]>
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> |
6aa8b732 | 54 | |
e495606d | 55 | #include <asm/processor.h> |
2ea75be3 | 56 | #include <asm/ioctl.h> |
7c0f6ba6 | 57 | #include <linux/uaccess.h> |
3e021bf5 | 58 | #include <asm/pgtable.h> |
6aa8b732 | 59 | |
5f94c174 | 60 | #include "coalesced_mmio.h" |
af585b92 | 61 | #include "async_pf.h" |
3c3c29fd | 62 | #include "vfio.h" |
5f94c174 | 63 | |
229456fc MT |
64 | #define CREATE_TRACE_POINTS |
65 | #include <trace/events/kvm.h> | |
66 | ||
536a6f88 JF |
67 | /* Worst case buffer size needed for holding an integer. */ |
68 | #define ITOA_MAX_LEN 12 | |
69 | ||
6aa8b732 AK |
70 | MODULE_AUTHOR("Qumranet"); |
71 | MODULE_LICENSE("GPL"); | |
72 | ||
920552b2 | 73 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 74 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
039c5d1b | 75 | module_param(halt_poll_ns, uint, 0644); |
ec76d819 | 76 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 77 | |
aca6ff29 | 78 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 79 | unsigned int halt_poll_ns_grow = 2; |
039c5d1b | 80 | module_param(halt_poll_ns_grow, uint, 0644); |
ec76d819 | 81 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 | 82 | |
49113d36 NW |
83 | /* The start value to grow halt_poll_ns from */ |
84 | unsigned int halt_poll_ns_grow_start = 10000; /* 10us */ | |
85 | module_param(halt_poll_ns_grow_start, uint, 0644); | |
86 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start); | |
87 | ||
aca6ff29 | 88 | /* Default resets per-vcpu halt_poll_ns . */ |
ec76d819 | 89 | unsigned int halt_poll_ns_shrink; |
039c5d1b | 90 | module_param(halt_poll_ns_shrink, uint, 0644); |
ec76d819 | 91 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 92 | |
fa40a821 MT |
93 | /* |
94 | * Ordering of locks: | |
95 | * | |
b7d409de | 96 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
97 | */ |
98 | ||
0d9ce162 | 99 | DEFINE_MUTEX(kvm_lock); |
4a937f96 | 100 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 101 | LIST_HEAD(vm_list); |
133de902 | 102 | |
7f59f492 | 103 | static cpumask_var_t cpus_hardware_enabled; |
f4fee932 | 104 | static int kvm_usage_count; |
10474ae8 | 105 | static atomic_t hardware_enable_failed; |
1b6c0168 | 106 | |
c16f862d RR |
107 | struct kmem_cache *kvm_vcpu_cache; |
108 | EXPORT_SYMBOL_GPL(kvm_vcpu_cache); | |
1165f5fe | 109 | |
15ad7146 AK |
110 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
111 | ||
76f7c879 | 112 | struct dentry *kvm_debugfs_dir; |
e23a808b | 113 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 114 | |
536a6f88 JF |
115 | static int kvm_debugfs_num_entries; |
116 | static const struct file_operations *stat_fops_per_vm[]; | |
117 | ||
59069e9e DM |
118 | static struct file_operations kvm_chardev_ops; |
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 | |
bc009e43 | 149 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn); |
e93f8a0f | 150 | |
52480137 | 151 | __visible bool kvm_rebooting; |
b7c4145b | 152 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 153 | |
54dee993 MT |
154 | static bool largepages_enabled = true; |
155 | ||
286de8f6 CI |
156 | #define KVM_EVENT_CREATE_VM 0 |
157 | #define KVM_EVENT_DESTROY_VM 1 | |
158 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); | |
159 | static unsigned long long kvm_createvm_count; | |
160 | static unsigned long long kvm_active_vms; | |
161 | ||
b512e631 ET |
162 | __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
163 | unsigned long start, unsigned long end) | |
b1394e74 RK |
164 | { |
165 | } | |
166 | ||
a78986aa SC |
167 | bool kvm_is_zone_device_pfn(kvm_pfn_t pfn) |
168 | { | |
169 | /* | |
170 | * The metadata used by is_zone_device_page() to determine whether or | |
171 | * not a page is ZONE_DEVICE is guaranteed to be valid if and only if | |
172 | * the device has been pinned, e.g. by get_user_pages(). WARN if the | |
173 | * page_count() is zero to help detect bad usage of this helper. | |
174 | */ | |
175 | if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn)))) | |
176 | return false; | |
177 | ||
178 | return is_zone_device_page(pfn_to_page(pfn)); | |
179 | } | |
180 | ||
ba049e93 | 181 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
cbff90a7 | 182 | { |
a78986aa SC |
183 | /* |
184 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting | |
185 | * perspective they are "normal" pages, albeit with slightly different | |
186 | * usage rules. | |
187 | */ | |
11feeb49 | 188 | if (pfn_valid(pfn)) |
a78986aa | 189 | return PageReserved(pfn_to_page(pfn)) && |
2ee36763 | 190 | !is_zero_pfn(pfn) && |
a78986aa | 191 | !kvm_is_zone_device_pfn(pfn); |
cbff90a7 BAY |
192 | |
193 | return true; | |
194 | } | |
195 | ||
bccf2150 AK |
196 | /* |
197 | * Switches to specified vcpu, until a matching vcpu_put() | |
198 | */ | |
ec7660cc | 199 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 200 | { |
ec7660cc | 201 | int cpu = get_cpu(); |
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 AK |
212 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
213 | preempt_enable(); | |
6aa8b732 | 214 | } |
2f1fe811 | 215 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 216 | |
7a97cec2 PB |
217 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
218 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
219 | { | |
220 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
221 | ||
222 | /* | |
223 | * We need to wait for the VCPU to reenable interrupts and get out of | |
224 | * READING_SHADOW_PAGE_TABLES mode. | |
225 | */ | |
226 | if (req & KVM_REQUEST_WAIT) | |
227 | return mode != OUTSIDE_GUEST_MODE; | |
228 | ||
229 | /* | |
230 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
231 | */ | |
232 | return mode == IN_GUEST_MODE; | |
233 | } | |
234 | ||
d9e368d6 AK |
235 | static void ack_flush(void *_completed) |
236 | { | |
d9e368d6 AK |
237 | } |
238 | ||
b49defe8 PB |
239 | static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait) |
240 | { | |
241 | if (unlikely(!cpus)) | |
242 | cpus = cpu_online_mask; | |
243 | ||
244 | if (cpumask_empty(cpus)) | |
245 | return false; | |
246 | ||
247 | smp_call_function_many(cpus, ack_flush, NULL, wait); | |
248 | return true; | |
249 | } | |
250 | ||
7053df4e VK |
251 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
252 | unsigned long *vcpu_bitmap, cpumask_var_t tmp) | |
d9e368d6 | 253 | { |
597a5f55 | 254 | int i, cpu, me; |
d9e368d6 | 255 | struct kvm_vcpu *vcpu; |
7053df4e | 256 | bool called; |
6ef7a1bc | 257 | |
3cba4130 | 258 | me = get_cpu(); |
7053df4e | 259 | |
988a2cae | 260 | kvm_for_each_vcpu(i, vcpu, kvm) { |
a812297c | 261 | if (vcpu_bitmap && !test_bit(i, vcpu_bitmap)) |
7053df4e VK |
262 | continue; |
263 | ||
3cba4130 | 264 | kvm_make_request(req, vcpu); |
d9e368d6 | 265 | cpu = vcpu->cpu; |
6b7e2d09 | 266 | |
178f02ff RK |
267 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) |
268 | continue; | |
6c6e8360 | 269 | |
7053df4e | 270 | if (tmp != NULL && cpu != -1 && cpu != me && |
7a97cec2 | 271 | kvm_request_needs_ipi(vcpu, req)) |
7053df4e | 272 | __cpumask_set_cpu(cpu, tmp); |
49846896 | 273 | } |
7053df4e VK |
274 | |
275 | called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT)); | |
3cba4130 | 276 | put_cpu(); |
7053df4e VK |
277 | |
278 | return called; | |
279 | } | |
280 | ||
281 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) | |
282 | { | |
283 | cpumask_var_t cpus; | |
284 | bool called; | |
7053df4e VK |
285 | |
286 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); | |
287 | ||
a812297c | 288 | called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus); |
7053df4e | 289 | |
6ef7a1bc | 290 | free_cpumask_var(cpus); |
49846896 | 291 | return called; |
d9e368d6 AK |
292 | } |
293 | ||
a6d51016 | 294 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 295 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 296 | { |
4ae3cb3a LT |
297 | /* |
298 | * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in | |
299 | * kvm_make_all_cpus_request. | |
300 | */ | |
301 | long dirty_count = smp_load_acquire(&kvm->tlbs_dirty); | |
302 | ||
303 | /* | |
304 | * We want to publish modifications to the page tables before reading | |
305 | * mode. Pairs with a memory barrier in arch-specific code. | |
306 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
307 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
308 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
309 | * | |
310 | * There is already an smp_mb__after_atomic() before | |
311 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
312 | * barrier here. | |
313 | */ | |
b08660e5 TL |
314 | if (!kvm_arch_flush_remote_tlb(kvm) |
315 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
49846896 | 316 | ++kvm->stat.remote_tlb_flush; |
a086f6a1 | 317 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 318 | } |
2ba9f0d8 | 319 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 320 | #endif |
2e53d63a | 321 | |
49846896 RR |
322 | void kvm_reload_remote_mmus(struct kvm *kvm) |
323 | { | |
445b8236 | 324 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 325 | } |
2e53d63a | 326 | |
fb3f0f51 RR |
327 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
328 | { | |
329 | struct page *page; | |
330 | int r; | |
331 | ||
332 | mutex_init(&vcpu->mutex); | |
333 | vcpu->cpu = -1; | |
fb3f0f51 RR |
334 | vcpu->kvm = kvm; |
335 | vcpu->vcpu_id = id; | |
34bb10b7 | 336 | vcpu->pid = NULL; |
8577370f | 337 | init_swait_queue_head(&vcpu->wq); |
af585b92 | 338 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 339 | |
bf9f6ac8 FW |
340 | vcpu->pre_pcpu = -1; |
341 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
342 | ||
fb3f0f51 RR |
343 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
344 | if (!page) { | |
345 | r = -ENOMEM; | |
346 | goto fail; | |
347 | } | |
348 | vcpu->run = page_address(page); | |
349 | ||
4c088493 R |
350 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
351 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 352 | vcpu->preempted = false; |
d73eb57b | 353 | vcpu->ready = false; |
4c088493 | 354 | |
e9b11c17 | 355 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 356 | if (r < 0) |
e9b11c17 | 357 | goto fail_free_run; |
fb3f0f51 RR |
358 | return 0; |
359 | ||
fb3f0f51 RR |
360 | fail_free_run: |
361 | free_page((unsigned long)vcpu->run); | |
362 | fail: | |
76fafa5e | 363 | return r; |
fb3f0f51 RR |
364 | } |
365 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
366 | ||
367 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
368 | { | |
0e4524a5 CB |
369 | /* |
370 | * no need for rcu_read_lock as VCPU_RUN is the only place that | |
371 | * will change the vcpu->pid pointer and on uninit all file | |
372 | * descriptors are already gone. | |
373 | */ | |
374 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
e9b11c17 | 375 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
376 | free_page((unsigned long)vcpu->run); |
377 | } | |
378 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
379 | ||
e930bffe AA |
380 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
381 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
382 | { | |
383 | return container_of(mn, struct kvm, mmu_notifier); | |
384 | } | |
385 | ||
b512e631 ET |
386 | static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn, |
387 | struct mm_struct *mm, | |
388 | unsigned long start, unsigned long end) | |
389 | { | |
390 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
391 | int idx; | |
392 | ||
393 | idx = srcu_read_lock(&kvm->srcu); | |
394 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
395 | srcu_read_unlock(&kvm->srcu, idx); | |
396 | } | |
397 | ||
3da0dd43 IE |
398 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
399 | struct mm_struct *mm, | |
400 | unsigned long address, | |
401 | pte_t pte) | |
402 | { | |
403 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 404 | int idx; |
3da0dd43 | 405 | |
bc6678a3 | 406 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
407 | spin_lock(&kvm->mmu_lock); |
408 | kvm->mmu_notifier_seq++; | |
0cf853c5 LT |
409 | |
410 | if (kvm_set_spte_hva(kvm, address, pte)) | |
411 | kvm_flush_remote_tlbs(kvm); | |
412 | ||
3da0dd43 | 413 | spin_unlock(&kvm->mmu_lock); |
bc6678a3 | 414 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
415 | } |
416 | ||
93065ac7 | 417 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
5d6527a7 | 418 | const struct mmu_notifier_range *range) |
e930bffe AA |
419 | { |
420 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 421 | int need_tlb_flush = 0, idx; |
e930bffe | 422 | |
bc6678a3 | 423 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
424 | spin_lock(&kvm->mmu_lock); |
425 | /* | |
426 | * The count increase must become visible at unlock time as no | |
427 | * spte can be established without taking the mmu_lock and | |
428 | * count is also read inside the mmu_lock critical section. | |
429 | */ | |
430 | kvm->mmu_notifier_count++; | |
5d890eb5 WD |
431 | need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end, |
432 | range->flags); | |
e930bffe | 433 | /* we've to flush the tlb before the pages can be freed */ |
69f05934 | 434 | if (need_tlb_flush || kvm->tlbs_dirty) |
e930bffe | 435 | kvm_flush_remote_tlbs(kvm); |
565f3be2 TY |
436 | |
437 | spin_unlock(&kvm->mmu_lock); | |
438 | srcu_read_unlock(&kvm->srcu, idx); | |
93065ac7 | 439 | |
b512e631 | 440 | return 0; |
e930bffe AA |
441 | } |
442 | ||
443 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
5d6527a7 | 444 | const struct mmu_notifier_range *range) |
e930bffe AA |
445 | { |
446 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
447 | ||
448 | spin_lock(&kvm->mmu_lock); | |
449 | /* | |
450 | * This sequence increase will notify the kvm page fault that | |
451 | * the page that is going to be mapped in the spte could have | |
452 | * been freed. | |
453 | */ | |
454 | kvm->mmu_notifier_seq++; | |
a355aa54 | 455 | smp_wmb(); |
e930bffe AA |
456 | /* |
457 | * The above sequence increase must be visible before the | |
a355aa54 PM |
458 | * below count decrease, which is ensured by the smp_wmb above |
459 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
460 | */ |
461 | kvm->mmu_notifier_count--; | |
462 | spin_unlock(&kvm->mmu_lock); | |
463 | ||
464 | BUG_ON(kvm->mmu_notifier_count < 0); | |
465 | } | |
466 | ||
467 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
468 | struct mm_struct *mm, | |
57128468 ALC |
469 | unsigned long start, |
470 | unsigned long end) | |
e930bffe AA |
471 | { |
472 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 473 | int young, idx; |
e930bffe | 474 | |
bc6678a3 | 475 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 476 | spin_lock(&kvm->mmu_lock); |
e930bffe | 477 | |
57128468 | 478 | young = kvm_age_hva(kvm, start, end); |
e930bffe AA |
479 | if (young) |
480 | kvm_flush_remote_tlbs(kvm); | |
481 | ||
565f3be2 TY |
482 | spin_unlock(&kvm->mmu_lock); |
483 | srcu_read_unlock(&kvm->srcu, idx); | |
484 | ||
e930bffe AA |
485 | return young; |
486 | } | |
487 | ||
1d7715c6 VD |
488 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
489 | struct mm_struct *mm, | |
490 | unsigned long start, | |
491 | unsigned long end) | |
492 | { | |
493 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
494 | int young, idx; | |
495 | ||
496 | idx = srcu_read_lock(&kvm->srcu); | |
497 | spin_lock(&kvm->mmu_lock); | |
498 | /* | |
499 | * Even though we do not flush TLB, this will still adversely | |
500 | * affect performance on pre-Haswell Intel EPT, where there is | |
501 | * no EPT Access Bit to clear so that we have to tear down EPT | |
502 | * tables instead. If we find this unacceptable, we can always | |
503 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
504 | * do anything on clear_young. | |
505 | * | |
506 | * Also note that currently we never issue secondary TLB flushes | |
507 | * from clear_young, leaving this job up to the regular system | |
508 | * cadence. If we find this inaccurate, we might come up with a | |
509 | * more sophisticated heuristic later. | |
510 | */ | |
511 | young = kvm_age_hva(kvm, start, end); | |
512 | spin_unlock(&kvm->mmu_lock); | |
513 | srcu_read_unlock(&kvm->srcu, idx); | |
514 | ||
515 | return young; | |
516 | } | |
517 | ||
8ee53820 AA |
518 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
519 | struct mm_struct *mm, | |
520 | unsigned long address) | |
521 | { | |
522 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
523 | int young, idx; | |
524 | ||
525 | idx = srcu_read_lock(&kvm->srcu); | |
526 | spin_lock(&kvm->mmu_lock); | |
527 | young = kvm_test_age_hva(kvm, address); | |
528 | spin_unlock(&kvm->mmu_lock); | |
529 | srcu_read_unlock(&kvm->srcu, idx); | |
530 | ||
531 | return young; | |
532 | } | |
533 | ||
85db06e5 MT |
534 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
535 | struct mm_struct *mm) | |
536 | { | |
537 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
538 | int idx; |
539 | ||
540 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 541 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 542 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
543 | } |
544 | ||
e930bffe | 545 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
b512e631 | 546 | .invalidate_range = kvm_mmu_notifier_invalidate_range, |
e930bffe AA |
547 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
548 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
549 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 550 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 551 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 552 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 553 | .release = kvm_mmu_notifier_release, |
e930bffe | 554 | }; |
4c07b0a4 AK |
555 | |
556 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
557 | { | |
558 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
559 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
560 | } | |
561 | ||
562 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
563 | ||
564 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
565 | { | |
566 | return 0; | |
567 | } | |
568 | ||
e930bffe AA |
569 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
570 | ||
a47d2b07 | 571 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
572 | { |
573 | int i; | |
a47d2b07 | 574 | struct kvm_memslots *slots; |
bf3e05bc | 575 | |
b12ce36a | 576 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
a47d2b07 PB |
577 | if (!slots) |
578 | return NULL; | |
579 | ||
bf3e05bc | 580 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
f85e2cb5 | 581 | slots->id_to_index[i] = slots->memslots[i].id = i; |
a47d2b07 PB |
582 | |
583 | return slots; | |
584 | } | |
585 | ||
586 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
587 | { | |
588 | if (!memslot->dirty_bitmap) | |
589 | return; | |
590 | ||
591 | kvfree(memslot->dirty_bitmap); | |
592 | memslot->dirty_bitmap = NULL; | |
593 | } | |
594 | ||
595 | /* | |
596 | * Free any memory in @free but not in @dont. | |
597 | */ | |
598 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
599 | struct kvm_memory_slot *dont) | |
600 | { | |
601 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) | |
602 | kvm_destroy_dirty_bitmap(free); | |
603 | ||
604 | kvm_arch_free_memslot(kvm, free, dont); | |
605 | ||
606 | free->npages = 0; | |
607 | } | |
608 | ||
609 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
610 | { | |
611 | struct kvm_memory_slot *memslot; | |
612 | ||
613 | if (!slots) | |
614 | return; | |
615 | ||
616 | kvm_for_each_memslot(memslot, slots) | |
617 | kvm_free_memslot(kvm, memslot, NULL); | |
618 | ||
619 | kvfree(slots); | |
bf3e05bc XG |
620 | } |
621 | ||
536a6f88 JF |
622 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
623 | { | |
624 | int i; | |
625 | ||
626 | if (!kvm->debugfs_dentry) | |
627 | return; | |
628 | ||
629 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
630 | ||
9d5a1dce LC |
631 | if (kvm->debugfs_stat_data) { |
632 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
633 | kfree(kvm->debugfs_stat_data[i]); | |
634 | kfree(kvm->debugfs_stat_data); | |
635 | } | |
536a6f88 JF |
636 | } |
637 | ||
638 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
639 | { | |
31542cd6 PB |
640 | static DEFINE_MUTEX(kvm_debugfs_lock); |
641 | struct dentry *dent; | |
536a6f88 JF |
642 | char dir_name[ITOA_MAX_LEN * 2]; |
643 | struct kvm_stat_data *stat_data; | |
644 | struct kvm_stats_debugfs_item *p; | |
645 | ||
646 | if (!debugfs_initialized()) | |
647 | return 0; | |
648 | ||
649 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
31542cd6 PB |
650 | mutex_lock(&kvm_debugfs_lock); |
651 | dent = debugfs_lookup(dir_name, kvm_debugfs_dir); | |
652 | if (dent) { | |
653 | pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name); | |
654 | dput(dent); | |
655 | mutex_unlock(&kvm_debugfs_lock); | |
656 | return 0; | |
657 | } | |
658 | dent = debugfs_create_dir(dir_name, kvm_debugfs_dir); | |
659 | mutex_unlock(&kvm_debugfs_lock); | |
660 | if (IS_ERR(dent)) | |
661 | return 0; | |
536a6f88 | 662 | |
31542cd6 | 663 | kvm->debugfs_dentry = dent; |
536a6f88 JF |
664 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
665 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 666 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
667 | if (!kvm->debugfs_stat_data) |
668 | return -ENOMEM; | |
669 | ||
670 | for (p = debugfs_entries; p->name; p++) { | |
b12ce36a | 671 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
672 | if (!stat_data) |
673 | return -ENOMEM; | |
674 | ||
675 | stat_data->kvm = kvm; | |
676 | stat_data->offset = p->offset; | |
833b45de | 677 | stat_data->mode = p->mode ? p->mode : 0644; |
536a6f88 | 678 | kvm->debugfs_stat_data[p - debugfs_entries] = stat_data; |
833b45de | 679 | debugfs_create_file(p->name, stat_data->mode, kvm->debugfs_dentry, |
929f45e3 | 680 | stat_data, stat_fops_per_vm[p->kind]); |
536a6f88 JF |
681 | } |
682 | return 0; | |
683 | } | |
684 | ||
1aa9b957 JS |
685 | /* |
686 | * Called after the VM is otherwise initialized, but just before adding it to | |
687 | * the vm_list. | |
688 | */ | |
689 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
690 | { | |
691 | return 0; | |
692 | } | |
693 | ||
694 | /* | |
695 | * Called just after removing the VM from the vm_list, but before doing any | |
696 | * other destruction. | |
697 | */ | |
698 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
699 | { | |
700 | } | |
701 | ||
e08b9637 | 702 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 703 | { |
d89f5eff | 704 | struct kvm *kvm = kvm_arch_alloc_vm(); |
9121923c JM |
705 | int r = -ENOMEM; |
706 | int i; | |
6aa8b732 | 707 | |
d89f5eff JK |
708 | if (!kvm) |
709 | return ERR_PTR(-ENOMEM); | |
710 | ||
e9ad4ec8 | 711 | spin_lock_init(&kvm->mmu_lock); |
f1f10076 | 712 | mmgrab(current->mm); |
e9ad4ec8 PB |
713 | kvm->mm = current->mm; |
714 | kvm_eventfd_init(kvm); | |
715 | mutex_init(&kvm->lock); | |
716 | mutex_init(&kvm->irq_lock); | |
717 | mutex_init(&kvm->slots_lock); | |
e9ad4ec8 PB |
718 | INIT_LIST_HEAD(&kvm->devices); |
719 | ||
9121923c JM |
720 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
721 | ||
8a44119a PB |
722 | if (init_srcu_struct(&kvm->srcu)) |
723 | goto out_err_no_srcu; | |
724 | if (init_srcu_struct(&kvm->irq_srcu)) | |
725 | goto out_err_no_irq_srcu; | |
726 | ||
e2d3fcaf | 727 | refcount_set(&kvm->users_count, 1); |
9121923c JM |
728 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
729 | struct kvm_memslots *slots = kvm_alloc_memslots(); | |
730 | ||
731 | if (!slots) | |
a97b0e77 | 732 | goto out_err_no_arch_destroy_vm; |
9121923c JM |
733 | /* Generations must be different for each address space. */ |
734 | slots->generation = i; | |
735 | rcu_assign_pointer(kvm->memslots[i], slots); | |
736 | } | |
737 | ||
738 | for (i = 0; i < KVM_NR_BUSES; i++) { | |
739 | rcu_assign_pointer(kvm->buses[i], | |
740 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); | |
741 | if (!kvm->buses[i]) | |
a97b0e77 | 742 | goto out_err_no_arch_destroy_vm; |
9121923c JM |
743 | } |
744 | ||
e08b9637 | 745 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 746 | if (r) |
a97b0e77 | 747 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
748 | |
749 | r = hardware_enable_all(); | |
750 | if (r) | |
719d93cd | 751 | goto out_err_no_disable; |
10474ae8 | 752 | |
c77dcacb | 753 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 754 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 755 | #endif |
6aa8b732 | 756 | |
74b5c5bf | 757 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
758 | if (r) |
759 | goto out_err_no_mmu_notifier; | |
760 | ||
761 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
762 | if (r) |
763 | goto out_err; | |
764 | ||
0d9ce162 | 765 | mutex_lock(&kvm_lock); |
5e58cfe4 | 766 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 767 | mutex_unlock(&kvm_lock); |
d89f5eff | 768 | |
2ecd9d29 PZ |
769 | preempt_notifier_inc(); |
770 | ||
59069e9e DM |
771 | /* |
772 | * When the fd passed to this ioctl() is opened it pins the module, | |
773 | * but try_module_get() also prevents getting a reference if the module | |
774 | * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait"). | |
775 | */ | |
776 | if (!try_module_get(kvm_chardev_ops.owner)) { | |
777 | r = -ENODEV; | |
778 | goto out_err; | |
779 | } | |
780 | ||
f17abe9a | 781 | return kvm; |
10474ae8 AG |
782 | |
783 | out_err: | |
1aa9b957 JS |
784 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
785 | if (kvm->mmu_notifier.ops) | |
786 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
787 | #endif | |
788 | out_err_no_mmu_notifier: | |
10474ae8 | 789 | hardware_disable_all(); |
719d93cd | 790 | out_err_no_disable: |
a97b0e77 | 791 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 792 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 793 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 794 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 795 | kfree(kvm_get_bus(kvm, i)); |
f481b069 | 796 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 797 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
8a44119a PB |
798 | cleanup_srcu_struct(&kvm->irq_srcu); |
799 | out_err_no_irq_srcu: | |
800 | cleanup_srcu_struct(&kvm->srcu); | |
801 | out_err_no_srcu: | |
d89f5eff | 802 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 803 | mmdrop(current->mm); |
10474ae8 | 804 | return ERR_PTR(r); |
f17abe9a AK |
805 | } |
806 | ||
07f0a7bd SW |
807 | static void kvm_destroy_devices(struct kvm *kvm) |
808 | { | |
e6e3b5a6 | 809 | struct kvm_device *dev, *tmp; |
07f0a7bd | 810 | |
a28ebea2 CD |
811 | /* |
812 | * We do not need to take the kvm->lock here, because nobody else | |
813 | * has a reference to the struct kvm at this point and therefore | |
814 | * cannot access the devices list anyhow. | |
815 | */ | |
e6e3b5a6 GT |
816 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
817 | list_del(&dev->vm_node); | |
07f0a7bd SW |
818 | dev->ops->destroy(dev); |
819 | } | |
820 | } | |
821 | ||
f17abe9a AK |
822 | static void kvm_destroy_vm(struct kvm *kvm) |
823 | { | |
e93f8a0f | 824 | int i; |
6d4e4c4f AK |
825 | struct mm_struct *mm = kvm->mm; |
826 | ||
286de8f6 | 827 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 828 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 829 | kvm_arch_sync_events(kvm); |
0d9ce162 | 830 | mutex_lock(&kvm_lock); |
133de902 | 831 | list_del(&kvm->vm_list); |
0d9ce162 | 832 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
833 | kvm_arch_pre_destroy_vm(kvm); |
834 | ||
399ec807 | 835 | kvm_free_irq_routing(kvm); |
df630b8c | 836 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 837 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 838 | |
4a12f951 CB |
839 | if (bus) |
840 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
841 | kvm->buses[i] = NULL; |
842 | } | |
980da6ce | 843 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
844 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
845 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 846 | #else |
2df72e9b | 847 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 848 | #endif |
d19a9cd2 | 849 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 850 | kvm_destroy_devices(kvm); |
f481b069 | 851 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 852 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
820b3fcd | 853 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
854 | cleanup_srcu_struct(&kvm->srcu); |
855 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 856 | preempt_notifier_dec(); |
10474ae8 | 857 | hardware_disable_all(); |
6d4e4c4f | 858 | mmdrop(mm); |
59069e9e | 859 | module_put(kvm_chardev_ops.owner); |
f17abe9a AK |
860 | } |
861 | ||
d39f13b0 IE |
862 | void kvm_get_kvm(struct kvm *kvm) |
863 | { | |
e3736c3e | 864 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
865 | } |
866 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
867 | ||
868 | void kvm_put_kvm(struct kvm *kvm) | |
869 | { | |
e3736c3e | 870 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
871 | kvm_destroy_vm(kvm); |
872 | } | |
873 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
874 | ||
875 | ||
f17abe9a AK |
876 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
877 | { | |
878 | struct kvm *kvm = filp->private_data; | |
879 | ||
721eecbf GH |
880 | kvm_irqfd_release(kvm); |
881 | ||
d39f13b0 | 882 | kvm_put_kvm(kvm); |
6aa8b732 AK |
883 | return 0; |
884 | } | |
885 | ||
515a0127 TY |
886 | /* |
887 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 888 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 889 | */ |
a36a57b1 TY |
890 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
891 | { | |
515a0127 | 892 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 893 | |
b12ce36a | 894 | memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
895 | if (!memslot->dirty_bitmap) |
896 | return -ENOMEM; | |
897 | ||
a36a57b1 TY |
898 | return 0; |
899 | } | |
900 | ||
bf3e05bc | 901 | /* |
0e60b079 IM |
902 | * Insert memslot and re-sort memslots based on their GFN, |
903 | * so binary search could be used to lookup GFN. | |
904 | * Sorting algorithm takes advantage of having initially | |
905 | * sorted array and known changed memslot position. | |
bf3e05bc | 906 | */ |
5cc15027 | 907 | static void update_memslots(struct kvm_memslots *slots, |
31fc4f95 WY |
908 | struct kvm_memory_slot *new, |
909 | enum kvm_mr_change change) | |
bf3e05bc | 910 | { |
8593176c PB |
911 | int id = new->id; |
912 | int i = slots->id_to_index[id]; | |
063584d4 | 913 | struct kvm_memory_slot *mslots = slots->memslots; |
f85e2cb5 | 914 | |
8593176c | 915 | WARN_ON(mslots[i].id != id); |
31fc4f95 WY |
916 | switch (change) { |
917 | case KVM_MR_CREATE: | |
918 | slots->used_slots++; | |
919 | WARN_ON(mslots[i].npages || !new->npages); | |
920 | break; | |
921 | case KVM_MR_DELETE: | |
922 | slots->used_slots--; | |
923 | WARN_ON(new->npages || !mslots[i].npages); | |
924 | break; | |
925 | default: | |
926 | break; | |
9c1a5d38 | 927 | } |
0e60b079 | 928 | |
7f379cff | 929 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
0e60b079 IM |
930 | new->base_gfn <= mslots[i + 1].base_gfn) { |
931 | if (!mslots[i + 1].npages) | |
932 | break; | |
7f379cff IM |
933 | mslots[i] = mslots[i + 1]; |
934 | slots->id_to_index[mslots[i].id] = i; | |
935 | i++; | |
936 | } | |
efbeec70 PB |
937 | |
938 | /* | |
939 | * The ">=" is needed when creating a slot with base_gfn == 0, | |
940 | * so that it moves before all those with base_gfn == npages == 0. | |
941 | * | |
942 | * On the other hand, if new->npages is zero, the above loop has | |
943 | * already left i pointing to the beginning of the empty part of | |
944 | * mslots, and the ">=" would move the hole backwards in this | |
945 | * case---which is wrong. So skip the loop when deleting a slot. | |
946 | */ | |
947 | if (new->npages) { | |
948 | while (i > 0 && | |
949 | new->base_gfn >= mslots[i - 1].base_gfn) { | |
950 | mslots[i] = mslots[i - 1]; | |
951 | slots->id_to_index[mslots[i].id] = i; | |
952 | i--; | |
953 | } | |
dbaff309 PB |
954 | } else |
955 | WARN_ON_ONCE(i != slots->used_slots); | |
f85e2cb5 | 956 | |
8593176c PB |
957 | mslots[i] = *new; |
958 | slots->id_to_index[mslots[i].id] = i; | |
bf3e05bc XG |
959 | } |
960 | ||
09170a49 | 961 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 962 | { |
4d8b81ab XG |
963 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
964 | ||
0f8a4de3 | 965 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
966 | valid_flags |= KVM_MEM_READONLY; |
967 | #endif | |
968 | ||
969 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
970 | return -EINVAL; |
971 | ||
972 | return 0; | |
973 | } | |
974 | ||
7ec4fb44 | 975 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 976 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 977 | { |
f481b069 | 978 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
361209e0 | 979 | u64 gen = old_memslots->generation; |
7ec4fb44 | 980 | |
361209e0 SC |
981 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
982 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 983 | |
f481b069 | 984 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
7ec4fb44 | 985 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 986 | |
ee3d1570 | 987 | /* |
361209e0 SC |
988 | * Increment the new memslot generation a second time, dropping the |
989 | * update in-progress flag and incrementing then generation based on | |
990 | * the number of address spaces. This provides a unique and easily | |
991 | * identifiable generation number while the memslots are in flux. | |
992 | */ | |
993 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
994 | ||
995 | /* | |
4bd518f1 PB |
996 | * Generations must be unique even across address spaces. We do not need |
997 | * a global counter for that, instead the generation space is evenly split | |
998 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
999 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1000 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1001 | */ |
164bf7e5 | 1002 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 1003 | |
15248258 | 1004 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1005 | |
15248258 | 1006 | slots->generation = gen; |
e59dbe09 TY |
1007 | |
1008 | return old_memslots; | |
7ec4fb44 GN |
1009 | } |
1010 | ||
6aa8b732 AK |
1011 | /* |
1012 | * Allocate some memory and give it an address in the guest physical address | |
1013 | * space. | |
1014 | * | |
1015 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1016 | * |
02d5d55b | 1017 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1018 | */ |
f78e0e2e | 1019 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 1020 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 1021 | { |
8234b22e | 1022 | int r; |
6aa8b732 | 1023 | gfn_t base_gfn; |
28bcb112 | 1024 | unsigned long npages; |
a843fac2 | 1025 | struct kvm_memory_slot *slot; |
6aa8b732 | 1026 | struct kvm_memory_slot old, new; |
b7f69c55 | 1027 | struct kvm_memslots *slots = NULL, *old_memslots; |
f481b069 | 1028 | int as_id, id; |
f64c0398 | 1029 | enum kvm_mr_change change; |
6aa8b732 | 1030 | |
a50d64d6 XG |
1031 | r = check_memory_region_flags(mem); |
1032 | if (r) | |
1033 | goto out; | |
1034 | ||
6aa8b732 | 1035 | r = -EINVAL; |
f481b069 PB |
1036 | as_id = mem->slot >> 16; |
1037 | id = (u16)mem->slot; | |
1038 | ||
6aa8b732 AK |
1039 | /* General sanity checks */ |
1040 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
1041 | goto out; | |
1042 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
1043 | goto out; | |
fa3d315a | 1044 | /* We can read the guest memory with __xxx_user() later on. */ |
f481b069 | 1045 | if ((id < KVM_USER_MEM_SLOTS) && |
fa3d315a | 1046 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
ce85d77c | 1047 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 1048 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
9e3bb6b6 | 1049 | mem->memory_size))) |
78749809 | 1050 | goto out; |
f481b069 | 1051 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
1052 | goto out; |
1053 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
1054 | goto out; | |
1055 | ||
f481b069 | 1056 | slot = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
6aa8b732 AK |
1057 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
1058 | npages = mem->memory_size >> PAGE_SHIFT; | |
1059 | ||
660c22c4 TY |
1060 | if (npages > KVM_MEM_MAX_NR_PAGES) |
1061 | goto out; | |
1062 | ||
a843fac2 | 1063 | new = old = *slot; |
6aa8b732 | 1064 | |
f481b069 | 1065 | new.id = id; |
6aa8b732 AK |
1066 | new.base_gfn = base_gfn; |
1067 | new.npages = npages; | |
1068 | new.flags = mem->flags; | |
1069 | ||
f64c0398 TY |
1070 | if (npages) { |
1071 | if (!old.npages) | |
1072 | change = KVM_MR_CREATE; | |
1073 | else { /* Modify an existing slot. */ | |
1074 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
1075 | (npages != old.npages) || |
1076 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
1077 | goto out; |
1078 | ||
1079 | if (base_gfn != old.base_gfn) | |
1080 | change = KVM_MR_MOVE; | |
1081 | else if (new.flags != old.flags) | |
1082 | change = KVM_MR_FLAGS_ONLY; | |
1083 | else { /* Nothing to change. */ | |
1084 | r = 0; | |
1085 | goto out; | |
1086 | } | |
1087 | } | |
09170a49 PB |
1088 | } else { |
1089 | if (!old.npages) | |
1090 | goto out; | |
1091 | ||
f64c0398 | 1092 | change = KVM_MR_DELETE; |
09170a49 PB |
1093 | new.base_gfn = 0; |
1094 | new.flags = 0; | |
1095 | } | |
6aa8b732 | 1096 | |
f64c0398 | 1097 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
1098 | /* Check for overlaps */ |
1099 | r = -EEXIST; | |
f481b069 | 1100 | kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) { |
b28676bb | 1101 | if (slot->id == id) |
0a706bee TY |
1102 | continue; |
1103 | if (!((base_gfn + npages <= slot->base_gfn) || | |
1104 | (base_gfn >= slot->base_gfn + slot->npages))) | |
1105 | goto out; | |
1106 | } | |
6aa8b732 | 1107 | } |
6aa8b732 | 1108 | |
6aa8b732 AK |
1109 | /* Free page dirty bitmap if unneeded */ |
1110 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 1111 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
1112 | |
1113 | r = -ENOMEM; | |
f64c0398 | 1114 | if (change == KVM_MR_CREATE) { |
189a2f7b | 1115 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 1116 | |
5587027c | 1117 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 1118 | goto out_free; |
6aa8b732 | 1119 | } |
ec04b260 | 1120 | |
6aa8b732 AK |
1121 | /* Allocate page dirty bitmap if needed */ |
1122 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 1123 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 1124 | goto out_free; |
6aa8b732 AK |
1125 | } |
1126 | ||
b12ce36a | 1127 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
f2a81036 PB |
1128 | if (!slots) |
1129 | goto out_free; | |
f481b069 | 1130 | memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots)); |
f2a81036 | 1131 | |
f64c0398 | 1132 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
f481b069 | 1133 | slot = id_to_memslot(slots, id); |
28a37544 XG |
1134 | slot->flags |= KVM_MEMSLOT_INVALID; |
1135 | ||
f481b069 | 1136 | old_memslots = install_new_memslots(kvm, as_id, slots); |
bc6678a3 | 1137 | |
12d6e753 MT |
1138 | /* From this point no new shadow pages pointing to a deleted, |
1139 | * or moved, memslot will be created. | |
bc6678a3 MT |
1140 | * |
1141 | * validation of sp->gfn happens in: | |
b7d409de XL |
1142 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) |
1143 | * - kvm_is_visible_gfn (mmu_check_roots) | |
bc6678a3 | 1144 | */ |
2df72e9b | 1145 | kvm_arch_flush_shadow_memslot(kvm, slot); |
f2a81036 PB |
1146 | |
1147 | /* | |
1148 | * We can re-use the old_memslots from above, the only difference | |
1149 | * from the currently installed memslots is the invalid flag. This | |
1150 | * will get overwritten by update_memslots anyway. | |
1151 | */ | |
b7f69c55 | 1152 | slots = old_memslots; |
bc6678a3 | 1153 | } |
34d4cb8f | 1154 | |
7b6195a9 | 1155 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 1156 | if (r) |
b7f69c55 | 1157 | goto out_slots; |
f7784b8e | 1158 | |
a47d2b07 | 1159 | /* actual memory is freed via old in kvm_free_memslot below */ |
f64c0398 | 1160 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 1161 | new.dirty_bitmap = NULL; |
db3fe4eb | 1162 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
1163 | } |
1164 | ||
31fc4f95 | 1165 | update_memslots(slots, &new, change); |
f481b069 | 1166 | old_memslots = install_new_memslots(kvm, as_id, slots); |
3ad82a7e | 1167 | |
f36f3f28 | 1168 | kvm_arch_commit_memory_region(kvm, mem, &old, &new, change); |
82ce2c96 | 1169 | |
a47d2b07 | 1170 | kvm_free_memslot(kvm, &old, &new); |
74496134 | 1171 | kvfree(old_memslots); |
6aa8b732 AK |
1172 | return 0; |
1173 | ||
e40f193f | 1174 | out_slots: |
74496134 | 1175 | kvfree(slots); |
f78e0e2e | 1176 | out_free: |
a47d2b07 | 1177 | kvm_free_memslot(kvm, &new, &old); |
6aa8b732 AK |
1178 | out: |
1179 | return r; | |
210c7c4d | 1180 | } |
f78e0e2e SY |
1181 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1182 | ||
1183 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1184 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1185 | { |
1186 | int r; | |
1187 | ||
79fac95e | 1188 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1189 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1190 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1191 | return r; |
1192 | } | |
210c7c4d IE |
1193 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1194 | ||
7940876e SH |
1195 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1196 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1197 | { |
f481b069 | 1198 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1199 | return -EINVAL; |
09170a49 | 1200 | |
47ae31e2 | 1201 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1202 | } |
1203 | ||
5bb064dc ZX |
1204 | int kvm_get_dirty_log(struct kvm *kvm, |
1205 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 | 1206 | { |
9f6b8029 | 1207 | struct kvm_memslots *slots; |
6aa8b732 | 1208 | struct kvm_memory_slot *memslot; |
843574a3 | 1209 | int i, as_id, id; |
87bf6e7d | 1210 | unsigned long n; |
6aa8b732 AK |
1211 | unsigned long any = 0; |
1212 | ||
f481b069 PB |
1213 | as_id = log->slot >> 16; |
1214 | id = (u16)log->slot; | |
1215 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 1216 | return -EINVAL; |
6aa8b732 | 1217 | |
f481b069 PB |
1218 | slots = __kvm_memslots(kvm, as_id); |
1219 | memslot = id_to_memslot(slots, id); | |
6aa8b732 | 1220 | if (!memslot->dirty_bitmap) |
843574a3 | 1221 | return -ENOENT; |
6aa8b732 | 1222 | |
87bf6e7d | 1223 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 1224 | |
cd1a4a98 | 1225 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
1226 | any = memslot->dirty_bitmap[i]; |
1227 | ||
6aa8b732 | 1228 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) |
843574a3 | 1229 | return -EFAULT; |
6aa8b732 | 1230 | |
5bb064dc ZX |
1231 | if (any) |
1232 | *is_dirty = 1; | |
843574a3 | 1233 | return 0; |
6aa8b732 | 1234 | } |
2ba9f0d8 | 1235 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1236 | |
ba0513b5 MS |
1237 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
1238 | /** | |
b8b00220 | 1239 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 1240 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
1241 | * @kvm: pointer to kvm instance |
1242 | * @log: slot id and address to which we copy the log | |
b8b00220 | 1243 | * @flush: true if TLB flush is needed by caller |
ba0513b5 MS |
1244 | * |
1245 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1246 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1247 | * following order: | |
1248 | * | |
1249 | * 1. Take a snapshot of the bit and clear it if needed. | |
1250 | * 2. Write protect the corresponding page. | |
1251 | * 3. Copy the snapshot to the userspace. | |
1252 | * 4. Upon return caller flushes TLB's if needed. | |
1253 | * | |
1254 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1255 | * entry. This is not a problem because the page is reported dirty using | |
1256 | * the snapshot taken before and step 4 ensures that writes done after | |
1257 | * exiting to userspace will be logged for the next call. | |
1258 | * | |
1259 | */ | |
1260 | int kvm_get_dirty_log_protect(struct kvm *kvm, | |
8fe65a82 | 1261 | struct kvm_dirty_log *log, bool *flush) |
ba0513b5 | 1262 | { |
9f6b8029 | 1263 | struct kvm_memslots *slots; |
ba0513b5 | 1264 | struct kvm_memory_slot *memslot; |
58d6db34 | 1265 | int i, as_id, id; |
ba0513b5 MS |
1266 | unsigned long n; |
1267 | unsigned long *dirty_bitmap; | |
1268 | unsigned long *dirty_bitmap_buffer; | |
1269 | ||
f481b069 PB |
1270 | as_id = log->slot >> 16; |
1271 | id = (u16)log->slot; | |
1272 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 1273 | return -EINVAL; |
ba0513b5 | 1274 | |
f481b069 PB |
1275 | slots = __kvm_memslots(kvm, as_id); |
1276 | memslot = id_to_memslot(slots, id); | |
ba0513b5 MS |
1277 | |
1278 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 1279 | if (!dirty_bitmap) |
58d6db34 | 1280 | return -ENOENT; |
ba0513b5 MS |
1281 | |
1282 | n = kvm_dirty_bitmap_bytes(memslot); | |
2a31b9db PB |
1283 | *flush = false; |
1284 | if (kvm->manual_dirty_log_protect) { | |
1285 | /* | |
1286 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
1287 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
1288 | * is some code duplication between this function and | |
1289 | * kvm_get_dirty_log, but hopefully all architecture | |
1290 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
1291 | * can be eliminated. | |
1292 | */ | |
1293 | dirty_bitmap_buffer = dirty_bitmap; | |
1294 | } else { | |
1295 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
1296 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 1297 | |
2a31b9db PB |
1298 | spin_lock(&kvm->mmu_lock); |
1299 | for (i = 0; i < n / sizeof(long); i++) { | |
1300 | unsigned long mask; | |
1301 | gfn_t offset; | |
ba0513b5 | 1302 | |
2a31b9db PB |
1303 | if (!dirty_bitmap[i]) |
1304 | continue; | |
1305 | ||
1306 | *flush = true; | |
1307 | mask = xchg(&dirty_bitmap[i], 0); | |
1308 | dirty_bitmap_buffer[i] = mask; | |
1309 | ||
a67794ca LT |
1310 | offset = i * BITS_PER_LONG; |
1311 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1312 | offset, mask); | |
2a31b9db PB |
1313 | } |
1314 | spin_unlock(&kvm->mmu_lock); | |
1315 | } | |
1316 | ||
1317 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) | |
1318 | return -EFAULT; | |
1319 | return 0; | |
1320 | } | |
1321 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); | |
1322 | ||
1323 | /** | |
1324 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
1325 | * and reenable dirty page tracking for the corresponding pages. | |
1326 | * @kvm: pointer to kvm instance | |
1327 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
b8b00220 | 1328 | * @flush: true if TLB flush is needed by caller |
2a31b9db PB |
1329 | */ |
1330 | int kvm_clear_dirty_log_protect(struct kvm *kvm, | |
1331 | struct kvm_clear_dirty_log *log, bool *flush) | |
1332 | { | |
1333 | struct kvm_memslots *slots; | |
1334 | struct kvm_memory_slot *memslot; | |
98938aa8 | 1335 | int as_id, id; |
2a31b9db | 1336 | gfn_t offset; |
98938aa8 | 1337 | unsigned long i, n; |
2a31b9db PB |
1338 | unsigned long *dirty_bitmap; |
1339 | unsigned long *dirty_bitmap_buffer; | |
1340 | ||
1341 | as_id = log->slot >> 16; | |
1342 | id = (u16)log->slot; | |
1343 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
1344 | return -EINVAL; | |
1345 | ||
76d58e0f | 1346 | if (log->first_page & 63) |
2a31b9db PB |
1347 | return -EINVAL; |
1348 | ||
1349 | slots = __kvm_memslots(kvm, as_id); | |
1350 | memslot = id_to_memslot(slots, id); | |
1351 | ||
1352 | dirty_bitmap = memslot->dirty_bitmap; | |
1353 | if (!dirty_bitmap) | |
1354 | return -ENOENT; | |
1355 | ||
4ddc9204 | 1356 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
1357 | |
1358 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
1359 | log->num_pages > memslot->npages - log->first_page || |
1360 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
1361 | return -EINVAL; | |
98938aa8 | 1362 | |
8fe65a82 | 1363 | *flush = false; |
2a31b9db PB |
1364 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
1365 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
1366 | return -EFAULT; | |
ba0513b5 | 1367 | |
2a31b9db | 1368 | spin_lock(&kvm->mmu_lock); |
53eac7a8 PX |
1369 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
1370 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
1371 | i++, offset += BITS_PER_LONG) { |
1372 | unsigned long mask = *dirty_bitmap_buffer++; | |
1373 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
1374 | if (!mask) | |
ba0513b5 MS |
1375 | continue; |
1376 | ||
2a31b9db | 1377 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 1378 | |
2a31b9db PB |
1379 | /* |
1380 | * mask contains the bits that really have been cleared. This | |
1381 | * never includes any bits beyond the length of the memslot (if | |
1382 | * the length is not aligned to 64 pages), therefore it is not | |
1383 | * a problem if userspace sets them in log->dirty_bitmap. | |
1384 | */ | |
58d2930f | 1385 | if (mask) { |
2a31b9db | 1386 | *flush = true; |
58d2930f TY |
1387 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
1388 | offset, mask); | |
1389 | } | |
ba0513b5 | 1390 | } |
ba0513b5 | 1391 | spin_unlock(&kvm->mmu_lock); |
2a31b9db | 1392 | |
58d6db34 | 1393 | return 0; |
ba0513b5 | 1394 | } |
2a31b9db | 1395 | EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect); |
ba0513b5 MS |
1396 | #endif |
1397 | ||
db3fe4eb TY |
1398 | bool kvm_largepages_enabled(void) |
1399 | { | |
1400 | return largepages_enabled; | |
1401 | } | |
1402 | ||
54dee993 MT |
1403 | void kvm_disable_largepages(void) |
1404 | { | |
1405 | largepages_enabled = false; | |
1406 | } | |
1407 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
1408 | ||
49c7754c GN |
1409 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1410 | { | |
1411 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1412 | } | |
a1f4d395 | 1413 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1414 | |
8e73485c PB |
1415 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
1416 | { | |
1417 | return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn); | |
1418 | } | |
1419 | ||
33e94154 | 1420 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 1421 | { |
bf3e05bc | 1422 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1423 | |
bbacc0c1 | 1424 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc | 1425 | memslot->flags & KVM_MEMSLOT_INVALID) |
33e94154 | 1426 | return false; |
e0d62c7f | 1427 | |
33e94154 | 1428 | return true; |
e0d62c7f IE |
1429 | } |
1430 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1431 | ||
5a50de0c | 1432 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
1433 | { |
1434 | struct vm_area_struct *vma; | |
1435 | unsigned long addr, size; | |
1436 | ||
1437 | size = PAGE_SIZE; | |
1438 | ||
43d37970 | 1439 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
1440 | if (kvm_is_error_hva(addr)) |
1441 | return PAGE_SIZE; | |
1442 | ||
1443 | down_read(¤t->mm->mmap_sem); | |
1444 | vma = find_vma(current->mm, addr); | |
1445 | if (!vma) | |
1446 | goto out; | |
1447 | ||
1448 | size = vma_kernel_pagesize(vma); | |
1449 | ||
1450 | out: | |
1451 | up_read(¤t->mm->mmap_sem); | |
1452 | ||
1453 | return size; | |
1454 | } | |
1455 | ||
4d8b81ab XG |
1456 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1457 | { | |
1458 | return slot->flags & KVM_MEM_READONLY; | |
1459 | } | |
1460 | ||
4d8b81ab XG |
1461 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1462 | gfn_t *nr_pages, bool write) | |
539cb660 | 1463 | { |
bc6678a3 | 1464 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1465 | return KVM_HVA_ERR_BAD; |
48987781 | 1466 | |
4d8b81ab XG |
1467 | if (memslot_is_readonly(slot) && write) |
1468 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1469 | |
1470 | if (nr_pages) | |
1471 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1472 | ||
4d8b81ab | 1473 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1474 | } |
48987781 | 1475 | |
4d8b81ab XG |
1476 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1477 | gfn_t *nr_pages) | |
1478 | { | |
1479 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1480 | } |
48987781 | 1481 | |
4d8b81ab | 1482 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1483 | gfn_t gfn) |
4d8b81ab XG |
1484 | { |
1485 | return gfn_to_hva_many(slot, gfn, NULL); | |
1486 | } | |
1487 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1488 | ||
48987781 XG |
1489 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1490 | { | |
49c7754c | 1491 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1492 | } |
0d150298 | 1493 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1494 | |
8e73485c PB |
1495 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
1496 | { | |
1497 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
1498 | } | |
1499 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
1500 | ||
86ab8cff | 1501 | /* |
970c0d4b WY |
1502 | * Return the hva of a @gfn and the R/W attribute if possible. |
1503 | * | |
1504 | * @slot: the kvm_memory_slot which contains @gfn | |
1505 | * @gfn: the gfn to be translated | |
1506 | * @writable: used to return the read/write attribute of the @slot if the hva | |
1507 | * is valid and @writable is not NULL | |
86ab8cff | 1508 | */ |
64d83126 CD |
1509 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
1510 | gfn_t gfn, bool *writable) | |
86ab8cff | 1511 | { |
a2ac07fe GN |
1512 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1513 | ||
1514 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1515 | *writable = !memslot_is_readonly(slot); |
1516 | ||
a2ac07fe | 1517 | return hva; |
86ab8cff XG |
1518 | } |
1519 | ||
64d83126 CD |
1520 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
1521 | { | |
1522 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1523 | ||
1524 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1525 | } | |
1526 | ||
8e73485c PB |
1527 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
1528 | { | |
1529 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1530 | ||
1531 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1532 | } | |
1533 | ||
fafc3dba HY |
1534 | static inline int check_user_page_hwpoison(unsigned long addr) |
1535 | { | |
0d731759 | 1536 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 1537 | |
0d731759 | 1538 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
1539 | return rc == -EHWPOISON; |
1540 | } | |
1541 | ||
2fc84311 | 1542 | /* |
b9b33da2 PB |
1543 | * The fast path to get the writable pfn which will be stored in @pfn, |
1544 | * true indicates success, otherwise false is returned. It's also the | |
1545 | * only part that runs if we can are in atomic context. | |
2fc84311 | 1546 | */ |
b9b33da2 PB |
1547 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
1548 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 1549 | { |
8d4e1288 | 1550 | struct page *page[1]; |
2fc84311 | 1551 | int npages; |
954bbbc2 | 1552 | |
12ce13fe XG |
1553 | /* |
1554 | * Fast pin a writable pfn only if it is a write fault request | |
1555 | * or the caller allows to map a writable pfn for a read fault | |
1556 | * request. | |
1557 | */ | |
1558 | if (!(write_fault || writable)) | |
1559 | return false; | |
612819c3 | 1560 | |
2fc84311 XG |
1561 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1562 | if (npages == 1) { | |
1563 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1564 | |
2fc84311 XG |
1565 | if (writable) |
1566 | *writable = true; | |
1567 | return true; | |
1568 | } | |
af585b92 | 1569 | |
2fc84311 XG |
1570 | return false; |
1571 | } | |
612819c3 | 1572 | |
2fc84311 XG |
1573 | /* |
1574 | * The slow path to get the pfn of the specified host virtual address, | |
1575 | * 1 indicates success, -errno is returned if error is detected. | |
1576 | */ | |
1577 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 1578 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 1579 | { |
ce53053c AV |
1580 | unsigned int flags = FOLL_HWPOISON; |
1581 | struct page *page; | |
2fc84311 | 1582 | int npages = 0; |
612819c3 | 1583 | |
2fc84311 XG |
1584 | might_sleep(); |
1585 | ||
1586 | if (writable) | |
1587 | *writable = write_fault; | |
1588 | ||
ce53053c AV |
1589 | if (write_fault) |
1590 | flags |= FOLL_WRITE; | |
1591 | if (async) | |
1592 | flags |= FOLL_NOWAIT; | |
d4944b0e | 1593 | |
ce53053c | 1594 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
1595 | if (npages != 1) |
1596 | return npages; | |
1597 | ||
1598 | /* map read fault as writable if possible */ | |
12ce13fe | 1599 | if (unlikely(!write_fault) && writable) { |
ce53053c | 1600 | struct page *wpage; |
2fc84311 | 1601 | |
ce53053c | 1602 | if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) { |
2fc84311 | 1603 | *writable = true; |
ce53053c AV |
1604 | put_page(page); |
1605 | page = wpage; | |
612819c3 | 1606 | } |
887c08ac | 1607 | } |
ce53053c | 1608 | *pfn = page_to_pfn(page); |
2fc84311 XG |
1609 | return npages; |
1610 | } | |
539cb660 | 1611 | |
4d8b81ab XG |
1612 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1613 | { | |
1614 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1615 | return false; | |
2e2e3738 | 1616 | |
4d8b81ab XG |
1617 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1618 | return false; | |
887c08ac | 1619 | |
4d8b81ab XG |
1620 | return true; |
1621 | } | |
bf998156 | 1622 | |
85f22599 NP |
1623 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
1624 | { | |
1625 | if (kvm_is_reserved_pfn(pfn)) | |
1626 | return 1; | |
1627 | return get_page_unless_zero(pfn_to_page(pfn)); | |
1628 | } | |
1629 | ||
92176a8e PB |
1630 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1631 | unsigned long addr, bool *async, | |
a340b3e2 KA |
1632 | bool write_fault, bool *writable, |
1633 | kvm_pfn_t *p_pfn) | |
92176a8e | 1634 | { |
1fcf209a | 1635 | kvm_pfn_t pfn; |
8d190fb2 PB |
1636 | pte_t *ptep; |
1637 | spinlock_t *ptl; | |
add6a0cd PB |
1638 | int r; |
1639 | ||
5f369a12 | 1640 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
1641 | if (r) { |
1642 | /* | |
1643 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
1644 | * not call the fault handler, so do it here. | |
1645 | */ | |
1646 | bool unlocked = false; | |
1647 | r = fixup_user_fault(current, current->mm, addr, | |
1648 | (write_fault ? FAULT_FLAG_WRITE : 0), | |
1649 | &unlocked); | |
1650 | if (unlocked) | |
1651 | return -EAGAIN; | |
1652 | if (r) | |
1653 | return r; | |
1654 | ||
5f369a12 | 1655 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
1656 | if (r) |
1657 | return r; | |
8d190fb2 | 1658 | } |
add6a0cd | 1659 | |
8d190fb2 PB |
1660 | if (write_fault && !pte_write(*ptep)) { |
1661 | pfn = KVM_PFN_ERR_RO_FAULT; | |
1662 | goto out; | |
add6a0cd PB |
1663 | } |
1664 | ||
a340b3e2 | 1665 | if (writable) |
8d190fb2 PB |
1666 | *writable = pte_write(*ptep); |
1667 | pfn = pte_pfn(*ptep); | |
add6a0cd PB |
1668 | |
1669 | /* | |
1670 | * Get a reference here because callers of *hva_to_pfn* and | |
1671 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
1672 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
1673 | * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will | |
1674 | * simply do nothing for reserved pfns. | |
1675 | * | |
1676 | * Whoever called remap_pfn_range is also going to call e.g. | |
1677 | * unmap_mapping_range before the underlying pages are freed, | |
1678 | * causing a call to our MMU notifier. | |
85f22599 NP |
1679 | * |
1680 | * Certain IO or PFNMAP mappings can be backed with valid | |
1681 | * struct pages, but be allocated without refcounting e.g., | |
1682 | * tail pages of non-compound higher order allocations, which | |
1683 | * would then underflow the refcount when the caller does the | |
1684 | * required put_page. Don't allow those pages here. | |
add6a0cd | 1685 | */ |
85f22599 NP |
1686 | if (!kvm_try_get_pfn(pfn)) |
1687 | r = -EFAULT; | |
add6a0cd | 1688 | |
8d190fb2 PB |
1689 | out: |
1690 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 1691 | *p_pfn = pfn; |
85f22599 NP |
1692 | |
1693 | return r; | |
92176a8e PB |
1694 | } |
1695 | ||
12ce13fe XG |
1696 | /* |
1697 | * Pin guest page in memory and return its pfn. | |
1698 | * @addr: host virtual address which maps memory to the guest | |
1699 | * @atomic: whether this function can sleep | |
1700 | * @async: whether this function need to wait IO complete if the | |
1701 | * host page is not in the memory | |
1702 | * @write_fault: whether we should get a writable host page | |
1703 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1704 | * | |
1705 | * The function will map a writable host page for these two cases: | |
1706 | * 1): @write_fault = true | |
1707 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1708 | * whether the mapping is writable. | |
1709 | */ | |
ba049e93 | 1710 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
1711 | bool write_fault, bool *writable) |
1712 | { | |
1713 | struct vm_area_struct *vma; | |
ba049e93 | 1714 | kvm_pfn_t pfn = 0; |
92176a8e | 1715 | int npages, r; |
2e2e3738 | 1716 | |
2fc84311 XG |
1717 | /* we can do it either atomically or asynchronously, not both */ |
1718 | BUG_ON(atomic && async); | |
8d4e1288 | 1719 | |
b9b33da2 | 1720 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
1721 | return pfn; |
1722 | ||
1723 | if (atomic) | |
1724 | return KVM_PFN_ERR_FAULT; | |
1725 | ||
1726 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1727 | if (npages == 1) | |
1728 | return pfn; | |
8d4e1288 | 1729 | |
2fc84311 XG |
1730 | down_read(¤t->mm->mmap_sem); |
1731 | if (npages == -EHWPOISON || | |
1732 | (!async && check_user_page_hwpoison(addr))) { | |
1733 | pfn = KVM_PFN_ERR_HWPOISON; | |
1734 | goto exit; | |
1735 | } | |
1736 | ||
add6a0cd | 1737 | retry: |
2fc84311 XG |
1738 | vma = find_vma_intersection(current->mm, addr, addr + 1); |
1739 | ||
1740 | if (vma == NULL) | |
1741 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 1742 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
a340b3e2 | 1743 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn); |
add6a0cd PB |
1744 | if (r == -EAGAIN) |
1745 | goto retry; | |
92176a8e PB |
1746 | if (r < 0) |
1747 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 1748 | } else { |
4d8b81ab | 1749 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1750 | *async = true; |
1751 | pfn = KVM_PFN_ERR_FAULT; | |
1752 | } | |
1753 | exit: | |
1754 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1755 | return pfn; |
35149e21 AL |
1756 | } |
1757 | ||
ba049e93 DW |
1758 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
1759 | bool atomic, bool *async, bool write_fault, | |
1760 | bool *writable) | |
887c08ac | 1761 | { |
4d8b81ab XG |
1762 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1763 | ||
b2740d35 PB |
1764 | if (addr == KVM_HVA_ERR_RO_BAD) { |
1765 | if (writable) | |
1766 | *writable = false; | |
4d8b81ab | 1767 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 1768 | } |
4d8b81ab | 1769 | |
b2740d35 PB |
1770 | if (kvm_is_error_hva(addr)) { |
1771 | if (writable) | |
1772 | *writable = false; | |
81c52c56 | 1773 | return KVM_PFN_NOSLOT; |
b2740d35 | 1774 | } |
4d8b81ab XG |
1775 | |
1776 | /* Do not map writable pfn in the readonly memslot. */ | |
1777 | if (writable && memslot_is_readonly(slot)) { | |
1778 | *writable = false; | |
1779 | writable = NULL; | |
1780 | } | |
1781 | ||
1782 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1783 | writable); | |
887c08ac | 1784 | } |
3520469d | 1785 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 1786 | |
ba049e93 | 1787 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
1788 | bool *writable) |
1789 | { | |
e37afc6e PB |
1790 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
1791 | write_fault, writable); | |
612819c3 MT |
1792 | } |
1793 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1794 | ||
ba049e93 | 1795 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1796 | { |
4d8b81ab | 1797 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f | 1798 | } |
e37afc6e | 1799 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 1800 | |
ba049e93 | 1801 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1802 | { |
4d8b81ab | 1803 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1804 | } |
037d92dc | 1805 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1806 | |
ba049e93 | 1807 | kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1808 | { |
1809 | return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn); | |
1810 | } | |
1811 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1812 | ||
ba049e93 | 1813 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1814 | { |
1815 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1816 | } | |
1817 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
1818 | ||
ba049e93 | 1819 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1820 | { |
1821 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
1822 | } | |
1823 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
1824 | ||
ba049e93 | 1825 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1826 | { |
1827 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1828 | } | |
1829 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
1830 | ||
d9ef13c2 PB |
1831 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1832 | struct page **pages, int nr_pages) | |
48987781 XG |
1833 | { |
1834 | unsigned long addr; | |
076b925d | 1835 | gfn_t entry = 0; |
48987781 | 1836 | |
d9ef13c2 | 1837 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
1838 | if (kvm_is_error_hva(addr)) |
1839 | return -1; | |
1840 | ||
1841 | if (entry < nr_pages) | |
1842 | return 0; | |
1843 | ||
1844 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1845 | } | |
1846 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1847 | ||
ba049e93 | 1848 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 1849 | { |
81c52c56 | 1850 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1851 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1852 | |
bf4bea8e | 1853 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 1854 | WARN_ON(1); |
6cede2e6 | 1855 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1856 | } |
a2766325 XG |
1857 | |
1858 | return pfn_to_page(pfn); | |
1859 | } | |
1860 | ||
35149e21 AL |
1861 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1862 | { | |
ba049e93 | 1863 | kvm_pfn_t pfn; |
2e2e3738 AL |
1864 | |
1865 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1866 | |
a2766325 | 1867 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
1868 | } |
1869 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
1870 | ||
7cf07ad4 BO |
1871 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache) |
1872 | { | |
1873 | if (pfn == 0) | |
1874 | return; | |
1875 | ||
1876 | if (cache) | |
1877 | cache->pfn = cache->gfn = 0; | |
1878 | ||
1879 | if (dirty) | |
1880 | kvm_release_pfn_dirty(pfn); | |
1881 | else | |
1882 | kvm_release_pfn_clean(pfn); | |
1883 | } | |
1884 | ||
1885 | static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn, | |
1886 | struct gfn_to_pfn_cache *cache, u64 gen) | |
1887 | { | |
1888 | kvm_release_pfn(cache->pfn, cache->dirty, cache); | |
1889 | ||
1890 | cache->pfn = gfn_to_pfn_memslot(slot, gfn); | |
1891 | cache->gfn = gfn; | |
1892 | cache->dirty = false; | |
1893 | cache->generation = gen; | |
1894 | } | |
1895 | ||
3490a3c3 | 1896 | static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn, |
7cf07ad4 BO |
1897 | struct kvm_host_map *map, |
1898 | struct gfn_to_pfn_cache *cache, | |
1899 | bool atomic) | |
e45adf66 KA |
1900 | { |
1901 | kvm_pfn_t pfn; | |
1902 | void *hva = NULL; | |
1903 | struct page *page = KVM_UNMAPPED_PAGE; | |
3490a3c3 | 1904 | struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn); |
7cf07ad4 | 1905 | u64 gen = slots->generation; |
e45adf66 KA |
1906 | |
1907 | if (!map) | |
1908 | return -EINVAL; | |
1909 | ||
7cf07ad4 BO |
1910 | if (cache) { |
1911 | if (!cache->pfn || cache->gfn != gfn || | |
1912 | cache->generation != gen) { | |
1913 | if (atomic) | |
1914 | return -EAGAIN; | |
1915 | kvm_cache_gfn_to_pfn(slot, gfn, cache, gen); | |
1916 | } | |
1917 | pfn = cache->pfn; | |
1918 | } else { | |
1919 | if (atomic) | |
1920 | return -EAGAIN; | |
1921 | pfn = gfn_to_pfn_memslot(slot, gfn); | |
1922 | } | |
e45adf66 KA |
1923 | if (is_error_noslot_pfn(pfn)) |
1924 | return -EINVAL; | |
1925 | ||
1926 | if (pfn_valid(pfn)) { | |
1927 | page = pfn_to_page(pfn); | |
7cf07ad4 BO |
1928 | if (atomic) |
1929 | hva = kmap_atomic(page); | |
1930 | else | |
1931 | hva = kmap(page); | |
d30b214d | 1932 | #ifdef CONFIG_HAS_IOMEM |
7cf07ad4 | 1933 | } else if (!atomic) { |
e45adf66 | 1934 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
7cf07ad4 BO |
1935 | } else { |
1936 | return -EINVAL; | |
d30b214d | 1937 | #endif |
e45adf66 KA |
1938 | } |
1939 | ||
1940 | if (!hva) | |
1941 | return -EFAULT; | |
1942 | ||
1943 | map->page = page; | |
1944 | map->hva = hva; | |
1945 | map->pfn = pfn; | |
1946 | map->gfn = gfn; | |
1947 | ||
1948 | return 0; | |
1949 | } | |
1950 | ||
7cf07ad4 BO |
1951 | int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map, |
1952 | struct gfn_to_pfn_cache *cache, bool atomic) | |
3490a3c3 | 1953 | { |
7cf07ad4 BO |
1954 | return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map, |
1955 | cache, atomic); | |
3490a3c3 BO |
1956 | } |
1957 | EXPORT_SYMBOL_GPL(kvm_map_gfn); | |
1958 | ||
e45adf66 KA |
1959 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
1960 | { | |
7cf07ad4 BO |
1961 | return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map, |
1962 | NULL, false); | |
e45adf66 KA |
1963 | } |
1964 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); | |
1965 | ||
3490a3c3 | 1966 | static void __kvm_unmap_gfn(struct kvm_memory_slot *memslot, |
7cf07ad4 BO |
1967 | struct kvm_host_map *map, |
1968 | struct gfn_to_pfn_cache *cache, | |
1969 | bool dirty, bool atomic) | |
e45adf66 KA |
1970 | { |
1971 | if (!map) | |
1972 | return; | |
1973 | ||
1974 | if (!map->hva) | |
1975 | return; | |
1976 | ||
7cf07ad4 BO |
1977 | if (map->page != KVM_UNMAPPED_PAGE) { |
1978 | if (atomic) | |
1979 | kunmap_atomic(map->hva); | |
1980 | else | |
1981 | kunmap(map->page); | |
1982 | } | |
eb1f2f38 | 1983 | #ifdef CONFIG_HAS_IOMEM |
7cf07ad4 | 1984 | else if (!atomic) |
e45adf66 | 1985 | memunmap(map->hva); |
7cf07ad4 BO |
1986 | else |
1987 | WARN_ONCE(1, "Unexpected unmapping in atomic context"); | |
eb1f2f38 | 1988 | #endif |
e45adf66 | 1989 | |
7cf07ad4 | 1990 | if (dirty) |
3490a3c3 | 1991 | mark_page_dirty_in_slot(memslot, map->gfn); |
7cf07ad4 BO |
1992 | |
1993 | if (cache) | |
1994 | cache->dirty |= dirty; | |
1995 | else | |
1996 | kvm_release_pfn(map->pfn, dirty, NULL); | |
e45adf66 KA |
1997 | |
1998 | map->hva = NULL; | |
1999 | map->page = NULL; | |
2000 | } | |
3490a3c3 | 2001 | |
7cf07ad4 BO |
2002 | int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map, |
2003 | struct gfn_to_pfn_cache *cache, bool dirty, bool atomic) | |
3490a3c3 | 2004 | { |
7cf07ad4 BO |
2005 | __kvm_unmap_gfn(gfn_to_memslot(vcpu->kvm, map->gfn), map, |
2006 | cache, dirty, atomic); | |
3490a3c3 BO |
2007 | return 0; |
2008 | } | |
2009 | EXPORT_SYMBOL_GPL(kvm_unmap_gfn); | |
2010 | ||
2011 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) | |
2012 | { | |
7cf07ad4 BO |
2013 | __kvm_unmap_gfn(kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), map, NULL, |
2014 | dirty, false); | |
3490a3c3 | 2015 | } |
e45adf66 KA |
2016 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); |
2017 | ||
8e73485c PB |
2018 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
2019 | { | |
ba049e93 | 2020 | kvm_pfn_t pfn; |
8e73485c PB |
2021 | |
2022 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
2023 | ||
2024 | return kvm_pfn_to_page(pfn); | |
2025 | } | |
2026 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
2027 | ||
b4231d61 IE |
2028 | void kvm_release_page_clean(struct page *page) |
2029 | { | |
32cad84f XG |
2030 | WARN_ON(is_error_page(page)); |
2031 | ||
35149e21 | 2032 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
2033 | } |
2034 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
2035 | ||
ba049e93 | 2036 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 2037 | { |
bf4bea8e | 2038 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 2039 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
2040 | } |
2041 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
2042 | ||
b4231d61 | 2043 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 2044 | { |
a2766325 XG |
2045 | WARN_ON(is_error_page(page)); |
2046 | ||
35149e21 AL |
2047 | kvm_release_pfn_dirty(page_to_pfn(page)); |
2048 | } | |
2049 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
2050 | ||
f7a6509f | 2051 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
2052 | { |
2053 | kvm_set_pfn_dirty(pfn); | |
2054 | kvm_release_pfn_clean(pfn); | |
2055 | } | |
f7a6509f | 2056 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 2057 | |
ba049e93 | 2058 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2059 | { |
a78986aa | 2060 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) { |
2e2e3738 | 2061 | struct page *page = pfn_to_page(pfn); |
f95ef0cd | 2062 | |
8f946da7 | 2063 | SetPageDirty(page); |
2e2e3738 | 2064 | } |
8a7ae055 | 2065 | } |
35149e21 AL |
2066 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
2067 | ||
ba049e93 | 2068 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 2069 | { |
a78986aa | 2070 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2e2e3738 | 2071 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
2072 | } |
2073 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
2074 | ||
ba049e93 | 2075 | void kvm_get_pfn(kvm_pfn_t pfn) |
35149e21 | 2076 | { |
bf4bea8e | 2077 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 2078 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
2079 | } |
2080 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 2081 | |
195aefde IE |
2082 | static int next_segment(unsigned long len, int offset) |
2083 | { | |
2084 | if (len > PAGE_SIZE - offset) | |
2085 | return PAGE_SIZE - offset; | |
2086 | else | |
2087 | return len; | |
2088 | } | |
2089 | ||
8e73485c PB |
2090 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
2091 | void *data, int offset, int len) | |
195aefde | 2092 | { |
e0506bcb IE |
2093 | int r; |
2094 | unsigned long addr; | |
195aefde | 2095 | |
8e73485c | 2096 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
2097 | if (kvm_is_error_hva(addr)) |
2098 | return -EFAULT; | |
3180a7fc | 2099 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 2100 | if (r) |
195aefde | 2101 | return -EFAULT; |
195aefde IE |
2102 | return 0; |
2103 | } | |
8e73485c PB |
2104 | |
2105 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
2106 | int len) | |
2107 | { | |
2108 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2109 | ||
2110 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2111 | } | |
195aefde IE |
2112 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
2113 | ||
8e73485c PB |
2114 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
2115 | int offset, int len) | |
2116 | { | |
2117 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2118 | ||
2119 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2120 | } | |
2121 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
2122 | ||
195aefde IE |
2123 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
2124 | { | |
2125 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2126 | int seg; | |
2127 | int offset = offset_in_page(gpa); | |
2128 | int ret; | |
2129 | ||
2130 | while ((seg = next_segment(len, offset)) != 0) { | |
2131 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
2132 | if (ret < 0) | |
2133 | return ret; | |
2134 | offset = 0; | |
2135 | len -= seg; | |
2136 | data += seg; | |
2137 | ++gfn; | |
2138 | } | |
2139 | return 0; | |
2140 | } | |
2141 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
2142 | ||
8e73485c | 2143 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 2144 | { |
7ec54588 | 2145 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 2146 | int seg; |
7ec54588 | 2147 | int offset = offset_in_page(gpa); |
8e73485c PB |
2148 | int ret; |
2149 | ||
2150 | while ((seg = next_segment(len, offset)) != 0) { | |
2151 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
2152 | if (ret < 0) | |
2153 | return ret; | |
2154 | offset = 0; | |
2155 | len -= seg; | |
2156 | data += seg; | |
2157 | ++gfn; | |
2158 | } | |
2159 | return 0; | |
2160 | } | |
2161 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 2162 | |
8e73485c PB |
2163 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2164 | void *data, int offset, unsigned long len) | |
2165 | { | |
2166 | int r; | |
2167 | unsigned long addr; | |
2168 | ||
2169 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
2170 | if (kvm_is_error_hva(addr)) |
2171 | return -EFAULT; | |
0aac03f0 | 2172 | pagefault_disable(); |
3180a7fc | 2173 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 2174 | pagefault_enable(); |
7ec54588 MT |
2175 | if (r) |
2176 | return -EFAULT; | |
2177 | return 0; | |
2178 | } | |
7ec54588 | 2179 | |
8e73485c PB |
2180 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
2181 | unsigned long len) | |
2182 | { | |
2183 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2184 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2185 | int offset = offset_in_page(gpa); | |
2186 | ||
2187 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2188 | } | |
2189 | EXPORT_SYMBOL_GPL(kvm_read_guest_atomic); | |
2190 | ||
2191 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, | |
2192 | void *data, unsigned long len) | |
2193 | { | |
2194 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2195 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2196 | int offset = offset_in_page(gpa); | |
2197 | ||
2198 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2199 | } | |
2200 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
2201 | ||
2202 | static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn, | |
2203 | const void *data, int offset, int len) | |
195aefde | 2204 | { |
e0506bcb IE |
2205 | int r; |
2206 | unsigned long addr; | |
195aefde | 2207 | |
251eb841 | 2208 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
2209 | if (kvm_is_error_hva(addr)) |
2210 | return -EFAULT; | |
8b0cedff | 2211 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 2212 | if (r) |
195aefde | 2213 | return -EFAULT; |
bc009e43 | 2214 | mark_page_dirty_in_slot(memslot, gfn); |
195aefde IE |
2215 | return 0; |
2216 | } | |
8e73485c PB |
2217 | |
2218 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
2219 | const void *data, int offset, int len) | |
2220 | { | |
2221 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2222 | ||
2223 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
2224 | } | |
195aefde IE |
2225 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
2226 | ||
8e73485c PB |
2227 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
2228 | const void *data, int offset, int len) | |
2229 | { | |
2230 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2231 | ||
2232 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
2233 | } | |
2234 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
2235 | ||
195aefde IE |
2236 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
2237 | unsigned long len) | |
2238 | { | |
2239 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2240 | int seg; | |
2241 | int offset = offset_in_page(gpa); | |
2242 | int ret; | |
2243 | ||
2244 | while ((seg = next_segment(len, offset)) != 0) { | |
2245 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
2246 | if (ret < 0) | |
2247 | return ret; | |
2248 | offset = 0; | |
2249 | len -= seg; | |
2250 | data += seg; | |
2251 | ++gfn; | |
2252 | } | |
2253 | return 0; | |
2254 | } | |
ff651cb6 | 2255 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 2256 | |
8e73485c PB |
2257 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
2258 | unsigned long len) | |
2259 | { | |
2260 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2261 | int seg; | |
2262 | int offset = offset_in_page(gpa); | |
2263 | int ret; | |
2264 | ||
2265 | while ((seg = next_segment(len, offset)) != 0) { | |
2266 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
2267 | if (ret < 0) | |
2268 | return ret; | |
2269 | offset = 0; | |
2270 | len -= seg; | |
2271 | data += seg; | |
2272 | ++gfn; | |
2273 | } | |
2274 | return 0; | |
2275 | } | |
2276 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
2277 | ||
5a2d4365 PB |
2278 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
2279 | struct gfn_to_hva_cache *ghc, | |
2280 | gpa_t gpa, unsigned long len) | |
49c7754c | 2281 | { |
49c7754c | 2282 | int offset = offset_in_page(gpa); |
8f964525 AH |
2283 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
2284 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
2285 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
2286 | gfn_t nr_pages_avail; | |
f1b9dd5e | 2287 | int r = start_gfn <= end_gfn ? 0 : -EINVAL; |
49c7754c GN |
2288 | |
2289 | ghc->gpa = gpa; | |
2290 | ghc->generation = slots->generation; | |
8f964525 | 2291 | ghc->len = len; |
f1b9dd5e JM |
2292 | ghc->hva = KVM_HVA_ERR_BAD; |
2293 | ||
2294 | /* | |
2295 | * If the requested region crosses two memslots, we still | |
2296 | * verify that the entire region is valid here. | |
2297 | */ | |
2298 | while (!r && start_gfn <= end_gfn) { | |
2299 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); | |
2300 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
2301 | &nr_pages_avail); | |
2302 | if (kvm_is_error_hva(ghc->hva)) | |
2303 | r = -EFAULT; | |
2304 | start_gfn += nr_pages_avail; | |
2305 | } | |
2306 | ||
2307 | /* Use the slow path for cross page reads and writes. */ | |
2308 | if (!r && nr_pages_needed == 1) | |
49c7754c | 2309 | ghc->hva += offset; |
f1b9dd5e | 2310 | else |
8f964525 | 2311 | ghc->memslot = NULL; |
f1b9dd5e JM |
2312 | |
2313 | return r; | |
49c7754c | 2314 | } |
5a2d4365 | 2315 | |
4e335d9e | 2316 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
2317 | gpa_t gpa, unsigned long len) |
2318 | { | |
4e335d9e | 2319 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
2320 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
2321 | } | |
4e335d9e | 2322 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 2323 | |
4e335d9e | 2324 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
2325 | void *data, unsigned int offset, |
2326 | unsigned long len) | |
49c7754c | 2327 | { |
4e335d9e | 2328 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 2329 | int r; |
4ec6e863 | 2330 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 2331 | |
4ec6e863 | 2332 | BUG_ON(len + offset > ghc->len); |
8f964525 | 2333 | |
49c7754c | 2334 | if (slots->generation != ghc->generation) |
5a2d4365 | 2335 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
8f964525 | 2336 | |
49c7754c GN |
2337 | if (kvm_is_error_hva(ghc->hva)) |
2338 | return -EFAULT; | |
2339 | ||
b0d47833 SC |
2340 | if (unlikely(!ghc->memslot)) |
2341 | return kvm_write_guest(kvm, gpa, data, len); | |
2342 | ||
4ec6e863 | 2343 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
2344 | if (r) |
2345 | return -EFAULT; | |
4ec6e863 | 2346 | mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
2347 | |
2348 | return 0; | |
2349 | } | |
4e335d9e | 2350 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 2351 | |
4e335d9e PB |
2352 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2353 | void *data, unsigned long len) | |
4ec6e863 | 2354 | { |
4e335d9e | 2355 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 2356 | } |
4e335d9e | 2357 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 2358 | |
4e335d9e PB |
2359 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2360 | void *data, unsigned long len) | |
e03b644f | 2361 | { |
4e335d9e | 2362 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f GN |
2363 | int r; |
2364 | ||
8f964525 AH |
2365 | BUG_ON(len > ghc->len); |
2366 | ||
e03b644f | 2367 | if (slots->generation != ghc->generation) |
5a2d4365 | 2368 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
8f964525 | 2369 | |
e03b644f GN |
2370 | if (kvm_is_error_hva(ghc->hva)) |
2371 | return -EFAULT; | |
2372 | ||
b0d47833 SC |
2373 | if (unlikely(!ghc->memslot)) |
2374 | return kvm_read_guest(kvm, ghc->gpa, data, len); | |
2375 | ||
e03b644f GN |
2376 | r = __copy_from_user(data, (void __user *)ghc->hva, len); |
2377 | if (r) | |
2378 | return -EFAULT; | |
2379 | ||
2380 | return 0; | |
2381 | } | |
4e335d9e | 2382 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 2383 | |
195aefde IE |
2384 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
2385 | { | |
8a3caa6d HC |
2386 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
2387 | ||
2388 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
2389 | } |
2390 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
2391 | ||
2392 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
2393 | { | |
2394 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2395 | int seg; | |
2396 | int offset = offset_in_page(gpa); | |
2397 | int ret; | |
2398 | ||
bfda0e84 | 2399 | while ((seg = next_segment(len, offset)) != 0) { |
195aefde IE |
2400 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); |
2401 | if (ret < 0) | |
2402 | return ret; | |
2403 | offset = 0; | |
2404 | len -= seg; | |
2405 | ++gfn; | |
2406 | } | |
2407 | return 0; | |
2408 | } | |
2409 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
2410 | ||
bc009e43 | 2411 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, |
7940876e | 2412 | gfn_t gfn) |
6aa8b732 | 2413 | { |
7e9d619d RR |
2414 | if (memslot && memslot->dirty_bitmap) { |
2415 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 2416 | |
b74ca3b3 | 2417 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
2418 | } |
2419 | } | |
2420 | ||
49c7754c GN |
2421 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
2422 | { | |
2423 | struct kvm_memory_slot *memslot; | |
2424 | ||
2425 | memslot = gfn_to_memslot(kvm, gfn); | |
bc009e43 | 2426 | mark_page_dirty_in_slot(memslot, gfn); |
49c7754c | 2427 | } |
2ba9f0d8 | 2428 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 2429 | |
8e73485c PB |
2430 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
2431 | { | |
2432 | struct kvm_memory_slot *memslot; | |
2433 | ||
2434 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2435 | mark_page_dirty_in_slot(memslot, gfn); | |
2436 | } | |
2437 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
2438 | ||
20b7035c JS |
2439 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
2440 | { | |
2441 | if (!vcpu->sigset_active) | |
2442 | return; | |
2443 | ||
2444 | /* | |
2445 | * This does a lockless modification of ->real_blocked, which is fine | |
2446 | * because, only current can change ->real_blocked and all readers of | |
2447 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
2448 | * of ->blocked. | |
2449 | */ | |
2450 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
2451 | } | |
2452 | ||
2453 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
2454 | { | |
2455 | if (!vcpu->sigset_active) | |
2456 | return; | |
2457 | ||
2458 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
2459 | sigemptyset(¤t->real_blocked); | |
2460 | } | |
2461 | ||
aca6ff29 WL |
2462 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
2463 | { | |
dee339b5 | 2464 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 2465 | |
2cbd7824 | 2466 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 2467 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 2468 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
2469 | if (!grow) |
2470 | goto out; | |
2471 | ||
dee339b5 NW |
2472 | val *= grow; |
2473 | if (val < grow_start) | |
2474 | val = grow_start; | |
aca6ff29 | 2475 | |
313f636d DM |
2476 | if (val > halt_poll_ns) |
2477 | val = halt_poll_ns; | |
2478 | ||
aca6ff29 | 2479 | vcpu->halt_poll_ns = val; |
7fa08e71 | 2480 | out: |
2cbd7824 | 2481 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2482 | } |
2483 | ||
2484 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
2485 | { | |
2b4bd831 | 2486 | unsigned int old, val, shrink, grow_start; |
aca6ff29 | 2487 | |
2cbd7824 | 2488 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 2489 | shrink = READ_ONCE(halt_poll_ns_shrink); |
2b4bd831 | 2490 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 2491 | if (shrink == 0) |
aca6ff29 WL |
2492 | val = 0; |
2493 | else | |
6b6de68c | 2494 | val /= shrink; |
aca6ff29 | 2495 | |
2b4bd831 SS |
2496 | if (val < grow_start) |
2497 | val = 0; | |
2498 | ||
aca6ff29 | 2499 | vcpu->halt_poll_ns = val; |
2cbd7824 | 2500 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2501 | } |
2502 | ||
f7819512 PB |
2503 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
2504 | { | |
50c28f21 JS |
2505 | int ret = -EINTR; |
2506 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
2507 | ||
f7819512 PB |
2508 | if (kvm_arch_vcpu_runnable(vcpu)) { |
2509 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 2510 | goto out; |
f7819512 PB |
2511 | } |
2512 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 2513 | goto out; |
f7819512 | 2514 | if (signal_pending(current)) |
50c28f21 | 2515 | goto out; |
f7819512 | 2516 | |
50c28f21 JS |
2517 | ret = 0; |
2518 | out: | |
2519 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
2520 | return ret; | |
f7819512 PB |
2521 | } |
2522 | ||
b6958ce4 ED |
2523 | /* |
2524 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
2525 | */ | |
8776e519 | 2526 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 2527 | { |
f7819512 | 2528 | ktime_t start, cur; |
8577370f | 2529 | DECLARE_SWAITQUEUE(wait); |
f7819512 | 2530 | bool waited = false; |
aca6ff29 | 2531 | u64 block_ns; |
f7819512 | 2532 | |
07ab0f8d MZ |
2533 | kvm_arch_vcpu_blocking(vcpu); |
2534 | ||
f7819512 | 2535 | start = cur = ktime_get(); |
cdd6ad3a | 2536 | if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) { |
19020f8a | 2537 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); |
f95ef0cd | 2538 | |
62bea5bf | 2539 | ++vcpu->stat.halt_attempted_poll; |
f7819512 PB |
2540 | do { |
2541 | /* | |
2542 | * This sets KVM_REQ_UNHALT if an interrupt | |
2543 | * arrives. | |
2544 | */ | |
2545 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
2546 | ++vcpu->stat.halt_successful_poll; | |
3491caf2 CB |
2547 | if (!vcpu_valid_wakeup(vcpu)) |
2548 | ++vcpu->stat.halt_poll_invalid; | |
f7819512 PB |
2549 | goto out; |
2550 | } | |
2551 | cur = ktime_get(); | |
2552 | } while (single_task_running() && ktime_before(cur, stop)); | |
2553 | } | |
e5c239cf MT |
2554 | |
2555 | for (;;) { | |
b3dae109 | 2556 | prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); |
e5c239cf | 2557 | |
f7819512 | 2558 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
2559 | break; |
2560 | ||
f7819512 | 2561 | waited = true; |
b6958ce4 | 2562 | schedule(); |
b6958ce4 | 2563 | } |
d3bef15f | 2564 | |
8577370f | 2565 | finish_swait(&vcpu->wq, &wait); |
f7819512 | 2566 | cur = ktime_get(); |
f7819512 | 2567 | out: |
07ab0f8d | 2568 | kvm_arch_vcpu_unblocking(vcpu); |
aca6ff29 WL |
2569 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
2570 | ||
44551b2f WL |
2571 | if (!kvm_arch_no_poll(vcpu)) { |
2572 | if (!vcpu_valid_wakeup(vcpu)) { | |
aca6ff29 | 2573 | shrink_halt_poll_ns(vcpu); |
44551b2f WL |
2574 | } else if (halt_poll_ns) { |
2575 | if (block_ns <= vcpu->halt_poll_ns) | |
2576 | ; | |
2577 | /* we had a long block, shrink polling */ | |
2578 | else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns) | |
2579 | shrink_halt_poll_ns(vcpu); | |
2580 | /* we had a short halt and our poll time is too small */ | |
2581 | else if (vcpu->halt_poll_ns < halt_poll_ns && | |
2582 | block_ns < halt_poll_ns) | |
2583 | grow_halt_poll_ns(vcpu); | |
2584 | } else { | |
2585 | vcpu->halt_poll_ns = 0; | |
2586 | } | |
2587 | } | |
aca6ff29 | 2588 | |
3491caf2 CB |
2589 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
2590 | kvm_arch_vcpu_block_finish(vcpu); | |
b6958ce4 | 2591 | } |
2ba9f0d8 | 2592 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 2593 | |
178f02ff | 2594 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 2595 | { |
8577370f | 2596 | struct swait_queue_head *wqp; |
b6d33834 CD |
2597 | |
2598 | wqp = kvm_arch_vcpu_wq(vcpu); | |
5e0018b3 | 2599 | if (swq_has_sleeper(wqp)) { |
b3dae109 | 2600 | swake_up_one(wqp); |
d73eb57b | 2601 | WRITE_ONCE(vcpu->ready, true); |
b6d33834 | 2602 | ++vcpu->stat.halt_wakeup; |
178f02ff | 2603 | return true; |
b6d33834 CD |
2604 | } |
2605 | ||
178f02ff | 2606 | return false; |
dd1a4cc1 RK |
2607 | } |
2608 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
2609 | ||
0266c894 | 2610 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
2611 | /* |
2612 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
2613 | */ | |
2614 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
2615 | { | |
2616 | int me; | |
2617 | int cpu = vcpu->cpu; | |
2618 | ||
178f02ff RK |
2619 | if (kvm_vcpu_wake_up(vcpu)) |
2620 | return; | |
2621 | ||
b6d33834 CD |
2622 | me = get_cpu(); |
2623 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
2624 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
2625 | smp_send_reschedule(cpu); | |
2626 | put_cpu(); | |
2627 | } | |
a20ed54d | 2628 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 2629 | #endif /* !CONFIG_S390 */ |
b6d33834 | 2630 | |
fa93384f | 2631 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
2632 | { |
2633 | struct pid *pid; | |
2634 | struct task_struct *task = NULL; | |
fa93384f | 2635 | int ret = 0; |
41628d33 KW |
2636 | |
2637 | rcu_read_lock(); | |
2638 | pid = rcu_dereference(target->pid); | |
2639 | if (pid) | |
27fbe64b | 2640 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
2641 | rcu_read_unlock(); |
2642 | if (!task) | |
c45c528e | 2643 | return ret; |
c45c528e | 2644 | ret = yield_to(task, 1); |
41628d33 | 2645 | put_task_struct(task); |
c45c528e R |
2646 | |
2647 | return ret; | |
41628d33 KW |
2648 | } |
2649 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
2650 | ||
06e48c51 R |
2651 | /* |
2652 | * Helper that checks whether a VCPU is eligible for directed yield. | |
2653 | * Most eligible candidate to yield is decided by following heuristics: | |
2654 | * | |
2655 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
2656 | * (preempted lock holder), indicated by @in_spin_loop. | |
2657 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
2658 | * | |
2659 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
2660 | * chance last time (mostly it has become eligible now since we have probably | |
2661 | * yielded to lockholder in last iteration. This is done by toggling | |
2662 | * @dy_eligible each time a VCPU checked for eligibility.) | |
2663 | * | |
2664 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
2665 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
2666 | * burning. Giving priority for a potential lock-holder increases lock | |
2667 | * progress. | |
2668 | * | |
2669 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
2670 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
2671 | * and continue with next VCPU and so on. | |
2672 | */ | |
7940876e | 2673 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 2674 | { |
4a55dd72 | 2675 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
2676 | bool eligible; |
2677 | ||
2678 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 2679 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
2680 | |
2681 | if (vcpu->spin_loop.in_spin_loop) | |
2682 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
2683 | ||
2684 | return eligible; | |
4a55dd72 SW |
2685 | #else |
2686 | return true; | |
06e48c51 | 2687 | #endif |
4a55dd72 | 2688 | } |
c45c528e | 2689 | |
17e433b5 WL |
2690 | /* |
2691 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
2692 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
2693 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
2694 | */ | |
2695 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
2696 | { | |
2697 | return kvm_arch_vcpu_runnable(vcpu); | |
2698 | } | |
2699 | ||
2700 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
2701 | { | |
2702 | if (kvm_arch_dy_runnable(vcpu)) | |
2703 | return true; | |
2704 | ||
2705 | #ifdef CONFIG_KVM_ASYNC_PF | |
2706 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
2707 | return true; | |
2708 | #endif | |
2709 | ||
2710 | return false; | |
2711 | } | |
2712 | ||
199b5763 | 2713 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 2714 | { |
217ece61 RR |
2715 | struct kvm *kvm = me->kvm; |
2716 | struct kvm_vcpu *vcpu; | |
2717 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
2718 | int yielded = 0; | |
c45c528e | 2719 | int try = 3; |
217ece61 RR |
2720 | int pass; |
2721 | int i; | |
d255f4f2 | 2722 | |
4c088493 | 2723 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
2724 | /* |
2725 | * We boost the priority of a VCPU that is runnable but not | |
2726 | * currently running, because it got preempted by something | |
2727 | * else and called schedule in __vcpu_run. Hopefully that | |
2728 | * VCPU is holding the lock that we need and will release it. | |
2729 | * We approximate round-robin by starting at the last boosted VCPU. | |
2730 | */ | |
c45c528e | 2731 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 2732 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 2733 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
2734 | i = last_boosted_vcpu; |
2735 | continue; | |
2736 | } else if (pass && i > last_boosted_vcpu) | |
2737 | break; | |
d73eb57b | 2738 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 2739 | continue; |
217ece61 RR |
2740 | if (vcpu == me) |
2741 | continue; | |
17e433b5 | 2742 | if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu)) |
217ece61 | 2743 | continue; |
046ddeed WL |
2744 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
2745 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 2746 | continue; |
06e48c51 R |
2747 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
2748 | continue; | |
c45c528e R |
2749 | |
2750 | yielded = kvm_vcpu_yield_to(vcpu); | |
2751 | if (yielded > 0) { | |
217ece61 | 2752 | kvm->last_boosted_vcpu = i; |
217ece61 | 2753 | break; |
c45c528e R |
2754 | } else if (yielded < 0) { |
2755 | try--; | |
2756 | if (!try) | |
2757 | break; | |
217ece61 | 2758 | } |
217ece61 RR |
2759 | } |
2760 | } | |
4c088493 | 2761 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
2762 | |
2763 | /* Ensure vcpu is not eligible during next spinloop */ | |
2764 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
2765 | } |
2766 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
2767 | ||
1499fa80 | 2768 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 2769 | { |
11bac800 | 2770 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
2771 | struct page *page; |
2772 | ||
e4a533a4 | 2773 | if (vmf->pgoff == 0) |
039576c0 | 2774 | page = virt_to_page(vcpu->run); |
09566765 | 2775 | #ifdef CONFIG_X86 |
e4a533a4 | 2776 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 2777 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 2778 | #endif |
4b4357e0 | 2779 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
2780 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
2781 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 2782 | #endif |
039576c0 | 2783 | else |
5b1c1493 | 2784 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 2785 | get_page(page); |
e4a533a4 | 2786 | vmf->page = page; |
2787 | return 0; | |
9a2bb7f4 AK |
2788 | } |
2789 | ||
f0f37e2f | 2790 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 2791 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
2792 | }; |
2793 | ||
2794 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
2795 | { | |
2796 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
2797 | return 0; | |
2798 | } | |
2799 | ||
bccf2150 AK |
2800 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
2801 | { | |
2802 | struct kvm_vcpu *vcpu = filp->private_data; | |
2803 | ||
45b5939e | 2804 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
66c0b394 | 2805 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
2806 | return 0; |
2807 | } | |
2808 | ||
3d3aab1b | 2809 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
2810 | .release = kvm_vcpu_release, |
2811 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 2812 | .mmap = kvm_vcpu_mmap, |
6038f373 | 2813 | .llseek = noop_llseek, |
7ddfd3e0 | 2814 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
2815 | }; |
2816 | ||
2817 | /* | |
2818 | * Allocates an inode for the vcpu. | |
2819 | */ | |
2820 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
2821 | { | |
e46b4692 MY |
2822 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
2823 | ||
2824 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
2825 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
2826 | } |
2827 | ||
3e7093d0 | 2828 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 2829 | { |
741cbbae | 2830 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
45b5939e | 2831 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 2832 | |
45b5939e | 2833 | if (!debugfs_initialized()) |
3e7093d0 | 2834 | return; |
45b5939e LC |
2835 | |
2836 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
2837 | vcpu->debugfs_dentry = debugfs_create_dir(dir_name, | |
3e7093d0 | 2838 | vcpu->kvm->debugfs_dentry); |
45b5939e | 2839 | |
3e7093d0 | 2840 | kvm_arch_create_vcpu_debugfs(vcpu); |
741cbbae | 2841 | #endif |
45b5939e LC |
2842 | } |
2843 | ||
c5ea7660 AK |
2844 | /* |
2845 | * Creates some virtual cpus. Good luck creating more than one. | |
2846 | */ | |
73880c80 | 2847 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
2848 | { |
2849 | int r; | |
e09fefde | 2850 | struct kvm_vcpu *vcpu; |
c5ea7660 | 2851 | |
0b1b1dfd | 2852 | if (id >= KVM_MAX_VCPU_ID) |
338c7dba AH |
2853 | return -EINVAL; |
2854 | ||
6c7caebc PB |
2855 | mutex_lock(&kvm->lock); |
2856 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
2857 | mutex_unlock(&kvm->lock); | |
2858 | return -EINVAL; | |
2859 | } | |
2860 | ||
2861 | kvm->created_vcpus++; | |
2862 | mutex_unlock(&kvm->lock); | |
2863 | ||
73880c80 | 2864 | vcpu = kvm_arch_vcpu_create(kvm, id); |
6c7caebc PB |
2865 | if (IS_ERR(vcpu)) { |
2866 | r = PTR_ERR(vcpu); | |
2867 | goto vcpu_decrement; | |
2868 | } | |
c5ea7660 | 2869 | |
15ad7146 AK |
2870 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
2871 | ||
26e5215f AK |
2872 | r = kvm_arch_vcpu_setup(vcpu); |
2873 | if (r) | |
d780592b | 2874 | goto vcpu_destroy; |
26e5215f | 2875 | |
3e7093d0 | 2876 | kvm_create_vcpu_debugfs(vcpu); |
45b5939e | 2877 | |
11ec2804 | 2878 | mutex_lock(&kvm->lock); |
e09fefde DH |
2879 | if (kvm_get_vcpu_by_id(kvm, id)) { |
2880 | r = -EEXIST; | |
2881 | goto unlock_vcpu_destroy; | |
2882 | } | |
73880c80 | 2883 | |
e9fe0d2c RK |
2884 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
2885 | BUG_ON(kvm->vcpus[vcpu->vcpu_idx]); | |
c5ea7660 | 2886 | |
fb3f0f51 | 2887 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 2888 | kvm_get_kvm(kvm); |
bccf2150 | 2889 | r = create_vcpu_fd(vcpu); |
73880c80 GN |
2890 | if (r < 0) { |
2891 | kvm_put_kvm(kvm); | |
d780592b | 2892 | goto unlock_vcpu_destroy; |
73880c80 GN |
2893 | } |
2894 | ||
e9fe0d2c | 2895 | kvm->vcpus[vcpu->vcpu_idx] = vcpu; |
dd489240 PB |
2896 | |
2897 | /* | |
2898 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
2899 | * before kvm->online_vcpu's incremented value. | |
2900 | */ | |
73880c80 GN |
2901 | smp_wmb(); |
2902 | atomic_inc(&kvm->online_vcpus); | |
2903 | ||
73880c80 | 2904 | mutex_unlock(&kvm->lock); |
42897d86 | 2905 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 2906 | return r; |
39c3b86e | 2907 | |
d780592b | 2908 | unlock_vcpu_destroy: |
7d8fece6 | 2909 | mutex_unlock(&kvm->lock); |
45b5939e | 2910 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
d780592b | 2911 | vcpu_destroy: |
d40ccc62 | 2912 | kvm_arch_vcpu_destroy(vcpu); |
6c7caebc PB |
2913 | vcpu_decrement: |
2914 | mutex_lock(&kvm->lock); | |
2915 | kvm->created_vcpus--; | |
2916 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
2917 | return r; |
2918 | } | |
2919 | ||
1961d276 AK |
2920 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
2921 | { | |
2922 | if (sigset) { | |
2923 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
2924 | vcpu->sigset_active = 1; | |
2925 | vcpu->sigset = *sigset; | |
2926 | } else | |
2927 | vcpu->sigset_active = 0; | |
2928 | return 0; | |
2929 | } | |
2930 | ||
bccf2150 AK |
2931 | static long kvm_vcpu_ioctl(struct file *filp, |
2932 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 2933 | { |
bccf2150 | 2934 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 2935 | void __user *argp = (void __user *)arg; |
313a3dc7 | 2936 | int r; |
fa3795a7 DH |
2937 | struct kvm_fpu *fpu = NULL; |
2938 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 2939 | |
6d4e4c4f AK |
2940 | if (vcpu->kvm->mm != current->mm) |
2941 | return -EIO; | |
2122ff5e | 2942 | |
2ea75be3 DM |
2943 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
2944 | return -EINVAL; | |
2945 | ||
2122ff5e | 2946 | /* |
5cb0944c PB |
2947 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
2948 | * execution; mutex_lock() would break them. | |
2122ff5e | 2949 | */ |
5cb0944c PB |
2950 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
2951 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 2952 | return r; |
2122ff5e | 2953 | |
ec7660cc CD |
2954 | if (mutex_lock_killable(&vcpu->mutex)) |
2955 | return -EINTR; | |
6aa8b732 | 2956 | switch (ioctl) { |
0e4524a5 CB |
2957 | case KVM_RUN: { |
2958 | struct pid *oldpid; | |
f0fe5108 AK |
2959 | r = -EINVAL; |
2960 | if (arg) | |
2961 | goto out; | |
0e4524a5 | 2962 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 2963 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 2964 | /* The thread running this VCPU changed. */ |
bd2a6394 | 2965 | struct pid *newpid; |
f95ef0cd | 2966 | |
bd2a6394 CD |
2967 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
2968 | if (r) | |
2969 | break; | |
2970 | ||
2971 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
2972 | rcu_assign_pointer(vcpu->pid, newpid); |
2973 | if (oldpid) | |
2974 | synchronize_rcu(); | |
2975 | put_pid(oldpid); | |
2976 | } | |
b6c7a5dc | 2977 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 2978 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 2979 | break; |
0e4524a5 | 2980 | } |
6aa8b732 | 2981 | case KVM_GET_REGS: { |
3e4bb3ac | 2982 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2983 | |
3e4bb3ac | 2984 | r = -ENOMEM; |
b12ce36a | 2985 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 2986 | if (!kvm_regs) |
6aa8b732 | 2987 | goto out; |
3e4bb3ac XZ |
2988 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2989 | if (r) | |
2990 | goto out_free1; | |
6aa8b732 | 2991 | r = -EFAULT; |
3e4bb3ac XZ |
2992 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2993 | goto out_free1; | |
6aa8b732 | 2994 | r = 0; |
3e4bb3ac XZ |
2995 | out_free1: |
2996 | kfree(kvm_regs); | |
6aa8b732 AK |
2997 | break; |
2998 | } | |
2999 | case KVM_SET_REGS: { | |
3e4bb3ac | 3000 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3001 | |
3e4bb3ac | 3002 | r = -ENOMEM; |
ff5c2c03 SL |
3003 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
3004 | if (IS_ERR(kvm_regs)) { | |
3005 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 3006 | goto out; |
ff5c2c03 | 3007 | } |
3e4bb3ac | 3008 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 3009 | kfree(kvm_regs); |
6aa8b732 AK |
3010 | break; |
3011 | } | |
3012 | case KVM_GET_SREGS: { | |
b12ce36a BG |
3013 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
3014 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
3015 | r = -ENOMEM; |
3016 | if (!kvm_sregs) | |
3017 | goto out; | |
3018 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
3019 | if (r) |
3020 | goto out; | |
3021 | r = -EFAULT; | |
fa3795a7 | 3022 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
3023 | goto out; |
3024 | r = 0; | |
3025 | break; | |
3026 | } | |
3027 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
3028 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
3029 | if (IS_ERR(kvm_sregs)) { | |
3030 | r = PTR_ERR(kvm_sregs); | |
18595411 | 3031 | kvm_sregs = NULL; |
6aa8b732 | 3032 | goto out; |
ff5c2c03 | 3033 | } |
fa3795a7 | 3034 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
3035 | break; |
3036 | } | |
62d9f0db MT |
3037 | case KVM_GET_MP_STATE: { |
3038 | struct kvm_mp_state mp_state; | |
3039 | ||
3040 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
3041 | if (r) | |
3042 | goto out; | |
3043 | r = -EFAULT; | |
893bdbf1 | 3044 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
3045 | goto out; |
3046 | r = 0; | |
3047 | break; | |
3048 | } | |
3049 | case KVM_SET_MP_STATE: { | |
3050 | struct kvm_mp_state mp_state; | |
3051 | ||
3052 | r = -EFAULT; | |
893bdbf1 | 3053 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
3054 | goto out; |
3055 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
3056 | break; |
3057 | } | |
6aa8b732 AK |
3058 | case KVM_TRANSLATE: { |
3059 | struct kvm_translation tr; | |
3060 | ||
3061 | r = -EFAULT; | |
893bdbf1 | 3062 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 3063 | goto out; |
8b006791 | 3064 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
3065 | if (r) |
3066 | goto out; | |
3067 | r = -EFAULT; | |
893bdbf1 | 3068 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
3069 | goto out; |
3070 | r = 0; | |
3071 | break; | |
3072 | } | |
d0bfb940 JK |
3073 | case KVM_SET_GUEST_DEBUG: { |
3074 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
3075 | |
3076 | r = -EFAULT; | |
893bdbf1 | 3077 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 3078 | goto out; |
d0bfb940 | 3079 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
3080 | break; |
3081 | } | |
1961d276 AK |
3082 | case KVM_SET_SIGNAL_MASK: { |
3083 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3084 | struct kvm_signal_mask kvm_sigmask; | |
3085 | sigset_t sigset, *p; | |
3086 | ||
3087 | p = NULL; | |
3088 | if (argp) { | |
3089 | r = -EFAULT; | |
3090 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3091 | sizeof(kvm_sigmask))) |
1961d276 AK |
3092 | goto out; |
3093 | r = -EINVAL; | |
893bdbf1 | 3094 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
3095 | goto out; |
3096 | r = -EFAULT; | |
3097 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 3098 | sizeof(sigset))) |
1961d276 AK |
3099 | goto out; |
3100 | p = &sigset; | |
3101 | } | |
376d41ff | 3102 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
3103 | break; |
3104 | } | |
b8836737 | 3105 | case KVM_GET_FPU: { |
b12ce36a | 3106 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
3107 | r = -ENOMEM; |
3108 | if (!fpu) | |
3109 | goto out; | |
3110 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
3111 | if (r) |
3112 | goto out; | |
3113 | r = -EFAULT; | |
fa3795a7 | 3114 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
3115 | goto out; |
3116 | r = 0; | |
3117 | break; | |
3118 | } | |
3119 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
3120 | fpu = memdup_user(argp, sizeof(*fpu)); |
3121 | if (IS_ERR(fpu)) { | |
3122 | r = PTR_ERR(fpu); | |
18595411 | 3123 | fpu = NULL; |
b8836737 | 3124 | goto out; |
ff5c2c03 | 3125 | } |
fa3795a7 | 3126 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
3127 | break; |
3128 | } | |
bccf2150 | 3129 | default: |
313a3dc7 | 3130 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
3131 | } |
3132 | out: | |
ec7660cc | 3133 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
3134 | kfree(fpu); |
3135 | kfree(kvm_sregs); | |
bccf2150 AK |
3136 | return r; |
3137 | } | |
3138 | ||
de8e5d74 | 3139 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
3140 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
3141 | unsigned int ioctl, unsigned long arg) | |
3142 | { | |
3143 | struct kvm_vcpu *vcpu = filp->private_data; | |
3144 | void __user *argp = compat_ptr(arg); | |
3145 | int r; | |
3146 | ||
3147 | if (vcpu->kvm->mm != current->mm) | |
3148 | return -EIO; | |
3149 | ||
3150 | switch (ioctl) { | |
3151 | case KVM_SET_SIGNAL_MASK: { | |
3152 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3153 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
3154 | sigset_t sigset; |
3155 | ||
3156 | if (argp) { | |
3157 | r = -EFAULT; | |
3158 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3159 | sizeof(kvm_sigmask))) |
1dda606c AG |
3160 | goto out; |
3161 | r = -EINVAL; | |
3968cf62 | 3162 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
3163 | goto out; |
3164 | r = -EFAULT; | |
3968cf62 | 3165 | if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset)) |
1dda606c | 3166 | goto out; |
760a9a30 AC |
3167 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
3168 | } else | |
3169 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
3170 | break; |
3171 | } | |
3172 | default: | |
3173 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
3174 | } | |
3175 | ||
3176 | out: | |
3177 | return r; | |
3178 | } | |
3179 | #endif | |
3180 | ||
a1cd3f08 CLG |
3181 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
3182 | { | |
3183 | struct kvm_device *dev = filp->private_data; | |
3184 | ||
3185 | if (dev->ops->mmap) | |
3186 | return dev->ops->mmap(dev, vma); | |
3187 | ||
3188 | return -ENODEV; | |
3189 | } | |
3190 | ||
852b6d57 SW |
3191 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
3192 | int (*accessor)(struct kvm_device *dev, | |
3193 | struct kvm_device_attr *attr), | |
3194 | unsigned long arg) | |
3195 | { | |
3196 | struct kvm_device_attr attr; | |
3197 | ||
3198 | if (!accessor) | |
3199 | return -EPERM; | |
3200 | ||
3201 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
3202 | return -EFAULT; | |
3203 | ||
3204 | return accessor(dev, &attr); | |
3205 | } | |
3206 | ||
3207 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
3208 | unsigned long arg) | |
3209 | { | |
3210 | struct kvm_device *dev = filp->private_data; | |
3211 | ||
ddba9180 SC |
3212 | if (dev->kvm->mm != current->mm) |
3213 | return -EIO; | |
3214 | ||
852b6d57 SW |
3215 | switch (ioctl) { |
3216 | case KVM_SET_DEVICE_ATTR: | |
3217 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
3218 | case KVM_GET_DEVICE_ATTR: | |
3219 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
3220 | case KVM_HAS_DEVICE_ATTR: | |
3221 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
3222 | default: | |
3223 | if (dev->ops->ioctl) | |
3224 | return dev->ops->ioctl(dev, ioctl, arg); | |
3225 | ||
3226 | return -ENOTTY; | |
3227 | } | |
3228 | } | |
3229 | ||
852b6d57 SW |
3230 | static int kvm_device_release(struct inode *inode, struct file *filp) |
3231 | { | |
3232 | struct kvm_device *dev = filp->private_data; | |
3233 | struct kvm *kvm = dev->kvm; | |
3234 | ||
2bde9b3e CLG |
3235 | if (dev->ops->release) { |
3236 | mutex_lock(&kvm->lock); | |
3237 | list_del(&dev->vm_node); | |
3238 | dev->ops->release(dev); | |
3239 | mutex_unlock(&kvm->lock); | |
3240 | } | |
3241 | ||
852b6d57 SW |
3242 | kvm_put_kvm(kvm); |
3243 | return 0; | |
3244 | } | |
3245 | ||
3246 | static const struct file_operations kvm_device_fops = { | |
3247 | .unlocked_ioctl = kvm_device_ioctl, | |
3248 | .release = kvm_device_release, | |
7ddfd3e0 | 3249 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 3250 | .mmap = kvm_device_mmap, |
852b6d57 SW |
3251 | }; |
3252 | ||
3253 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
3254 | { | |
3255 | if (filp->f_op != &kvm_device_fops) | |
3256 | return NULL; | |
3257 | ||
3258 | return filp->private_data; | |
3259 | } | |
3260 | ||
d60eacb0 | 3261 | static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 3262 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
3263 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
3264 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 3265 | #endif |
d60eacb0 WD |
3266 | }; |
3267 | ||
3268 | int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type) | |
3269 | { | |
3270 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
3271 | return -ENOSPC; | |
3272 | ||
3273 | if (kvm_device_ops_table[type] != NULL) | |
3274 | return -EEXIST; | |
3275 | ||
3276 | kvm_device_ops_table[type] = ops; | |
3277 | return 0; | |
3278 | } | |
3279 | ||
571ee1b6 WL |
3280 | void kvm_unregister_device_ops(u32 type) |
3281 | { | |
3282 | if (kvm_device_ops_table[type] != NULL) | |
3283 | kvm_device_ops_table[type] = NULL; | |
3284 | } | |
3285 | ||
852b6d57 SW |
3286 | static int kvm_ioctl_create_device(struct kvm *kvm, |
3287 | struct kvm_create_device *cd) | |
3288 | { | |
3289 | struct kvm_device_ops *ops = NULL; | |
3290 | struct kvm_device *dev; | |
3291 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 3292 | int type; |
852b6d57 SW |
3293 | int ret; |
3294 | ||
d60eacb0 WD |
3295 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
3296 | return -ENODEV; | |
3297 | ||
1d487e9b PB |
3298 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
3299 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 3300 | if (ops == NULL) |
852b6d57 | 3301 | return -ENODEV; |
852b6d57 SW |
3302 | |
3303 | if (test) | |
3304 | return 0; | |
3305 | ||
b12ce36a | 3306 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
3307 | if (!dev) |
3308 | return -ENOMEM; | |
3309 | ||
3310 | dev->ops = ops; | |
3311 | dev->kvm = kvm; | |
852b6d57 | 3312 | |
a28ebea2 | 3313 | mutex_lock(&kvm->lock); |
1d487e9b | 3314 | ret = ops->create(dev, type); |
852b6d57 | 3315 | if (ret < 0) { |
a28ebea2 | 3316 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
3317 | kfree(dev); |
3318 | return ret; | |
3319 | } | |
a28ebea2 CD |
3320 | list_add(&dev->vm_node, &kvm->devices); |
3321 | mutex_unlock(&kvm->lock); | |
852b6d57 | 3322 | |
023e9fdd CD |
3323 | if (ops->init) |
3324 | ops->init(dev); | |
3325 | ||
cfa39381 | 3326 | kvm_get_kvm(kvm); |
24009b05 | 3327 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 3328 | if (ret < 0) { |
cfa39381 | 3329 | kvm_put_kvm(kvm); |
a28ebea2 CD |
3330 | mutex_lock(&kvm->lock); |
3331 | list_del(&dev->vm_node); | |
3332 | mutex_unlock(&kvm->lock); | |
a0f1d21c | 3333 | ops->destroy(dev); |
852b6d57 SW |
3334 | return ret; |
3335 | } | |
3336 | ||
852b6d57 SW |
3337 | cd->fd = ret; |
3338 | return 0; | |
3339 | } | |
3340 | ||
92b591a4 AG |
3341 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
3342 | { | |
3343 | switch (arg) { | |
3344 | case KVM_CAP_USER_MEMORY: | |
3345 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
3346 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
3347 | case KVM_CAP_INTERNAL_ERROR_DATA: |
3348 | #ifdef CONFIG_HAVE_KVM_MSI | |
3349 | case KVM_CAP_SIGNAL_MSI: | |
3350 | #endif | |
297e2105 | 3351 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 3352 | case KVM_CAP_IRQFD: |
92b591a4 AG |
3353 | case KVM_CAP_IRQFD_RESAMPLE: |
3354 | #endif | |
e9ea5069 | 3355 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 3356 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 3357 | case KVM_CAP_ENABLE_CAP_VM: |
2a31b9db | 3358 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
d7547c55 | 3359 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: |
2a31b9db | 3360 | #endif |
92b591a4 | 3361 | return 1; |
4b4357e0 | 3362 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
3363 | case KVM_CAP_COALESCED_MMIO: |
3364 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
3365 | case KVM_CAP_COALESCED_PIO: |
3366 | return 1; | |
30422558 | 3367 | #endif |
92b591a4 AG |
3368 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
3369 | case KVM_CAP_IRQ_ROUTING: | |
3370 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
3371 | #endif |
3372 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
3373 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
3374 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 3375 | #endif |
c110ae57 PB |
3376 | case KVM_CAP_NR_MEMSLOTS: |
3377 | return KVM_USER_MEM_SLOTS; | |
92b591a4 AG |
3378 | default: |
3379 | break; | |
3380 | } | |
3381 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
3382 | } | |
3383 | ||
e5d83c74 PB |
3384 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
3385 | struct kvm_enable_cap *cap) | |
3386 | { | |
3387 | return -EINVAL; | |
3388 | } | |
3389 | ||
3390 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
3391 | struct kvm_enable_cap *cap) | |
3392 | { | |
3393 | switch (cap->cap) { | |
2a31b9db | 3394 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
d7547c55 | 3395 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: |
2a31b9db PB |
3396 | if (cap->flags || (cap->args[0] & ~1)) |
3397 | return -EINVAL; | |
3398 | kvm->manual_dirty_log_protect = cap->args[0]; | |
3399 | return 0; | |
3400 | #endif | |
e5d83c74 PB |
3401 | default: |
3402 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
3403 | } | |
3404 | } | |
3405 | ||
bccf2150 AK |
3406 | static long kvm_vm_ioctl(struct file *filp, |
3407 | unsigned int ioctl, unsigned long arg) | |
3408 | { | |
3409 | struct kvm *kvm = filp->private_data; | |
3410 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 3411 | int r; |
bccf2150 | 3412 | |
6d4e4c4f AK |
3413 | if (kvm->mm != current->mm) |
3414 | return -EIO; | |
bccf2150 AK |
3415 | switch (ioctl) { |
3416 | case KVM_CREATE_VCPU: | |
3417 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 3418 | break; |
e5d83c74 PB |
3419 | case KVM_ENABLE_CAP: { |
3420 | struct kvm_enable_cap cap; | |
3421 | ||
3422 | r = -EFAULT; | |
3423 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
3424 | goto out; | |
3425 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
3426 | break; | |
3427 | } | |
6fc138d2 IE |
3428 | case KVM_SET_USER_MEMORY_REGION: { |
3429 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
3430 | ||
3431 | r = -EFAULT; | |
3432 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 3433 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
3434 | goto out; |
3435 | ||
47ae31e2 | 3436 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
3437 | break; |
3438 | } | |
3439 | case KVM_GET_DIRTY_LOG: { | |
3440 | struct kvm_dirty_log log; | |
3441 | ||
3442 | r = -EFAULT; | |
893bdbf1 | 3443 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 3444 | goto out; |
2c6f5df9 | 3445 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
3446 | break; |
3447 | } | |
2a31b9db PB |
3448 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3449 | case KVM_CLEAR_DIRTY_LOG: { | |
3450 | struct kvm_clear_dirty_log log; | |
3451 | ||
3452 | r = -EFAULT; | |
3453 | if (copy_from_user(&log, argp, sizeof(log))) | |
3454 | goto out; | |
3455 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
3456 | break; | |
3457 | } | |
3458 | #endif | |
4b4357e0 | 3459 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3460 | case KVM_REGISTER_COALESCED_MMIO: { |
3461 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 3462 | |
5f94c174 | 3463 | r = -EFAULT; |
893bdbf1 | 3464 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 3465 | goto out; |
5f94c174 | 3466 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
3467 | break; |
3468 | } | |
3469 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
3470 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 3471 | |
5f94c174 | 3472 | r = -EFAULT; |
893bdbf1 | 3473 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 3474 | goto out; |
5f94c174 | 3475 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
3476 | break; |
3477 | } | |
3478 | #endif | |
721eecbf GH |
3479 | case KVM_IRQFD: { |
3480 | struct kvm_irqfd data; | |
3481 | ||
3482 | r = -EFAULT; | |
893bdbf1 | 3483 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 3484 | goto out; |
d4db2935 | 3485 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
3486 | break; |
3487 | } | |
d34e6b17 GH |
3488 | case KVM_IOEVENTFD: { |
3489 | struct kvm_ioeventfd data; | |
3490 | ||
3491 | r = -EFAULT; | |
893bdbf1 | 3492 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
3493 | goto out; |
3494 | r = kvm_ioeventfd(kvm, &data); | |
3495 | break; | |
3496 | } | |
07975ad3 JK |
3497 | #ifdef CONFIG_HAVE_KVM_MSI |
3498 | case KVM_SIGNAL_MSI: { | |
3499 | struct kvm_msi msi; | |
3500 | ||
3501 | r = -EFAULT; | |
893bdbf1 | 3502 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
3503 | goto out; |
3504 | r = kvm_send_userspace_msi(kvm, &msi); | |
3505 | break; | |
3506 | } | |
23d43cf9 CD |
3507 | #endif |
3508 | #ifdef __KVM_HAVE_IRQ_LINE | |
3509 | case KVM_IRQ_LINE_STATUS: | |
3510 | case KVM_IRQ_LINE: { | |
3511 | struct kvm_irq_level irq_event; | |
3512 | ||
3513 | r = -EFAULT; | |
893bdbf1 | 3514 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
3515 | goto out; |
3516 | ||
aa2fbe6d YZ |
3517 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
3518 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
3519 | if (r) |
3520 | goto out; | |
3521 | ||
3522 | r = -EFAULT; | |
3523 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 3524 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
3525 | goto out; |
3526 | } | |
3527 | ||
3528 | r = 0; | |
3529 | break; | |
3530 | } | |
73880c80 | 3531 | #endif |
aa8d5944 AG |
3532 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
3533 | case KVM_SET_GSI_ROUTING: { | |
3534 | struct kvm_irq_routing routing; | |
3535 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 3536 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
3537 | |
3538 | r = -EFAULT; | |
3539 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
3540 | goto out; | |
3541 | r = -EINVAL; | |
5c0aea0e DH |
3542 | if (!kvm_arch_can_set_irq_routing(kvm)) |
3543 | goto out; | |
caf1ff26 | 3544 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
3545 | goto out; |
3546 | if (routing.flags) | |
3547 | goto out; | |
f8c1b85b PB |
3548 | if (routing.nr) { |
3549 | r = -ENOMEM; | |
42bc47b3 KC |
3550 | entries = vmalloc(array_size(sizeof(*entries), |
3551 | routing.nr)); | |
f8c1b85b PB |
3552 | if (!entries) |
3553 | goto out; | |
3554 | r = -EFAULT; | |
3555 | urouting = argp; | |
3556 | if (copy_from_user(entries, urouting->entries, | |
3557 | routing.nr * sizeof(*entries))) | |
3558 | goto out_free_irq_routing; | |
3559 | } | |
aa8d5944 AG |
3560 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
3561 | routing.flags); | |
a642a175 | 3562 | out_free_irq_routing: |
aa8d5944 AG |
3563 | vfree(entries); |
3564 | break; | |
3565 | } | |
3566 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
3567 | case KVM_CREATE_DEVICE: { |
3568 | struct kvm_create_device cd; | |
3569 | ||
3570 | r = -EFAULT; | |
3571 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
3572 | goto out; | |
3573 | ||
3574 | r = kvm_ioctl_create_device(kvm, &cd); | |
3575 | if (r) | |
3576 | goto out; | |
3577 | ||
3578 | r = -EFAULT; | |
3579 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
3580 | goto out; | |
3581 | ||
3582 | r = 0; | |
3583 | break; | |
3584 | } | |
92b591a4 AG |
3585 | case KVM_CHECK_EXTENSION: |
3586 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
3587 | break; | |
f17abe9a | 3588 | default: |
1fe779f8 | 3589 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
3590 | } |
3591 | out: | |
3592 | return r; | |
3593 | } | |
3594 | ||
de8e5d74 | 3595 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
3596 | struct compat_kvm_dirty_log { |
3597 | __u32 slot; | |
3598 | __u32 padding1; | |
3599 | union { | |
3600 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
3601 | __u64 padding2; | |
3602 | }; | |
3603 | }; | |
3604 | ||
104db231 PB |
3605 | struct compat_kvm_clear_dirty_log { |
3606 | __u32 slot; | |
3607 | __u32 num_pages; | |
3608 | __u64 first_page; | |
3609 | union { | |
3610 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
3611 | __u64 padding2; | |
3612 | }; | |
3613 | }; | |
3614 | ||
6ff5894c AB |
3615 | static long kvm_vm_compat_ioctl(struct file *filp, |
3616 | unsigned int ioctl, unsigned long arg) | |
3617 | { | |
3618 | struct kvm *kvm = filp->private_data; | |
3619 | int r; | |
3620 | ||
3621 | if (kvm->mm != current->mm) | |
3622 | return -EIO; | |
3623 | switch (ioctl) { | |
104db231 PB |
3624 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3625 | case KVM_CLEAR_DIRTY_LOG: { | |
3626 | struct compat_kvm_clear_dirty_log compat_log; | |
3627 | struct kvm_clear_dirty_log log; | |
3628 | ||
3629 | if (copy_from_user(&compat_log, (void __user *)arg, | |
3630 | sizeof(compat_log))) | |
3631 | return -EFAULT; | |
3632 | log.slot = compat_log.slot; | |
3633 | log.num_pages = compat_log.num_pages; | |
3634 | log.first_page = compat_log.first_page; | |
3635 | log.padding2 = compat_log.padding2; | |
3636 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
3637 | ||
3638 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
3639 | break; | |
3640 | } | |
3641 | #endif | |
6ff5894c AB |
3642 | case KVM_GET_DIRTY_LOG: { |
3643 | struct compat_kvm_dirty_log compat_log; | |
3644 | struct kvm_dirty_log log; | |
3645 | ||
6ff5894c AB |
3646 | if (copy_from_user(&compat_log, (void __user *)arg, |
3647 | sizeof(compat_log))) | |
f6a3b168 | 3648 | return -EFAULT; |
6ff5894c AB |
3649 | log.slot = compat_log.slot; |
3650 | log.padding1 = compat_log.padding1; | |
3651 | log.padding2 = compat_log.padding2; | |
3652 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
3653 | ||
3654 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
3655 | break; |
3656 | } | |
3657 | default: | |
3658 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
3659 | } | |
6ff5894c AB |
3660 | return r; |
3661 | } | |
3662 | #endif | |
3663 | ||
3d3aab1b | 3664 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
3665 | .release = kvm_vm_release, |
3666 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 3667 | .llseek = noop_llseek, |
7ddfd3e0 | 3668 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
3669 | }; |
3670 | ||
e08b9637 | 3671 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 3672 | { |
aac87636 | 3673 | int r; |
f17abe9a | 3674 | struct kvm *kvm; |
506cfba9 | 3675 | struct file *file; |
f17abe9a | 3676 | |
e08b9637 | 3677 | kvm = kvm_create_vm(type); |
d6d28168 AK |
3678 | if (IS_ERR(kvm)) |
3679 | return PTR_ERR(kvm); | |
4b4357e0 | 3680 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 3681 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
3682 | if (r < 0) |
3683 | goto put_kvm; | |
6ce5a090 | 3684 | #endif |
506cfba9 | 3685 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
3686 | if (r < 0) |
3687 | goto put_kvm; | |
3688 | ||
506cfba9 AV |
3689 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
3690 | if (IS_ERR(file)) { | |
3691 | put_unused_fd(r); | |
78588335 ME |
3692 | r = PTR_ERR(file); |
3693 | goto put_kvm; | |
506cfba9 | 3694 | } |
536a6f88 | 3695 | |
525df861 PB |
3696 | /* |
3697 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
3698 | * already set, with ->release() being kvm_vm_release(). In error | |
3699 | * cases it will be called by the final fput(file) and will take | |
3700 | * care of doing kvm_put_kvm(kvm). | |
3701 | */ | |
536a6f88 | 3702 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
3703 | put_unused_fd(r); |
3704 | fput(file); | |
536a6f88 JF |
3705 | return -ENOMEM; |
3706 | } | |
286de8f6 | 3707 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 3708 | |
506cfba9 | 3709 | fd_install(r, file); |
aac87636 | 3710 | return r; |
78588335 ME |
3711 | |
3712 | put_kvm: | |
3713 | kvm_put_kvm(kvm); | |
3714 | return r; | |
f17abe9a AK |
3715 | } |
3716 | ||
3717 | static long kvm_dev_ioctl(struct file *filp, | |
3718 | unsigned int ioctl, unsigned long arg) | |
3719 | { | |
07c45a36 | 3720 | long r = -EINVAL; |
f17abe9a AK |
3721 | |
3722 | switch (ioctl) { | |
3723 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
3724 | if (arg) |
3725 | goto out; | |
f17abe9a AK |
3726 | r = KVM_API_VERSION; |
3727 | break; | |
3728 | case KVM_CREATE_VM: | |
e08b9637 | 3729 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 3730 | break; |
018d00d2 | 3731 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 3732 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 3733 | break; |
07c45a36 | 3734 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
3735 | if (arg) |
3736 | goto out; | |
adb1ff46 AK |
3737 | r = PAGE_SIZE; /* struct kvm_run */ |
3738 | #ifdef CONFIG_X86 | |
3739 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 3740 | #endif |
4b4357e0 | 3741 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 3742 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 3743 | #endif |
07c45a36 | 3744 | break; |
d4c9ff2d FEL |
3745 | case KVM_TRACE_ENABLE: |
3746 | case KVM_TRACE_PAUSE: | |
3747 | case KVM_TRACE_DISABLE: | |
2023a29c | 3748 | r = -EOPNOTSUPP; |
d4c9ff2d | 3749 | break; |
6aa8b732 | 3750 | default: |
043405e1 | 3751 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
3752 | } |
3753 | out: | |
3754 | return r; | |
3755 | } | |
3756 | ||
6aa8b732 | 3757 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 3758 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 3759 | .llseek = noop_llseek, |
7ddfd3e0 | 3760 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
3761 | }; |
3762 | ||
3763 | static struct miscdevice kvm_dev = { | |
bbe4432e | 3764 | KVM_MINOR, |
6aa8b732 AK |
3765 | "kvm", |
3766 | &kvm_chardev_ops, | |
3767 | }; | |
3768 | ||
75b7127c | 3769 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
3770 | { |
3771 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 3772 | int r; |
1b6c0168 | 3773 | |
7f59f492 | 3774 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3775 | return; |
10474ae8 | 3776 | |
7f59f492 | 3777 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 3778 | |
13a34e06 | 3779 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
3780 | |
3781 | if (r) { | |
3782 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
3783 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 3784 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 3785 | } |
1b6c0168 AK |
3786 | } |
3787 | ||
8c18b2d2 | 3788 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 3789 | { |
4a937f96 | 3790 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3791 | if (kvm_usage_count) |
3792 | hardware_enable_nolock(NULL); | |
4a937f96 | 3793 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3794 | return 0; |
75b7127c TY |
3795 | } |
3796 | ||
3797 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
3798 | { |
3799 | int cpu = raw_smp_processor_id(); | |
3800 | ||
7f59f492 | 3801 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3802 | return; |
7f59f492 | 3803 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 3804 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
3805 | } |
3806 | ||
8c18b2d2 | 3807 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 3808 | { |
4a937f96 | 3809 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3810 | if (kvm_usage_count) |
3811 | hardware_disable_nolock(NULL); | |
4a937f96 | 3812 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3813 | return 0; |
75b7127c TY |
3814 | } |
3815 | ||
10474ae8 AG |
3816 | static void hardware_disable_all_nolock(void) |
3817 | { | |
3818 | BUG_ON(!kvm_usage_count); | |
3819 | ||
3820 | kvm_usage_count--; | |
3821 | if (!kvm_usage_count) | |
75b7127c | 3822 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
3823 | } |
3824 | ||
3825 | static void hardware_disable_all(void) | |
3826 | { | |
4a937f96 | 3827 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 3828 | hardware_disable_all_nolock(); |
4a937f96 | 3829 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3830 | } |
3831 | ||
3832 | static int hardware_enable_all(void) | |
3833 | { | |
3834 | int r = 0; | |
3835 | ||
4a937f96 | 3836 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
3837 | |
3838 | kvm_usage_count++; | |
3839 | if (kvm_usage_count == 1) { | |
3840 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 3841 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
3842 | |
3843 | if (atomic_read(&hardware_enable_failed)) { | |
3844 | hardware_disable_all_nolock(); | |
3845 | r = -EBUSY; | |
3846 | } | |
3847 | } | |
3848 | ||
4a937f96 | 3849 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3850 | |
3851 | return r; | |
3852 | } | |
3853 | ||
9a2b85c6 | 3854 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 3855 | void *v) |
9a2b85c6 | 3856 | { |
8e1c1815 SY |
3857 | /* |
3858 | * Some (well, at least mine) BIOSes hang on reboot if | |
3859 | * in vmx root mode. | |
3860 | * | |
3861 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
3862 | */ | |
1170adc6 | 3863 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 3864 | kvm_rebooting = true; |
75b7127c | 3865 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
3866 | return NOTIFY_OK; |
3867 | } | |
3868 | ||
3869 | static struct notifier_block kvm_reboot_notifier = { | |
3870 | .notifier_call = kvm_reboot, | |
3871 | .priority = 0, | |
3872 | }; | |
3873 | ||
e93f8a0f | 3874 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
3875 | { |
3876 | int i; | |
3877 | ||
3878 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 3879 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
3880 | |
3881 | kvm_iodevice_destructor(pos); | |
3882 | } | |
e93f8a0f | 3883 | kfree(bus); |
2eeb2e94 GH |
3884 | } |
3885 | ||
c21fbff1 | 3886 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 3887 | const struct kvm_io_range *r2) |
743eeb0b | 3888 | { |
8f4216c7 JW |
3889 | gpa_t addr1 = r1->addr; |
3890 | gpa_t addr2 = r2->addr; | |
3891 | ||
3892 | if (addr1 < addr2) | |
743eeb0b | 3893 | return -1; |
8f4216c7 JW |
3894 | |
3895 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
3896 | * accept any overlapping write. Any order is acceptable for | |
3897 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
3898 | * we process all of them. | |
3899 | */ | |
3900 | if (r2->len) { | |
3901 | addr1 += r1->len; | |
3902 | addr2 += r2->len; | |
3903 | } | |
3904 | ||
3905 | if (addr1 > addr2) | |
743eeb0b | 3906 | return 1; |
8f4216c7 | 3907 | |
743eeb0b SL |
3908 | return 0; |
3909 | } | |
3910 | ||
a343c9b7 PB |
3911 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
3912 | { | |
c21fbff1 | 3913 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
3914 | } |
3915 | ||
39369f7a | 3916 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
3917 | gpa_t addr, int len) |
3918 | { | |
3919 | struct kvm_io_range *range, key; | |
3920 | int off; | |
3921 | ||
3922 | key = (struct kvm_io_range) { | |
3923 | .addr = addr, | |
3924 | .len = len, | |
3925 | }; | |
3926 | ||
3927 | range = bsearch(&key, bus->range, bus->dev_count, | |
3928 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
3929 | if (range == NULL) | |
3930 | return -ENOENT; | |
3931 | ||
3932 | off = range - bus->range; | |
3933 | ||
c21fbff1 | 3934 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
3935 | off--; |
3936 | ||
3937 | return off; | |
3938 | } | |
3939 | ||
e32edf4f | 3940 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
3941 | struct kvm_io_range *range, const void *val) |
3942 | { | |
3943 | int idx; | |
3944 | ||
3945 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
3946 | if (idx < 0) | |
3947 | return -EOPNOTSUPP; | |
3948 | ||
3949 | while (idx < bus->dev_count && | |
c21fbff1 | 3950 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3951 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3952 | range->len, val)) |
3953 | return idx; | |
3954 | idx++; | |
3955 | } | |
3956 | ||
3957 | return -EOPNOTSUPP; | |
3958 | } | |
3959 | ||
bda9020e | 3960 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 3961 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 3962 | int len, const void *val) |
2eeb2e94 | 3963 | { |
90d83dc3 | 3964 | struct kvm_io_bus *bus; |
743eeb0b | 3965 | struct kvm_io_range range; |
126a5af5 | 3966 | int r; |
743eeb0b SL |
3967 | |
3968 | range = (struct kvm_io_range) { | |
3969 | .addr = addr, | |
3970 | .len = len, | |
3971 | }; | |
90d83dc3 | 3972 | |
e32edf4f | 3973 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
3974 | if (!bus) |
3975 | return -ENOMEM; | |
e32edf4f | 3976 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3977 | return r < 0 ? r : 0; |
3978 | } | |
a2420107 | 3979 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
3980 | |
3981 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
3982 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
3983 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
3984 | { |
3985 | struct kvm_io_bus *bus; | |
3986 | struct kvm_io_range range; | |
3987 | ||
3988 | range = (struct kvm_io_range) { | |
3989 | .addr = addr, | |
3990 | .len = len, | |
3991 | }; | |
3992 | ||
e32edf4f | 3993 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
3994 | if (!bus) |
3995 | return -ENOMEM; | |
126a5af5 CH |
3996 | |
3997 | /* First try the device referenced by cookie. */ | |
3998 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 3999 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 4000 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
4001 | val)) |
4002 | return cookie; | |
4003 | ||
4004 | /* | |
4005 | * cookie contained garbage; fall back to search and return the | |
4006 | * correct cookie value. | |
4007 | */ | |
e32edf4f | 4008 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4009 | } |
4010 | ||
e32edf4f NN |
4011 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
4012 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
4013 | { |
4014 | int idx; | |
4015 | ||
4016 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
4017 | if (idx < 0) |
4018 | return -EOPNOTSUPP; | |
4019 | ||
4020 | while (idx < bus->dev_count && | |
c21fbff1 | 4021 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4022 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4023 | range->len, val)) |
4024 | return idx; | |
743eeb0b SL |
4025 | idx++; |
4026 | } | |
4027 | ||
bda9020e MT |
4028 | return -EOPNOTSUPP; |
4029 | } | |
2eeb2e94 | 4030 | |
bda9020e | 4031 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 4032 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 4033 | int len, void *val) |
bda9020e | 4034 | { |
90d83dc3 | 4035 | struct kvm_io_bus *bus; |
743eeb0b | 4036 | struct kvm_io_range range; |
126a5af5 | 4037 | int r; |
743eeb0b SL |
4038 | |
4039 | range = (struct kvm_io_range) { | |
4040 | .addr = addr, | |
4041 | .len = len, | |
4042 | }; | |
e93f8a0f | 4043 | |
e32edf4f | 4044 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4045 | if (!bus) |
4046 | return -ENOMEM; | |
e32edf4f | 4047 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
4048 | return r < 0 ? r : 0; |
4049 | } | |
743eeb0b | 4050 | |
79fac95e | 4051 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
4052 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
4053 | int len, struct kvm_io_device *dev) | |
6c474694 | 4054 | { |
d4c67a7a | 4055 | int i; |
e93f8a0f | 4056 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 4057 | struct kvm_io_range range; |
090b7aff | 4058 | |
4a12f951 | 4059 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
4060 | if (!bus) |
4061 | return -ENOMEM; | |
4062 | ||
6ea34c9b AK |
4063 | /* exclude ioeventfd which is limited by maximum fd */ |
4064 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 4065 | return -ENOSPC; |
2eeb2e94 | 4066 | |
90952cd3 | 4067 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 4068 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
4069 | if (!new_bus) |
4070 | return -ENOMEM; | |
d4c67a7a GH |
4071 | |
4072 | range = (struct kvm_io_range) { | |
4073 | .addr = addr, | |
4074 | .len = len, | |
4075 | .dev = dev, | |
4076 | }; | |
4077 | ||
4078 | for (i = 0; i < bus->dev_count; i++) | |
4079 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
4080 | break; | |
4081 | ||
4082 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
4083 | new_bus->dev_count++; | |
4084 | new_bus->range[i] = range; | |
4085 | memcpy(new_bus->range + i + 1, bus->range + i, | |
4086 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
4087 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
4088 | synchronize_srcu_expedited(&kvm->srcu); | |
4089 | kfree(bus); | |
090b7aff GH |
4090 | |
4091 | return 0; | |
4092 | } | |
4093 | ||
79fac95e | 4094 | /* Caller must hold slots_lock. */ |
6205476b SC |
4095 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
4096 | struct kvm_io_device *dev) | |
090b7aff | 4097 | { |
6a622979 | 4098 | int i, j; |
e93f8a0f | 4099 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 4100 | |
4a12f951 | 4101 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 4102 | if (!bus) |
6205476b | 4103 | return 0; |
df630b8c | 4104 | |
a1300716 AK |
4105 | for (i = 0; i < bus->dev_count; i++) |
4106 | if (bus->range[i].dev == dev) { | |
090b7aff GH |
4107 | break; |
4108 | } | |
e93f8a0f | 4109 | |
90db1043 | 4110 | if (i == bus->dev_count) |
6205476b | 4111 | return 0; |
a1300716 | 4112 | |
90952cd3 | 4113 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 4114 | GFP_KERNEL_ACCOUNT); |
6a622979 RK |
4115 | if (new_bus) { |
4116 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
4117 | new_bus->dev_count--; | |
4118 | memcpy(new_bus->range + i, bus->range + i + 1, | |
4119 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
4120 | } else { | |
90db1043 | 4121 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
6a622979 RK |
4122 | for (j = 0; j < bus->dev_count; j++) { |
4123 | if (j == i) | |
4124 | continue; | |
4125 | kvm_iodevice_destructor(bus->range[j].dev); | |
4126 | } | |
90db1043 | 4127 | } |
a1300716 | 4128 | |
e93f8a0f MT |
4129 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
4130 | synchronize_srcu_expedited(&kvm->srcu); | |
4131 | kfree(bus); | |
6205476b | 4132 | return new_bus ? 0 : -ENOMEM; |
2eeb2e94 GH |
4133 | } |
4134 | ||
8a39d006 AP |
4135 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
4136 | gpa_t addr) | |
4137 | { | |
4138 | struct kvm_io_bus *bus; | |
4139 | int dev_idx, srcu_idx; | |
4140 | struct kvm_io_device *iodev = NULL; | |
4141 | ||
4142 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
4143 | ||
4144 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
4145 | if (!bus) |
4146 | goto out_unlock; | |
8a39d006 AP |
4147 | |
4148 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
4149 | if (dev_idx < 0) | |
4150 | goto out_unlock; | |
4151 | ||
4152 | iodev = bus->range[dev_idx].dev; | |
4153 | ||
4154 | out_unlock: | |
4155 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
4156 | ||
4157 | return iodev; | |
4158 | } | |
4159 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
4160 | ||
536a6f88 JF |
4161 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
4162 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
4163 | const char *fmt) | |
4164 | { | |
4165 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
4166 | inode->i_private; | |
4167 | ||
4168 | /* The debugfs files are a reference to the kvm struct which | |
4169 | * is still valid when kvm_destroy_vm is called. | |
4170 | * To avoid the race between open and the removal of the debugfs | |
4171 | * directory we test against the users count. | |
4172 | */ | |
e3736c3e | 4173 | if (!refcount_inc_not_zero(&stat_data->kvm->users_count)) |
536a6f88 JF |
4174 | return -ENOENT; |
4175 | ||
833b45de PB |
4176 | if (simple_attr_open(inode, file, get, |
4177 | stat_data->mode & S_IWUGO ? set : NULL, | |
4178 | fmt)) { | |
536a6f88 JF |
4179 | kvm_put_kvm(stat_data->kvm); |
4180 | return -ENOMEM; | |
4181 | } | |
4182 | ||
4183 | return 0; | |
4184 | } | |
4185 | ||
4186 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
4187 | { | |
4188 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
4189 | inode->i_private; | |
4190 | ||
4191 | simple_attr_release(inode, file); | |
4192 | kvm_put_kvm(stat_data->kvm); | |
4193 | ||
4194 | return 0; | |
4195 | } | |
4196 | ||
4197 | static int vm_stat_get_per_vm(void *data, u64 *val) | |
4198 | { | |
4199 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
4200 | ||
8a7e75d4 | 4201 | *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset); |
536a6f88 JF |
4202 | |
4203 | return 0; | |
4204 | } | |
4205 | ||
ce35ef27 SJS |
4206 | static int vm_stat_clear_per_vm(void *data, u64 val) |
4207 | { | |
4208 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
4209 | ||
4210 | if (val) | |
4211 | return -EINVAL; | |
4212 | ||
4213 | *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0; | |
4214 | ||
4215 | return 0; | |
4216 | } | |
4217 | ||
536a6f88 JF |
4218 | static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file) |
4219 | { | |
4220 | __simple_attr_check_format("%llu\n", 0ull); | |
4221 | return kvm_debugfs_open(inode, file, vm_stat_get_per_vm, | |
ce35ef27 | 4222 | vm_stat_clear_per_vm, "%llu\n"); |
536a6f88 JF |
4223 | } |
4224 | ||
4225 | static const struct file_operations vm_stat_get_per_vm_fops = { | |
4226 | .owner = THIS_MODULE, | |
4227 | .open = vm_stat_get_per_vm_open, | |
4228 | .release = kvm_debugfs_release, | |
4229 | .read = simple_attr_read, | |
4230 | .write = simple_attr_write, | |
3bed8888 | 4231 | .llseek = no_llseek, |
536a6f88 JF |
4232 | }; |
4233 | ||
4234 | static int vcpu_stat_get_per_vm(void *data, u64 *val) | |
4235 | { | |
4236 | int i; | |
4237 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
4238 | struct kvm_vcpu *vcpu; | |
4239 | ||
4240 | *val = 0; | |
4241 | ||
4242 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) | |
8a7e75d4 | 4243 | *val += *(u64 *)((void *)vcpu + stat_data->offset); |
536a6f88 JF |
4244 | |
4245 | return 0; | |
4246 | } | |
4247 | ||
ce35ef27 SJS |
4248 | static int vcpu_stat_clear_per_vm(void *data, u64 val) |
4249 | { | |
4250 | int i; | |
4251 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
4252 | struct kvm_vcpu *vcpu; | |
4253 | ||
4254 | if (val) | |
4255 | return -EINVAL; | |
4256 | ||
4257 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) | |
4258 | *(u64 *)((void *)vcpu + stat_data->offset) = 0; | |
4259 | ||
4260 | return 0; | |
4261 | } | |
4262 | ||
536a6f88 JF |
4263 | static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file) |
4264 | { | |
4265 | __simple_attr_check_format("%llu\n", 0ull); | |
4266 | return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm, | |
ce35ef27 | 4267 | vcpu_stat_clear_per_vm, "%llu\n"); |
536a6f88 JF |
4268 | } |
4269 | ||
4270 | static const struct file_operations vcpu_stat_get_per_vm_fops = { | |
4271 | .owner = THIS_MODULE, | |
4272 | .open = vcpu_stat_get_per_vm_open, | |
4273 | .release = kvm_debugfs_release, | |
4274 | .read = simple_attr_read, | |
4275 | .write = simple_attr_write, | |
3bed8888 | 4276 | .llseek = no_llseek, |
536a6f88 JF |
4277 | }; |
4278 | ||
4279 | static const struct file_operations *stat_fops_per_vm[] = { | |
4280 | [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops, | |
4281 | [KVM_STAT_VM] = &vm_stat_get_per_vm_fops, | |
4282 | }; | |
4283 | ||
8b88b099 | 4284 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
4285 | { |
4286 | unsigned offset = (long)_offset; | |
ba1389b7 | 4287 | struct kvm *kvm; |
536a6f88 JF |
4288 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
4289 | u64 tmp_val; | |
ba1389b7 | 4290 | |
8b88b099 | 4291 | *val = 0; |
0d9ce162 | 4292 | mutex_lock(&kvm_lock); |
536a6f88 JF |
4293 | list_for_each_entry(kvm, &vm_list, vm_list) { |
4294 | stat_tmp.kvm = kvm; | |
4295 | vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val); | |
4296 | *val += tmp_val; | |
4297 | } | |
0d9ce162 | 4298 | mutex_unlock(&kvm_lock); |
8b88b099 | 4299 | return 0; |
ba1389b7 AK |
4300 | } |
4301 | ||
ce35ef27 SJS |
4302 | static int vm_stat_clear(void *_offset, u64 val) |
4303 | { | |
4304 | unsigned offset = (long)_offset; | |
4305 | struct kvm *kvm; | |
4306 | struct kvm_stat_data stat_tmp = {.offset = offset}; | |
4307 | ||
4308 | if (val) | |
4309 | return -EINVAL; | |
4310 | ||
0d9ce162 | 4311 | mutex_lock(&kvm_lock); |
ce35ef27 SJS |
4312 | list_for_each_entry(kvm, &vm_list, vm_list) { |
4313 | stat_tmp.kvm = kvm; | |
4314 | vm_stat_clear_per_vm((void *)&stat_tmp, 0); | |
4315 | } | |
0d9ce162 | 4316 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
4317 | |
4318 | return 0; | |
4319 | } | |
4320 | ||
4321 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
ba1389b7 | 4322 | |
8b88b099 | 4323 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
4324 | { |
4325 | unsigned offset = (long)_offset; | |
1165f5fe | 4326 | struct kvm *kvm; |
536a6f88 JF |
4327 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
4328 | u64 tmp_val; | |
1165f5fe | 4329 | |
8b88b099 | 4330 | *val = 0; |
0d9ce162 | 4331 | mutex_lock(&kvm_lock); |
536a6f88 JF |
4332 | list_for_each_entry(kvm, &vm_list, vm_list) { |
4333 | stat_tmp.kvm = kvm; | |
4334 | vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val); | |
4335 | *val += tmp_val; | |
4336 | } | |
0d9ce162 | 4337 | mutex_unlock(&kvm_lock); |
8b88b099 | 4338 | return 0; |
1165f5fe AK |
4339 | } |
4340 | ||
ce35ef27 SJS |
4341 | static int vcpu_stat_clear(void *_offset, u64 val) |
4342 | { | |
4343 | unsigned offset = (long)_offset; | |
4344 | struct kvm *kvm; | |
4345 | struct kvm_stat_data stat_tmp = {.offset = offset}; | |
4346 | ||
4347 | if (val) | |
4348 | return -EINVAL; | |
4349 | ||
0d9ce162 | 4350 | mutex_lock(&kvm_lock); |
ce35ef27 SJS |
4351 | list_for_each_entry(kvm, &vm_list, vm_list) { |
4352 | stat_tmp.kvm = kvm; | |
4353 | vcpu_stat_clear_per_vm((void *)&stat_tmp, 0); | |
4354 | } | |
0d9ce162 | 4355 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
4356 | |
4357 | return 0; | |
4358 | } | |
4359 | ||
4360 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
4361 | "%llu\n"); | |
ba1389b7 | 4362 | |
828c0950 | 4363 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
4364 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
4365 | [KVM_STAT_VM] = &vm_stat_fops, | |
4366 | }; | |
1165f5fe | 4367 | |
286de8f6 CI |
4368 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
4369 | { | |
4370 | struct kobj_uevent_env *env; | |
286de8f6 CI |
4371 | unsigned long long created, active; |
4372 | ||
4373 | if (!kvm_dev.this_device || !kvm) | |
4374 | return; | |
4375 | ||
0d9ce162 | 4376 | mutex_lock(&kvm_lock); |
286de8f6 CI |
4377 | if (type == KVM_EVENT_CREATE_VM) { |
4378 | kvm_createvm_count++; | |
4379 | kvm_active_vms++; | |
4380 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
4381 | kvm_active_vms--; | |
4382 | } | |
4383 | created = kvm_createvm_count; | |
4384 | active = kvm_active_vms; | |
0d9ce162 | 4385 | mutex_unlock(&kvm_lock); |
286de8f6 | 4386 | |
b12ce36a | 4387 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
4388 | if (!env) |
4389 | return; | |
4390 | ||
4391 | add_uevent_var(env, "CREATED=%llu", created); | |
4392 | add_uevent_var(env, "COUNT=%llu", active); | |
4393 | ||
fdeaf7e3 | 4394 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 4395 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
4396 | kvm->userspace_pid = task_pid_nr(current); |
4397 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 4398 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
4399 | } |
4400 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 4401 | |
31542cd6 | 4402 | if (kvm->debugfs_dentry) { |
b12ce36a | 4403 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
4404 | |
4405 | if (p) { | |
4406 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
4407 | if (!IS_ERR(tmp)) | |
4408 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
4409 | kfree(p); | |
286de8f6 CI |
4410 | } |
4411 | } | |
4412 | /* no need for checks, since we are adding at most only 5 keys */ | |
4413 | env->envp[env->envp_idx++] = NULL; | |
4414 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
4415 | kfree(env); | |
286de8f6 CI |
4416 | } |
4417 | ||
929f45e3 | 4418 | static void kvm_init_debug(void) |
6aa8b732 AK |
4419 | { |
4420 | struct kvm_stats_debugfs_item *p; | |
4421 | ||
76f7c879 | 4422 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 4423 | |
536a6f88 JF |
4424 | kvm_debugfs_num_entries = 0; |
4425 | for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) { | |
833b45de PB |
4426 | int mode = p->mode ? p->mode : 0644; |
4427 | debugfs_create_file(p->name, mode, kvm_debugfs_dir, | |
929f45e3 GKH |
4428 | (void *)(long)p->offset, |
4429 | stat_fops[p->kind]); | |
4f69b680 | 4430 | } |
6aa8b732 AK |
4431 | } |
4432 | ||
fb3600cc | 4433 | static int kvm_suspend(void) |
59ae6c6b | 4434 | { |
10474ae8 | 4435 | if (kvm_usage_count) |
75b7127c | 4436 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
4437 | return 0; |
4438 | } | |
4439 | ||
fb3600cc | 4440 | static void kvm_resume(void) |
59ae6c6b | 4441 | { |
ca84d1a2 | 4442 | if (kvm_usage_count) { |
2eb06c30 WL |
4443 | #ifdef CONFIG_LOCKDEP |
4444 | WARN_ON(lockdep_is_held(&kvm_count_lock)); | |
4445 | #endif | |
75b7127c | 4446 | hardware_enable_nolock(NULL); |
ca84d1a2 | 4447 | } |
59ae6c6b AK |
4448 | } |
4449 | ||
fb3600cc | 4450 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
4451 | .suspend = kvm_suspend, |
4452 | .resume = kvm_resume, | |
4453 | }; | |
4454 | ||
15ad7146 AK |
4455 | static inline |
4456 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
4457 | { | |
4458 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
4459 | } | |
4460 | ||
4461 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
4462 | { | |
4463 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 4464 | |
046ddeed | 4465 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 4466 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 4467 | |
e790d9ef RK |
4468 | kvm_arch_sched_in(vcpu, cpu); |
4469 | ||
e9b11c17 | 4470 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
4471 | } |
4472 | ||
4473 | static void kvm_sched_out(struct preempt_notifier *pn, | |
4474 | struct task_struct *next) | |
4475 | { | |
4476 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
4477 | ||
d73eb57b | 4478 | if (current->state == TASK_RUNNING) { |
046ddeed | 4479 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
4480 | WRITE_ONCE(vcpu->ready, true); |
4481 | } | |
e9b11c17 | 4482 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
4483 | } |
4484 | ||
f257d6dc SC |
4485 | static void check_processor_compat(void *rtn) |
4486 | { | |
4487 | *(int *)rtn = kvm_arch_check_processor_compat(); | |
4488 | } | |
4489 | ||
0ee75bea | 4490 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 4491 | struct module *module) |
6aa8b732 AK |
4492 | { |
4493 | int r; | |
002c7f7c | 4494 | int cpu; |
6aa8b732 | 4495 | |
f8c16bba ZX |
4496 | r = kvm_arch_init(opaque); |
4497 | if (r) | |
d2308784 | 4498 | goto out_fail; |
cb498ea2 | 4499 | |
7dac16c3 AH |
4500 | /* |
4501 | * kvm_arch_init makes sure there's at most one caller | |
4502 | * for architectures that support multiple implementations, | |
4503 | * like intel and amd on x86. | |
36343f6e PB |
4504 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
4505 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 4506 | */ |
36343f6e PB |
4507 | r = kvm_irqfd_init(); |
4508 | if (r) | |
4509 | goto out_irqfd; | |
7dac16c3 | 4510 | |
8437a617 | 4511 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
4512 | r = -ENOMEM; |
4513 | goto out_free_0; | |
4514 | } | |
4515 | ||
e9b11c17 | 4516 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 4517 | if (r < 0) |
7f59f492 | 4518 | goto out_free_0a; |
6aa8b732 | 4519 | |
002c7f7c | 4520 | for_each_online_cpu(cpu) { |
f257d6dc | 4521 | smp_call_function_single(cpu, check_processor_compat, &r, 1); |
002c7f7c | 4522 | if (r < 0) |
d2308784 | 4523 | goto out_free_1; |
002c7f7c YS |
4524 | } |
4525 | ||
73c1b41e | 4526 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 4527 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 4528 | if (r) |
d2308784 | 4529 | goto out_free_2; |
6aa8b732 AK |
4530 | register_reboot_notifier(&kvm_reboot_notifier); |
4531 | ||
c16f862d | 4532 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
4533 | if (!vcpu_align) |
4534 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
4535 | kvm_vcpu_cache = |
4536 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
4537 | SLAB_ACCOUNT, | |
4538 | offsetof(struct kvm_vcpu, arch), | |
4539 | sizeof_field(struct kvm_vcpu, arch), | |
4540 | NULL); | |
c16f862d RR |
4541 | if (!kvm_vcpu_cache) { |
4542 | r = -ENOMEM; | |
fb3600cc | 4543 | goto out_free_3; |
c16f862d RR |
4544 | } |
4545 | ||
af585b92 GN |
4546 | r = kvm_async_pf_init(); |
4547 | if (r) | |
4548 | goto out_free; | |
4549 | ||
6aa8b732 | 4550 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
4551 | kvm_vm_fops.owner = module; |
4552 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
4553 | |
4554 | r = misc_register(&kvm_dev); | |
4555 | if (r) { | |
1170adc6 | 4556 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 4557 | goto out_unreg; |
6aa8b732 AK |
4558 | } |
4559 | ||
fb3600cc RW |
4560 | register_syscore_ops(&kvm_syscore_ops); |
4561 | ||
15ad7146 AK |
4562 | kvm_preempt_ops.sched_in = kvm_sched_in; |
4563 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
4564 | ||
929f45e3 | 4565 | kvm_init_debug(); |
0ea4ed8e | 4566 | |
3c3c29fd PB |
4567 | r = kvm_vfio_ops_init(); |
4568 | WARN_ON(r); | |
4569 | ||
c7addb90 | 4570 | return 0; |
6aa8b732 | 4571 | |
af585b92 GN |
4572 | out_unreg: |
4573 | kvm_async_pf_deinit(); | |
6aa8b732 | 4574 | out_free: |
c16f862d | 4575 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 4576 | out_free_3: |
6aa8b732 | 4577 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4578 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 4579 | out_free_2: |
d2308784 | 4580 | out_free_1: |
e9b11c17 | 4581 | kvm_arch_hardware_unsetup(); |
7f59f492 RR |
4582 | out_free_0a: |
4583 | free_cpumask_var(cpus_hardware_enabled); | |
d2308784 | 4584 | out_free_0: |
a0f155e9 | 4585 | kvm_irqfd_exit(); |
36343f6e | 4586 | out_irqfd: |
7dac16c3 AH |
4587 | kvm_arch_exit(); |
4588 | out_fail: | |
6aa8b732 AK |
4589 | return r; |
4590 | } | |
cb498ea2 | 4591 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 4592 | |
cb498ea2 | 4593 | void kvm_exit(void) |
6aa8b732 | 4594 | { |
4bd33b56 | 4595 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 4596 | misc_deregister(&kvm_dev); |
c16f862d | 4597 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 4598 | kvm_async_pf_deinit(); |
fb3600cc | 4599 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 4600 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4601 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 4602 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 4603 | kvm_arch_hardware_unsetup(); |
f8c16bba | 4604 | kvm_arch_exit(); |
a0f155e9 | 4605 | kvm_irqfd_exit(); |
7f59f492 | 4606 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 4607 | kvm_vfio_ops_exit(); |
6aa8b732 | 4608 | } |
cb498ea2 | 4609 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
4610 | |
4611 | struct kvm_vm_worker_thread_context { | |
4612 | struct kvm *kvm; | |
4613 | struct task_struct *parent; | |
4614 | struct completion init_done; | |
4615 | kvm_vm_thread_fn_t thread_fn; | |
4616 | uintptr_t data; | |
4617 | int err; | |
4618 | }; | |
4619 | ||
4620 | static int kvm_vm_worker_thread(void *context) | |
4621 | { | |
4622 | /* | |
4623 | * The init_context is allocated on the stack of the parent thread, so | |
4624 | * we have to locally copy anything that is needed beyond initialization | |
4625 | */ | |
4626 | struct kvm_vm_worker_thread_context *init_context = context; | |
4627 | struct kvm *kvm = init_context->kvm; | |
4628 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
4629 | uintptr_t data = init_context->data; | |
4630 | int err; | |
4631 | ||
4632 | err = kthread_park(current); | |
4633 | /* kthread_park(current) is never supposed to return an error */ | |
4634 | WARN_ON(err != 0); | |
4635 | if (err) | |
4636 | goto init_complete; | |
4637 | ||
4638 | err = cgroup_attach_task_all(init_context->parent, current); | |
4639 | if (err) { | |
4640 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
4641 | __func__, err); | |
4642 | goto init_complete; | |
4643 | } | |
4644 | ||
4645 | set_user_nice(current, task_nice(init_context->parent)); | |
4646 | ||
4647 | init_complete: | |
4648 | init_context->err = err; | |
4649 | complete(&init_context->init_done); | |
4650 | init_context = NULL; | |
4651 | ||
4652 | if (err) | |
4653 | return err; | |
4654 | ||
4655 | /* Wait to be woken up by the spawner before proceeding. */ | |
4656 | kthread_parkme(); | |
4657 | ||
4658 | if (!kthread_should_stop()) | |
4659 | err = thread_fn(kvm, data); | |
4660 | ||
4661 | return err; | |
4662 | } | |
4663 | ||
4664 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
4665 | uintptr_t data, const char *name, | |
4666 | struct task_struct **thread_ptr) | |
4667 | { | |
4668 | struct kvm_vm_worker_thread_context init_context = {}; | |
4669 | struct task_struct *thread; | |
4670 | ||
4671 | *thread_ptr = NULL; | |
4672 | init_context.kvm = kvm; | |
4673 | init_context.parent = current; | |
4674 | init_context.thread_fn = thread_fn; | |
4675 | init_context.data = data; | |
4676 | init_completion(&init_context.init_done); | |
4677 | ||
4678 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
4679 | "%s-%d", name, task_pid_nr(current)); | |
4680 | if (IS_ERR(thread)) | |
4681 | return PTR_ERR(thread); | |
4682 | ||
4683 | /* kthread_run is never supposed to return NULL */ | |
4684 | WARN_ON(thread == NULL); | |
4685 | ||
4686 | wait_for_completion(&init_context.init_done); | |
4687 | ||
4688 | if (!init_context.err) | |
4689 | *thread_ptr = thread; | |
4690 | ||
4691 | return init_context.err; | |
4692 | } |