |
2214 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
2215 | ||
86ab8cff | 2216 | /* |
970c0d4b WY |
2217 | * Return the hva of a @gfn and the R/W attribute if possible. |
2218 | * | |
2219 | * @slot: the kvm_memory_slot which contains @gfn | |
2220 | * @gfn: the gfn to be translated | |
2221 | * @writable: used to return the read/write attribute of the @slot if the hva | |
2222 | * is valid and @writable is not NULL | |
86ab8cff | 2223 | */ |
64d83126 CD |
2224 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
2225 | gfn_t gfn, bool *writable) | |
86ab8cff | 2226 | { |
a2ac07fe GN |
2227 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
2228 | ||
2229 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
2230 | *writable = !memslot_is_readonly(slot); |
2231 | ||
a2ac07fe | 2232 | return hva; |
86ab8cff XG |
2233 | } |
2234 | ||
64d83126 CD |
2235 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
2236 | { | |
2237 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2238 | ||
2239 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2240 | } | |
2241 | ||
8e73485c PB |
2242 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
2243 | { | |
2244 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2245 | ||
2246 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2247 | } | |
2248 | ||
fafc3dba HY |
2249 | static inline int check_user_page_hwpoison(unsigned long addr) |
2250 | { | |
0d731759 | 2251 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 2252 | |
0d731759 | 2253 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
2254 | return rc == -EHWPOISON; |
2255 | } | |
2256 | ||
2fc84311 | 2257 | /* |
b9b33da2 PB |
2258 | * The fast path to get the writable pfn which will be stored in @pfn, |
2259 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 2260 | * only part that runs if we can in atomic context. |
2fc84311 | 2261 | */ |
b9b33da2 PB |
2262 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
2263 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 2264 | { |
8d4e1288 | 2265 | struct page *page[1]; |
954bbbc2 | 2266 | |
12ce13fe XG |
2267 | /* |
2268 | * Fast pin a writable pfn only if it is a write fault request | |
2269 | * or the caller allows to map a writable pfn for a read fault | |
2270 | * request. | |
2271 | */ | |
2272 | if (!(write_fault || writable)) | |
2273 | return false; | |
612819c3 | 2274 | |
dadbb612 | 2275 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 2276 | *pfn = page_to_pfn(page[0]); |
612819c3 | 2277 | |
2fc84311 XG |
2278 | if (writable) |
2279 | *writable = true; | |
2280 | return true; | |
2281 | } | |
af585b92 | 2282 | |
2fc84311 XG |
2283 | return false; |
2284 | } | |
612819c3 | 2285 | |
2fc84311 XG |
2286 | /* |
2287 | * The slow path to get the pfn of the specified host virtual address, | |
2288 | * 1 indicates success, -errno is returned if error is detected. | |
2289 | */ | |
2290 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 2291 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 2292 | { |
ce53053c AV |
2293 | unsigned int flags = FOLL_HWPOISON; |
2294 | struct page *page; | |
2fc84311 | 2295 | int npages = 0; |
612819c3 | 2296 | |
2fc84311 XG |
2297 | might_sleep(); |
2298 | ||
2299 | if (writable) | |
2300 | *writable = write_fault; | |
2301 | ||
ce53053c AV |
2302 | if (write_fault) |
2303 | flags |= FOLL_WRITE; | |
2304 | if (async) | |
2305 | flags |= FOLL_NOWAIT; | |
d4944b0e | 2306 | |
ce53053c | 2307 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
2308 | if (npages != 1) |
2309 | return npages; | |
2310 | ||
2311 | /* map read fault as writable if possible */ | |
12ce13fe | 2312 | if (unlikely(!write_fault) && writable) { |
ce53053c | 2313 | struct page *wpage; |
2fc84311 | 2314 | |
dadbb612 | 2315 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 2316 | *writable = true; |
ce53053c AV |
2317 | put_page(page); |
2318 | page = wpage; | |
612819c3 | 2319 | } |
887c08ac | 2320 | } |
ce53053c | 2321 | *pfn = page_to_pfn(page); |
2fc84311 XG |
2322 | return npages; |
2323 | } | |
539cb660 | 2324 | |
4d8b81ab XG |
2325 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
2326 | { | |
2327 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
2328 | return false; | |
2e2e3738 | 2329 | |
4d8b81ab XG |
2330 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
2331 | return false; | |
887c08ac | 2332 | |
4d8b81ab XG |
2333 | return true; |
2334 | } | |
bf998156 | 2335 | |
f8be156b NP |
2336 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
2337 | { | |
2338 | if (kvm_is_reserved_pfn(pfn)) | |
2339 | return 1; | |
2340 | return get_page_unless_zero(pfn_to_page(pfn)); | |
2341 | } | |
2342 | ||
92176a8e PB |
2343 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
2344 | unsigned long addr, bool *async, | |
a340b3e2 KA |
2345 | bool write_fault, bool *writable, |
2346 | kvm_pfn_t *p_pfn) | |
92176a8e | 2347 | { |
a9545779 | 2348 | kvm_pfn_t pfn; |
bd2fae8d PB |
2349 | pte_t *ptep; |
2350 | spinlock_t *ptl; | |
add6a0cd PB |
2351 | int r; |
2352 | ||
9fd6dad1 | 2353 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2354 | if (r) { |
2355 | /* | |
2356 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
2357 | * not call the fault handler, so do it here. | |
2358 | */ | |
2359 | bool unlocked = false; | |
64019a2e | 2360 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
2361 | (write_fault ? FAULT_FLAG_WRITE : 0), |
2362 | &unlocked); | |
a8387d0b PB |
2363 | if (unlocked) |
2364 | return -EAGAIN; | |
add6a0cd PB |
2365 | if (r) |
2366 | return r; | |
2367 | ||
9fd6dad1 | 2368 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2369 | if (r) |
2370 | return r; | |
bd2fae8d | 2371 | } |
add6a0cd | 2372 | |
bd2fae8d PB |
2373 | if (write_fault && !pte_write(*ptep)) { |
2374 | pfn = KVM_PFN_ERR_RO_FAULT; | |
2375 | goto out; | |
add6a0cd PB |
2376 | } |
2377 | ||
a340b3e2 | 2378 | if (writable) |
bd2fae8d PB |
2379 | *writable = pte_write(*ptep); |
2380 | pfn = pte_pfn(*ptep); | |
add6a0cd PB |
2381 | |
2382 | /* | |
2383 | * Get a reference here because callers of *hva_to_pfn* and | |
2384 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
2385 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
36c3ce6c | 2386 | * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will |
add6a0cd PB |
2387 | * simply do nothing for reserved pfns. |
2388 | * | |
2389 | * Whoever called remap_pfn_range is also going to call e.g. | |
2390 | * unmap_mapping_range before the underlying pages are freed, | |
2391 | * causing a call to our MMU notifier. | |
f8be156b NP |
2392 | * |
2393 | * Certain IO or PFNMAP mappings can be backed with valid | |
2394 | * struct pages, but be allocated without refcounting e.g., | |
2395 | * tail pages of non-compound higher order allocations, which | |
2396 | * would then underflow the refcount when the caller does the | |
2397 | * required put_page. Don't allow those pages here. | |
add6a0cd | 2398 | */ |
f8be156b NP |
2399 | if (!kvm_try_get_pfn(pfn)) |
2400 | r = -EFAULT; | |
add6a0cd | 2401 | |
bd2fae8d PB |
2402 | out: |
2403 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 2404 | *p_pfn = pfn; |
f8be156b NP |
2405 | |
2406 | return r; | |
92176a8e PB |
2407 | } |
2408 | ||
12ce13fe XG |
2409 | /* |
2410 | * Pin guest page in memory and return its pfn. | |
2411 | * @addr: host virtual address which maps memory to the guest | |
2412 | * @atomic: whether this function can sleep | |
2413 | * @async: whether this function need to wait IO complete if the | |
2414 | * host page is not in the memory | |
2415 | * @write_fault: whether we should get a writable host page | |
2416 | * @writable: whether it allows to map a writable host page for !@write_fault | |
2417 | * | |
2418 | * The function will map a writable host page for these two cases: | |
2419 | * 1): @write_fault = true | |
2420 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
2421 | * whether the mapping is writable. | |
2422 | */ | |
ba049e93 | 2423 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
2424 | bool write_fault, bool *writable) |
2425 | { | |
2426 | struct vm_area_struct *vma; | |
ba049e93 | 2427 | kvm_pfn_t pfn = 0; |
92176a8e | 2428 | int npages, r; |
2e2e3738 | 2429 | |
2fc84311 XG |
2430 | /* we can do it either atomically or asynchronously, not both */ |
2431 | BUG_ON(atomic && async); | |
8d4e1288 | 2432 | |
b9b33da2 | 2433 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
2434 | return pfn; |
2435 | ||
2436 | if (atomic) | |
2437 | return KVM_PFN_ERR_FAULT; | |
2438 | ||
2439 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
2440 | if (npages == 1) | |
2441 | return pfn; | |
8d4e1288 | 2442 | |
d8ed45c5 | 2443 | mmap_read_lock(current->mm); |
2fc84311 XG |
2444 | if (npages == -EHWPOISON || |
2445 | (!async && check_user_page_hwpoison(addr))) { | |
2446 | pfn = KVM_PFN_ERR_HWPOISON; | |
2447 | goto exit; | |
2448 | } | |
2449 | ||
a8387d0b | 2450 | retry: |
fc98c03b | 2451 | vma = vma_lookup(current->mm, addr); |
2fc84311 XG |
2452 | |
2453 | if (vma == NULL) | |
2454 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 2455 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
a340b3e2 | 2456 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn); |
a8387d0b PB |
2457 | if (r == -EAGAIN) |
2458 | goto retry; | |
92176a8e PB |
2459 | if (r < 0) |
2460 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 2461 | } else { |
4d8b81ab | 2462 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
2463 | *async = true; |
2464 | pfn = KVM_PFN_ERR_FAULT; | |
2465 | } | |
2466 | exit: | |
d8ed45c5 | 2467 | mmap_read_unlock(current->mm); |
2e2e3738 | 2468 | return pfn; |
35149e21 AL |
2469 | } |
2470 | ||
ba049e93 DW |
2471 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
2472 | bool atomic, bool *async, bool write_fault, | |
4a42d848 | 2473 | bool *writable, hva_t *hva) |
887c08ac | 2474 | { |
4d8b81ab XG |
2475 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
2476 | ||
4a42d848 DS |
2477 | if (hva) |
2478 | *hva = addr; | |
2479 | ||
b2740d35 PB |
2480 | if (addr == KVM_HVA_ERR_RO_BAD) { |
2481 | if (writable) | |
2482 | *writable = false; | |
4d8b81ab | 2483 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 2484 | } |
4d8b81ab | 2485 | |
b2740d35 PB |
2486 | if (kvm_is_error_hva(addr)) { |
2487 | if (writable) | |
2488 | *writable = false; | |
81c52c56 | 2489 | return KVM_PFN_NOSLOT; |
b2740d35 | 2490 | } |
4d8b81ab XG |
2491 | |
2492 | /* Do not map writable pfn in the readonly memslot. */ | |
2493 | if (writable && memslot_is_readonly(slot)) { | |
2494 | *writable = false; | |
2495 | writable = NULL; | |
2496 | } | |
2497 | ||
2498 | return hva_to_pfn(addr, atomic, async, write_fault, | |
2499 | writable); | |
887c08ac | 2500 | } |
3520469d | 2501 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 2502 | |
ba049e93 | 2503 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
2504 | bool *writable) |
2505 | { | |
e37afc6e | 2506 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
4a42d848 | 2507 | write_fault, writable, NULL); |
612819c3 MT |
2508 | } |
2509 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
2510 | ||
ba049e93 | 2511 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2512 | { |
4a42d848 | 2513 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL); |
506f0d6f | 2514 | } |
e37afc6e | 2515 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 2516 | |
ba049e93 | 2517 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2518 | { |
4a42d848 | 2519 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL); |
506f0d6f | 2520 | } |
037d92dc | 2521 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 2522 | |
ba049e93 | 2523 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2524 | { |
2525 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2526 | } | |
2527 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
2528 | ||
ba049e93 | 2529 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
2530 | { |
2531 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
2532 | } | |
2533 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
2534 | ||
ba049e93 | 2535 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2536 | { |
2537 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2538 | } | |
2539 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
2540 | ||
d9ef13c2 PB |
2541 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2542 | struct page **pages, int nr_pages) | |
48987781 XG |
2543 | { |
2544 | unsigned long addr; | |
076b925d | 2545 | gfn_t entry = 0; |
48987781 | 2546 | |
d9ef13c2 | 2547 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
2548 | if (kvm_is_error_hva(addr)) |
2549 | return -1; | |
2550 | ||
2551 | if (entry < nr_pages) | |
2552 | return 0; | |
2553 | ||
dadbb612 | 2554 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
2555 | } |
2556 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
2557 | ||
ba049e93 | 2558 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 2559 | { |
81c52c56 | 2560 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 2561 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2562 | |
bf4bea8e | 2563 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 2564 | WARN_ON(1); |
6cede2e6 | 2565 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 2566 | } |
a2766325 XG |
2567 | |
2568 | return pfn_to_page(pfn); | |
2569 | } | |
2570 | ||
35149e21 AL |
2571 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
2572 | { | |
ba049e93 | 2573 | kvm_pfn_t pfn; |
2e2e3738 AL |
2574 | |
2575 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 2576 | |
a2766325 | 2577 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
2578 | } |
2579 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
2580 | ||
357a18ad | 2581 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty) |
91724814 BO |
2582 | { |
2583 | if (pfn == 0) | |
2584 | return; | |
2585 | ||
91724814 BO |
2586 | if (dirty) |
2587 | kvm_release_pfn_dirty(pfn); | |
2588 | else | |
2589 | kvm_release_pfn_clean(pfn); | |
2590 | } | |
2591 | ||
357a18ad | 2592 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
e45adf66 KA |
2593 | { |
2594 | kvm_pfn_t pfn; | |
2595 | void *hva = NULL; | |
2596 | struct page *page = KVM_UNMAPPED_PAGE; | |
2597 | ||
2598 | if (!map) | |
2599 | return -EINVAL; | |
2600 | ||
357a18ad | 2601 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
e45adf66 KA |
2602 | if (is_error_noslot_pfn(pfn)) |
2603 | return -EINVAL; | |
2604 | ||
2605 | if (pfn_valid(pfn)) { | |
2606 | page = pfn_to_page(pfn); | |
357a18ad | 2607 | hva = kmap(page); |
d30b214d | 2608 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2609 | } else { |
357a18ad | 2610 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
d30b214d | 2611 | #endif |
e45adf66 KA |
2612 | } |
2613 | ||
2614 | if (!hva) | |
2615 | return -EFAULT; | |
2616 | ||
2617 | map->page = page; | |
2618 | map->hva = hva; | |
2619 | map->pfn = pfn; | |
2620 | map->gfn = gfn; | |
2621 | ||
2622 | return 0; | |
2623 | } | |
e45adf66 KA |
2624 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); |
2625 | ||
357a18ad | 2626 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) |
e45adf66 KA |
2627 | { |
2628 | if (!map) | |
2629 | return; | |
2630 | ||
2631 | if (!map->hva) | |
2632 | return; | |
2633 | ||
357a18ad DW |
2634 | if (map->page != KVM_UNMAPPED_PAGE) |
2635 | kunmap(map->page); | |
eb1f2f38 | 2636 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2637 | else |
357a18ad | 2638 | memunmap(map->hva); |
eb1f2f38 | 2639 | #endif |
e45adf66 | 2640 | |
91724814 | 2641 | if (dirty) |
357a18ad | 2642 | kvm_vcpu_mark_page_dirty(vcpu, map->gfn); |
91724814 | 2643 | |
357a18ad | 2644 | kvm_release_pfn(map->pfn, dirty); |
e45adf66 KA |
2645 | |
2646 | map->hva = NULL; | |
2647 | map->page = NULL; | |
2648 | } | |
2649 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); | |
2650 | ||
8e73485c PB |
2651 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
2652 | { | |
ba049e93 | 2653 | kvm_pfn_t pfn; |
8e73485c PB |
2654 | |
2655 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
2656 | ||
2657 | return kvm_pfn_to_page(pfn); | |
2658 | } | |
2659 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
2660 | ||
b4231d61 IE |
2661 | void kvm_release_page_clean(struct page *page) |
2662 | { | |
32cad84f XG |
2663 | WARN_ON(is_error_page(page)); |
2664 | ||
35149e21 | 2665 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
2666 | } |
2667 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
2668 | ||
ba049e93 | 2669 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 2670 | { |
bf4bea8e | 2671 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 2672 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
2673 | } |
2674 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
2675 | ||
b4231d61 | 2676 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 2677 | { |
a2766325 XG |
2678 | WARN_ON(is_error_page(page)); |
2679 | ||
35149e21 AL |
2680 | kvm_release_pfn_dirty(page_to_pfn(page)); |
2681 | } | |
2682 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
2683 | ||
f7a6509f | 2684 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
2685 | { |
2686 | kvm_set_pfn_dirty(pfn); | |
2687 | kvm_release_pfn_clean(pfn); | |
2688 | } | |
f7a6509f | 2689 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 2690 | |
ba049e93 | 2691 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2692 | { |
d29c03a5 ML |
2693 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2694 | SetPageDirty(pfn_to_page(pfn)); | |
8a7ae055 | 2695 | } |
35149e21 AL |
2696 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
2697 | ||
ba049e93 | 2698 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 2699 | { |
a78986aa | 2700 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2e2e3738 | 2701 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
2702 | } |
2703 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
2704 | ||
195aefde IE |
2705 | static int next_segment(unsigned long len, int offset) |
2706 | { | |
2707 | if (len > PAGE_SIZE - offset) | |
2708 | return PAGE_SIZE - offset; | |
2709 | else | |
2710 | return len; | |
2711 | } | |
2712 | ||
8e73485c PB |
2713 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
2714 | void *data, int offset, int len) | |
195aefde | 2715 | { |
e0506bcb IE |
2716 | int r; |
2717 | unsigned long addr; | |
195aefde | 2718 | |
8e73485c | 2719 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
2720 | if (kvm_is_error_hva(addr)) |
2721 | return -EFAULT; | |
3180a7fc | 2722 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 2723 | if (r) |
195aefde | 2724 | return -EFAULT; |
195aefde IE |
2725 | return 0; |
2726 | } | |
8e73485c PB |
2727 | |
2728 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
2729 | int len) | |
2730 | { | |
2731 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2732 | ||
2733 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2734 | } | |
195aefde IE |
2735 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
2736 | ||
8e73485c PB |
2737 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
2738 | int offset, int len) | |
2739 | { | |
2740 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2741 | ||
2742 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2743 | } | |
2744 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
2745 | ||
195aefde IE |
2746 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
2747 | { | |
2748 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2749 | int seg; | |
2750 | int offset = offset_in_page(gpa); | |
2751 | int ret; | |
2752 | ||
2753 | while ((seg = next_segment(len, offset)) != 0) { | |
2754 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
2755 | if (ret < 0) | |
2756 | return ret; | |
2757 | offset = 0; | |
2758 | len -= seg; | |
2759 | data += seg; | |
2760 | ++gfn; | |
2761 | } | |
2762 | return 0; | |
2763 | } | |
2764 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
2765 | ||
8e73485c | 2766 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 2767 | { |
7ec54588 | 2768 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 2769 | int seg; |
7ec54588 | 2770 | int offset = offset_in_page(gpa); |
8e73485c PB |
2771 | int ret; |
2772 | ||
2773 | while ((seg = next_segment(len, offset)) != 0) { | |
2774 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
2775 | if (ret < 0) | |
2776 | return ret; | |
2777 | offset = 0; | |
2778 | len -= seg; | |
2779 | data += seg; | |
2780 | ++gfn; | |
2781 | } | |
2782 | return 0; | |
2783 | } | |
2784 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 2785 | |
8e73485c PB |
2786 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2787 | void *data, int offset, unsigned long len) | |
2788 | { | |
2789 | int r; | |
2790 | unsigned long addr; | |
2791 | ||
2792 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
2793 | if (kvm_is_error_hva(addr)) |
2794 | return -EFAULT; | |
0aac03f0 | 2795 | pagefault_disable(); |
3180a7fc | 2796 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 2797 | pagefault_enable(); |
7ec54588 MT |
2798 | if (r) |
2799 | return -EFAULT; | |
2800 | return 0; | |
2801 | } | |
7ec54588 | 2802 | |
8e73485c PB |
2803 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
2804 | void *data, unsigned long len) | |
2805 | { | |
2806 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2807 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2808 | int offset = offset_in_page(gpa); | |
2809 | ||
2810 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2811 | } | |
2812 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
2813 | ||
28bd726a PX |
2814 | static int __kvm_write_guest_page(struct kvm *kvm, |
2815 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 2816 | const void *data, int offset, int len) |
195aefde | 2817 | { |
e0506bcb IE |
2818 | int r; |
2819 | unsigned long addr; | |
195aefde | 2820 | |
251eb841 | 2821 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
2822 | if (kvm_is_error_hva(addr)) |
2823 | return -EFAULT; | |
8b0cedff | 2824 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 2825 | if (r) |
195aefde | 2826 | return -EFAULT; |
28bd726a | 2827 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
2828 | return 0; |
2829 | } | |
8e73485c PB |
2830 | |
2831 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
2832 | const void *data, int offset, int len) | |
2833 | { | |
2834 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2835 | ||
28bd726a | 2836 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 2837 | } |
195aefde IE |
2838 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
2839 | ||
8e73485c PB |
2840 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
2841 | const void *data, int offset, int len) | |
2842 | { | |
2843 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2844 | ||
28bd726a | 2845 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
2846 | } |
2847 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
2848 | ||
195aefde IE |
2849 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
2850 | unsigned long len) | |
2851 | { | |
2852 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2853 | int seg; | |
2854 | int offset = offset_in_page(gpa); | |
2855 | int ret; | |
2856 | ||
2857 | while ((seg = next_segment(len, offset)) != 0) { | |
2858 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
2859 | if (ret < 0) | |
2860 | return ret; | |
2861 | offset = 0; | |
2862 | len -= seg; | |
2863 | data += seg; | |
2864 | ++gfn; | |
2865 | } | |
2866 | return 0; | |
2867 | } | |
ff651cb6 | 2868 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 2869 | |
8e73485c PB |
2870 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
2871 | unsigned long len) | |
2872 | { | |
2873 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2874 | int seg; | |
2875 | int offset = offset_in_page(gpa); | |
2876 | int ret; | |
2877 | ||
2878 | while ((seg = next_segment(len, offset)) != 0) { | |
2879 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
2880 | if (ret < 0) | |
2881 | return ret; | |
2882 | offset = 0; | |
2883 | len -= seg; | |
2884 | data += seg; | |
2885 | ++gfn; | |
2886 | } | |
2887 | return 0; | |
2888 | } | |
2889 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
2890 | ||
5a2d4365 PB |
2891 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
2892 | struct gfn_to_hva_cache *ghc, | |
2893 | gpa_t gpa, unsigned long len) | |
49c7754c | 2894 | { |
49c7754c | 2895 | int offset = offset_in_page(gpa); |
8f964525 AH |
2896 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
2897 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
2898 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
2899 | gfn_t nr_pages_avail; | |
49c7754c | 2900 | |
6ad1e29f | 2901 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 2902 | ghc->generation = slots->generation; |
6ad1e29f SC |
2903 | |
2904 | if (start_gfn > end_gfn) { | |
2905 | ghc->hva = KVM_HVA_ERR_BAD; | |
2906 | return -EINVAL; | |
2907 | } | |
f1b9dd5e JM |
2908 | |
2909 | /* | |
2910 | * If the requested region crosses two memslots, we still | |
2911 | * verify that the entire region is valid here. | |
2912 | */ | |
6ad1e29f | 2913 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
2914 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
2915 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
2916 | &nr_pages_avail); | |
2917 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 2918 | return -EFAULT; |
f1b9dd5e JM |
2919 | } |
2920 | ||
2921 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 2922 | if (nr_pages_needed == 1) |
49c7754c | 2923 | ghc->hva += offset; |
f1b9dd5e | 2924 | else |
8f964525 | 2925 | ghc->memslot = NULL; |
f1b9dd5e | 2926 | |
6ad1e29f SC |
2927 | ghc->gpa = gpa; |
2928 | ghc->len = len; | |
2929 | return 0; | |
49c7754c | 2930 | } |
5a2d4365 | 2931 | |
4e335d9e | 2932 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
2933 | gpa_t gpa, unsigned long len) |
2934 | { | |
4e335d9e | 2935 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
2936 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
2937 | } | |
4e335d9e | 2938 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 2939 | |
4e335d9e | 2940 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
2941 | void *data, unsigned int offset, |
2942 | unsigned long len) | |
49c7754c | 2943 | { |
4e335d9e | 2944 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 2945 | int r; |
4ec6e863 | 2946 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 2947 | |
5f25e71e PB |
2948 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
2949 | return -EINVAL; | |
8f964525 | 2950 | |
dc9ce71e SC |
2951 | if (slots->generation != ghc->generation) { |
2952 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
2953 | return -EFAULT; | |
2954 | } | |
8f964525 | 2955 | |
49c7754c GN |
2956 | if (kvm_is_error_hva(ghc->hva)) |
2957 | return -EFAULT; | |
2958 | ||
fcfbc617 SC |
2959 | if (unlikely(!ghc->memslot)) |
2960 | return kvm_write_guest(kvm, gpa, data, len); | |
2961 | ||
4ec6e863 | 2962 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
2963 | if (r) |
2964 | return -EFAULT; | |
28bd726a | 2965 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
2966 | |
2967 | return 0; | |
2968 | } | |
4e335d9e | 2969 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 2970 | |
4e335d9e PB |
2971 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2972 | void *data, unsigned long len) | |
4ec6e863 | 2973 | { |
4e335d9e | 2974 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 2975 | } |
4e335d9e | 2976 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 2977 | |
0958f0ce VK |
2978 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2979 | void *data, unsigned int offset, | |
2980 | unsigned long len) | |
e03b644f | 2981 | { |
4e335d9e | 2982 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 2983 | int r; |
0958f0ce | 2984 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 2985 | |
5f25e71e PB |
2986 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
2987 | return -EINVAL; | |
8f964525 | 2988 | |
dc9ce71e SC |
2989 | if (slots->generation != ghc->generation) { |
2990 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
2991 | return -EFAULT; | |
2992 | } | |
8f964525 | 2993 | |
e03b644f GN |
2994 | if (kvm_is_error_hva(ghc->hva)) |
2995 | return -EFAULT; | |
2996 | ||
fcfbc617 | 2997 | if (unlikely(!ghc->memslot)) |
0958f0ce | 2998 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 2999 | |
0958f0ce | 3000 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
3001 | if (r) |
3002 | return -EFAULT; | |
3003 | ||
3004 | return 0; | |
3005 | } | |
0958f0ce VK |
3006 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
3007 | ||
3008 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
3009 | void *data, unsigned long len) | |
3010 | { | |
3011 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
3012 | } | |
4e335d9e | 3013 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 3014 | |
195aefde IE |
3015 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
3016 | { | |
2f541442 | 3017 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
3018 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3019 | int seg; | |
3020 | int offset = offset_in_page(gpa); | |
3021 | int ret; | |
3022 | ||
bfda0e84 | 3023 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 3024 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
3025 | if (ret < 0) |
3026 | return ret; | |
3027 | offset = 0; | |
3028 | len -= seg; | |
3029 | ++gfn; | |
3030 | } | |
3031 | return 0; | |
3032 | } | |
3033 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
3034 | ||
28bd726a PX |
3035 | void mark_page_dirty_in_slot(struct kvm *kvm, |
3036 | struct kvm_memory_slot *memslot, | |
3037 | gfn_t gfn) | |
6aa8b732 | 3038 | { |
044c59c4 | 3039 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 3040 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 3041 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 3042 | |
fb04a1ed PX |
3043 | if (kvm->dirty_ring_size) |
3044 | kvm_dirty_ring_push(kvm_dirty_ring_get(kvm), | |
3045 | slot, rel_gfn); | |
3046 | else | |
3047 | set_bit_le(rel_gfn, memslot->dirty_bitmap); | |
6aa8b732 AK |
3048 | } |
3049 | } | |
a6a0b05d | 3050 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 3051 | |
49c7754c GN |
3052 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
3053 | { | |
3054 | struct kvm_memory_slot *memslot; | |
3055 | ||
3056 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 3057 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 3058 | } |
2ba9f0d8 | 3059 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 3060 | |
8e73485c PB |
3061 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
3062 | { | |
3063 | struct kvm_memory_slot *memslot; | |
3064 | ||
3065 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 3066 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
3067 | } |
3068 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
3069 | ||
20b7035c JS |
3070 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
3071 | { | |
3072 | if (!vcpu->sigset_active) | |
3073 | return; | |
3074 | ||
3075 | /* | |
3076 | * This does a lockless modification of ->real_blocked, which is fine | |
3077 | * because, only current can change ->real_blocked and all readers of | |
3078 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
3079 | * of ->blocked. | |
3080 | */ | |
3081 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
3082 | } | |
3083 | ||
3084 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
3085 | { | |
3086 | if (!vcpu->sigset_active) | |
3087 | return; | |
3088 | ||
3089 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
3090 | sigemptyset(¤t->real_blocked); | |
3091 | } | |
3092 | ||
aca6ff29 WL |
3093 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
3094 | { | |
dee339b5 | 3095 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 3096 | |
2cbd7824 | 3097 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 3098 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3099 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
3100 | if (!grow) |
3101 | goto out; | |
3102 | ||
dee339b5 NW |
3103 | val *= grow; |
3104 | if (val < grow_start) | |
3105 | val = grow_start; | |
aca6ff29 | 3106 | |
258785ef DM |
3107 | if (val > vcpu->kvm->max_halt_poll_ns) |
3108 | val = vcpu->kvm->max_halt_poll_ns; | |
313f636d | 3109 | |
aca6ff29 | 3110 | vcpu->halt_poll_ns = val; |
7fa08e71 | 3111 | out: |
2cbd7824 | 3112 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3113 | } |
3114 | ||
3115 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
3116 | { | |
ae232ea4 | 3117 | unsigned int old, val, shrink, grow_start; |
aca6ff29 | 3118 | |
2cbd7824 | 3119 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 3120 | shrink = READ_ONCE(halt_poll_ns_shrink); |
ae232ea4 | 3121 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3122 | if (shrink == 0) |
aca6ff29 WL |
3123 | val = 0; |
3124 | else | |
6b6de68c | 3125 | val /= shrink; |
aca6ff29 | 3126 | |
ae232ea4 SS |
3127 | if (val < grow_start) |
3128 | val = 0; | |
3129 | ||
aca6ff29 | 3130 | vcpu->halt_poll_ns = val; |
2cbd7824 | 3131 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3132 | } |
3133 | ||
f7819512 PB |
3134 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
3135 | { | |
50c28f21 JS |
3136 | int ret = -EINTR; |
3137 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
3138 | ||
f7819512 PB |
3139 | if (kvm_arch_vcpu_runnable(vcpu)) { |
3140 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 3141 | goto out; |
f7819512 PB |
3142 | } |
3143 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 3144 | goto out; |
f7819512 | 3145 | if (signal_pending(current)) |
50c28f21 | 3146 | goto out; |
084071d5 MT |
3147 | if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu)) |
3148 | goto out; | |
f7819512 | 3149 | |
50c28f21 JS |
3150 | ret = 0; |
3151 | out: | |
3152 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
3153 | return ret; | |
f7819512 PB |
3154 | } |
3155 | ||
cb953129 DM |
3156 | static inline void |
3157 | update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited) | |
3158 | { | |
3159 | if (waited) | |
0193cc90 | 3160 | vcpu->stat.generic.halt_poll_fail_ns += poll_ns; |
cb953129 | 3161 | else |
0193cc90 | 3162 | vcpu->stat.generic.halt_poll_success_ns += poll_ns; |
cb953129 DM |
3163 | } |
3164 | ||
b6958ce4 ED |
3165 | /* |
3166 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
3167 | */ | |
8776e519 | 3168 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 3169 | { |
cb953129 | 3170 | ktime_t start, cur, poll_end; |
f7819512 | 3171 | bool waited = false; |
aca6ff29 | 3172 | u64 block_ns; |
f7819512 | 3173 | |
07ab0f8d MZ |
3174 | kvm_arch_vcpu_blocking(vcpu); |
3175 | ||
cb953129 | 3176 | start = cur = poll_end = ktime_get(); |
cdd6ad3a | 3177 | if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) { |
19020f8a | 3178 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); |
f95ef0cd | 3179 | |
0193cc90 | 3180 | ++vcpu->stat.generic.halt_attempted_poll; |
f7819512 PB |
3181 | do { |
3182 | /* | |
3183 | * This sets KVM_REQ_UNHALT if an interrupt | |
3184 | * arrives. | |
3185 | */ | |
3186 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
0193cc90 | 3187 | ++vcpu->stat.generic.halt_successful_poll; |
3491caf2 | 3188 | if (!vcpu_valid_wakeup(vcpu)) |
0193cc90 | 3189 | ++vcpu->stat.generic.halt_poll_invalid; |
8ccba534 JZ |
3190 | |
3191 | KVM_STATS_LOG_HIST_UPDATE( | |
3192 | vcpu->stat.generic.halt_poll_success_hist, | |
3193 | ktime_to_ns(ktime_get()) - | |
3194 | ktime_to_ns(start)); | |
f7819512 PB |
3195 | goto out; |
3196 | } | |
74775654 | 3197 | cpu_relax(); |
cb953129 | 3198 | poll_end = cur = ktime_get(); |
6bd5b743 | 3199 | } while (kvm_vcpu_can_poll(cur, stop)); |
8ccba534 JZ |
3200 | |
3201 | KVM_STATS_LOG_HIST_UPDATE( | |
3202 | vcpu->stat.generic.halt_poll_fail_hist, | |
3203 | ktime_to_ns(ktime_get()) - ktime_to_ns(start)); | |
f7819512 | 3204 | } |
e5c239cf | 3205 | |
8ccba534 | 3206 | |
da4ad88c | 3207 | prepare_to_rcuwait(&vcpu->wait); |
e5c239cf | 3208 | for (;;) { |
da4ad88c | 3209 | set_current_state(TASK_INTERRUPTIBLE); |
e5c239cf | 3210 | |
f7819512 | 3211 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
3212 | break; |
3213 | ||
f7819512 | 3214 | waited = true; |
b6958ce4 | 3215 | schedule(); |
b6958ce4 | 3216 | } |
da4ad88c | 3217 | finish_rcuwait(&vcpu->wait); |
f7819512 | 3218 | cur = ktime_get(); |
87bcc5fa JZ |
3219 | if (waited) { |
3220 | vcpu->stat.generic.halt_wait_ns += | |
3221 | ktime_to_ns(cur) - ktime_to_ns(poll_end); | |
8ccba534 JZ |
3222 | KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist, |
3223 | ktime_to_ns(cur) - ktime_to_ns(poll_end)); | |
87bcc5fa | 3224 | } |
f7819512 | 3225 | out: |
07ab0f8d | 3226 | kvm_arch_vcpu_unblocking(vcpu); |
aca6ff29 WL |
3227 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
3228 | ||
cb953129 DM |
3229 | update_halt_poll_stats( |
3230 | vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited); | |
3231 | ||
44551b2f WL |
3232 | if (!kvm_arch_no_poll(vcpu)) { |
3233 | if (!vcpu_valid_wakeup(vcpu)) { | |
aca6ff29 | 3234 | shrink_halt_poll_ns(vcpu); |
acd05785 | 3235 | } else if (vcpu->kvm->max_halt_poll_ns) { |
44551b2f WL |
3236 | if (block_ns <= vcpu->halt_poll_ns) |
3237 | ; | |
3238 | /* we had a long block, shrink polling */ | |
acd05785 DM |
3239 | else if (vcpu->halt_poll_ns && |
3240 | block_ns > vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
3241 | shrink_halt_poll_ns(vcpu); |
3242 | /* we had a short halt and our poll time is too small */ | |
acd05785 DM |
3243 | else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && |
3244 | block_ns < vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
3245 | grow_halt_poll_ns(vcpu); |
3246 | } else { | |
3247 | vcpu->halt_poll_ns = 0; | |
3248 | } | |
3249 | } | |
aca6ff29 | 3250 | |
3491caf2 CB |
3251 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
3252 | kvm_arch_vcpu_block_finish(vcpu); | |
b6958ce4 | 3253 | } |
2ba9f0d8 | 3254 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 3255 | |
178f02ff | 3256 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 3257 | { |
da4ad88c | 3258 | struct rcuwait *waitp; |
b6d33834 | 3259 | |
da4ad88c DB |
3260 | waitp = kvm_arch_vcpu_get_wait(vcpu); |
3261 | if (rcuwait_wake_up(waitp)) { | |
d73eb57b | 3262 | WRITE_ONCE(vcpu->ready, true); |
0193cc90 | 3263 | ++vcpu->stat.generic.halt_wakeup; |
178f02ff | 3264 | return true; |
b6d33834 CD |
3265 | } |
3266 | ||
178f02ff | 3267 | return false; |
dd1a4cc1 RK |
3268 | } |
3269 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
3270 | ||
0266c894 | 3271 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
3272 | /* |
3273 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
3274 | */ | |
3275 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
3276 | { | |
85b64045 | 3277 | int me, cpu; |
dd1a4cc1 | 3278 | |
178f02ff RK |
3279 | if (kvm_vcpu_wake_up(vcpu)) |
3280 | return; | |
3281 | ||
85b64045 SC |
3282 | /* |
3283 | * Note, the vCPU could get migrated to a different pCPU at any point | |
3284 | * after kvm_arch_vcpu_should_kick(), which could result in sending an | |
3285 | * IPI to the previous pCPU. But, that's ok because the purpose of the | |
3286 | * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the | |
3287 | * vCPU also requires it to leave IN_GUEST_MODE. | |
3288 | */ | |
b6d33834 | 3289 | me = get_cpu(); |
85b64045 SC |
3290 | if (kvm_arch_vcpu_should_kick(vcpu)) { |
3291 | cpu = READ_ONCE(vcpu->cpu); | |
3292 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
b6d33834 | 3293 | smp_send_reschedule(cpu); |
85b64045 | 3294 | } |
b6d33834 CD |
3295 | put_cpu(); |
3296 | } | |
a20ed54d | 3297 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 3298 | #endif /* !CONFIG_S390 */ |
b6d33834 | 3299 | |
fa93384f | 3300 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
3301 | { |
3302 | struct pid *pid; | |
3303 | struct task_struct *task = NULL; | |
fa93384f | 3304 | int ret = 0; |
41628d33 KW |
3305 | |
3306 | rcu_read_lock(); | |
3307 | pid = rcu_dereference(target->pid); | |
3308 | if (pid) | |
27fbe64b | 3309 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
3310 | rcu_read_unlock(); |
3311 | if (!task) | |
c45c528e | 3312 | return ret; |
c45c528e | 3313 | ret = yield_to(task, 1); |
41628d33 | 3314 | put_task_struct(task); |
c45c528e R |
3315 | |
3316 | return ret; | |
41628d33 KW |
3317 | } |
3318 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
3319 | ||
06e48c51 R |
3320 | /* |
3321 | * Helper that checks whether a VCPU is eligible for directed yield. | |
3322 | * Most eligible candidate to yield is decided by following heuristics: | |
3323 | * | |
3324 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
3325 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 3326 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
3327 | * |
3328 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
3329 | * chance last time (mostly it has become eligible now since we have probably | |
3330 | * yielded to lockholder in last iteration. This is done by toggling | |
3331 | * @dy_eligible each time a VCPU checked for eligibility.) | |
3332 | * | |
3333 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
3334 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
3335 | * burning. Giving priority for a potential lock-holder increases lock | |
3336 | * progress. | |
3337 | * | |
3338 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
3339 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
3340 | * and continue with next VCPU and so on. | |
3341 | */ | |
7940876e | 3342 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 3343 | { |
4a55dd72 | 3344 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
3345 | bool eligible; |
3346 | ||
3347 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 3348 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
3349 | |
3350 | if (vcpu->spin_loop.in_spin_loop) | |
3351 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
3352 | ||
3353 | return eligible; | |
4a55dd72 SW |
3354 | #else |
3355 | return true; | |
06e48c51 | 3356 | #endif |
4a55dd72 | 3357 | } |
c45c528e | 3358 | |
17e433b5 WL |
3359 | /* |
3360 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
3361 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
3362 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
3363 | */ | |
3364 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
3365 | { | |
3366 | return kvm_arch_vcpu_runnable(vcpu); | |
3367 | } | |
3368 | ||
3369 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
3370 | { | |
3371 | if (kvm_arch_dy_runnable(vcpu)) | |
3372 | return true; | |
3373 | ||
3374 | #ifdef CONFIG_KVM_ASYNC_PF | |
3375 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
3376 | return true; | |
3377 | #endif | |
3378 | ||
3379 | return false; | |
3380 | } | |
3381 | ||
52acd22f WL |
3382 | bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
3383 | { | |
3384 | return false; | |
3385 | } | |
3386 | ||
199b5763 | 3387 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 3388 | { |
217ece61 RR |
3389 | struct kvm *kvm = me->kvm; |
3390 | struct kvm_vcpu *vcpu; | |
3391 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
3392 | int yielded = 0; | |
c45c528e | 3393 | int try = 3; |
217ece61 RR |
3394 | int pass; |
3395 | int i; | |
d255f4f2 | 3396 | |
4c088493 | 3397 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
3398 | /* |
3399 | * We boost the priority of a VCPU that is runnable but not | |
3400 | * currently running, because it got preempted by something | |
3401 | * else and called schedule in __vcpu_run. Hopefully that | |
3402 | * VCPU is holding the lock that we need and will release it. | |
3403 | * We approximate round-robin by starting at the last boosted VCPU. | |
3404 | */ | |
c45c528e | 3405 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 3406 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 3407 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
3408 | i = last_boosted_vcpu; |
3409 | continue; | |
3410 | } else if (pass && i > last_boosted_vcpu) | |
3411 | break; | |
d73eb57b | 3412 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 3413 | continue; |
217ece61 RR |
3414 | if (vcpu == me) |
3415 | continue; | |
da4ad88c DB |
3416 | if (rcuwait_active(&vcpu->wait) && |
3417 | !vcpu_dy_runnable(vcpu)) | |
217ece61 | 3418 | continue; |
046ddeed | 3419 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
52acd22f WL |
3420 | !kvm_arch_dy_has_pending_interrupt(vcpu) && |
3421 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 3422 | continue; |
06e48c51 R |
3423 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
3424 | continue; | |
c45c528e R |
3425 | |
3426 | yielded = kvm_vcpu_yield_to(vcpu); | |
3427 | if (yielded > 0) { | |
217ece61 | 3428 | kvm->last_boosted_vcpu = i; |
217ece61 | 3429 | break; |
c45c528e R |
3430 | } else if (yielded < 0) { |
3431 | try--; | |
3432 | if (!try) | |
3433 | break; | |
217ece61 | 3434 | } |
217ece61 RR |
3435 | } |
3436 | } | |
4c088493 | 3437 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
3438 | |
3439 | /* Ensure vcpu is not eligible during next spinloop */ | |
3440 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
3441 | } |
3442 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
3443 | ||
fb04a1ed PX |
3444 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
3445 | { | |
3446 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
3447 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && | |
3448 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
3449 | kvm->dirty_ring_size / PAGE_SIZE); | |
3450 | #else | |
3451 | return false; | |
3452 | #endif | |
3453 | } | |
3454 | ||
1499fa80 | 3455 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 3456 | { |
11bac800 | 3457 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
3458 | struct page *page; |
3459 | ||
e4a533a4 | 3460 | if (vmf->pgoff == 0) |
039576c0 | 3461 | page = virt_to_page(vcpu->run); |
09566765 | 3462 | #ifdef CONFIG_X86 |
e4a533a4 | 3463 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 3464 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 3465 | #endif |
4b4357e0 | 3466 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3467 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
3468 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 3469 | #endif |
fb04a1ed PX |
3470 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
3471 | page = kvm_dirty_ring_get_page( | |
3472 | &vcpu->dirty_ring, | |
3473 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 3474 | else |
5b1c1493 | 3475 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 3476 | get_page(page); |
e4a533a4 NP |
3477 | vmf->page = page; |
3478 | return 0; | |
9a2bb7f4 AK |
3479 | } |
3480 | ||
f0f37e2f | 3481 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 3482 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
3483 | }; |
3484 | ||
3485 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
3486 | { | |
fb04a1ed | 3487 | struct kvm_vcpu *vcpu = file->private_data; |
11476d27 | 3488 | unsigned long pages = vma_pages(vma); |
fb04a1ed PX |
3489 | |
3490 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
3491 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
3492 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
3493 | return -EINVAL; | |
3494 | ||
9a2bb7f4 AK |
3495 | vma->vm_ops = &kvm_vcpu_vm_ops; |
3496 | return 0; | |
3497 | } | |
3498 | ||
bccf2150 AK |
3499 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
3500 | { | |
3501 | struct kvm_vcpu *vcpu = filp->private_data; | |
3502 | ||
66c0b394 | 3503 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
3504 | return 0; |
3505 | } | |
3506 | ||
3d3aab1b | 3507 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
3508 | .release = kvm_vcpu_release, |
3509 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 3510 | .mmap = kvm_vcpu_mmap, |
6038f373 | 3511 | .llseek = noop_llseek, |
7ddfd3e0 | 3512 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
3513 | }; |
3514 | ||
3515 | /* | |
3516 | * Allocates an inode for the vcpu. | |
3517 | */ | |
3518 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
3519 | { | |
e46b4692 MY |
3520 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
3521 | ||
3522 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
3523 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
3524 | } |
3525 | ||
3e7093d0 | 3526 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 3527 | { |
741cbbae | 3528 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
d56f5136 | 3529 | struct dentry *debugfs_dentry; |
45b5939e | 3530 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 3531 | |
45b5939e | 3532 | if (!debugfs_initialized()) |
3e7093d0 | 3533 | return; |
45b5939e LC |
3534 | |
3535 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
3536 | debugfs_dentry = debugfs_create_dir(dir_name, |
3537 | vcpu->kvm->debugfs_dentry); | |
45b5939e | 3538 | |
d56f5136 | 3539 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
741cbbae | 3540 | #endif |
45b5939e LC |
3541 | } |
3542 | ||
c5ea7660 AK |
3543 | /* |
3544 | * Creates some virtual cpus. Good luck creating more than one. | |
3545 | */ | |
73880c80 | 3546 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
3547 | { |
3548 | int r; | |
e09fefde | 3549 | struct kvm_vcpu *vcpu; |
8bd826d6 | 3550 | struct page *page; |
c5ea7660 | 3551 | |
a1c42dde | 3552 | if (id >= KVM_MAX_VCPU_IDS) |
338c7dba AH |
3553 | return -EINVAL; |
3554 | ||
6c7caebc PB |
3555 | mutex_lock(&kvm->lock); |
3556 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
3557 | mutex_unlock(&kvm->lock); | |
3558 | return -EINVAL; | |
3559 | } | |
3560 | ||
3561 | kvm->created_vcpus++; | |
3562 | mutex_unlock(&kvm->lock); | |
3563 | ||
897cc38e SC |
3564 | r = kvm_arch_vcpu_precreate(kvm, id); |
3565 | if (r) | |
3566 | goto vcpu_decrement; | |
3567 | ||
85f47930 | 3568 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
e529ef66 SC |
3569 | if (!vcpu) { |
3570 | r = -ENOMEM; | |
6c7caebc PB |
3571 | goto vcpu_decrement; |
3572 | } | |
c5ea7660 | 3573 | |
fcd97ad5 | 3574 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 3575 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
3576 | if (!page) { |
3577 | r = -ENOMEM; | |
e529ef66 | 3578 | goto vcpu_free; |
8bd826d6 SC |
3579 | } |
3580 | vcpu->run = page_address(page); | |
3581 | ||
3582 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
3583 | |
3584 | r = kvm_arch_vcpu_create(vcpu); | |
3585 | if (r) | |
8bd826d6 | 3586 | goto vcpu_free_run_page; |
e529ef66 | 3587 | |
fb04a1ed PX |
3588 | if (kvm->dirty_ring_size) { |
3589 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
3590 | id, kvm->dirty_ring_size); | |
3591 | if (r) | |
3592 | goto arch_vcpu_destroy; | |
3593 | } | |
3594 | ||
11ec2804 | 3595 | mutex_lock(&kvm->lock); |
e09fefde DH |
3596 | if (kvm_get_vcpu_by_id(kvm, id)) { |
3597 | r = -EEXIST; | |
3598 | goto unlock_vcpu_destroy; | |
3599 | } | |
73880c80 | 3600 | |
8750e72a | 3601 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
c5b07754 MZ |
3602 | r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT); |
3603 | BUG_ON(r == -EBUSY); | |
3604 | if (r) | |
3605 | goto unlock_vcpu_destroy; | |
c5ea7660 | 3606 | |
ce55c049 JZ |
3607 | /* Fill the stats id string for the vcpu */ |
3608 | snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d", | |
3609 | task_pid_nr(current), id); | |
3610 | ||
fb3f0f51 | 3611 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 3612 | kvm_get_kvm(kvm); |
bccf2150 | 3613 | r = create_vcpu_fd(vcpu); |
73880c80 | 3614 | if (r < 0) { |
c5b07754 | 3615 | xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx); |
149487bd | 3616 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 3617 | goto unlock_vcpu_destroy; |
73880c80 GN |
3618 | } |
3619 | ||
dd489240 | 3620 | /* |
c5b07754 MZ |
3621 | * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu |
3622 | * pointer before kvm->online_vcpu's incremented value. | |
dd489240 | 3623 | */ |
73880c80 GN |
3624 | smp_wmb(); |
3625 | atomic_inc(&kvm->online_vcpus); | |
3626 | ||
73880c80 | 3627 | mutex_unlock(&kvm->lock); |
42897d86 | 3628 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 3629 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 3630 | return r; |
39c3b86e | 3631 | |
d780592b | 3632 | unlock_vcpu_destroy: |
7d8fece6 | 3633 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
3634 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
3635 | arch_vcpu_destroy: | |
d40ccc62 | 3636 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
3637 | vcpu_free_run_page: |
3638 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
3639 | vcpu_free: |
3640 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
3641 | vcpu_decrement: |
3642 | mutex_lock(&kvm->lock); | |
3643 | kvm->created_vcpus--; | |
3644 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
3645 | return r; |
3646 | } | |
3647 | ||
1961d276 AK |
3648 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
3649 | { | |
3650 | if (sigset) { | |
3651 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
3652 | vcpu->sigset_active = 1; | |
3653 | vcpu->sigset = *sigset; | |
3654 | } else | |
3655 | vcpu->sigset_active = 0; | |
3656 | return 0; | |
3657 | } | |
3658 | ||
ce55c049 JZ |
3659 | static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer, |
3660 | size_t size, loff_t *offset) | |
3661 | { | |
3662 | struct kvm_vcpu *vcpu = file->private_data; | |
3663 | ||
3664 | return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header, | |
3665 | &kvm_vcpu_stats_desc[0], &vcpu->stat, | |
3666 | sizeof(vcpu->stat), user_buffer, size, offset); | |
3667 | } | |
3668 | ||
3669 | static const struct file_operations kvm_vcpu_stats_fops = { | |
3670 | .read = kvm_vcpu_stats_read, | |
3671 | .llseek = noop_llseek, | |
3672 | }; | |
3673 | ||
3674 | static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu) | |
3675 | { | |
3676 | int fd; | |
3677 | struct file *file; | |
3678 | char name[15 + ITOA_MAX_LEN + 1]; | |
3679 | ||
3680 | snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id); | |
3681 | ||
3682 | fd = get_unused_fd_flags(O_CLOEXEC); | |
3683 | if (fd < 0) | |
3684 | return fd; | |
3685 | ||
3686 | file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY); | |
3687 | if (IS_ERR(file)) { | |
3688 | put_unused_fd(fd); | |
3689 | return PTR_ERR(file); | |
3690 | } | |
3691 | file->f_mode |= FMODE_PREAD; | |
3692 | fd_install(fd, file); | |
3693 | ||
3694 | return fd; | |
3695 | } | |
3696 | ||
bccf2150 AK |
3697 | static long kvm_vcpu_ioctl(struct file *filp, |
3698 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 3699 | { |
bccf2150 | 3700 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 3701 | void __user *argp = (void __user *)arg; |
313a3dc7 | 3702 | int r; |
fa3795a7 DH |
3703 | struct kvm_fpu *fpu = NULL; |
3704 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 3705 | |
f4d31653 | 3706 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
6d4e4c4f | 3707 | return -EIO; |
2122ff5e | 3708 | |
2ea75be3 DM |
3709 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
3710 | return -EINVAL; | |
3711 | ||
2122ff5e | 3712 | /* |
5cb0944c PB |
3713 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
3714 | * execution; mutex_lock() would break them. | |
2122ff5e | 3715 | */ |
5cb0944c PB |
3716 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
3717 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 3718 | return r; |
2122ff5e | 3719 | |
ec7660cc CD |
3720 | if (mutex_lock_killable(&vcpu->mutex)) |
3721 | return -EINTR; | |
6aa8b732 | 3722 | switch (ioctl) { |
0e4524a5 CB |
3723 | case KVM_RUN: { |
3724 | struct pid *oldpid; | |
f0fe5108 AK |
3725 | r = -EINVAL; |
3726 | if (arg) | |
3727 | goto out; | |
0e4524a5 | 3728 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 3729 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 3730 | /* The thread running this VCPU changed. */ |
bd2a6394 | 3731 | struct pid *newpid; |
f95ef0cd | 3732 | |
bd2a6394 CD |
3733 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
3734 | if (r) | |
3735 | break; | |
3736 | ||
3737 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
3738 | rcu_assign_pointer(vcpu->pid, newpid); |
3739 | if (oldpid) | |
3740 | synchronize_rcu(); | |
3741 | put_pid(oldpid); | |
3742 | } | |
1b94f6f8 | 3743 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 3744 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 3745 | break; |
0e4524a5 | 3746 | } |
6aa8b732 | 3747 | case KVM_GET_REGS: { |
3e4bb3ac | 3748 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3749 | |
3e4bb3ac | 3750 | r = -ENOMEM; |
b12ce36a | 3751 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 3752 | if (!kvm_regs) |
6aa8b732 | 3753 | goto out; |
3e4bb3ac XZ |
3754 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
3755 | if (r) | |
3756 | goto out_free1; | |
6aa8b732 | 3757 | r = -EFAULT; |
3e4bb3ac XZ |
3758 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
3759 | goto out_free1; | |
6aa8b732 | 3760 | r = 0; |
3e4bb3ac XZ |
3761 | out_free1: |
3762 | kfree(kvm_regs); | |
6aa8b732 AK |
3763 | break; |
3764 | } | |
3765 | case KVM_SET_REGS: { | |
3e4bb3ac | 3766 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3767 | |
ff5c2c03 SL |
3768 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
3769 | if (IS_ERR(kvm_regs)) { | |
3770 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 3771 | goto out; |
ff5c2c03 | 3772 | } |
3e4bb3ac | 3773 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 3774 | kfree(kvm_regs); |
6aa8b732 AK |
3775 | break; |
3776 | } | |
3777 | case KVM_GET_SREGS: { | |
b12ce36a BG |
3778 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
3779 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
3780 | r = -ENOMEM; |
3781 | if (!kvm_sregs) | |
3782 | goto out; | |
3783 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
3784 | if (r) |
3785 | goto out; | |
3786 | r = -EFAULT; | |
fa3795a7 | 3787 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
3788 | goto out; |
3789 | r = 0; | |
3790 | break; | |
3791 | } | |
3792 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
3793 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
3794 | if (IS_ERR(kvm_sregs)) { | |
3795 | r = PTR_ERR(kvm_sregs); | |
18595411 | 3796 | kvm_sregs = NULL; |
6aa8b732 | 3797 | goto out; |
ff5c2c03 | 3798 | } |
fa3795a7 | 3799 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
3800 | break; |
3801 | } | |
62d9f0db MT |
3802 | case KVM_GET_MP_STATE: { |
3803 | struct kvm_mp_state mp_state; | |
3804 | ||
3805 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
3806 | if (r) | |
3807 | goto out; | |
3808 | r = -EFAULT; | |
893bdbf1 | 3809 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
3810 | goto out; |
3811 | r = 0; | |
3812 | break; | |
3813 | } | |
3814 | case KVM_SET_MP_STATE: { | |
3815 | struct kvm_mp_state mp_state; | |
3816 | ||
3817 | r = -EFAULT; | |
893bdbf1 | 3818 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
3819 | goto out; |
3820 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
3821 | break; |
3822 | } | |
6aa8b732 AK |
3823 | case KVM_TRANSLATE: { |
3824 | struct kvm_translation tr; | |
3825 | ||
3826 | r = -EFAULT; | |
893bdbf1 | 3827 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 3828 | goto out; |
8b006791 | 3829 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
3830 | if (r) |
3831 | goto out; | |
3832 | r = -EFAULT; | |
893bdbf1 | 3833 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
3834 | goto out; |
3835 | r = 0; | |
3836 | break; | |
3837 | } | |
d0bfb940 JK |
3838 | case KVM_SET_GUEST_DEBUG: { |
3839 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
3840 | |
3841 | r = -EFAULT; | |
893bdbf1 | 3842 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 3843 | goto out; |
d0bfb940 | 3844 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
3845 | break; |
3846 | } | |
1961d276 AK |
3847 | case KVM_SET_SIGNAL_MASK: { |
3848 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3849 | struct kvm_signal_mask kvm_sigmask; | |
3850 | sigset_t sigset, *p; | |
3851 | ||
3852 | p = NULL; | |
3853 | if (argp) { | |
3854 | r = -EFAULT; | |
3855 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3856 | sizeof(kvm_sigmask))) |
1961d276 AK |
3857 | goto out; |
3858 | r = -EINVAL; | |
893bdbf1 | 3859 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
3860 | goto out; |
3861 | r = -EFAULT; | |
3862 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 3863 | sizeof(sigset))) |
1961d276 AK |
3864 | goto out; |
3865 | p = &sigset; | |
3866 | } | |
376d41ff | 3867 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
3868 | break; |
3869 | } | |
b8836737 | 3870 | case KVM_GET_FPU: { |
b12ce36a | 3871 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
3872 | r = -ENOMEM; |
3873 | if (!fpu) | |
3874 | goto out; | |
3875 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
3876 | if (r) |
3877 | goto out; | |
3878 | r = -EFAULT; | |
fa3795a7 | 3879 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
3880 | goto out; |
3881 | r = 0; | |
3882 | break; | |
3883 | } | |
3884 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
3885 | fpu = memdup_user(argp, sizeof(*fpu)); |
3886 | if (IS_ERR(fpu)) { | |
3887 | r = PTR_ERR(fpu); | |
18595411 | 3888 | fpu = NULL; |
b8836737 | 3889 | goto out; |
ff5c2c03 | 3890 | } |
fa3795a7 | 3891 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
3892 | break; |
3893 | } | |
ce55c049 JZ |
3894 | case KVM_GET_STATS_FD: { |
3895 | r = kvm_vcpu_ioctl_get_stats_fd(vcpu); | |
3896 | break; | |
3897 | } | |
bccf2150 | 3898 | default: |
313a3dc7 | 3899 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
3900 | } |
3901 | out: | |
ec7660cc | 3902 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
3903 | kfree(fpu); |
3904 | kfree(kvm_sregs); | |
bccf2150 AK |
3905 | return r; |
3906 | } | |
3907 | ||
de8e5d74 | 3908 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
3909 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
3910 | unsigned int ioctl, unsigned long arg) | |
3911 | { | |
3912 | struct kvm_vcpu *vcpu = filp->private_data; | |
3913 | void __user *argp = compat_ptr(arg); | |
3914 | int r; | |
3915 | ||
f4d31653 | 3916 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
1dda606c AG |
3917 | return -EIO; |
3918 | ||
3919 | switch (ioctl) { | |
3920 | case KVM_SET_SIGNAL_MASK: { | |
3921 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3922 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
3923 | sigset_t sigset; |
3924 | ||
3925 | if (argp) { | |
3926 | r = -EFAULT; | |
3927 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3928 | sizeof(kvm_sigmask))) |
1dda606c AG |
3929 | goto out; |
3930 | r = -EINVAL; | |
3968cf62 | 3931 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
3932 | goto out; |
3933 | r = -EFAULT; | |
1393b4aa PB |
3934 | if (get_compat_sigset(&sigset, |
3935 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 3936 | goto out; |
760a9a30 AC |
3937 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
3938 | } else | |
3939 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
3940 | break; |
3941 | } | |
3942 | default: | |
3943 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
3944 | } | |
3945 | ||
3946 | out: | |
3947 | return r; | |
3948 | } | |
3949 | #endif | |
3950 | ||
a1cd3f08 CLG |
3951 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
3952 | { | |
3953 | struct kvm_device *dev = filp->private_data; | |
3954 | ||
3955 | if (dev->ops->mmap) | |
3956 | return dev->ops->mmap(dev, vma); | |
3957 | ||
3958 | return -ENODEV; | |
3959 | } | |
3960 | ||
852b6d57 SW |
3961 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
3962 | int (*accessor)(struct kvm_device *dev, | |
3963 | struct kvm_device_attr *attr), | |
3964 | unsigned long arg) | |
3965 | { | |
3966 | struct kvm_device_attr attr; | |
3967 | ||
3968 | if (!accessor) | |
3969 | return -EPERM; | |
3970 | ||
3971 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
3972 | return -EFAULT; | |
3973 | ||
3974 | return accessor(dev, &attr); | |
3975 | } | |
3976 | ||
3977 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
3978 | unsigned long arg) | |
3979 | { | |
3980 | struct kvm_device *dev = filp->private_data; | |
3981 | ||
f4d31653 | 3982 | if (dev->kvm->mm != current->mm || dev->kvm->vm_dead) |
ddba9180 SC |
3983 | return -EIO; |
3984 | ||
852b6d57 SW |
3985 | switch (ioctl) { |
3986 | case KVM_SET_DEVICE_ATTR: | |
3987 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
3988 | case KVM_GET_DEVICE_ATTR: | |
3989 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
3990 | case KVM_HAS_DEVICE_ATTR: | |
3991 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
3992 | default: | |
3993 | if (dev->ops->ioctl) | |
3994 | return dev->ops->ioctl(dev, ioctl, arg); | |
3995 | ||
3996 | return -ENOTTY; | |
3997 | } | |
3998 | } | |
3999 | ||
852b6d57 SW |
4000 | static int kvm_device_release(struct inode *inode, struct file *filp) |
4001 | { | |
4002 | struct kvm_device *dev = filp->private_data; | |
4003 | struct kvm *kvm = dev->kvm; | |
4004 | ||
2bde9b3e CLG |
4005 | if (dev->ops->release) { |
4006 | mutex_lock(&kvm->lock); | |
4007 | list_del(&dev->vm_node); | |
4008 | dev->ops->release(dev); | |
4009 | mutex_unlock(&kvm->lock); | |
4010 | } | |
4011 | ||
852b6d57 SW |
4012 | kvm_put_kvm(kvm); |
4013 | return 0; | |
4014 | } | |
4015 | ||
4016 | static const struct file_operations kvm_device_fops = { | |
4017 | .unlocked_ioctl = kvm_device_ioctl, | |
4018 | .release = kvm_device_release, | |
7ddfd3e0 | 4019 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 4020 | .mmap = kvm_device_mmap, |
852b6d57 SW |
4021 | }; |
4022 | ||
4023 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
4024 | { | |
4025 | if (filp->f_op != &kvm_device_fops) | |
4026 | return NULL; | |
4027 | ||
4028 | return filp->private_data; | |
4029 | } | |
4030 | ||
8538cb22 | 4031 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 4032 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
4033 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
4034 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 4035 | #endif |
d60eacb0 WD |
4036 | }; |
4037 | ||
8538cb22 | 4038 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
4039 | { |
4040 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
4041 | return -ENOSPC; | |
4042 | ||
4043 | if (kvm_device_ops_table[type] != NULL) | |
4044 | return -EEXIST; | |
4045 | ||
4046 | kvm_device_ops_table[type] = ops; | |
4047 | return 0; | |
4048 | } | |
4049 | ||
571ee1b6 WL |
4050 | void kvm_unregister_device_ops(u32 type) |
4051 | { | |
4052 | if (kvm_device_ops_table[type] != NULL) | |
4053 | kvm_device_ops_table[type] = NULL; | |
4054 | } | |
4055 | ||
852b6d57 SW |
4056 | static int kvm_ioctl_create_device(struct kvm *kvm, |
4057 | struct kvm_create_device *cd) | |
4058 | { | |
8538cb22 | 4059 | const struct kvm_device_ops *ops = NULL; |
852b6d57 SW |
4060 | struct kvm_device *dev; |
4061 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 4062 | int type; |
852b6d57 SW |
4063 | int ret; |
4064 | ||
d60eacb0 WD |
4065 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
4066 | return -ENODEV; | |
4067 | ||
1d487e9b PB |
4068 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
4069 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 4070 | if (ops == NULL) |
852b6d57 | 4071 | return -ENODEV; |
852b6d57 SW |
4072 | |
4073 | if (test) | |
4074 | return 0; | |
4075 | ||
b12ce36a | 4076 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
4077 | if (!dev) |
4078 | return -ENOMEM; | |
4079 | ||
4080 | dev->ops = ops; | |
4081 | dev->kvm = kvm; | |
852b6d57 | 4082 | |
a28ebea2 | 4083 | mutex_lock(&kvm->lock); |
1d487e9b | 4084 | ret = ops->create(dev, type); |
852b6d57 | 4085 | if (ret < 0) { |
a28ebea2 | 4086 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
4087 | kfree(dev); |
4088 | return ret; | |
4089 | } | |
a28ebea2 CD |
4090 | list_add(&dev->vm_node, &kvm->devices); |
4091 | mutex_unlock(&kvm->lock); | |
852b6d57 | 4092 | |
023e9fdd CD |
4093 | if (ops->init) |
4094 | ops->init(dev); | |
4095 | ||
cfa39381 | 4096 | kvm_get_kvm(kvm); |
24009b05 | 4097 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 4098 | if (ret < 0) { |
149487bd | 4099 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
4100 | mutex_lock(&kvm->lock); |
4101 | list_del(&dev->vm_node); | |
4102 | mutex_unlock(&kvm->lock); | |
a0f1d21c | 4103 | ops->destroy(dev); |
852b6d57 SW |
4104 | return ret; |
4105 | } | |
4106 | ||
852b6d57 SW |
4107 | cd->fd = ret; |
4108 | return 0; | |
4109 | } | |
4110 | ||
92b591a4 AG |
4111 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
4112 | { | |
4113 | switch (arg) { | |
4114 | case KVM_CAP_USER_MEMORY: | |
4115 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
4116 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
4117 | case KVM_CAP_INTERNAL_ERROR_DATA: |
4118 | #ifdef CONFIG_HAVE_KVM_MSI | |
4119 | case KVM_CAP_SIGNAL_MSI: | |
4120 | #endif | |
297e2105 | 4121 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 4122 | case KVM_CAP_IRQFD: |
92b591a4 AG |
4123 | case KVM_CAP_IRQFD_RESAMPLE: |
4124 | #endif | |
e9ea5069 | 4125 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 4126 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 4127 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 4128 | case KVM_CAP_HALT_POLL: |
92b591a4 | 4129 | return 1; |
4b4357e0 | 4130 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
4131 | case KVM_CAP_COALESCED_MMIO: |
4132 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
4133 | case KVM_CAP_COALESCED_PIO: |
4134 | return 1; | |
30422558 | 4135 | #endif |
3c9bd400 JZ |
4136 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4137 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
4138 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
4139 | #endif | |
92b591a4 AG |
4140 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4141 | case KVM_CAP_IRQ_ROUTING: | |
4142 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
4143 | #endif |
4144 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
4145 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
4146 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 4147 | #endif |
c110ae57 PB |
4148 | case KVM_CAP_NR_MEMSLOTS: |
4149 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed PX |
4150 | case KVM_CAP_DIRTY_LOG_RING: |
4151 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
4152 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); | |
4153 | #else | |
4154 | return 0; | |
4155 | #endif | |
ce55c049 JZ |
4156 | case KVM_CAP_BINARY_STATS_FD: |
4157 | return 1; | |
92b591a4 AG |
4158 | default: |
4159 | break; | |
4160 | } | |
4161 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
4162 | } | |
4163 | ||
fb04a1ed PX |
4164 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
4165 | { | |
4166 | int r; | |
4167 | ||
4168 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
4169 | return -EINVAL; | |
4170 | ||
4171 | /* the size should be power of 2 */ | |
4172 | if (!size || (size & (size - 1))) | |
4173 | return -EINVAL; | |
4174 | ||
4175 | /* Should be bigger to keep the reserved entries, or a page */ | |
4176 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
4177 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
4178 | return -EINVAL; | |
4179 | ||
4180 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
4181 | sizeof(struct kvm_dirty_gfn)) | |
4182 | return -E2BIG; | |
4183 | ||
4184 | /* We only allow it to set once */ | |
4185 | if (kvm->dirty_ring_size) | |
4186 | return -EINVAL; | |
4187 | ||
4188 | mutex_lock(&kvm->lock); | |
4189 | ||
4190 | if (kvm->created_vcpus) { | |
4191 | /* We don't allow to change this value after vcpu created */ | |
4192 | r = -EINVAL; | |
4193 | } else { | |
4194 | kvm->dirty_ring_size = size; | |
4195 | r = 0; | |
4196 | } | |
4197 | ||
4198 | mutex_unlock(&kvm->lock); | |
4199 | return r; | |
4200 | } | |
4201 | ||
4202 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
4203 | { | |
4204 | int i; | |
4205 | struct kvm_vcpu *vcpu; | |
4206 | int cleared = 0; | |
4207 | ||
4208 | if (!kvm->dirty_ring_size) | |
4209 | return -EINVAL; | |
4210 | ||
4211 | mutex_lock(&kvm->slots_lock); | |
4212 | ||
4213 | kvm_for_each_vcpu(i, vcpu, kvm) | |
4214 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
4215 | ||
4216 | mutex_unlock(&kvm->slots_lock); | |
4217 | ||
4218 | if (cleared) | |
4219 | kvm_flush_remote_tlbs(kvm); | |
4220 | ||
4221 | return cleared; | |
4222 | } | |
4223 | ||
e5d83c74 PB |
4224 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4225 | struct kvm_enable_cap *cap) | |
4226 | { | |
4227 | return -EINVAL; | |
4228 | } | |
4229 | ||
4230 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
4231 | struct kvm_enable_cap *cap) | |
4232 | { | |
4233 | switch (cap->cap) { | |
2a31b9db | 4234 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
4235 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
4236 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
4237 | ||
4238 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
4239 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
4240 | ||
4241 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
4242 | return -EINVAL; |
4243 | kvm->manual_dirty_log_protect = cap->args[0]; | |
4244 | return 0; | |
3c9bd400 | 4245 | } |
2a31b9db | 4246 | #endif |
acd05785 DM |
4247 | case KVM_CAP_HALT_POLL: { |
4248 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
4249 | return -EINVAL; | |
4250 | ||
4251 | kvm->max_halt_poll_ns = cap->args[0]; | |
4252 | return 0; | |
4253 | } | |
fb04a1ed PX |
4254 | case KVM_CAP_DIRTY_LOG_RING: |
4255 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); | |
e5d83c74 PB |
4256 | default: |
4257 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
4258 | } | |
4259 | } | |
4260 | ||
fcfe1bae JZ |
4261 | static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer, |
4262 | size_t size, loff_t *offset) | |
4263 | { | |
4264 | struct kvm *kvm = file->private_data; | |
4265 | ||
4266 | return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header, | |
4267 | &kvm_vm_stats_desc[0], &kvm->stat, | |
4268 | sizeof(kvm->stat), user_buffer, size, offset); | |
4269 | } | |
4270 | ||
4271 | static const struct file_operations kvm_vm_stats_fops = { | |
4272 | .read = kvm_vm_stats_read, | |
4273 | .llseek = noop_llseek, | |
4274 | }; | |
4275 | ||
4276 | static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm) | |
4277 | { | |
4278 | int fd; | |
4279 | struct file *file; | |
4280 | ||
4281 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4282 | if (fd < 0) | |
4283 | return fd; | |
4284 | ||
4285 | file = anon_inode_getfile("kvm-vm-stats", | |
4286 | &kvm_vm_stats_fops, kvm, O_RDONLY); | |
4287 | if (IS_ERR(file)) { | |
4288 | put_unused_fd(fd); | |
4289 | return PTR_ERR(file); | |
4290 | } | |
4291 | file->f_mode |= FMODE_PREAD; | |
4292 | fd_install(fd, file); | |
4293 | ||
4294 | return fd; | |
4295 | } | |
4296 | ||
bccf2150 AK |
4297 | static long kvm_vm_ioctl(struct file *filp, |
4298 | unsigned int ioctl, unsigned long arg) | |
4299 | { | |
4300 | struct kvm *kvm = filp->private_data; | |
4301 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 4302 | int r; |
bccf2150 | 4303 | |
f4d31653 | 4304 | if (kvm->mm != current->mm || kvm->vm_dead) |
6d4e4c4f | 4305 | return -EIO; |
bccf2150 AK |
4306 | switch (ioctl) { |
4307 | case KVM_CREATE_VCPU: | |
4308 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 4309 | break; |
e5d83c74 PB |
4310 | case KVM_ENABLE_CAP: { |
4311 | struct kvm_enable_cap cap; | |
4312 | ||
4313 | r = -EFAULT; | |
4314 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4315 | goto out; | |
4316 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
4317 | break; | |
4318 | } | |
6fc138d2 IE |
4319 | case KVM_SET_USER_MEMORY_REGION: { |
4320 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
4321 | ||
4322 | r = -EFAULT; | |
4323 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 4324 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
4325 | goto out; |
4326 | ||
47ae31e2 | 4327 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
4328 | break; |
4329 | } | |
4330 | case KVM_GET_DIRTY_LOG: { | |
4331 | struct kvm_dirty_log log; | |
4332 | ||
4333 | r = -EFAULT; | |
893bdbf1 | 4334 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 4335 | goto out; |
2c6f5df9 | 4336 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
4337 | break; |
4338 | } | |
2a31b9db PB |
4339 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4340 | case KVM_CLEAR_DIRTY_LOG: { | |
4341 | struct kvm_clear_dirty_log log; | |
4342 | ||
4343 | r = -EFAULT; | |
4344 | if (copy_from_user(&log, argp, sizeof(log))) | |
4345 | goto out; | |
4346 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4347 | break; | |
4348 | } | |
4349 | #endif | |
4b4357e0 | 4350 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
4351 | case KVM_REGISTER_COALESCED_MMIO: { |
4352 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4353 | |
5f94c174 | 4354 | r = -EFAULT; |
893bdbf1 | 4355 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4356 | goto out; |
5f94c174 | 4357 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4358 | break; |
4359 | } | |
4360 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
4361 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4362 | |
5f94c174 | 4363 | r = -EFAULT; |
893bdbf1 | 4364 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4365 | goto out; |
5f94c174 | 4366 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4367 | break; |
4368 | } | |
4369 | #endif | |
721eecbf GH |
4370 | case KVM_IRQFD: { |
4371 | struct kvm_irqfd data; | |
4372 | ||
4373 | r = -EFAULT; | |
893bdbf1 | 4374 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 4375 | goto out; |
d4db2935 | 4376 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
4377 | break; |
4378 | } | |
d34e6b17 GH |
4379 | case KVM_IOEVENTFD: { |
4380 | struct kvm_ioeventfd data; | |
4381 | ||
4382 | r = -EFAULT; | |
893bdbf1 | 4383 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
4384 | goto out; |
4385 | r = kvm_ioeventfd(kvm, &data); | |
4386 | break; | |
4387 | } | |
07975ad3 JK |
4388 | #ifdef CONFIG_HAVE_KVM_MSI |
4389 | case KVM_SIGNAL_MSI: { | |
4390 | struct kvm_msi msi; | |
4391 | ||
4392 | r = -EFAULT; | |
893bdbf1 | 4393 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
4394 | goto out; |
4395 | r = kvm_send_userspace_msi(kvm, &msi); | |
4396 | break; | |
4397 | } | |
23d43cf9 CD |
4398 | #endif |
4399 | #ifdef __KVM_HAVE_IRQ_LINE | |
4400 | case KVM_IRQ_LINE_STATUS: | |
4401 | case KVM_IRQ_LINE: { | |
4402 | struct kvm_irq_level irq_event; | |
4403 | ||
4404 | r = -EFAULT; | |
893bdbf1 | 4405 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
4406 | goto out; |
4407 | ||
aa2fbe6d YZ |
4408 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
4409 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
4410 | if (r) |
4411 | goto out; | |
4412 | ||
4413 | r = -EFAULT; | |
4414 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 4415 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
4416 | goto out; |
4417 | } | |
4418 | ||
4419 | r = 0; | |
4420 | break; | |
4421 | } | |
73880c80 | 4422 | #endif |
aa8d5944 AG |
4423 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4424 | case KVM_SET_GSI_ROUTING: { | |
4425 | struct kvm_irq_routing routing; | |
4426 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 4427 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
4428 | |
4429 | r = -EFAULT; | |
4430 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
4431 | goto out; | |
4432 | r = -EINVAL; | |
5c0aea0e DH |
4433 | if (!kvm_arch_can_set_irq_routing(kvm)) |
4434 | goto out; | |
caf1ff26 | 4435 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
4436 | goto out; |
4437 | if (routing.flags) | |
4438 | goto out; | |
f8c1b85b | 4439 | if (routing.nr) { |
f8c1b85b | 4440 | urouting = argp; |
7ec28e26 DE |
4441 | entries = vmemdup_user(urouting->entries, |
4442 | array_size(sizeof(*entries), | |
4443 | routing.nr)); | |
4444 | if (IS_ERR(entries)) { | |
4445 | r = PTR_ERR(entries); | |
4446 | goto out; | |
4447 | } | |
f8c1b85b | 4448 | } |
aa8d5944 AG |
4449 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
4450 | routing.flags); | |
7ec28e26 | 4451 | kvfree(entries); |
aa8d5944 AG |
4452 | break; |
4453 | } | |
4454 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
4455 | case KVM_CREATE_DEVICE: { |
4456 | struct kvm_create_device cd; | |
4457 | ||
4458 | r = -EFAULT; | |
4459 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
4460 | goto out; | |
4461 | ||
4462 | r = kvm_ioctl_create_device(kvm, &cd); | |
4463 | if (r) | |
4464 | goto out; | |
4465 | ||
4466 | r = -EFAULT; | |
4467 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
4468 | goto out; | |
4469 | ||
4470 | r = 0; | |
4471 | break; | |
4472 | } | |
92b591a4 AG |
4473 | case KVM_CHECK_EXTENSION: |
4474 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
4475 | break; | |
fb04a1ed PX |
4476 | case KVM_RESET_DIRTY_RINGS: |
4477 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
4478 | break; | |
fcfe1bae JZ |
4479 | case KVM_GET_STATS_FD: |
4480 | r = kvm_vm_ioctl_get_stats_fd(kvm); | |
4481 | break; | |
f17abe9a | 4482 | default: |
1fe779f8 | 4483 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
4484 | } |
4485 | out: | |
4486 | return r; | |
4487 | } | |
4488 | ||
de8e5d74 | 4489 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
4490 | struct compat_kvm_dirty_log { |
4491 | __u32 slot; | |
4492 | __u32 padding1; | |
4493 | union { | |
4494 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4495 | __u64 padding2; | |
4496 | }; | |
4497 | }; | |
4498 | ||
8750f9bb PB |
4499 | struct compat_kvm_clear_dirty_log { |
4500 | __u32 slot; | |
4501 | __u32 num_pages; | |
4502 | __u64 first_page; | |
4503 | union { | |
4504 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4505 | __u64 padding2; | |
4506 | }; | |
4507 | }; | |
4508 | ||
6ff5894c AB |
4509 | static long kvm_vm_compat_ioctl(struct file *filp, |
4510 | unsigned int ioctl, unsigned long arg) | |
4511 | { | |
4512 | struct kvm *kvm = filp->private_data; | |
4513 | int r; | |
4514 | ||
f4d31653 | 4515 | if (kvm->mm != current->mm || kvm->vm_dead) |
6ff5894c AB |
4516 | return -EIO; |
4517 | switch (ioctl) { | |
8750f9bb PB |
4518 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4519 | case KVM_CLEAR_DIRTY_LOG: { | |
4520 | struct compat_kvm_clear_dirty_log compat_log; | |
4521 | struct kvm_clear_dirty_log log; | |
4522 | ||
4523 | if (copy_from_user(&compat_log, (void __user *)arg, | |
4524 | sizeof(compat_log))) | |
4525 | return -EFAULT; | |
4526 | log.slot = compat_log.slot; | |
4527 | log.num_pages = compat_log.num_pages; | |
4528 | log.first_page = compat_log.first_page; | |
4529 | log.padding2 = compat_log.padding2; | |
4530 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4531 | ||
4532 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4533 | break; | |
4534 | } | |
4535 | #endif | |
6ff5894c AB |
4536 | case KVM_GET_DIRTY_LOG: { |
4537 | struct compat_kvm_dirty_log compat_log; | |
4538 | struct kvm_dirty_log log; | |
4539 | ||
6ff5894c AB |
4540 | if (copy_from_user(&compat_log, (void __user *)arg, |
4541 | sizeof(compat_log))) | |
f6a3b168 | 4542 | return -EFAULT; |
6ff5894c AB |
4543 | log.slot = compat_log.slot; |
4544 | log.padding1 = compat_log.padding1; | |
4545 | log.padding2 = compat_log.padding2; | |
4546 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4547 | ||
4548 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
4549 | break; |
4550 | } | |
4551 | default: | |
4552 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
4553 | } | |
6ff5894c AB |
4554 | return r; |
4555 | } | |
4556 | #endif | |
4557 | ||
3d3aab1b | 4558 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
4559 | .release = kvm_vm_release, |
4560 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 4561 | .llseek = noop_llseek, |
7ddfd3e0 | 4562 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
4563 | }; |
4564 | ||
54526d1f NT |
4565 | bool file_is_kvm(struct file *file) |
4566 | { | |
4567 | return file && file->f_op == &kvm_vm_fops; | |
4568 | } | |
4569 | EXPORT_SYMBOL_GPL(file_is_kvm); | |
4570 | ||
e08b9637 | 4571 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 4572 | { |
aac87636 | 4573 | int r; |
f17abe9a | 4574 | struct kvm *kvm; |
506cfba9 | 4575 | struct file *file; |
f17abe9a | 4576 | |
e08b9637 | 4577 | kvm = kvm_create_vm(type); |
d6d28168 AK |
4578 | if (IS_ERR(kvm)) |
4579 | return PTR_ERR(kvm); | |
4b4357e0 | 4580 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 4581 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
4582 | if (r < 0) |
4583 | goto put_kvm; | |
6ce5a090 | 4584 | #endif |
506cfba9 | 4585 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
4586 | if (r < 0) |
4587 | goto put_kvm; | |
4588 | ||
fcfe1bae JZ |
4589 | snprintf(kvm->stats_id, sizeof(kvm->stats_id), |
4590 | "kvm-%d", task_pid_nr(current)); | |
4591 | ||
506cfba9 AV |
4592 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
4593 | if (IS_ERR(file)) { | |
4594 | put_unused_fd(r); | |
78588335 ME |
4595 | r = PTR_ERR(file); |
4596 | goto put_kvm; | |
506cfba9 | 4597 | } |
536a6f88 | 4598 | |
525df861 PB |
4599 | /* |
4600 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
4601 | * already set, with ->release() being kvm_vm_release(). In error | |
4602 | * cases it will be called by the final fput(file) and will take | |
4603 | * care of doing kvm_put_kvm(kvm). | |
4604 | */ | |
536a6f88 | 4605 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
4606 | put_unused_fd(r); |
4607 | fput(file); | |
536a6f88 JF |
4608 | return -ENOMEM; |
4609 | } | |
286de8f6 | 4610 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 4611 | |
506cfba9 | 4612 | fd_install(r, file); |
aac87636 | 4613 | return r; |
78588335 ME |
4614 | |
4615 | put_kvm: | |
4616 | kvm_put_kvm(kvm); | |
4617 | return r; | |
f17abe9a AK |
4618 | } |
4619 | ||
4620 | static long kvm_dev_ioctl(struct file *filp, | |
4621 | unsigned int ioctl, unsigned long arg) | |
4622 | { | |
07c45a36 | 4623 | long r = -EINVAL; |
f17abe9a AK |
4624 | |
4625 | switch (ioctl) { | |
4626 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
4627 | if (arg) |
4628 | goto out; | |
f17abe9a AK |
4629 | r = KVM_API_VERSION; |
4630 | break; | |
4631 | case KVM_CREATE_VM: | |
e08b9637 | 4632 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 4633 | break; |
018d00d2 | 4634 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 4635 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 4636 | break; |
07c45a36 | 4637 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
4638 | if (arg) |
4639 | goto out; | |
adb1ff46 AK |
4640 | r = PAGE_SIZE; /* struct kvm_run */ |
4641 | #ifdef CONFIG_X86 | |
4642 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 4643 | #endif |
4b4357e0 | 4644 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 4645 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 4646 | #endif |
07c45a36 | 4647 | break; |
d4c9ff2d FEL |
4648 | case KVM_TRACE_ENABLE: |
4649 | case KVM_TRACE_PAUSE: | |
4650 | case KVM_TRACE_DISABLE: | |
2023a29c | 4651 | r = -EOPNOTSUPP; |
d4c9ff2d | 4652 | break; |
6aa8b732 | 4653 | default: |
043405e1 | 4654 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
4655 | } |
4656 | out: | |
4657 | return r; | |
4658 | } | |
4659 | ||
6aa8b732 | 4660 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 4661 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 4662 | .llseek = noop_llseek, |
7ddfd3e0 | 4663 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
4664 | }; |
4665 | ||
4666 | static struct miscdevice kvm_dev = { | |
bbe4432e | 4667 | KVM_MINOR, |
6aa8b732 AK |
4668 | "kvm", |
4669 | &kvm_chardev_ops, | |
4670 | }; | |
4671 | ||
75b7127c | 4672 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
4673 | { |
4674 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 4675 | int r; |
1b6c0168 | 4676 | |
7f59f492 | 4677 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4678 | return; |
10474ae8 | 4679 | |
7f59f492 | 4680 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 4681 | |
13a34e06 | 4682 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
4683 | |
4684 | if (r) { | |
4685 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
4686 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 4687 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 4688 | } |
1b6c0168 AK |
4689 | } |
4690 | ||
8c18b2d2 | 4691 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 4692 | { |
4a937f96 | 4693 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4694 | if (kvm_usage_count) |
4695 | hardware_enable_nolock(NULL); | |
4a937f96 | 4696 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4697 | return 0; |
75b7127c TY |
4698 | } |
4699 | ||
4700 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
4701 | { |
4702 | int cpu = raw_smp_processor_id(); | |
4703 | ||
7f59f492 | 4704 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4705 | return; |
7f59f492 | 4706 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 4707 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
4708 | } |
4709 | ||
8c18b2d2 | 4710 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 4711 | { |
4a937f96 | 4712 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4713 | if (kvm_usage_count) |
4714 | hardware_disable_nolock(NULL); | |
4a937f96 | 4715 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4716 | return 0; |
75b7127c TY |
4717 | } |
4718 | ||
10474ae8 AG |
4719 | static void hardware_disable_all_nolock(void) |
4720 | { | |
4721 | BUG_ON(!kvm_usage_count); | |
4722 | ||
4723 | kvm_usage_count--; | |
4724 | if (!kvm_usage_count) | |
75b7127c | 4725 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
4726 | } |
4727 | ||
4728 | static void hardware_disable_all(void) | |
4729 | { | |
4a937f96 | 4730 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 4731 | hardware_disable_all_nolock(); |
4a937f96 | 4732 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4733 | } |
4734 | ||
4735 | static int hardware_enable_all(void) | |
4736 | { | |
4737 | int r = 0; | |
4738 | ||
4a937f96 | 4739 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
4740 | |
4741 | kvm_usage_count++; | |
4742 | if (kvm_usage_count == 1) { | |
4743 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 4744 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
4745 | |
4746 | if (atomic_read(&hardware_enable_failed)) { | |
4747 | hardware_disable_all_nolock(); | |
4748 | r = -EBUSY; | |
4749 | } | |
4750 | } | |
4751 | ||
4a937f96 | 4752 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4753 | |
4754 | return r; | |
4755 | } | |
4756 | ||
9a2b85c6 | 4757 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 4758 | void *v) |
9a2b85c6 | 4759 | { |
8e1c1815 SY |
4760 | /* |
4761 | * Some (well, at least mine) BIOSes hang on reboot if | |
4762 | * in vmx root mode. | |
4763 | * | |
4764 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
4765 | */ | |
1170adc6 | 4766 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 4767 | kvm_rebooting = true; |
75b7127c | 4768 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
4769 | return NOTIFY_OK; |
4770 | } | |
4771 | ||
4772 | static struct notifier_block kvm_reboot_notifier = { | |
4773 | .notifier_call = kvm_reboot, | |
4774 | .priority = 0, | |
4775 | }; | |
4776 | ||
e93f8a0f | 4777 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
4778 | { |
4779 | int i; | |
4780 | ||
4781 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 4782 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
4783 | |
4784 | kvm_iodevice_destructor(pos); | |
4785 | } | |
e93f8a0f | 4786 | kfree(bus); |
2eeb2e94 GH |
4787 | } |
4788 | ||
c21fbff1 | 4789 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 4790 | const struct kvm_io_range *r2) |
743eeb0b | 4791 | { |
8f4216c7 JW |
4792 | gpa_t addr1 = r1->addr; |
4793 | gpa_t addr2 = r2->addr; | |
4794 | ||
4795 | if (addr1 < addr2) | |
743eeb0b | 4796 | return -1; |
8f4216c7 JW |
4797 | |
4798 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
4799 | * accept any overlapping write. Any order is acceptable for | |
4800 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
4801 | * we process all of them. | |
4802 | */ | |
4803 | if (r2->len) { | |
4804 | addr1 += r1->len; | |
4805 | addr2 += r2->len; | |
4806 | } | |
4807 | ||
4808 | if (addr1 > addr2) | |
743eeb0b | 4809 | return 1; |
8f4216c7 | 4810 | |
743eeb0b SL |
4811 | return 0; |
4812 | } | |
4813 | ||
a343c9b7 PB |
4814 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
4815 | { | |
c21fbff1 | 4816 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
4817 | } |
4818 | ||
39369f7a | 4819 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
4820 | gpa_t addr, int len) |
4821 | { | |
4822 | struct kvm_io_range *range, key; | |
4823 | int off; | |
4824 | ||
4825 | key = (struct kvm_io_range) { | |
4826 | .addr = addr, | |
4827 | .len = len, | |
4828 | }; | |
4829 | ||
4830 | range = bsearch(&key, bus->range, bus->dev_count, | |
4831 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
4832 | if (range == NULL) | |
4833 | return -ENOENT; | |
4834 | ||
4835 | off = range - bus->range; | |
4836 | ||
c21fbff1 | 4837 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
4838 | off--; |
4839 | ||
4840 | return off; | |
4841 | } | |
4842 | ||
e32edf4f | 4843 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
4844 | struct kvm_io_range *range, const void *val) |
4845 | { | |
4846 | int idx; | |
4847 | ||
4848 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
4849 | if (idx < 0) | |
4850 | return -EOPNOTSUPP; | |
4851 | ||
4852 | while (idx < bus->dev_count && | |
c21fbff1 | 4853 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4854 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4855 | range->len, val)) |
4856 | return idx; | |
4857 | idx++; | |
4858 | } | |
4859 | ||
4860 | return -EOPNOTSUPP; | |
4861 | } | |
4862 | ||
bda9020e | 4863 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 4864 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 4865 | int len, const void *val) |
2eeb2e94 | 4866 | { |
90d83dc3 | 4867 | struct kvm_io_bus *bus; |
743eeb0b | 4868 | struct kvm_io_range range; |
126a5af5 | 4869 | int r; |
743eeb0b SL |
4870 | |
4871 | range = (struct kvm_io_range) { | |
4872 | .addr = addr, | |
4873 | .len = len, | |
4874 | }; | |
90d83dc3 | 4875 | |
e32edf4f | 4876 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4877 | if (!bus) |
4878 | return -ENOMEM; | |
e32edf4f | 4879 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4880 | return r < 0 ? r : 0; |
4881 | } | |
a2420107 | 4882 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
4883 | |
4884 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
4885 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
4886 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
4887 | { |
4888 | struct kvm_io_bus *bus; | |
4889 | struct kvm_io_range range; | |
4890 | ||
4891 | range = (struct kvm_io_range) { | |
4892 | .addr = addr, | |
4893 | .len = len, | |
4894 | }; | |
4895 | ||
e32edf4f | 4896 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4897 | if (!bus) |
4898 | return -ENOMEM; | |
126a5af5 CH |
4899 | |
4900 | /* First try the device referenced by cookie. */ | |
4901 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 4902 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 4903 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
4904 | val)) |
4905 | return cookie; | |
4906 | ||
4907 | /* | |
4908 | * cookie contained garbage; fall back to search and return the | |
4909 | * correct cookie value. | |
4910 | */ | |
e32edf4f | 4911 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4912 | } |
4913 | ||
e32edf4f NN |
4914 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
4915 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
4916 | { |
4917 | int idx; | |
4918 | ||
4919 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
4920 | if (idx < 0) |
4921 | return -EOPNOTSUPP; | |
4922 | ||
4923 | while (idx < bus->dev_count && | |
c21fbff1 | 4924 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4925 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4926 | range->len, val)) |
4927 | return idx; | |
743eeb0b SL |
4928 | idx++; |
4929 | } | |
4930 | ||
bda9020e MT |
4931 | return -EOPNOTSUPP; |
4932 | } | |
2eeb2e94 | 4933 | |
bda9020e | 4934 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 4935 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 4936 | int len, void *val) |
bda9020e | 4937 | { |
90d83dc3 | 4938 | struct kvm_io_bus *bus; |
743eeb0b | 4939 | struct kvm_io_range range; |
126a5af5 | 4940 | int r; |
743eeb0b SL |
4941 | |
4942 | range = (struct kvm_io_range) { | |
4943 | .addr = addr, | |
4944 | .len = len, | |
4945 | }; | |
e93f8a0f | 4946 | |
e32edf4f | 4947 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4948 | if (!bus) |
4949 | return -ENOMEM; | |
e32edf4f | 4950 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
4951 | return r < 0 ? r : 0; |
4952 | } | |
743eeb0b | 4953 | |
79fac95e | 4954 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
4955 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
4956 | int len, struct kvm_io_device *dev) | |
6c474694 | 4957 | { |
d4c67a7a | 4958 | int i; |
e93f8a0f | 4959 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 4960 | struct kvm_io_range range; |
090b7aff | 4961 | |
4a12f951 | 4962 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
4963 | if (!bus) |
4964 | return -ENOMEM; | |
4965 | ||
6ea34c9b AK |
4966 | /* exclude ioeventfd which is limited by maximum fd */ |
4967 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 4968 | return -ENOSPC; |
2eeb2e94 | 4969 | |
90952cd3 | 4970 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 4971 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
4972 | if (!new_bus) |
4973 | return -ENOMEM; | |
d4c67a7a GH |
4974 | |
4975 | range = (struct kvm_io_range) { | |
4976 | .addr = addr, | |
4977 | .len = len, | |
4978 | .dev = dev, | |
4979 | }; | |
4980 | ||
4981 | for (i = 0; i < bus->dev_count; i++) | |
4982 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
4983 | break; | |
4984 | ||
4985 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
4986 | new_bus->dev_count++; | |
4987 | new_bus->range[i] = range; | |
4988 | memcpy(new_bus->range + i + 1, bus->range + i, | |
4989 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
4990 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
4991 | synchronize_srcu_expedited(&kvm->srcu); | |
4992 | kfree(bus); | |
090b7aff GH |
4993 | |
4994 | return 0; | |
4995 | } | |
4996 | ||
5d3c4c79 SC |
4997 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
4998 | struct kvm_io_device *dev) | |
090b7aff | 4999 | { |
f6588660 | 5000 | int i, j; |
e93f8a0f | 5001 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 5002 | |
7c896d37 SC |
5003 | lockdep_assert_held(&kvm->slots_lock); |
5004 | ||
4a12f951 | 5005 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 5006 | if (!bus) |
5d3c4c79 | 5007 | return 0; |
df630b8c | 5008 | |
7c896d37 | 5009 | for (i = 0; i < bus->dev_count; i++) { |
a1300716 | 5010 | if (bus->range[i].dev == dev) { |
090b7aff GH |
5011 | break; |
5012 | } | |
7c896d37 | 5013 | } |
e93f8a0f | 5014 | |
90db1043 | 5015 | if (i == bus->dev_count) |
5d3c4c79 | 5016 | return 0; |
a1300716 | 5017 | |
90952cd3 | 5018 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 5019 | GFP_KERNEL_ACCOUNT); |
f6588660 | 5020 | if (new_bus) { |
871c433b | 5021 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
5022 | new_bus->dev_count--; |
5023 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 5024 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
2ee37574 SC |
5025 | } |
5026 | ||
5027 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
5028 | synchronize_srcu_expedited(&kvm->srcu); | |
5029 | ||
5030 | /* Destroy the old bus _after_ installing the (null) bus. */ | |
5031 | if (!new_bus) { | |
90db1043 | 5032 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
f6588660 RK |
5033 | for (j = 0; j < bus->dev_count; j++) { |
5034 | if (j == i) | |
5035 | continue; | |
5036 | kvm_iodevice_destructor(bus->range[j].dev); | |
5037 | } | |
90db1043 | 5038 | } |
a1300716 | 5039 | |
e93f8a0f | 5040 | kfree(bus); |
5d3c4c79 | 5041 | return new_bus ? 0 : -ENOMEM; |
2eeb2e94 GH |
5042 | } |
5043 | ||
8a39d006 AP |
5044 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5045 | gpa_t addr) | |
5046 | { | |
5047 | struct kvm_io_bus *bus; | |
5048 | int dev_idx, srcu_idx; | |
5049 | struct kvm_io_device *iodev = NULL; | |
5050 | ||
5051 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
5052 | ||
5053 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
5054 | if (!bus) |
5055 | goto out_unlock; | |
8a39d006 AP |
5056 | |
5057 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
5058 | if (dev_idx < 0) | |
5059 | goto out_unlock; | |
5060 | ||
5061 | iodev = bus->range[dev_idx].dev; | |
5062 | ||
5063 | out_unlock: | |
5064 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
5065 | ||
5066 | return iodev; | |
5067 | } | |
5068 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
5069 | ||
536a6f88 JF |
5070 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
5071 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
5072 | const char *fmt) | |
5073 | { | |
5074 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5075 | inode->i_private; | |
5076 | ||
605c7130 PX |
5077 | /* |
5078 | * The debugfs files are a reference to the kvm struct which | |
5079 | * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe | |
5080 | * avoids the race between open and the removal of the debugfs directory. | |
536a6f88 | 5081 | */ |
605c7130 | 5082 | if (!kvm_get_kvm_safe(stat_data->kvm)) |
536a6f88 JF |
5083 | return -ENOENT; |
5084 | ||
833b45de | 5085 | if (simple_attr_open(inode, file, get, |
bc9e9e67 | 5086 | kvm_stats_debugfs_mode(stat_data->desc) & 0222 |
09cbcef6 MP |
5087 | ? set : NULL, |
5088 | fmt)) { | |
536a6f88 JF |
5089 | kvm_put_kvm(stat_data->kvm); |
5090 | return -ENOMEM; | |
5091 | } | |
5092 | ||
5093 | return 0; | |
5094 | } | |
5095 | ||
5096 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
5097 | { | |
5098 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5099 | inode->i_private; | |
5100 | ||
5101 | simple_attr_release(inode, file); | |
5102 | kvm_put_kvm(stat_data->kvm); | |
5103 | ||
5104 | return 0; | |
5105 | } | |
5106 | ||
09cbcef6 | 5107 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 5108 | { |
bc9e9e67 | 5109 | *val = *(u64 *)((void *)(&kvm->stat) + offset); |
536a6f88 | 5110 | |
09cbcef6 MP |
5111 | return 0; |
5112 | } | |
5113 | ||
5114 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
5115 | { | |
bc9e9e67 | 5116 | *(u64 *)((void *)(&kvm->stat) + offset) = 0; |
536a6f88 JF |
5117 | |
5118 | return 0; | |
5119 | } | |
5120 | ||
09cbcef6 | 5121 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 5122 | { |
09cbcef6 MP |
5123 | int i; |
5124 | struct kvm_vcpu *vcpu; | |
ce35ef27 | 5125 | |
09cbcef6 | 5126 | *val = 0; |
ce35ef27 | 5127 | |
09cbcef6 | 5128 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5129 | *val += *(u64 *)((void *)(&vcpu->stat) + offset); |
ce35ef27 SJS |
5130 | |
5131 | return 0; | |
5132 | } | |
5133 | ||
09cbcef6 | 5134 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 5135 | { |
09cbcef6 MP |
5136 | int i; |
5137 | struct kvm_vcpu *vcpu; | |
536a6f88 | 5138 | |
09cbcef6 | 5139 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5140 | *(u64 *)((void *)(&vcpu->stat) + offset) = 0; |
09cbcef6 MP |
5141 | |
5142 | return 0; | |
5143 | } | |
536a6f88 | 5144 | |
09cbcef6 | 5145 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 5146 | { |
09cbcef6 | 5147 | int r = -EFAULT; |
536a6f88 | 5148 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
536a6f88 | 5149 | |
bc9e9e67 | 5150 | switch (stat_data->kind) { |
09cbcef6 MP |
5151 | case KVM_STAT_VM: |
5152 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5153 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5154 | break; |
5155 | case KVM_STAT_VCPU: | |
5156 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5157 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5158 | break; |
5159 | } | |
536a6f88 | 5160 | |
09cbcef6 | 5161 | return r; |
536a6f88 JF |
5162 | } |
5163 | ||
09cbcef6 | 5164 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 5165 | { |
09cbcef6 | 5166 | int r = -EFAULT; |
ce35ef27 | 5167 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
ce35ef27 SJS |
5168 | |
5169 | if (val) | |
5170 | return -EINVAL; | |
5171 | ||
bc9e9e67 | 5172 | switch (stat_data->kind) { |
09cbcef6 MP |
5173 | case KVM_STAT_VM: |
5174 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5175 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5176 | break; |
5177 | case KVM_STAT_VCPU: | |
5178 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5179 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5180 | break; |
5181 | } | |
ce35ef27 | 5182 | |
09cbcef6 | 5183 | return r; |
ce35ef27 SJS |
5184 | } |
5185 | ||
09cbcef6 | 5186 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
5187 | { |
5188 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
5189 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
5190 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
5191 | } |
5192 | ||
09cbcef6 MP |
5193 | static const struct file_operations stat_fops_per_vm = { |
5194 | .owner = THIS_MODULE, | |
5195 | .open = kvm_stat_data_open, | |
536a6f88 | 5196 | .release = kvm_debugfs_release, |
09cbcef6 MP |
5197 | .read = simple_attr_read, |
5198 | .write = simple_attr_write, | |
5199 | .llseek = no_llseek, | |
536a6f88 JF |
5200 | }; |
5201 | ||
8b88b099 | 5202 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
5203 | { |
5204 | unsigned offset = (long)_offset; | |
ba1389b7 | 5205 | struct kvm *kvm; |
536a6f88 | 5206 | u64 tmp_val; |
ba1389b7 | 5207 | |
8b88b099 | 5208 | *val = 0; |
0d9ce162 | 5209 | mutex_lock(&kvm_lock); |
536a6f88 | 5210 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5211 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
5212 | *val += tmp_val; |
5213 | } | |
0d9ce162 | 5214 | mutex_unlock(&kvm_lock); |
8b88b099 | 5215 | return 0; |
ba1389b7 AK |
5216 | } |
5217 | ||
ce35ef27 SJS |
5218 | static int vm_stat_clear(void *_offset, u64 val) |
5219 | { | |
5220 | unsigned offset = (long)_offset; | |
5221 | struct kvm *kvm; | |
ce35ef27 SJS |
5222 | |
5223 | if (val) | |
5224 | return -EINVAL; | |
5225 | ||
0d9ce162 | 5226 | mutex_lock(&kvm_lock); |
ce35ef27 | 5227 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5228 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 5229 | } |
0d9ce162 | 5230 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5231 | |
5232 | return 0; | |
5233 | } | |
5234 | ||
5235 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
bc9e9e67 | 5236 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n"); |
ba1389b7 | 5237 | |
8b88b099 | 5238 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
5239 | { |
5240 | unsigned offset = (long)_offset; | |
1165f5fe | 5241 | struct kvm *kvm; |
536a6f88 | 5242 | u64 tmp_val; |
1165f5fe | 5243 | |
8b88b099 | 5244 | *val = 0; |
0d9ce162 | 5245 | mutex_lock(&kvm_lock); |
536a6f88 | 5246 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5247 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
5248 | *val += tmp_val; |
5249 | } | |
0d9ce162 | 5250 | mutex_unlock(&kvm_lock); |
8b88b099 | 5251 | return 0; |
1165f5fe AK |
5252 | } |
5253 | ||
ce35ef27 SJS |
5254 | static int vcpu_stat_clear(void *_offset, u64 val) |
5255 | { | |
5256 | unsigned offset = (long)_offset; | |
5257 | struct kvm *kvm; | |
ce35ef27 SJS |
5258 | |
5259 | if (val) | |
5260 | return -EINVAL; | |
5261 | ||
0d9ce162 | 5262 | mutex_lock(&kvm_lock); |
ce35ef27 | 5263 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5264 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 5265 | } |
0d9ce162 | 5266 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5267 | |
5268 | return 0; | |
5269 | } | |
5270 | ||
5271 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
5272 | "%llu\n"); | |
bc9e9e67 | 5273 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n"); |
1165f5fe | 5274 | |
286de8f6 CI |
5275 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
5276 | { | |
5277 | struct kobj_uevent_env *env; | |
286de8f6 CI |
5278 | unsigned long long created, active; |
5279 | ||
5280 | if (!kvm_dev.this_device || !kvm) | |
5281 | return; | |
5282 | ||
0d9ce162 | 5283 | mutex_lock(&kvm_lock); |
286de8f6 CI |
5284 | if (type == KVM_EVENT_CREATE_VM) { |
5285 | kvm_createvm_count++; | |
5286 | kvm_active_vms++; | |
5287 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
5288 | kvm_active_vms--; | |
5289 | } | |
5290 | created = kvm_createvm_count; | |
5291 | active = kvm_active_vms; | |
0d9ce162 | 5292 | mutex_unlock(&kvm_lock); |
286de8f6 | 5293 | |
b12ce36a | 5294 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
5295 | if (!env) |
5296 | return; | |
5297 | ||
5298 | add_uevent_var(env, "CREATED=%llu", created); | |
5299 | add_uevent_var(env, "COUNT=%llu", active); | |
5300 | ||
fdeaf7e3 | 5301 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 5302 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
5303 | kvm->userspace_pid = task_pid_nr(current); |
5304 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 5305 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
5306 | } |
5307 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 5308 | |
85cd39af | 5309 | if (kvm->debugfs_dentry) { |
b12ce36a | 5310 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
5311 | |
5312 | if (p) { | |
5313 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
5314 | if (!IS_ERR(tmp)) | |
5315 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
5316 | kfree(p); | |
286de8f6 CI |
5317 | } |
5318 | } | |
5319 | /* no need for checks, since we are adding at most only 5 keys */ | |
5320 | env->envp[env->envp_idx++] = NULL; | |
5321 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
5322 | kfree(env); | |
286de8f6 CI |
5323 | } |
5324 | ||
929f45e3 | 5325 | static void kvm_init_debug(void) |
6aa8b732 | 5326 | { |
bc9e9e67 JZ |
5327 | const struct file_operations *fops; |
5328 | const struct _kvm_stats_desc *pdesc; | |
5329 | int i; | |
6aa8b732 | 5330 | |
76f7c879 | 5331 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 5332 | |
bc9e9e67 JZ |
5333 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
5334 | pdesc = &kvm_vm_stats_desc[i]; | |
5335 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5336 | fops = &vm_stat_fops; | |
5337 | else | |
5338 | fops = &vm_stat_readonly_fops; | |
5339 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5340 | kvm_debugfs_dir, | |
5341 | (void *)(long)pdesc->desc.offset, fops); | |
5342 | } | |
5343 | ||
5344 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
5345 | pdesc = &kvm_vcpu_stats_desc[i]; | |
5346 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5347 | fops = &vcpu_stat_fops; | |
5348 | else | |
5349 | fops = &vcpu_stat_readonly_fops; | |
5350 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5351 | kvm_debugfs_dir, | |
5352 | (void *)(long)pdesc->desc.offset, fops); | |
4f69b680 | 5353 | } |
6aa8b732 AK |
5354 | } |
5355 | ||
fb3600cc | 5356 | static int kvm_suspend(void) |
59ae6c6b | 5357 | { |
10474ae8 | 5358 | if (kvm_usage_count) |
75b7127c | 5359 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
5360 | return 0; |
5361 | } | |
5362 | ||
fb3600cc | 5363 | static void kvm_resume(void) |
59ae6c6b | 5364 | { |
ca84d1a2 | 5365 | if (kvm_usage_count) { |
2eb06c30 WL |
5366 | #ifdef CONFIG_LOCKDEP |
5367 | WARN_ON(lockdep_is_held(&kvm_count_lock)); | |
5368 | #endif | |
75b7127c | 5369 | hardware_enable_nolock(NULL); |
ca84d1a2 | 5370 | } |
59ae6c6b AK |
5371 | } |
5372 | ||
fb3600cc | 5373 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
5374 | .suspend = kvm_suspend, |
5375 | .resume = kvm_resume, | |
5376 | }; | |
5377 | ||
15ad7146 AK |
5378 | static inline |
5379 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
5380 | { | |
5381 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
5382 | } | |
5383 | ||
5384 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
5385 | { | |
5386 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 5387 | |
046ddeed | 5388 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 5389 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 5390 | |
7495e22b | 5391 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 5392 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 5393 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
5394 | } |
5395 | ||
5396 | static void kvm_sched_out(struct preempt_notifier *pn, | |
5397 | struct task_struct *next) | |
5398 | { | |
5399 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
5400 | ||
3ba9f93b | 5401 | if (current->on_rq) { |
046ddeed | 5402 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
5403 | WRITE_ONCE(vcpu->ready, true); |
5404 | } | |
e9b11c17 | 5405 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
5406 | __this_cpu_write(kvm_running_vcpu, NULL); |
5407 | } | |
5408 | ||
5409 | /** | |
5410 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
5411 | * |
5412 | * We can disable preemption locally around accessing the per-CPU variable, | |
5413 | * and use the resolved vcpu pointer after enabling preemption again, | |
5414 | * because even if the current thread is migrated to another CPU, reading | |
5415 | * the per-CPU value later will give us the same value as we update the | |
5416 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
5417 | */ |
5418 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
5419 | { | |
1f03b2bc MZ |
5420 | struct kvm_vcpu *vcpu; |
5421 | ||
5422 | preempt_disable(); | |
5423 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
5424 | preempt_enable(); | |
5425 | ||
5426 | return vcpu; | |
7495e22b | 5427 | } |
379a3c8e | 5428 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
5429 | |
5430 | /** | |
5431 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
5432 | */ | |
5433 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
5434 | { | |
5435 | return &kvm_running_vcpu; | |
15ad7146 AK |
5436 | } |
5437 | ||
b9904085 SC |
5438 | struct kvm_cpu_compat_check { |
5439 | void *opaque; | |
5440 | int *ret; | |
5441 | }; | |
5442 | ||
5443 | static void check_processor_compat(void *data) | |
f257d6dc | 5444 | { |
b9904085 SC |
5445 | struct kvm_cpu_compat_check *c = data; |
5446 | ||
5447 | *c->ret = kvm_arch_check_processor_compat(c->opaque); | |
f257d6dc SC |
5448 | } |
5449 | ||
0ee75bea | 5450 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 5451 | struct module *module) |
6aa8b732 | 5452 | { |
b9904085 | 5453 | struct kvm_cpu_compat_check c; |
6aa8b732 | 5454 | int r; |
002c7f7c | 5455 | int cpu; |
6aa8b732 | 5456 | |
f8c16bba ZX |
5457 | r = kvm_arch_init(opaque); |
5458 | if (r) | |
d2308784 | 5459 | goto out_fail; |
cb498ea2 | 5460 | |
7dac16c3 AH |
5461 | /* |
5462 | * kvm_arch_init makes sure there's at most one caller | |
5463 | * for architectures that support multiple implementations, | |
5464 | * like intel and amd on x86. | |
36343f6e PB |
5465 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
5466 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 5467 | */ |
36343f6e PB |
5468 | r = kvm_irqfd_init(); |
5469 | if (r) | |
5470 | goto out_irqfd; | |
7dac16c3 | 5471 | |
8437a617 | 5472 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
5473 | r = -ENOMEM; |
5474 | goto out_free_0; | |
5475 | } | |
5476 | ||
b9904085 | 5477 | r = kvm_arch_hardware_setup(opaque); |
6aa8b732 | 5478 | if (r < 0) |
faf0be22 | 5479 | goto out_free_1; |
6aa8b732 | 5480 | |
b9904085 SC |
5481 | c.ret = &r; |
5482 | c.opaque = opaque; | |
002c7f7c | 5483 | for_each_online_cpu(cpu) { |
b9904085 | 5484 | smp_call_function_single(cpu, check_processor_compat, &c, 1); |
002c7f7c | 5485 | if (r < 0) |
faf0be22 | 5486 | goto out_free_2; |
002c7f7c YS |
5487 | } |
5488 | ||
73c1b41e | 5489 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 5490 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 5491 | if (r) |
d2308784 | 5492 | goto out_free_2; |
6aa8b732 AK |
5493 | register_reboot_notifier(&kvm_reboot_notifier); |
5494 | ||
c16f862d | 5495 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
5496 | if (!vcpu_align) |
5497 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
5498 | kvm_vcpu_cache = |
5499 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
5500 | SLAB_ACCOUNT, | |
5501 | offsetof(struct kvm_vcpu, arch), | |
ce55c049 JZ |
5502 | offsetofend(struct kvm_vcpu, stats_id) |
5503 | - offsetof(struct kvm_vcpu, arch), | |
46515736 | 5504 | NULL); |
c16f862d RR |
5505 | if (!kvm_vcpu_cache) { |
5506 | r = -ENOMEM; | |
fb3600cc | 5507 | goto out_free_3; |
c16f862d RR |
5508 | } |
5509 | ||
baff59cc VK |
5510 | for_each_possible_cpu(cpu) { |
5511 | if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu), | |
5512 | GFP_KERNEL, cpu_to_node(cpu))) { | |
5513 | r = -ENOMEM; | |
5514 | goto out_free_4; | |
5515 | } | |
5516 | } | |
5517 | ||
af585b92 GN |
5518 | r = kvm_async_pf_init(); |
5519 | if (r) | |
baff59cc | 5520 | goto out_free_5; |
af585b92 | 5521 | |
6aa8b732 | 5522 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
5523 | kvm_vm_fops.owner = module; |
5524 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
5525 | |
5526 | r = misc_register(&kvm_dev); | |
5527 | if (r) { | |
1170adc6 | 5528 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 5529 | goto out_unreg; |
6aa8b732 AK |
5530 | } |
5531 | ||
fb3600cc RW |
5532 | register_syscore_ops(&kvm_syscore_ops); |
5533 | ||
15ad7146 AK |
5534 | kvm_preempt_ops.sched_in = kvm_sched_in; |
5535 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
5536 | ||
929f45e3 | 5537 | kvm_init_debug(); |
0ea4ed8e | 5538 | |
3c3c29fd PB |
5539 | r = kvm_vfio_ops_init(); |
5540 | WARN_ON(r); | |
5541 | ||
c7addb90 | 5542 | return 0; |
6aa8b732 | 5543 | |
af585b92 GN |
5544 | out_unreg: |
5545 | kvm_async_pf_deinit(); | |
baff59cc VK |
5546 | out_free_5: |
5547 | for_each_possible_cpu(cpu) | |
5548 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
5549 | out_free_4: | |
c16f862d | 5550 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 5551 | out_free_3: |
6aa8b732 | 5552 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5553 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 5554 | out_free_2: |
e9b11c17 | 5555 | kvm_arch_hardware_unsetup(); |
faf0be22 | 5556 | out_free_1: |
7f59f492 | 5557 | free_cpumask_var(cpus_hardware_enabled); |
d2308784 | 5558 | out_free_0: |
a0f155e9 | 5559 | kvm_irqfd_exit(); |
36343f6e | 5560 | out_irqfd: |
7dac16c3 AH |
5561 | kvm_arch_exit(); |
5562 | out_fail: | |
6aa8b732 AK |
5563 | return r; |
5564 | } | |
cb498ea2 | 5565 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 5566 | |
cb498ea2 | 5567 | void kvm_exit(void) |
6aa8b732 | 5568 | { |
baff59cc VK |
5569 | int cpu; |
5570 | ||
4bd33b56 | 5571 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 5572 | misc_deregister(&kvm_dev); |
baff59cc VK |
5573 | for_each_possible_cpu(cpu) |
5574 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
c16f862d | 5575 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 5576 | kvm_async_pf_deinit(); |
fb3600cc | 5577 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 5578 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5579 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 5580 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 5581 | kvm_arch_hardware_unsetup(); |
f8c16bba | 5582 | kvm_arch_exit(); |
a0f155e9 | 5583 | kvm_irqfd_exit(); |
7f59f492 | 5584 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 5585 | kvm_vfio_ops_exit(); |
6aa8b732 | 5586 | } |
cb498ea2 | 5587 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
5588 | |
5589 | struct kvm_vm_worker_thread_context { | |
5590 | struct kvm *kvm; | |
5591 | struct task_struct *parent; | |
5592 | struct completion init_done; | |
5593 | kvm_vm_thread_fn_t thread_fn; | |
5594 | uintptr_t data; | |
5595 | int err; | |
5596 | }; | |
5597 | ||
5598 | static int kvm_vm_worker_thread(void *context) | |
5599 | { | |
5600 | /* | |
5601 | * The init_context is allocated on the stack of the parent thread, so | |
5602 | * we have to locally copy anything that is needed beyond initialization | |
5603 | */ | |
5604 | struct kvm_vm_worker_thread_context *init_context = context; | |
5605 | struct kvm *kvm = init_context->kvm; | |
5606 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
5607 | uintptr_t data = init_context->data; | |
5608 | int err; | |
5609 | ||
5610 | err = kthread_park(current); | |
5611 | /* kthread_park(current) is never supposed to return an error */ | |
5612 | WARN_ON(err != 0); | |
5613 | if (err) | |
5614 | goto init_complete; | |
5615 | ||
5616 | err = cgroup_attach_task_all(init_context->parent, current); | |
5617 | if (err) { | |
5618 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
5619 | __func__, err); | |
5620 | goto init_complete; | |
5621 | } | |
5622 | ||
5623 | set_user_nice(current, task_nice(init_context->parent)); | |
5624 | ||
5625 | init_complete: | |
5626 | init_context->err = err; | |
5627 | complete(&init_context->init_done); | |
5628 | init_context = NULL; | |
5629 | ||
5630 | if (err) | |
5631 | return err; | |
5632 | ||
5633 | /* Wait to be woken up by the spawner before proceeding. */ | |
5634 | kthread_parkme(); | |
5635 | ||
5636 | if (!kthread_should_stop()) | |
5637 | err = thread_fn(kvm, data); | |
5638 | ||
5639 | return err; | |
5640 | } | |
5641 | ||
5642 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
5643 | uintptr_t data, const char *name, | |
5644 | struct task_struct **thread_ptr) | |
5645 | { | |
5646 | struct kvm_vm_worker_thread_context init_context = {}; | |
5647 | struct task_struct *thread; | |
5648 | ||
5649 | *thread_ptr = NULL; | |
5650 | init_context.kvm = kvm; | |
5651 | init_context.parent = current; | |
5652 | init_context.thread_fn = thread_fn; | |
5653 | init_context.data = data; | |
5654 | init_completion(&init_context.init_done); | |
5655 | ||
5656 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
5657 | "%s-%d", name, task_pid_nr(current)); | |
5658 | if (IS_ERR(thread)) | |
5659 | return PTR_ERR(thread); | |
5660 | ||
5661 | /* kthread_run is never supposed to return NULL */ | |
5662 | WARN_ON(thread == NULL); | |
5663 | ||
5664 | wait_for_completion(&init_context.init_done); | |
5665 | ||
5666 | if (!init_context.err) | |
5667 | *thread_ptr = thread; | |
5668 | ||
5669 | return init_context.err; | |
5670 | } |