]>
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 | ||
bccf2150 AK |
118 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
119 | unsigned long arg); | |
de8e5d74 | 120 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
121 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
122 | unsigned long arg); | |
7ddfd3e0 MZ |
123 | #define KVM_COMPAT(c) .compat_ioctl = (c) |
124 | #else | |
9cb09e7c MZ |
125 | /* |
126 | * For architectures that don't implement a compat infrastructure, | |
127 | * adopt a double line of defense: | |
128 | * - Prevent a compat task from opening /dev/kvm | |
129 | * - If the open has been done by a 64bit task, and the KVM fd | |
130 | * passed to a compat task, let the ioctls fail. | |
131 | */ | |
7ddfd3e0 MZ |
132 | static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl, |
133 | unsigned long arg) { return -EINVAL; } | |
b9876e6d MZ |
134 | |
135 | static int kvm_no_compat_open(struct inode *inode, struct file *file) | |
136 | { | |
137 | return is_compat_task() ? -ENODEV : 0; | |
138 | } | |
139 | #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \ | |
140 | .open = kvm_no_compat_open | |
1dda606c | 141 | #endif |
10474ae8 AG |
142 | static int hardware_enable_all(void); |
143 | static void hardware_disable_all(void); | |
bccf2150 | 144 | |
e93f8a0f | 145 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 146 | |
bc009e43 | 147 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn); |
e93f8a0f | 148 | |
52480137 | 149 | __visible bool kvm_rebooting; |
b7c4145b | 150 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 151 | |
54dee993 MT |
152 | static bool largepages_enabled = true; |
153 | ||
286de8f6 CI |
154 | #define KVM_EVENT_CREATE_VM 0 |
155 | #define KVM_EVENT_DESTROY_VM 1 | |
156 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); | |
157 | static unsigned long long kvm_createvm_count; | |
158 | static unsigned long long kvm_active_vms; | |
159 | ||
93065ac7 MH |
160 | __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
161 | unsigned long start, unsigned long end, bool blockable) | |
b1394e74 | 162 | { |
93065ac7 | 163 | return 0; |
b1394e74 RK |
164 | } |
165 | ||
a78986aa SC |
166 | bool kvm_is_zone_device_pfn(kvm_pfn_t pfn) |
167 | { | |
168 | /* | |
169 | * The metadata used by is_zone_device_page() to determine whether or | |
170 | * not a page is ZONE_DEVICE is guaranteed to be valid if and only if | |
171 | * the device has been pinned, e.g. by get_user_pages(). WARN if the | |
172 | * page_count() is zero to help detect bad usage of this helper. | |
173 | */ | |
174 | if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn)))) | |
175 | return false; | |
176 | ||
177 | return is_zone_device_page(pfn_to_page(pfn)); | |
178 | } | |
179 | ||
ba049e93 | 180 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
cbff90a7 | 181 | { |
a78986aa SC |
182 | /* |
183 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting | |
184 | * perspective they are "normal" pages, albeit with slightly different | |
185 | * usage rules. | |
186 | */ | |
11feeb49 | 187 | if (pfn_valid(pfn)) |
a78986aa SC |
188 | return PageReserved(pfn_to_page(pfn)) && |
189 | !kvm_is_zone_device_pfn(pfn); | |
cbff90a7 BAY |
190 | |
191 | return true; | |
192 | } | |
193 | ||
bccf2150 AK |
194 | /* |
195 | * Switches to specified vcpu, until a matching vcpu_put() | |
196 | */ | |
ec7660cc | 197 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 198 | { |
ec7660cc | 199 | int cpu = get_cpu(); |
15ad7146 | 200 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 201 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 202 | put_cpu(); |
6aa8b732 | 203 | } |
2f1fe811 | 204 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 205 | |
313a3dc7 | 206 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 207 | { |
15ad7146 | 208 | preempt_disable(); |
313a3dc7 | 209 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
210 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
211 | preempt_enable(); | |
6aa8b732 | 212 | } |
2f1fe811 | 213 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 214 | |
7a97cec2 PB |
215 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
216 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
217 | { | |
218 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
219 | ||
220 | /* | |
221 | * We need to wait for the VCPU to reenable interrupts and get out of | |
222 | * READING_SHADOW_PAGE_TABLES mode. | |
223 | */ | |
224 | if (req & KVM_REQUEST_WAIT) | |
225 | return mode != OUTSIDE_GUEST_MODE; | |
226 | ||
227 | /* | |
228 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
229 | */ | |
230 | return mode == IN_GUEST_MODE; | |
231 | } | |
232 | ||
d9e368d6 AK |
233 | static void ack_flush(void *_completed) |
234 | { | |
d9e368d6 AK |
235 | } |
236 | ||
b49defe8 PB |
237 | static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait) |
238 | { | |
239 | if (unlikely(!cpus)) | |
240 | cpus = cpu_online_mask; | |
241 | ||
242 | if (cpumask_empty(cpus)) | |
243 | return false; | |
244 | ||
245 | smp_call_function_many(cpus, ack_flush, NULL, wait); | |
246 | return true; | |
247 | } | |
248 | ||
7053df4e VK |
249 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
250 | unsigned long *vcpu_bitmap, cpumask_var_t tmp) | |
d9e368d6 | 251 | { |
597a5f55 | 252 | int i, cpu, me; |
d9e368d6 | 253 | struct kvm_vcpu *vcpu; |
7053df4e | 254 | bool called; |
6ef7a1bc | 255 | |
3cba4130 | 256 | me = get_cpu(); |
7053df4e | 257 | |
988a2cae | 258 | kvm_for_each_vcpu(i, vcpu, kvm) { |
a812297c | 259 | if (vcpu_bitmap && !test_bit(i, vcpu_bitmap)) |
7053df4e VK |
260 | continue; |
261 | ||
3cba4130 | 262 | kvm_make_request(req, vcpu); |
d9e368d6 | 263 | cpu = vcpu->cpu; |
6b7e2d09 | 264 | |
178f02ff RK |
265 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) |
266 | continue; | |
6c6e8360 | 267 | |
7053df4e | 268 | if (tmp != NULL && cpu != -1 && cpu != me && |
7a97cec2 | 269 | kvm_request_needs_ipi(vcpu, req)) |
7053df4e | 270 | __cpumask_set_cpu(cpu, tmp); |
49846896 | 271 | } |
7053df4e VK |
272 | |
273 | called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT)); | |
3cba4130 | 274 | put_cpu(); |
7053df4e VK |
275 | |
276 | return called; | |
277 | } | |
278 | ||
279 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) | |
280 | { | |
281 | cpumask_var_t cpus; | |
282 | bool called; | |
7053df4e VK |
283 | |
284 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); | |
285 | ||
a812297c | 286 | called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus); |
7053df4e | 287 | |
6ef7a1bc | 288 | free_cpumask_var(cpus); |
49846896 | 289 | return called; |
d9e368d6 AK |
290 | } |
291 | ||
a6d51016 | 292 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 293 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 294 | { |
4ae3cb3a LT |
295 | /* |
296 | * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in | |
297 | * kvm_make_all_cpus_request. | |
298 | */ | |
299 | long dirty_count = smp_load_acquire(&kvm->tlbs_dirty); | |
300 | ||
301 | /* | |
302 | * We want to publish modifications to the page tables before reading | |
303 | * mode. Pairs with a memory barrier in arch-specific code. | |
304 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
305 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
306 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
307 | * | |
308 | * There is already an smp_mb__after_atomic() before | |
309 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
310 | * barrier here. | |
311 | */ | |
b08660e5 TL |
312 | if (!kvm_arch_flush_remote_tlb(kvm) |
313 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
49846896 | 314 | ++kvm->stat.remote_tlb_flush; |
a086f6a1 | 315 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 316 | } |
2ba9f0d8 | 317 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 318 | #endif |
2e53d63a | 319 | |
49846896 RR |
320 | void kvm_reload_remote_mmus(struct kvm *kvm) |
321 | { | |
445b8236 | 322 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 323 | } |
2e53d63a | 324 | |
fb3f0f51 RR |
325 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
326 | { | |
327 | struct page *page; | |
328 | int r; | |
329 | ||
330 | mutex_init(&vcpu->mutex); | |
331 | vcpu->cpu = -1; | |
fb3f0f51 RR |
332 | vcpu->kvm = kvm; |
333 | vcpu->vcpu_id = id; | |
34bb10b7 | 334 | vcpu->pid = NULL; |
8577370f | 335 | init_swait_queue_head(&vcpu->wq); |
af585b92 | 336 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 337 | |
bf9f6ac8 FW |
338 | vcpu->pre_pcpu = -1; |
339 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
340 | ||
fb3f0f51 RR |
341 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
342 | if (!page) { | |
343 | r = -ENOMEM; | |
344 | goto fail; | |
345 | } | |
346 | vcpu->run = page_address(page); | |
347 | ||
4c088493 R |
348 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
349 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 350 | vcpu->preempted = false; |
d73eb57b | 351 | vcpu->ready = false; |
4c088493 | 352 | |
e9b11c17 | 353 | r = kvm_arch_vcpu_init(vcpu); |
fb3f0f51 | 354 | if (r < 0) |
e9b11c17 | 355 | goto fail_free_run; |
fb3f0f51 RR |
356 | return 0; |
357 | ||
fb3f0f51 RR |
358 | fail_free_run: |
359 | free_page((unsigned long)vcpu->run); | |
360 | fail: | |
76fafa5e | 361 | return r; |
fb3f0f51 RR |
362 | } |
363 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); | |
364 | ||
365 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) | |
366 | { | |
0e4524a5 CB |
367 | /* |
368 | * no need for rcu_read_lock as VCPU_RUN is the only place that | |
369 | * will change the vcpu->pid pointer and on uninit all file | |
370 | * descriptors are already gone. | |
371 | */ | |
372 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
e9b11c17 | 373 | kvm_arch_vcpu_uninit(vcpu); |
fb3f0f51 RR |
374 | free_page((unsigned long)vcpu->run); |
375 | } | |
376 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); | |
377 | ||
e930bffe AA |
378 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
379 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
380 | { | |
381 | return container_of(mn, struct kvm, mmu_notifier); | |
382 | } | |
383 | ||
3da0dd43 IE |
384 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
385 | struct mm_struct *mm, | |
386 | unsigned long address, | |
387 | pte_t pte) | |
388 | { | |
389 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 390 | int idx; |
3da0dd43 | 391 | |
bc6678a3 | 392 | idx = srcu_read_lock(&kvm->srcu); |
3da0dd43 IE |
393 | spin_lock(&kvm->mmu_lock); |
394 | kvm->mmu_notifier_seq++; | |
0cf853c5 LT |
395 | |
396 | if (kvm_set_spte_hva(kvm, address, pte)) | |
397 | kvm_flush_remote_tlbs(kvm); | |
398 | ||
3da0dd43 | 399 | spin_unlock(&kvm->mmu_lock); |
bc6678a3 | 400 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
401 | } |
402 | ||
93065ac7 | 403 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
5d6527a7 | 404 | const struct mmu_notifier_range *range) |
e930bffe AA |
405 | { |
406 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 407 | int need_tlb_flush = 0, idx; |
93065ac7 | 408 | int ret; |
e930bffe | 409 | |
bc6678a3 | 410 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe AA |
411 | spin_lock(&kvm->mmu_lock); |
412 | /* | |
413 | * The count increase must become visible at unlock time as no | |
414 | * spte can be established without taking the mmu_lock and | |
415 | * count is also read inside the mmu_lock critical section. | |
416 | */ | |
417 | kvm->mmu_notifier_count++; | |
5d6527a7 | 418 | need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end); |
a4ee1ca4 | 419 | need_tlb_flush |= kvm->tlbs_dirty; |
e930bffe AA |
420 | /* we've to flush the tlb before the pages can be freed */ |
421 | if (need_tlb_flush) | |
422 | kvm_flush_remote_tlbs(kvm); | |
565f3be2 TY |
423 | |
424 | spin_unlock(&kvm->mmu_lock); | |
b1394e74 | 425 | |
5d6527a7 | 426 | ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start, |
dfcd6660 JG |
427 | range->end, |
428 | mmu_notifier_range_blockable(range)); | |
b1394e74 | 429 | |
565f3be2 | 430 | srcu_read_unlock(&kvm->srcu, idx); |
93065ac7 MH |
431 | |
432 | return ret; | |
e930bffe AA |
433 | } |
434 | ||
435 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
5d6527a7 | 436 | const struct mmu_notifier_range *range) |
e930bffe AA |
437 | { |
438 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
439 | ||
440 | spin_lock(&kvm->mmu_lock); | |
441 | /* | |
442 | * This sequence increase will notify the kvm page fault that | |
443 | * the page that is going to be mapped in the spte could have | |
444 | * been freed. | |
445 | */ | |
446 | kvm->mmu_notifier_seq++; | |
a355aa54 | 447 | smp_wmb(); |
e930bffe AA |
448 | /* |
449 | * The above sequence increase must be visible before the | |
a355aa54 PM |
450 | * below count decrease, which is ensured by the smp_wmb above |
451 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
452 | */ |
453 | kvm->mmu_notifier_count--; | |
454 | spin_unlock(&kvm->mmu_lock); | |
455 | ||
456 | BUG_ON(kvm->mmu_notifier_count < 0); | |
457 | } | |
458 | ||
459 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
460 | struct mm_struct *mm, | |
57128468 ALC |
461 | unsigned long start, |
462 | unsigned long end) | |
e930bffe AA |
463 | { |
464 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 465 | int young, idx; |
e930bffe | 466 | |
bc6678a3 | 467 | idx = srcu_read_lock(&kvm->srcu); |
e930bffe | 468 | spin_lock(&kvm->mmu_lock); |
e930bffe | 469 | |
57128468 | 470 | young = kvm_age_hva(kvm, start, end); |
e930bffe AA |
471 | if (young) |
472 | kvm_flush_remote_tlbs(kvm); | |
473 | ||
565f3be2 TY |
474 | spin_unlock(&kvm->mmu_lock); |
475 | srcu_read_unlock(&kvm->srcu, idx); | |
476 | ||
e930bffe AA |
477 | return young; |
478 | } | |
479 | ||
1d7715c6 VD |
480 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
481 | struct mm_struct *mm, | |
482 | unsigned long start, | |
483 | unsigned long end) | |
484 | { | |
485 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
486 | int young, idx; | |
487 | ||
488 | idx = srcu_read_lock(&kvm->srcu); | |
489 | spin_lock(&kvm->mmu_lock); | |
490 | /* | |
491 | * Even though we do not flush TLB, this will still adversely | |
492 | * affect performance on pre-Haswell Intel EPT, where there is | |
493 | * no EPT Access Bit to clear so that we have to tear down EPT | |
494 | * tables instead. If we find this unacceptable, we can always | |
495 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
496 | * do anything on clear_young. | |
497 | * | |
498 | * Also note that currently we never issue secondary TLB flushes | |
499 | * from clear_young, leaving this job up to the regular system | |
500 | * cadence. If we find this inaccurate, we might come up with a | |
501 | * more sophisticated heuristic later. | |
502 | */ | |
503 | young = kvm_age_hva(kvm, start, end); | |
504 | spin_unlock(&kvm->mmu_lock); | |
505 | srcu_read_unlock(&kvm->srcu, idx); | |
506 | ||
507 | return young; | |
508 | } | |
509 | ||
8ee53820 AA |
510 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
511 | struct mm_struct *mm, | |
512 | unsigned long address) | |
513 | { | |
514 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
515 | int young, idx; | |
516 | ||
517 | idx = srcu_read_lock(&kvm->srcu); | |
518 | spin_lock(&kvm->mmu_lock); | |
519 | young = kvm_test_age_hva(kvm, address); | |
520 | spin_unlock(&kvm->mmu_lock); | |
521 | srcu_read_unlock(&kvm->srcu, idx); | |
522 | ||
523 | return young; | |
524 | } | |
525 | ||
85db06e5 MT |
526 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
527 | struct mm_struct *mm) | |
528 | { | |
529 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
530 | int idx; |
531 | ||
532 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 533 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 534 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
535 | } |
536 | ||
e930bffe | 537 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e930bffe AA |
538 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
539 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
540 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 541 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 542 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 543 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 544 | .release = kvm_mmu_notifier_release, |
e930bffe | 545 | }; |
4c07b0a4 AK |
546 | |
547 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
548 | { | |
549 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
550 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
551 | } | |
552 | ||
553 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
554 | ||
555 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
556 | { | |
557 | return 0; | |
558 | } | |
559 | ||
e930bffe AA |
560 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
561 | ||
a47d2b07 | 562 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
563 | { |
564 | int i; | |
a47d2b07 | 565 | struct kvm_memslots *slots; |
bf3e05bc | 566 | |
b12ce36a | 567 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
a47d2b07 PB |
568 | if (!slots) |
569 | return NULL; | |
570 | ||
bf3e05bc | 571 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
f85e2cb5 | 572 | slots->id_to_index[i] = slots->memslots[i].id = i; |
a47d2b07 PB |
573 | |
574 | return slots; | |
575 | } | |
576 | ||
577 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
578 | { | |
579 | if (!memslot->dirty_bitmap) | |
580 | return; | |
581 | ||
582 | kvfree(memslot->dirty_bitmap); | |
583 | memslot->dirty_bitmap = NULL; | |
584 | } | |
585 | ||
586 | /* | |
587 | * Free any memory in @free but not in @dont. | |
588 | */ | |
589 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
590 | struct kvm_memory_slot *dont) | |
591 | { | |
592 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) | |
593 | kvm_destroy_dirty_bitmap(free); | |
594 | ||
595 | kvm_arch_free_memslot(kvm, free, dont); | |
596 | ||
597 | free->npages = 0; | |
598 | } | |
599 | ||
600 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
601 | { | |
602 | struct kvm_memory_slot *memslot; | |
603 | ||
604 | if (!slots) | |
605 | return; | |
606 | ||
607 | kvm_for_each_memslot(memslot, slots) | |
608 | kvm_free_memslot(kvm, memslot, NULL); | |
609 | ||
610 | kvfree(slots); | |
bf3e05bc XG |
611 | } |
612 | ||
536a6f88 JF |
613 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
614 | { | |
615 | int i; | |
616 | ||
617 | if (!kvm->debugfs_dentry) | |
618 | return; | |
619 | ||
620 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
621 | ||
9d5a1dce LC |
622 | if (kvm->debugfs_stat_data) { |
623 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
624 | kfree(kvm->debugfs_stat_data[i]); | |
625 | kfree(kvm->debugfs_stat_data); | |
626 | } | |
536a6f88 JF |
627 | } |
628 | ||
629 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
630 | { | |
631 | char dir_name[ITOA_MAX_LEN * 2]; | |
632 | struct kvm_stat_data *stat_data; | |
633 | struct kvm_stats_debugfs_item *p; | |
634 | ||
635 | if (!debugfs_initialized()) | |
636 | return 0; | |
637 | ||
638 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
929f45e3 | 639 | kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir); |
536a6f88 JF |
640 | |
641 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, | |
642 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 643 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
644 | if (!kvm->debugfs_stat_data) |
645 | return -ENOMEM; | |
646 | ||
647 | for (p = debugfs_entries; p->name; p++) { | |
b12ce36a | 648 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
649 | if (!stat_data) |
650 | return -ENOMEM; | |
651 | ||
652 | stat_data->kvm = kvm; | |
653 | stat_data->offset = p->offset; | |
833b45de | 654 | stat_data->mode = p->mode ? p->mode : 0644; |
536a6f88 | 655 | kvm->debugfs_stat_data[p - debugfs_entries] = stat_data; |
833b45de | 656 | debugfs_create_file(p->name, stat_data->mode, kvm->debugfs_dentry, |
929f45e3 | 657 | stat_data, stat_fops_per_vm[p->kind]); |
536a6f88 JF |
658 | } |
659 | return 0; | |
660 | } | |
661 | ||
1aa9b957 JS |
662 | /* |
663 | * Called after the VM is otherwise initialized, but just before adding it to | |
664 | * the vm_list. | |
665 | */ | |
666 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
667 | { | |
668 | return 0; | |
669 | } | |
670 | ||
671 | /* | |
672 | * Called just after removing the VM from the vm_list, but before doing any | |
673 | * other destruction. | |
674 | */ | |
675 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
676 | { | |
677 | } | |
678 | ||
e08b9637 | 679 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 680 | { |
d89f5eff | 681 | struct kvm *kvm = kvm_arch_alloc_vm(); |
9121923c JM |
682 | int r = -ENOMEM; |
683 | int i; | |
6aa8b732 | 684 | |
d89f5eff JK |
685 | if (!kvm) |
686 | return ERR_PTR(-ENOMEM); | |
687 | ||
e9ad4ec8 | 688 | spin_lock_init(&kvm->mmu_lock); |
f1f10076 | 689 | mmgrab(current->mm); |
e9ad4ec8 PB |
690 | kvm->mm = current->mm; |
691 | kvm_eventfd_init(kvm); | |
692 | mutex_init(&kvm->lock); | |
693 | mutex_init(&kvm->irq_lock); | |
694 | mutex_init(&kvm->slots_lock); | |
e9ad4ec8 PB |
695 | INIT_LIST_HEAD(&kvm->devices); |
696 | ||
1e702d9a AW |
697 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
698 | ||
8a44119a PB |
699 | if (init_srcu_struct(&kvm->srcu)) |
700 | goto out_err_no_srcu; | |
701 | if (init_srcu_struct(&kvm->irq_srcu)) | |
702 | goto out_err_no_irq_srcu; | |
703 | ||
e2d3fcaf | 704 | refcount_set(&kvm->users_count, 1); |
f481b069 | 705 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4bd518f1 | 706 | struct kvm_memslots *slots = kvm_alloc_memslots(); |
9121923c | 707 | |
4bd518f1 | 708 | if (!slots) |
a97b0e77 | 709 | goto out_err_no_arch_destroy_vm; |
0e32958e | 710 | /* Generations must be different for each address space. */ |
164bf7e5 | 711 | slots->generation = i; |
4bd518f1 | 712 | rcu_assign_pointer(kvm->memslots[i], slots); |
f481b069 | 713 | } |
00f034a1 | 714 | |
e93f8a0f | 715 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 716 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 717 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 718 | if (!kvm->buses[i]) |
a97b0e77 | 719 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 720 | } |
e930bffe | 721 | |
e08b9637 | 722 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 723 | if (r) |
a97b0e77 | 724 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
725 | |
726 | r = hardware_enable_all(); | |
727 | if (r) | |
719d93cd | 728 | goto out_err_no_disable; |
10474ae8 | 729 | |
c77dcacb | 730 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 731 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 732 | #endif |
6aa8b732 | 733 | |
74b5c5bf | 734 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
735 | if (r) |
736 | goto out_err_no_mmu_notifier; | |
737 | ||
738 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
739 | if (r) |
740 | goto out_err; | |
741 | ||
0d9ce162 | 742 | mutex_lock(&kvm_lock); |
5e58cfe4 | 743 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 744 | mutex_unlock(&kvm_lock); |
d89f5eff | 745 | |
2ecd9d29 PZ |
746 | preempt_notifier_inc(); |
747 | ||
f17abe9a | 748 | return kvm; |
10474ae8 AG |
749 | |
750 | out_err: | |
1aa9b957 JS |
751 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
752 | if (kvm->mmu_notifier.ops) | |
753 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
754 | #endif | |
755 | out_err_no_mmu_notifier: | |
10474ae8 | 756 | hardware_disable_all(); |
719d93cd | 757 | out_err_no_disable: |
a97b0e77 | 758 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 759 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 760 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 761 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 762 | kfree(kvm_get_bus(kvm, i)); |
f481b069 | 763 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 764 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
8a44119a PB |
765 | cleanup_srcu_struct(&kvm->irq_srcu); |
766 | out_err_no_irq_srcu: | |
767 | cleanup_srcu_struct(&kvm->srcu); | |
768 | out_err_no_srcu: | |
d89f5eff | 769 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 770 | mmdrop(current->mm); |
10474ae8 | 771 | return ERR_PTR(r); |
f17abe9a AK |
772 | } |
773 | ||
07f0a7bd SW |
774 | static void kvm_destroy_devices(struct kvm *kvm) |
775 | { | |
e6e3b5a6 | 776 | struct kvm_device *dev, *tmp; |
07f0a7bd | 777 | |
a28ebea2 CD |
778 | /* |
779 | * We do not need to take the kvm->lock here, because nobody else | |
780 | * has a reference to the struct kvm at this point and therefore | |
781 | * cannot access the devices list anyhow. | |
782 | */ | |
e6e3b5a6 GT |
783 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
784 | list_del(&dev->vm_node); | |
07f0a7bd SW |
785 | dev->ops->destroy(dev); |
786 | } | |
787 | } | |
788 | ||
f17abe9a AK |
789 | static void kvm_destroy_vm(struct kvm *kvm) |
790 | { | |
e93f8a0f | 791 | int i; |
6d4e4c4f AK |
792 | struct mm_struct *mm = kvm->mm; |
793 | ||
286de8f6 | 794 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 795 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 796 | kvm_arch_sync_events(kvm); |
0d9ce162 | 797 | mutex_lock(&kvm_lock); |
133de902 | 798 | list_del(&kvm->vm_list); |
0d9ce162 | 799 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
800 | kvm_arch_pre_destroy_vm(kvm); |
801 | ||
399ec807 | 802 | kvm_free_irq_routing(kvm); |
df630b8c | 803 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 804 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 805 | |
4a12f951 CB |
806 | if (bus) |
807 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
808 | kvm->buses[i] = NULL; |
809 | } | |
980da6ce | 810 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
811 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
812 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 813 | #else |
2df72e9b | 814 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 815 | #endif |
d19a9cd2 | 816 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 817 | kvm_destroy_devices(kvm); |
f481b069 | 818 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 819 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
820b3fcd | 820 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
821 | cleanup_srcu_struct(&kvm->srcu); |
822 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 823 | preempt_notifier_dec(); |
10474ae8 | 824 | hardware_disable_all(); |
6d4e4c4f | 825 | mmdrop(mm); |
f17abe9a AK |
826 | } |
827 | ||
d39f13b0 IE |
828 | void kvm_get_kvm(struct kvm *kvm) |
829 | { | |
e3736c3e | 830 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
831 | } |
832 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
833 | ||
834 | void kvm_put_kvm(struct kvm *kvm) | |
835 | { | |
e3736c3e | 836 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
837 | kvm_destroy_vm(kvm); |
838 | } | |
839 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
840 | ||
149487bd SC |
841 | /* |
842 | * Used to put a reference that was taken on behalf of an object associated | |
843 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
844 | * of the new file descriptor fails and the reference cannot be transferred to | |
845 | * its final owner. In such cases, the caller is still actively using @kvm and | |
846 | * will fail miserably if the refcount unexpectedly hits zero. | |
847 | */ | |
848 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
849 | { | |
850 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
851 | } | |
852 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 853 | |
f17abe9a AK |
854 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
855 | { | |
856 | struct kvm *kvm = filp->private_data; | |
857 | ||
721eecbf GH |
858 | kvm_irqfd_release(kvm); |
859 | ||
d39f13b0 | 860 | kvm_put_kvm(kvm); |
6aa8b732 AK |
861 | return 0; |
862 | } | |
863 | ||
515a0127 TY |
864 | /* |
865 | * Allocation size is twice as large as the actual dirty bitmap size. | |
93474b25 | 866 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 867 | */ |
a36a57b1 TY |
868 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
869 | { | |
515a0127 | 870 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 871 | |
b12ce36a | 872 | memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
873 | if (!memslot->dirty_bitmap) |
874 | return -ENOMEM; | |
875 | ||
a36a57b1 TY |
876 | return 0; |
877 | } | |
878 | ||
bf3e05bc | 879 | /* |
0e60b079 IM |
880 | * Insert memslot and re-sort memslots based on their GFN, |
881 | * so binary search could be used to lookup GFN. | |
882 | * Sorting algorithm takes advantage of having initially | |
883 | * sorted array and known changed memslot position. | |
bf3e05bc | 884 | */ |
5cc15027 | 885 | static void update_memslots(struct kvm_memslots *slots, |
31fc4f95 WY |
886 | struct kvm_memory_slot *new, |
887 | enum kvm_mr_change change) | |
bf3e05bc | 888 | { |
8593176c PB |
889 | int id = new->id; |
890 | int i = slots->id_to_index[id]; | |
063584d4 | 891 | struct kvm_memory_slot *mslots = slots->memslots; |
f85e2cb5 | 892 | |
8593176c | 893 | WARN_ON(mslots[i].id != id); |
31fc4f95 WY |
894 | switch (change) { |
895 | case KVM_MR_CREATE: | |
896 | slots->used_slots++; | |
897 | WARN_ON(mslots[i].npages || !new->npages); | |
898 | break; | |
899 | case KVM_MR_DELETE: | |
900 | slots->used_slots--; | |
901 | WARN_ON(new->npages || !mslots[i].npages); | |
902 | break; | |
903 | default: | |
904 | break; | |
9c1a5d38 | 905 | } |
0e60b079 | 906 | |
7f379cff | 907 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
0e60b079 IM |
908 | new->base_gfn <= mslots[i + 1].base_gfn) { |
909 | if (!mslots[i + 1].npages) | |
910 | break; | |
7f379cff IM |
911 | mslots[i] = mslots[i + 1]; |
912 | slots->id_to_index[mslots[i].id] = i; | |
913 | i++; | |
914 | } | |
efbeec70 PB |
915 | |
916 | /* | |
917 | * The ">=" is needed when creating a slot with base_gfn == 0, | |
918 | * so that it moves before all those with base_gfn == npages == 0. | |
919 | * | |
920 | * On the other hand, if new->npages is zero, the above loop has | |
921 | * already left i pointing to the beginning of the empty part of | |
922 | * mslots, and the ">=" would move the hole backwards in this | |
923 | * case---which is wrong. So skip the loop when deleting a slot. | |
924 | */ | |
925 | if (new->npages) { | |
926 | while (i > 0 && | |
927 | new->base_gfn >= mslots[i - 1].base_gfn) { | |
928 | mslots[i] = mslots[i - 1]; | |
929 | slots->id_to_index[mslots[i].id] = i; | |
930 | i--; | |
931 | } | |
dbaff309 PB |
932 | } else |
933 | WARN_ON_ONCE(i != slots->used_slots); | |
f85e2cb5 | 934 | |
8593176c PB |
935 | mslots[i] = *new; |
936 | slots->id_to_index[mslots[i].id] = i; | |
bf3e05bc XG |
937 | } |
938 | ||
09170a49 | 939 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 940 | { |
4d8b81ab XG |
941 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
942 | ||
0f8a4de3 | 943 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
944 | valid_flags |= KVM_MEM_READONLY; |
945 | #endif | |
946 | ||
947 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
948 | return -EINVAL; |
949 | ||
950 | return 0; | |
951 | } | |
952 | ||
7ec4fb44 | 953 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 954 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 955 | { |
f481b069 | 956 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
361209e0 | 957 | u64 gen = old_memslots->generation; |
7ec4fb44 | 958 | |
361209e0 SC |
959 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
960 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 961 | |
f481b069 | 962 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
7ec4fb44 | 963 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 964 | |
ee3d1570 | 965 | /* |
361209e0 SC |
966 | * Increment the new memslot generation a second time, dropping the |
967 | * update in-progress flag and incrementing then generation based on | |
968 | * the number of address spaces. This provides a unique and easily | |
969 | * identifiable generation number while the memslots are in flux. | |
970 | */ | |
971 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
972 | ||
973 | /* | |
4bd518f1 PB |
974 | * Generations must be unique even across address spaces. We do not need |
975 | * a global counter for that, instead the generation space is evenly split | |
976 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
977 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
978 | * use generations 1, 3, 5, ... | |
ee3d1570 | 979 | */ |
164bf7e5 | 980 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 981 | |
15248258 | 982 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 983 | |
15248258 | 984 | slots->generation = gen; |
e59dbe09 TY |
985 | |
986 | return old_memslots; | |
7ec4fb44 GN |
987 | } |
988 | ||
6aa8b732 AK |
989 | /* |
990 | * Allocate some memory and give it an address in the guest physical address | |
991 | * space. | |
992 | * | |
993 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 994 | * |
02d5d55b | 995 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 996 | */ |
f78e0e2e | 997 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 998 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 999 | { |
8234b22e | 1000 | int r; |
6aa8b732 | 1001 | gfn_t base_gfn; |
28bcb112 | 1002 | unsigned long npages; |
a843fac2 | 1003 | struct kvm_memory_slot *slot; |
6aa8b732 | 1004 | struct kvm_memory_slot old, new; |
b7f69c55 | 1005 | struct kvm_memslots *slots = NULL, *old_memslots; |
f481b069 | 1006 | int as_id, id; |
f64c0398 | 1007 | enum kvm_mr_change change; |
6aa8b732 | 1008 | |
a50d64d6 XG |
1009 | r = check_memory_region_flags(mem); |
1010 | if (r) | |
1011 | goto out; | |
1012 | ||
6aa8b732 | 1013 | r = -EINVAL; |
f481b069 PB |
1014 | as_id = mem->slot >> 16; |
1015 | id = (u16)mem->slot; | |
1016 | ||
6aa8b732 AK |
1017 | /* General sanity checks */ |
1018 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
1019 | goto out; | |
1020 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) | |
1021 | goto out; | |
fa3d315a | 1022 | /* We can read the guest memory with __xxx_user() later on. */ |
f481b069 | 1023 | if ((id < KVM_USER_MEM_SLOTS) && |
fa3d315a | 1024 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
96d4f267 | 1025 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
9e3bb6b6 | 1026 | mem->memory_size))) |
78749809 | 1027 | goto out; |
f481b069 | 1028 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
6aa8b732 AK |
1029 | goto out; |
1030 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) | |
1031 | goto out; | |
1032 | ||
f481b069 | 1033 | slot = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
6aa8b732 AK |
1034 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
1035 | npages = mem->memory_size >> PAGE_SHIFT; | |
1036 | ||
660c22c4 TY |
1037 | if (npages > KVM_MEM_MAX_NR_PAGES) |
1038 | goto out; | |
1039 | ||
a843fac2 | 1040 | new = old = *slot; |
6aa8b732 | 1041 | |
f481b069 | 1042 | new.id = id; |
6aa8b732 AK |
1043 | new.base_gfn = base_gfn; |
1044 | new.npages = npages; | |
1045 | new.flags = mem->flags; | |
1046 | ||
f64c0398 TY |
1047 | if (npages) { |
1048 | if (!old.npages) | |
1049 | change = KVM_MR_CREATE; | |
1050 | else { /* Modify an existing slot. */ | |
1051 | if ((mem->userspace_addr != old.userspace_addr) || | |
75d61fbc TY |
1052 | (npages != old.npages) || |
1053 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) | |
f64c0398 TY |
1054 | goto out; |
1055 | ||
1056 | if (base_gfn != old.base_gfn) | |
1057 | change = KVM_MR_MOVE; | |
1058 | else if (new.flags != old.flags) | |
1059 | change = KVM_MR_FLAGS_ONLY; | |
1060 | else { /* Nothing to change. */ | |
1061 | r = 0; | |
1062 | goto out; | |
1063 | } | |
1064 | } | |
09170a49 PB |
1065 | } else { |
1066 | if (!old.npages) | |
1067 | goto out; | |
1068 | ||
f64c0398 | 1069 | change = KVM_MR_DELETE; |
09170a49 PB |
1070 | new.base_gfn = 0; |
1071 | new.flags = 0; | |
1072 | } | |
6aa8b732 | 1073 | |
f64c0398 | 1074 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee TY |
1075 | /* Check for overlaps */ |
1076 | r = -EEXIST; | |
f481b069 | 1077 | kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) { |
b28676bb | 1078 | if (slot->id == id) |
0a706bee TY |
1079 | continue; |
1080 | if (!((base_gfn + npages <= slot->base_gfn) || | |
1081 | (base_gfn >= slot->base_gfn + slot->npages))) | |
1082 | goto out; | |
1083 | } | |
6aa8b732 | 1084 | } |
6aa8b732 | 1085 | |
6aa8b732 AK |
1086 | /* Free page dirty bitmap if unneeded */ |
1087 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
8b6d44c7 | 1088 | new.dirty_bitmap = NULL; |
6aa8b732 AK |
1089 | |
1090 | r = -ENOMEM; | |
f64c0398 | 1091 | if (change == KVM_MR_CREATE) { |
189a2f7b | 1092 | new.userspace_addr = mem->userspace_addr; |
d89cc617 | 1093 | |
5587027c | 1094 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
db3fe4eb | 1095 | goto out_free; |
6aa8b732 | 1096 | } |
ec04b260 | 1097 | |
6aa8b732 AK |
1098 | /* Allocate page dirty bitmap if needed */ |
1099 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { | |
a36a57b1 | 1100 | if (kvm_create_dirty_bitmap(&new) < 0) |
f78e0e2e | 1101 | goto out_free; |
6aa8b732 AK |
1102 | } |
1103 | ||
b12ce36a | 1104 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
f2a81036 PB |
1105 | if (!slots) |
1106 | goto out_free; | |
f481b069 | 1107 | memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots)); |
f2a81036 | 1108 | |
f64c0398 | 1109 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
f481b069 | 1110 | slot = id_to_memslot(slots, id); |
28a37544 XG |
1111 | slot->flags |= KVM_MEMSLOT_INVALID; |
1112 | ||
f481b069 | 1113 | old_memslots = install_new_memslots(kvm, as_id, slots); |
bc6678a3 | 1114 | |
12d6e753 MT |
1115 | /* From this point no new shadow pages pointing to a deleted, |
1116 | * or moved, memslot will be created. | |
bc6678a3 MT |
1117 | * |
1118 | * validation of sp->gfn happens in: | |
b7d409de XL |
1119 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) |
1120 | * - kvm_is_visible_gfn (mmu_check_roots) | |
bc6678a3 | 1121 | */ |
2df72e9b | 1122 | kvm_arch_flush_shadow_memslot(kvm, slot); |
f2a81036 PB |
1123 | |
1124 | /* | |
1125 | * We can re-use the old_memslots from above, the only difference | |
1126 | * from the currently installed memslots is the invalid flag. This | |
1127 | * will get overwritten by update_memslots anyway. | |
1128 | */ | |
b7f69c55 | 1129 | slots = old_memslots; |
bc6678a3 | 1130 | } |
34d4cb8f | 1131 | |
7b6195a9 | 1132 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
f7784b8e | 1133 | if (r) |
b7f69c55 | 1134 | goto out_slots; |
f7784b8e | 1135 | |
a47d2b07 | 1136 | /* actual memory is freed via old in kvm_free_memslot below */ |
f64c0398 | 1137 | if (change == KVM_MR_DELETE) { |
bc6678a3 | 1138 | new.dirty_bitmap = NULL; |
db3fe4eb | 1139 | memset(&new.arch, 0, sizeof(new.arch)); |
bc6678a3 MT |
1140 | } |
1141 | ||
31fc4f95 | 1142 | update_memslots(slots, &new, change); |
f481b069 | 1143 | old_memslots = install_new_memslots(kvm, as_id, slots); |
3ad82a7e | 1144 | |
f36f3f28 | 1145 | kvm_arch_commit_memory_region(kvm, mem, &old, &new, change); |
82ce2c96 | 1146 | |
a47d2b07 | 1147 | kvm_free_memslot(kvm, &old, &new); |
74496134 | 1148 | kvfree(old_memslots); |
6aa8b732 AK |
1149 | return 0; |
1150 | ||
e40f193f | 1151 | out_slots: |
74496134 | 1152 | kvfree(slots); |
f78e0e2e | 1153 | out_free: |
a47d2b07 | 1154 | kvm_free_memslot(kvm, &new, &old); |
6aa8b732 AK |
1155 | out: |
1156 | return r; | |
210c7c4d | 1157 | } |
f78e0e2e SY |
1158 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1159 | ||
1160 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1161 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1162 | { |
1163 | int r; | |
1164 | ||
79fac95e | 1165 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1166 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1167 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1168 | return r; |
1169 | } | |
210c7c4d IE |
1170 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1171 | ||
7940876e SH |
1172 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1173 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1174 | { |
f481b069 | 1175 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1176 | return -EINVAL; |
09170a49 | 1177 | |
47ae31e2 | 1178 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1179 | } |
1180 | ||
5bb064dc ZX |
1181 | int kvm_get_dirty_log(struct kvm *kvm, |
1182 | struct kvm_dirty_log *log, int *is_dirty) | |
6aa8b732 | 1183 | { |
9f6b8029 | 1184 | struct kvm_memslots *slots; |
6aa8b732 | 1185 | struct kvm_memory_slot *memslot; |
843574a3 | 1186 | int i, as_id, id; |
87bf6e7d | 1187 | unsigned long n; |
6aa8b732 AK |
1188 | unsigned long any = 0; |
1189 | ||
f481b069 PB |
1190 | as_id = log->slot >> 16; |
1191 | id = (u16)log->slot; | |
1192 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 1193 | return -EINVAL; |
6aa8b732 | 1194 | |
f481b069 PB |
1195 | slots = __kvm_memslots(kvm, as_id); |
1196 | memslot = id_to_memslot(slots, id); | |
6aa8b732 | 1197 | if (!memslot->dirty_bitmap) |
843574a3 | 1198 | return -ENOENT; |
6aa8b732 | 1199 | |
87bf6e7d | 1200 | n = kvm_dirty_bitmap_bytes(memslot); |
6aa8b732 | 1201 | |
cd1a4a98 | 1202 | for (i = 0; !any && i < n/sizeof(long); ++i) |
6aa8b732 AK |
1203 | any = memslot->dirty_bitmap[i]; |
1204 | ||
6aa8b732 | 1205 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) |
843574a3 | 1206 | return -EFAULT; |
6aa8b732 | 1207 | |
5bb064dc ZX |
1208 | if (any) |
1209 | *is_dirty = 1; | |
843574a3 | 1210 | return 0; |
6aa8b732 | 1211 | } |
2ba9f0d8 | 1212 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1213 | |
ba0513b5 MS |
1214 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
1215 | /** | |
b8b00220 | 1216 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 1217 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
1218 | * @kvm: pointer to kvm instance |
1219 | * @log: slot id and address to which we copy the log | |
b8b00220 | 1220 | * @flush: true if TLB flush is needed by caller |
ba0513b5 MS |
1221 | * |
1222 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1223 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1224 | * following order: | |
1225 | * | |
1226 | * 1. Take a snapshot of the bit and clear it if needed. | |
1227 | * 2. Write protect the corresponding page. | |
1228 | * 3. Copy the snapshot to the userspace. | |
1229 | * 4. Upon return caller flushes TLB's if needed. | |
1230 | * | |
1231 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1232 | * entry. This is not a problem because the page is reported dirty using | |
1233 | * the snapshot taken before and step 4 ensures that writes done after | |
1234 | * exiting to userspace will be logged for the next call. | |
1235 | * | |
1236 | */ | |
1237 | int kvm_get_dirty_log_protect(struct kvm *kvm, | |
8fe65a82 | 1238 | struct kvm_dirty_log *log, bool *flush) |
ba0513b5 | 1239 | { |
9f6b8029 | 1240 | struct kvm_memslots *slots; |
ba0513b5 | 1241 | struct kvm_memory_slot *memslot; |
58d6db34 | 1242 | int i, as_id, id; |
ba0513b5 MS |
1243 | unsigned long n; |
1244 | unsigned long *dirty_bitmap; | |
1245 | unsigned long *dirty_bitmap_buffer; | |
1246 | ||
f481b069 PB |
1247 | as_id = log->slot >> 16; |
1248 | id = (u16)log->slot; | |
1249 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 1250 | return -EINVAL; |
ba0513b5 | 1251 | |
f481b069 PB |
1252 | slots = __kvm_memslots(kvm, as_id); |
1253 | memslot = id_to_memslot(slots, id); | |
ba0513b5 MS |
1254 | |
1255 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 1256 | if (!dirty_bitmap) |
58d6db34 | 1257 | return -ENOENT; |
ba0513b5 MS |
1258 | |
1259 | n = kvm_dirty_bitmap_bytes(memslot); | |
2a31b9db PB |
1260 | *flush = false; |
1261 | if (kvm->manual_dirty_log_protect) { | |
1262 | /* | |
1263 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
1264 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
1265 | * is some code duplication between this function and | |
1266 | * kvm_get_dirty_log, but hopefully all architecture | |
1267 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
1268 | * can be eliminated. | |
1269 | */ | |
1270 | dirty_bitmap_buffer = dirty_bitmap; | |
1271 | } else { | |
1272 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
1273 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 1274 | |
2a31b9db PB |
1275 | spin_lock(&kvm->mmu_lock); |
1276 | for (i = 0; i < n / sizeof(long); i++) { | |
1277 | unsigned long mask; | |
1278 | gfn_t offset; | |
ba0513b5 | 1279 | |
2a31b9db PB |
1280 | if (!dirty_bitmap[i]) |
1281 | continue; | |
1282 | ||
1283 | *flush = true; | |
1284 | mask = xchg(&dirty_bitmap[i], 0); | |
1285 | dirty_bitmap_buffer[i] = mask; | |
1286 | ||
a67794ca LT |
1287 | offset = i * BITS_PER_LONG; |
1288 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1289 | offset, mask); | |
2a31b9db PB |
1290 | } |
1291 | spin_unlock(&kvm->mmu_lock); | |
1292 | } | |
1293 | ||
1294 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) | |
1295 | return -EFAULT; | |
1296 | return 0; | |
1297 | } | |
1298 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); | |
1299 | ||
1300 | /** | |
1301 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
1302 | * and reenable dirty page tracking for the corresponding pages. | |
1303 | * @kvm: pointer to kvm instance | |
1304 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
b8b00220 | 1305 | * @flush: true if TLB flush is needed by caller |
2a31b9db PB |
1306 | */ |
1307 | int kvm_clear_dirty_log_protect(struct kvm *kvm, | |
1308 | struct kvm_clear_dirty_log *log, bool *flush) | |
1309 | { | |
1310 | struct kvm_memslots *slots; | |
1311 | struct kvm_memory_slot *memslot; | |
98938aa8 | 1312 | int as_id, id; |
2a31b9db | 1313 | gfn_t offset; |
98938aa8 | 1314 | unsigned long i, n; |
2a31b9db PB |
1315 | unsigned long *dirty_bitmap; |
1316 | unsigned long *dirty_bitmap_buffer; | |
1317 | ||
1318 | as_id = log->slot >> 16; | |
1319 | id = (u16)log->slot; | |
1320 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
1321 | return -EINVAL; | |
1322 | ||
76d58e0f | 1323 | if (log->first_page & 63) |
2a31b9db PB |
1324 | return -EINVAL; |
1325 | ||
1326 | slots = __kvm_memslots(kvm, as_id); | |
1327 | memslot = id_to_memslot(slots, id); | |
1328 | ||
1329 | dirty_bitmap = memslot->dirty_bitmap; | |
1330 | if (!dirty_bitmap) | |
1331 | return -ENOENT; | |
1332 | ||
4ddc9204 | 1333 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
1334 | |
1335 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
1336 | log->num_pages > memslot->npages - log->first_page || |
1337 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
1338 | return -EINVAL; | |
98938aa8 | 1339 | |
8fe65a82 | 1340 | *flush = false; |
2a31b9db PB |
1341 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
1342 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
1343 | return -EFAULT; | |
ba0513b5 | 1344 | |
2a31b9db | 1345 | spin_lock(&kvm->mmu_lock); |
53eac7a8 PX |
1346 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
1347 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
1348 | i++, offset += BITS_PER_LONG) { |
1349 | unsigned long mask = *dirty_bitmap_buffer++; | |
1350 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
1351 | if (!mask) | |
ba0513b5 MS |
1352 | continue; |
1353 | ||
2a31b9db | 1354 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 1355 | |
2a31b9db PB |
1356 | /* |
1357 | * mask contains the bits that really have been cleared. This | |
1358 | * never includes any bits beyond the length of the memslot (if | |
1359 | * the length is not aligned to 64 pages), therefore it is not | |
1360 | * a problem if userspace sets them in log->dirty_bitmap. | |
1361 | */ | |
58d2930f | 1362 | if (mask) { |
2a31b9db | 1363 | *flush = true; |
58d2930f TY |
1364 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
1365 | offset, mask); | |
1366 | } | |
ba0513b5 | 1367 | } |
ba0513b5 | 1368 | spin_unlock(&kvm->mmu_lock); |
2a31b9db | 1369 | |
58d6db34 | 1370 | return 0; |
ba0513b5 | 1371 | } |
2a31b9db | 1372 | EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect); |
ba0513b5 MS |
1373 | #endif |
1374 | ||
db3fe4eb TY |
1375 | bool kvm_largepages_enabled(void) |
1376 | { | |
1377 | return largepages_enabled; | |
1378 | } | |
1379 | ||
54dee993 MT |
1380 | void kvm_disable_largepages(void) |
1381 | { | |
1382 | largepages_enabled = false; | |
1383 | } | |
1384 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); | |
1385 | ||
49c7754c GN |
1386 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1387 | { | |
1388 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1389 | } | |
a1f4d395 | 1390 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1391 | |
8e73485c PB |
1392 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
1393 | { | |
1394 | return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn); | |
1395 | } | |
1396 | ||
33e94154 | 1397 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 1398 | { |
bf3e05bc | 1399 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1400 | |
bbacc0c1 | 1401 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
bf3e05bc | 1402 | memslot->flags & KVM_MEMSLOT_INVALID) |
33e94154 | 1403 | return false; |
e0d62c7f | 1404 | |
33e94154 | 1405 | return true; |
e0d62c7f IE |
1406 | } |
1407 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1408 | ||
8f0b1ab6 JR |
1409 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
1410 | { | |
1411 | struct vm_area_struct *vma; | |
1412 | unsigned long addr, size; | |
1413 | ||
1414 | size = PAGE_SIZE; | |
1415 | ||
1416 | addr = gfn_to_hva(kvm, gfn); | |
1417 | if (kvm_is_error_hva(addr)) | |
1418 | return PAGE_SIZE; | |
1419 | ||
1420 | down_read(¤t->mm->mmap_sem); | |
1421 | vma = find_vma(current->mm, addr); | |
1422 | if (!vma) | |
1423 | goto out; | |
1424 | ||
1425 | size = vma_kernel_pagesize(vma); | |
1426 | ||
1427 | out: | |
1428 | up_read(¤t->mm->mmap_sem); | |
1429 | ||
1430 | return size; | |
1431 | } | |
1432 | ||
4d8b81ab XG |
1433 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1434 | { | |
1435 | return slot->flags & KVM_MEM_READONLY; | |
1436 | } | |
1437 | ||
4d8b81ab XG |
1438 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1439 | gfn_t *nr_pages, bool write) | |
539cb660 | 1440 | { |
bc6678a3 | 1441 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1442 | return KVM_HVA_ERR_BAD; |
48987781 | 1443 | |
4d8b81ab XG |
1444 | if (memslot_is_readonly(slot) && write) |
1445 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1446 | |
1447 | if (nr_pages) | |
1448 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1449 | ||
4d8b81ab | 1450 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1451 | } |
48987781 | 1452 | |
4d8b81ab XG |
1453 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1454 | gfn_t *nr_pages) | |
1455 | { | |
1456 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1457 | } |
48987781 | 1458 | |
4d8b81ab | 1459 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1460 | gfn_t gfn) |
4d8b81ab XG |
1461 | { |
1462 | return gfn_to_hva_many(slot, gfn, NULL); | |
1463 | } | |
1464 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1465 | ||
48987781 XG |
1466 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1467 | { | |
49c7754c | 1468 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1469 | } |
0d150298 | 1470 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1471 | |
8e73485c PB |
1472 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
1473 | { | |
1474 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
1475 | } | |
1476 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
1477 | ||
86ab8cff | 1478 | /* |
970c0d4b WY |
1479 | * Return the hva of a @gfn and the R/W attribute if possible. |
1480 | * | |
1481 | * @slot: the kvm_memory_slot which contains @gfn | |
1482 | * @gfn: the gfn to be translated | |
1483 | * @writable: used to return the read/write attribute of the @slot if the hva | |
1484 | * is valid and @writable is not NULL | |
86ab8cff | 1485 | */ |
64d83126 CD |
1486 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
1487 | gfn_t gfn, bool *writable) | |
86ab8cff | 1488 | { |
a2ac07fe GN |
1489 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1490 | ||
1491 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1492 | *writable = !memslot_is_readonly(slot); |
1493 | ||
a2ac07fe | 1494 | return hva; |
86ab8cff XG |
1495 | } |
1496 | ||
64d83126 CD |
1497 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
1498 | { | |
1499 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1500 | ||
1501 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1502 | } | |
1503 | ||
8e73485c PB |
1504 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
1505 | { | |
1506 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1507 | ||
1508 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1509 | } | |
1510 | ||
fafc3dba HY |
1511 | static inline int check_user_page_hwpoison(unsigned long addr) |
1512 | { | |
0d731759 | 1513 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 1514 | |
0d731759 | 1515 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
1516 | return rc == -EHWPOISON; |
1517 | } | |
1518 | ||
2fc84311 | 1519 | /* |
b9b33da2 PB |
1520 | * The fast path to get the writable pfn which will be stored in @pfn, |
1521 | * true indicates success, otherwise false is returned. It's also the | |
1522 | * only part that runs if we can are in atomic context. | |
2fc84311 | 1523 | */ |
b9b33da2 PB |
1524 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
1525 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 1526 | { |
8d4e1288 | 1527 | struct page *page[1]; |
2fc84311 | 1528 | int npages; |
954bbbc2 | 1529 | |
12ce13fe XG |
1530 | /* |
1531 | * Fast pin a writable pfn only if it is a write fault request | |
1532 | * or the caller allows to map a writable pfn for a read fault | |
1533 | * request. | |
1534 | */ | |
1535 | if (!(write_fault || writable)) | |
1536 | return false; | |
612819c3 | 1537 | |
2fc84311 XG |
1538 | npages = __get_user_pages_fast(addr, 1, 1, page); |
1539 | if (npages == 1) { | |
1540 | *pfn = page_to_pfn(page[0]); | |
612819c3 | 1541 | |
2fc84311 XG |
1542 | if (writable) |
1543 | *writable = true; | |
1544 | return true; | |
1545 | } | |
af585b92 | 1546 | |
2fc84311 XG |
1547 | return false; |
1548 | } | |
612819c3 | 1549 | |
2fc84311 XG |
1550 | /* |
1551 | * The slow path to get the pfn of the specified host virtual address, | |
1552 | * 1 indicates success, -errno is returned if error is detected. | |
1553 | */ | |
1554 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 1555 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 1556 | { |
ce53053c AV |
1557 | unsigned int flags = FOLL_HWPOISON; |
1558 | struct page *page; | |
2fc84311 | 1559 | int npages = 0; |
612819c3 | 1560 | |
2fc84311 XG |
1561 | might_sleep(); |
1562 | ||
1563 | if (writable) | |
1564 | *writable = write_fault; | |
1565 | ||
ce53053c AV |
1566 | if (write_fault) |
1567 | flags |= FOLL_WRITE; | |
1568 | if (async) | |
1569 | flags |= FOLL_NOWAIT; | |
d4944b0e | 1570 | |
ce53053c | 1571 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
1572 | if (npages != 1) |
1573 | return npages; | |
1574 | ||
1575 | /* map read fault as writable if possible */ | |
12ce13fe | 1576 | if (unlikely(!write_fault) && writable) { |
ce53053c | 1577 | struct page *wpage; |
2fc84311 | 1578 | |
ce53053c | 1579 | if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) { |
2fc84311 | 1580 | *writable = true; |
ce53053c AV |
1581 | put_page(page); |
1582 | page = wpage; | |
612819c3 | 1583 | } |
887c08ac | 1584 | } |
ce53053c | 1585 | *pfn = page_to_pfn(page); |
2fc84311 XG |
1586 | return npages; |
1587 | } | |
539cb660 | 1588 | |
4d8b81ab XG |
1589 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1590 | { | |
1591 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1592 | return false; | |
2e2e3738 | 1593 | |
4d8b81ab XG |
1594 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1595 | return false; | |
887c08ac | 1596 | |
4d8b81ab XG |
1597 | return true; |
1598 | } | |
bf998156 | 1599 | |
92176a8e PB |
1600 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1601 | unsigned long addr, bool *async, | |
a340b3e2 KA |
1602 | bool write_fault, bool *writable, |
1603 | kvm_pfn_t *p_pfn) | |
92176a8e | 1604 | { |
add6a0cd PB |
1605 | unsigned long pfn; |
1606 | int r; | |
1607 | ||
1608 | r = follow_pfn(vma, addr, &pfn); | |
1609 | if (r) { | |
1610 | /* | |
1611 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
1612 | * not call the fault handler, so do it here. | |
1613 | */ | |
1614 | bool unlocked = false; | |
1615 | r = fixup_user_fault(current, current->mm, addr, | |
1616 | (write_fault ? FAULT_FLAG_WRITE : 0), | |
1617 | &unlocked); | |
1618 | if (unlocked) | |
1619 | return -EAGAIN; | |
1620 | if (r) | |
1621 | return r; | |
1622 | ||
1623 | r = follow_pfn(vma, addr, &pfn); | |
1624 | if (r) | |
1625 | return r; | |
1626 | ||
1627 | } | |
1628 | ||
a340b3e2 KA |
1629 | if (writable) |
1630 | *writable = true; | |
add6a0cd PB |
1631 | |
1632 | /* | |
1633 | * Get a reference here because callers of *hva_to_pfn* and | |
1634 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
1635 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
1636 | * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will | |
1637 | * simply do nothing for reserved pfns. | |
1638 | * | |
1639 | * Whoever called remap_pfn_range is also going to call e.g. | |
1640 | * unmap_mapping_range before the underlying pages are freed, | |
1641 | * causing a call to our MMU notifier. | |
1642 | */ | |
1643 | kvm_get_pfn(pfn); | |
1644 | ||
1645 | *p_pfn = pfn; | |
92176a8e PB |
1646 | return 0; |
1647 | } | |
1648 | ||
12ce13fe XG |
1649 | /* |
1650 | * Pin guest page in memory and return its pfn. | |
1651 | * @addr: host virtual address which maps memory to the guest | |
1652 | * @atomic: whether this function can sleep | |
1653 | * @async: whether this function need to wait IO complete if the | |
1654 | * host page is not in the memory | |
1655 | * @write_fault: whether we should get a writable host page | |
1656 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1657 | * | |
1658 | * The function will map a writable host page for these two cases: | |
1659 | * 1): @write_fault = true | |
1660 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1661 | * whether the mapping is writable. | |
1662 | */ | |
ba049e93 | 1663 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
1664 | bool write_fault, bool *writable) |
1665 | { | |
1666 | struct vm_area_struct *vma; | |
ba049e93 | 1667 | kvm_pfn_t pfn = 0; |
92176a8e | 1668 | int npages, r; |
2e2e3738 | 1669 | |
2fc84311 XG |
1670 | /* we can do it either atomically or asynchronously, not both */ |
1671 | BUG_ON(atomic && async); | |
8d4e1288 | 1672 | |
b9b33da2 | 1673 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
1674 | return pfn; |
1675 | ||
1676 | if (atomic) | |
1677 | return KVM_PFN_ERR_FAULT; | |
1678 | ||
1679 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1680 | if (npages == 1) | |
1681 | return pfn; | |
8d4e1288 | 1682 | |
2fc84311 XG |
1683 | down_read(¤t->mm->mmap_sem); |
1684 | if (npages == -EHWPOISON || | |
1685 | (!async && check_user_page_hwpoison(addr))) { | |
1686 | pfn = KVM_PFN_ERR_HWPOISON; | |
1687 | goto exit; | |
1688 | } | |
1689 | ||
add6a0cd | 1690 | retry: |
2fc84311 XG |
1691 | vma = find_vma_intersection(current->mm, addr, addr + 1); |
1692 | ||
1693 | if (vma == NULL) | |
1694 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 1695 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
a340b3e2 | 1696 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn); |
add6a0cd PB |
1697 | if (r == -EAGAIN) |
1698 | goto retry; | |
92176a8e PB |
1699 | if (r < 0) |
1700 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 1701 | } else { |
4d8b81ab | 1702 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
1703 | *async = true; |
1704 | pfn = KVM_PFN_ERR_FAULT; | |
1705 | } | |
1706 | exit: | |
1707 | up_read(¤t->mm->mmap_sem); | |
2e2e3738 | 1708 | return pfn; |
35149e21 AL |
1709 | } |
1710 | ||
ba049e93 DW |
1711 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
1712 | bool atomic, bool *async, bool write_fault, | |
1713 | bool *writable) | |
887c08ac | 1714 | { |
4d8b81ab XG |
1715 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
1716 | ||
b2740d35 PB |
1717 | if (addr == KVM_HVA_ERR_RO_BAD) { |
1718 | if (writable) | |
1719 | *writable = false; | |
4d8b81ab | 1720 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 1721 | } |
4d8b81ab | 1722 | |
b2740d35 PB |
1723 | if (kvm_is_error_hva(addr)) { |
1724 | if (writable) | |
1725 | *writable = false; | |
81c52c56 | 1726 | return KVM_PFN_NOSLOT; |
b2740d35 | 1727 | } |
4d8b81ab XG |
1728 | |
1729 | /* Do not map writable pfn in the readonly memslot. */ | |
1730 | if (writable && memslot_is_readonly(slot)) { | |
1731 | *writable = false; | |
1732 | writable = NULL; | |
1733 | } | |
1734 | ||
1735 | return hva_to_pfn(addr, atomic, async, write_fault, | |
1736 | writable); | |
887c08ac | 1737 | } |
3520469d | 1738 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 1739 | |
ba049e93 | 1740 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
1741 | bool *writable) |
1742 | { | |
e37afc6e PB |
1743 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
1744 | write_fault, writable); | |
612819c3 MT |
1745 | } |
1746 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
1747 | ||
ba049e93 | 1748 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1749 | { |
4d8b81ab | 1750 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f | 1751 | } |
e37afc6e | 1752 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 1753 | |
ba049e93 | 1754 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 1755 | { |
4d8b81ab | 1756 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 1757 | } |
037d92dc | 1758 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 1759 | |
ba049e93 | 1760 | kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1761 | { |
1762 | return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn); | |
1763 | } | |
1764 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); | |
1765 | ||
ba049e93 | 1766 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1767 | { |
1768 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1769 | } | |
1770 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
1771 | ||
ba049e93 | 1772 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
1773 | { |
1774 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
1775 | } | |
1776 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
1777 | ||
ba049e93 | 1778 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
1779 | { |
1780 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
1781 | } | |
1782 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
1783 | ||
d9ef13c2 PB |
1784 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
1785 | struct page **pages, int nr_pages) | |
48987781 XG |
1786 | { |
1787 | unsigned long addr; | |
076b925d | 1788 | gfn_t entry = 0; |
48987781 | 1789 | |
d9ef13c2 | 1790 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
1791 | if (kvm_is_error_hva(addr)) |
1792 | return -1; | |
1793 | ||
1794 | if (entry < nr_pages) | |
1795 | return 0; | |
1796 | ||
1797 | return __get_user_pages_fast(addr, nr_pages, 1, pages); | |
1798 | } | |
1799 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
1800 | ||
ba049e93 | 1801 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 1802 | { |
81c52c56 | 1803 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 1804 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 1805 | |
bf4bea8e | 1806 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 1807 | WARN_ON(1); |
6cede2e6 | 1808 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 1809 | } |
a2766325 XG |
1810 | |
1811 | return pfn_to_page(pfn); | |
1812 | } | |
1813 | ||
35149e21 AL |
1814 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
1815 | { | |
ba049e93 | 1816 | kvm_pfn_t pfn; |
2e2e3738 AL |
1817 | |
1818 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 1819 | |
a2766325 | 1820 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
1821 | } |
1822 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
1823 | ||
e45adf66 KA |
1824 | static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn, |
1825 | struct kvm_host_map *map) | |
1826 | { | |
1827 | kvm_pfn_t pfn; | |
1828 | void *hva = NULL; | |
1829 | struct page *page = KVM_UNMAPPED_PAGE; | |
1830 | ||
1831 | if (!map) | |
1832 | return -EINVAL; | |
1833 | ||
1834 | pfn = gfn_to_pfn_memslot(slot, gfn); | |
1835 | if (is_error_noslot_pfn(pfn)) | |
1836 | return -EINVAL; | |
1837 | ||
1838 | if (pfn_valid(pfn)) { | |
1839 | page = pfn_to_page(pfn); | |
1840 | hva = kmap(page); | |
d30b214d | 1841 | #ifdef CONFIG_HAS_IOMEM |
e45adf66 KA |
1842 | } else { |
1843 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); | |
d30b214d | 1844 | #endif |
e45adf66 KA |
1845 | } |
1846 | ||
1847 | if (!hva) | |
1848 | return -EFAULT; | |
1849 | ||
1850 | map->page = page; | |
1851 | map->hva = hva; | |
1852 | map->pfn = pfn; | |
1853 | map->gfn = gfn; | |
1854 | ||
1855 | return 0; | |
1856 | } | |
1857 | ||
1858 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) | |
1859 | { | |
1860 | return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map); | |
1861 | } | |
1862 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); | |
1863 | ||
1864 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, | |
1865 | bool dirty) | |
1866 | { | |
1867 | if (!map) | |
1868 | return; | |
1869 | ||
1870 | if (!map->hva) | |
1871 | return; | |
1872 | ||
b614c602 | 1873 | if (map->page != KVM_UNMAPPED_PAGE) |
e45adf66 | 1874 | kunmap(map->page); |
eb1f2f38 | 1875 | #ifdef CONFIG_HAS_IOMEM |
e45adf66 KA |
1876 | else |
1877 | memunmap(map->hva); | |
eb1f2f38 | 1878 | #endif |
e45adf66 KA |
1879 | |
1880 | if (dirty) { | |
1881 | kvm_vcpu_mark_page_dirty(vcpu, map->gfn); | |
1882 | kvm_release_pfn_dirty(map->pfn); | |
1883 | } else { | |
1884 | kvm_release_pfn_clean(map->pfn); | |
1885 | } | |
1886 | ||
1887 | map->hva = NULL; | |
1888 | map->page = NULL; | |
1889 | } | |
1890 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); | |
1891 | ||
8e73485c PB |
1892 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
1893 | { | |
ba049e93 | 1894 | kvm_pfn_t pfn; |
8e73485c PB |
1895 | |
1896 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
1897 | ||
1898 | return kvm_pfn_to_page(pfn); | |
1899 | } | |
1900 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
1901 | ||
b4231d61 IE |
1902 | void kvm_release_page_clean(struct page *page) |
1903 | { | |
32cad84f XG |
1904 | WARN_ON(is_error_page(page)); |
1905 | ||
35149e21 | 1906 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
1907 | } |
1908 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
1909 | ||
ba049e93 | 1910 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 1911 | { |
bf4bea8e | 1912 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1913 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
1914 | } |
1915 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
1916 | ||
b4231d61 | 1917 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 1918 | { |
a2766325 XG |
1919 | WARN_ON(is_error_page(page)); |
1920 | ||
35149e21 AL |
1921 | kvm_release_pfn_dirty(page_to_pfn(page)); |
1922 | } | |
1923 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
1924 | ||
f7a6509f | 1925 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
1926 | { |
1927 | kvm_set_pfn_dirty(pfn); | |
1928 | kvm_release_pfn_clean(pfn); | |
1929 | } | |
f7a6509f | 1930 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 1931 | |
ba049e93 | 1932 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 1933 | { |
a78986aa | 1934 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) { |
2e2e3738 | 1935 | struct page *page = pfn_to_page(pfn); |
f95ef0cd | 1936 | |
8f946da7 | 1937 | SetPageDirty(page); |
2e2e3738 | 1938 | } |
8a7ae055 | 1939 | } |
35149e21 AL |
1940 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
1941 | ||
ba049e93 | 1942 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 1943 | { |
a78986aa | 1944 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2e2e3738 | 1945 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
1946 | } |
1947 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
1948 | ||
ba049e93 | 1949 | void kvm_get_pfn(kvm_pfn_t pfn) |
35149e21 | 1950 | { |
bf4bea8e | 1951 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 1952 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
1953 | } |
1954 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 1955 | |
195aefde IE |
1956 | static int next_segment(unsigned long len, int offset) |
1957 | { | |
1958 | if (len > PAGE_SIZE - offset) | |
1959 | return PAGE_SIZE - offset; | |
1960 | else | |
1961 | return len; | |
1962 | } | |
1963 | ||
8e73485c PB |
1964 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
1965 | void *data, int offset, int len) | |
195aefde | 1966 | { |
e0506bcb IE |
1967 | int r; |
1968 | unsigned long addr; | |
195aefde | 1969 | |
8e73485c | 1970 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
1971 | if (kvm_is_error_hva(addr)) |
1972 | return -EFAULT; | |
3180a7fc | 1973 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 1974 | if (r) |
195aefde | 1975 | return -EFAULT; |
195aefde IE |
1976 | return 0; |
1977 | } | |
8e73485c PB |
1978 | |
1979 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
1980 | int len) | |
1981 | { | |
1982 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1983 | ||
1984 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1985 | } | |
195aefde IE |
1986 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
1987 | ||
8e73485c PB |
1988 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
1989 | int offset, int len) | |
1990 | { | |
1991 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1992 | ||
1993 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
1994 | } | |
1995 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
1996 | ||
195aefde IE |
1997 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
1998 | { | |
1999 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2000 | int seg; | |
2001 | int offset = offset_in_page(gpa); | |
2002 | int ret; | |
2003 | ||
2004 | while ((seg = next_segment(len, offset)) != 0) { | |
2005 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
2006 | if (ret < 0) | |
2007 | return ret; | |
2008 | offset = 0; | |
2009 | len -= seg; | |
2010 | data += seg; | |
2011 | ++gfn; | |
2012 | } | |
2013 | return 0; | |
2014 | } | |
2015 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
2016 | ||
8e73485c | 2017 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 2018 | { |
7ec54588 | 2019 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 2020 | int seg; |
7ec54588 | 2021 | int offset = offset_in_page(gpa); |
8e73485c PB |
2022 | int ret; |
2023 | ||
2024 | while ((seg = next_segment(len, offset)) != 0) { | |
2025 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
2026 | if (ret < 0) | |
2027 | return ret; | |
2028 | offset = 0; | |
2029 | len -= seg; | |
2030 | data += seg; | |
2031 | ++gfn; | |
2032 | } | |
2033 | return 0; | |
2034 | } | |
2035 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 2036 | |
8e73485c PB |
2037 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2038 | void *data, int offset, unsigned long len) | |
2039 | { | |
2040 | int r; | |
2041 | unsigned long addr; | |
2042 | ||
2043 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
2044 | if (kvm_is_error_hva(addr)) |
2045 | return -EFAULT; | |
0aac03f0 | 2046 | pagefault_disable(); |
3180a7fc | 2047 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 2048 | pagefault_enable(); |
7ec54588 MT |
2049 | if (r) |
2050 | return -EFAULT; | |
2051 | return 0; | |
2052 | } | |
7ec54588 | 2053 | |
8e73485c PB |
2054 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
2055 | unsigned long len) | |
2056 | { | |
2057 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2058 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2059 | int offset = offset_in_page(gpa); | |
2060 | ||
2061 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2062 | } | |
2063 | EXPORT_SYMBOL_GPL(kvm_read_guest_atomic); | |
2064 | ||
2065 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, | |
2066 | void *data, unsigned long len) | |
2067 | { | |
2068 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2069 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2070 | int offset = offset_in_page(gpa); | |
2071 | ||
2072 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2073 | } | |
2074 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
2075 | ||
2076 | static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn, | |
2077 | const void *data, int offset, int len) | |
195aefde | 2078 | { |
e0506bcb IE |
2079 | int r; |
2080 | unsigned long addr; | |
195aefde | 2081 | |
251eb841 | 2082 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
2083 | if (kvm_is_error_hva(addr)) |
2084 | return -EFAULT; | |
8b0cedff | 2085 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 2086 | if (r) |
195aefde | 2087 | return -EFAULT; |
bc009e43 | 2088 | mark_page_dirty_in_slot(memslot, gfn); |
195aefde IE |
2089 | return 0; |
2090 | } | |
8e73485c PB |
2091 | |
2092 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
2093 | const void *data, int offset, int len) | |
2094 | { | |
2095 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2096 | ||
2097 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
2098 | } | |
195aefde IE |
2099 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
2100 | ||
8e73485c PB |
2101 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
2102 | const void *data, int offset, int len) | |
2103 | { | |
2104 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2105 | ||
2106 | return __kvm_write_guest_page(slot, gfn, data, offset, len); | |
2107 | } | |
2108 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
2109 | ||
195aefde IE |
2110 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
2111 | unsigned long len) | |
2112 | { | |
2113 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2114 | int seg; | |
2115 | int offset = offset_in_page(gpa); | |
2116 | int ret; | |
2117 | ||
2118 | while ((seg = next_segment(len, offset)) != 0) { | |
2119 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
2120 | if (ret < 0) | |
2121 | return ret; | |
2122 | offset = 0; | |
2123 | len -= seg; | |
2124 | data += seg; | |
2125 | ++gfn; | |
2126 | } | |
2127 | return 0; | |
2128 | } | |
ff651cb6 | 2129 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 2130 | |
8e73485c PB |
2131 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
2132 | unsigned long len) | |
2133 | { | |
2134 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2135 | int seg; | |
2136 | int offset = offset_in_page(gpa); | |
2137 | int ret; | |
2138 | ||
2139 | while ((seg = next_segment(len, offset)) != 0) { | |
2140 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
2141 | if (ret < 0) | |
2142 | return ret; | |
2143 | offset = 0; | |
2144 | len -= seg; | |
2145 | data += seg; | |
2146 | ++gfn; | |
2147 | } | |
2148 | return 0; | |
2149 | } | |
2150 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
2151 | ||
5a2d4365 PB |
2152 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
2153 | struct gfn_to_hva_cache *ghc, | |
2154 | gpa_t gpa, unsigned long len) | |
49c7754c | 2155 | { |
49c7754c | 2156 | int offset = offset_in_page(gpa); |
8f964525 AH |
2157 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
2158 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
2159 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
2160 | gfn_t nr_pages_avail; | |
f1b9dd5e | 2161 | int r = start_gfn <= end_gfn ? 0 : -EINVAL; |
49c7754c GN |
2162 | |
2163 | ghc->gpa = gpa; | |
2164 | ghc->generation = slots->generation; | |
8f964525 | 2165 | ghc->len = len; |
f1b9dd5e JM |
2166 | ghc->hva = KVM_HVA_ERR_BAD; |
2167 | ||
2168 | /* | |
2169 | * If the requested region crosses two memslots, we still | |
2170 | * verify that the entire region is valid here. | |
2171 | */ | |
2172 | while (!r && start_gfn <= end_gfn) { | |
2173 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); | |
2174 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
2175 | &nr_pages_avail); | |
2176 | if (kvm_is_error_hva(ghc->hva)) | |
2177 | r = -EFAULT; | |
2178 | start_gfn += nr_pages_avail; | |
2179 | } | |
2180 | ||
2181 | /* Use the slow path for cross page reads and writes. */ | |
2182 | if (!r && nr_pages_needed == 1) | |
49c7754c | 2183 | ghc->hva += offset; |
f1b9dd5e | 2184 | else |
8f964525 | 2185 | ghc->memslot = NULL; |
f1b9dd5e JM |
2186 | |
2187 | return r; | |
49c7754c | 2188 | } |
5a2d4365 | 2189 | |
4e335d9e | 2190 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
2191 | gpa_t gpa, unsigned long len) |
2192 | { | |
4e335d9e | 2193 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
2194 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
2195 | } | |
4e335d9e | 2196 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 2197 | |
4e335d9e | 2198 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
2199 | void *data, unsigned int offset, |
2200 | unsigned long len) | |
49c7754c | 2201 | { |
4e335d9e | 2202 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 2203 | int r; |
4ec6e863 | 2204 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 2205 | |
4ec6e863 | 2206 | BUG_ON(len + offset > ghc->len); |
8f964525 | 2207 | |
49c7754c | 2208 | if (slots->generation != ghc->generation) |
5a2d4365 | 2209 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
8f964525 AH |
2210 | |
2211 | if (unlikely(!ghc->memslot)) | |
4e335d9e | 2212 | return kvm_write_guest(kvm, gpa, data, len); |
49c7754c GN |
2213 | |
2214 | if (kvm_is_error_hva(ghc->hva)) | |
2215 | return -EFAULT; | |
2216 | ||
4ec6e863 | 2217 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
2218 | if (r) |
2219 | return -EFAULT; | |
4ec6e863 | 2220 | mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
2221 | |
2222 | return 0; | |
2223 | } | |
4e335d9e | 2224 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 2225 | |
4e335d9e PB |
2226 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2227 | void *data, unsigned long len) | |
4ec6e863 | 2228 | { |
4e335d9e | 2229 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 2230 | } |
4e335d9e | 2231 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 2232 | |
4e335d9e PB |
2233 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2234 | void *data, unsigned long len) | |
e03b644f | 2235 | { |
4e335d9e | 2236 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f GN |
2237 | int r; |
2238 | ||
8f964525 AH |
2239 | BUG_ON(len > ghc->len); |
2240 | ||
e03b644f | 2241 | if (slots->generation != ghc->generation) |
5a2d4365 | 2242 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
8f964525 AH |
2243 | |
2244 | if (unlikely(!ghc->memslot)) | |
4e335d9e | 2245 | return kvm_read_guest(kvm, ghc->gpa, data, len); |
e03b644f GN |
2246 | |
2247 | if (kvm_is_error_hva(ghc->hva)) | |
2248 | return -EFAULT; | |
2249 | ||
2250 | r = __copy_from_user(data, (void __user *)ghc->hva, len); | |
2251 | if (r) | |
2252 | return -EFAULT; | |
2253 | ||
2254 | return 0; | |
2255 | } | |
4e335d9e | 2256 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 2257 | |
195aefde IE |
2258 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
2259 | { | |
8a3caa6d HC |
2260 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
2261 | ||
2262 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); | |
195aefde IE |
2263 | } |
2264 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); | |
2265 | ||
2266 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) | |
2267 | { | |
2268 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2269 | int seg; | |
2270 | int offset = offset_in_page(gpa); | |
2271 | int ret; | |
2272 | ||
bfda0e84 | 2273 | while ((seg = next_segment(len, offset)) != 0) { |
195aefde IE |
2274 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); |
2275 | if (ret < 0) | |
2276 | return ret; | |
2277 | offset = 0; | |
2278 | len -= seg; | |
2279 | ++gfn; | |
2280 | } | |
2281 | return 0; | |
2282 | } | |
2283 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
2284 | ||
bc009e43 | 2285 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, |
7940876e | 2286 | gfn_t gfn) |
6aa8b732 | 2287 | { |
7e9d619d RR |
2288 | if (memslot && memslot->dirty_bitmap) { |
2289 | unsigned long rel_gfn = gfn - memslot->base_gfn; | |
6aa8b732 | 2290 | |
b74ca3b3 | 2291 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
2292 | } |
2293 | } | |
2294 | ||
49c7754c GN |
2295 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
2296 | { | |
2297 | struct kvm_memory_slot *memslot; | |
2298 | ||
2299 | memslot = gfn_to_memslot(kvm, gfn); | |
bc009e43 | 2300 | mark_page_dirty_in_slot(memslot, gfn); |
49c7754c | 2301 | } |
2ba9f0d8 | 2302 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 2303 | |
8e73485c PB |
2304 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
2305 | { | |
2306 | struct kvm_memory_slot *memslot; | |
2307 | ||
2308 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2309 | mark_page_dirty_in_slot(memslot, gfn); | |
2310 | } | |
2311 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
2312 | ||
20b7035c JS |
2313 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
2314 | { | |
2315 | if (!vcpu->sigset_active) | |
2316 | return; | |
2317 | ||
2318 | /* | |
2319 | * This does a lockless modification of ->real_blocked, which is fine | |
2320 | * because, only current can change ->real_blocked and all readers of | |
2321 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
2322 | * of ->blocked. | |
2323 | */ | |
2324 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
2325 | } | |
2326 | ||
2327 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
2328 | { | |
2329 | if (!vcpu->sigset_active) | |
2330 | return; | |
2331 | ||
2332 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
2333 | sigemptyset(¤t->real_blocked); | |
2334 | } | |
2335 | ||
aca6ff29 WL |
2336 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
2337 | { | |
dee339b5 | 2338 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 2339 | |
2cbd7824 | 2340 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 2341 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 2342 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
2343 | if (!grow) |
2344 | goto out; | |
2345 | ||
dee339b5 NW |
2346 | val *= grow; |
2347 | if (val < grow_start) | |
2348 | val = grow_start; | |
aca6ff29 | 2349 | |
313f636d DM |
2350 | if (val > halt_poll_ns) |
2351 | val = halt_poll_ns; | |
2352 | ||
aca6ff29 | 2353 | vcpu->halt_poll_ns = val; |
7fa08e71 | 2354 | out: |
2cbd7824 | 2355 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2356 | } |
2357 | ||
2358 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
2359 | { | |
6b6de68c | 2360 | unsigned int old, val, shrink; |
aca6ff29 | 2361 | |
2cbd7824 | 2362 | old = val = vcpu->halt_poll_ns; |
6b6de68c CB |
2363 | shrink = READ_ONCE(halt_poll_ns_shrink); |
2364 | if (shrink == 0) | |
aca6ff29 WL |
2365 | val = 0; |
2366 | else | |
6b6de68c | 2367 | val /= shrink; |
aca6ff29 WL |
2368 | |
2369 | vcpu->halt_poll_ns = val; | |
2cbd7824 | 2370 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2371 | } |
2372 | ||
f7819512 PB |
2373 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
2374 | { | |
50c28f21 JS |
2375 | int ret = -EINTR; |
2376 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
2377 | ||
f7819512 PB |
2378 | if (kvm_arch_vcpu_runnable(vcpu)) { |
2379 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 2380 | goto out; |
f7819512 PB |
2381 | } |
2382 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 2383 | goto out; |
f7819512 | 2384 | if (signal_pending(current)) |
50c28f21 | 2385 | goto out; |
f7819512 | 2386 | |
50c28f21 JS |
2387 | ret = 0; |
2388 | out: | |
2389 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
2390 | return ret; | |
f7819512 PB |
2391 | } |
2392 | ||
b6958ce4 ED |
2393 | /* |
2394 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
2395 | */ | |
8776e519 | 2396 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 2397 | { |
f7819512 | 2398 | ktime_t start, cur; |
8577370f | 2399 | DECLARE_SWAITQUEUE(wait); |
f7819512 | 2400 | bool waited = false; |
aca6ff29 | 2401 | u64 block_ns; |
f7819512 | 2402 | |
07ab0f8d MZ |
2403 | kvm_arch_vcpu_blocking(vcpu); |
2404 | ||
f7819512 | 2405 | start = cur = ktime_get(); |
cdd6ad3a | 2406 | if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) { |
19020f8a | 2407 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); |
f95ef0cd | 2408 | |
62bea5bf | 2409 | ++vcpu->stat.halt_attempted_poll; |
f7819512 PB |
2410 | do { |
2411 | /* | |
2412 | * This sets KVM_REQ_UNHALT if an interrupt | |
2413 | * arrives. | |
2414 | */ | |
2415 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
2416 | ++vcpu->stat.halt_successful_poll; | |
3491caf2 CB |
2417 | if (!vcpu_valid_wakeup(vcpu)) |
2418 | ++vcpu->stat.halt_poll_invalid; | |
f7819512 PB |
2419 | goto out; |
2420 | } | |
2421 | cur = ktime_get(); | |
2422 | } while (single_task_running() && ktime_before(cur, stop)); | |
2423 | } | |
e5c239cf MT |
2424 | |
2425 | for (;;) { | |
b3dae109 | 2426 | prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); |
e5c239cf | 2427 | |
f7819512 | 2428 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
2429 | break; |
2430 | ||
f7819512 | 2431 | waited = true; |
b6958ce4 | 2432 | schedule(); |
b6958ce4 | 2433 | } |
d3bef15f | 2434 | |
8577370f | 2435 | finish_swait(&vcpu->wq, &wait); |
f7819512 | 2436 | cur = ktime_get(); |
f7819512 | 2437 | out: |
07ab0f8d | 2438 | kvm_arch_vcpu_unblocking(vcpu); |
aca6ff29 WL |
2439 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
2440 | ||
44551b2f WL |
2441 | if (!kvm_arch_no_poll(vcpu)) { |
2442 | if (!vcpu_valid_wakeup(vcpu)) { | |
aca6ff29 | 2443 | shrink_halt_poll_ns(vcpu); |
44551b2f WL |
2444 | } else if (halt_poll_ns) { |
2445 | if (block_ns <= vcpu->halt_poll_ns) | |
2446 | ; | |
2447 | /* we had a long block, shrink polling */ | |
2448 | else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns) | |
2449 | shrink_halt_poll_ns(vcpu); | |
2450 | /* we had a short halt and our poll time is too small */ | |
2451 | else if (vcpu->halt_poll_ns < halt_poll_ns && | |
2452 | block_ns < halt_poll_ns) | |
2453 | grow_halt_poll_ns(vcpu); | |
2454 | } else { | |
2455 | vcpu->halt_poll_ns = 0; | |
2456 | } | |
2457 | } | |
aca6ff29 | 2458 | |
3491caf2 CB |
2459 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
2460 | kvm_arch_vcpu_block_finish(vcpu); | |
b6958ce4 | 2461 | } |
2ba9f0d8 | 2462 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 2463 | |
178f02ff | 2464 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 2465 | { |
8577370f | 2466 | struct swait_queue_head *wqp; |
b6d33834 CD |
2467 | |
2468 | wqp = kvm_arch_vcpu_wq(vcpu); | |
5e0018b3 | 2469 | if (swq_has_sleeper(wqp)) { |
b3dae109 | 2470 | swake_up_one(wqp); |
d73eb57b | 2471 | WRITE_ONCE(vcpu->ready, true); |
b6d33834 | 2472 | ++vcpu->stat.halt_wakeup; |
178f02ff | 2473 | return true; |
b6d33834 CD |
2474 | } |
2475 | ||
178f02ff | 2476 | return false; |
dd1a4cc1 RK |
2477 | } |
2478 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
2479 | ||
0266c894 | 2480 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
2481 | /* |
2482 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
2483 | */ | |
2484 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
2485 | { | |
2486 | int me; | |
2487 | int cpu = vcpu->cpu; | |
2488 | ||
178f02ff RK |
2489 | if (kvm_vcpu_wake_up(vcpu)) |
2490 | return; | |
2491 | ||
b6d33834 CD |
2492 | me = get_cpu(); |
2493 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
2494 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
2495 | smp_send_reschedule(cpu); | |
2496 | put_cpu(); | |
2497 | } | |
a20ed54d | 2498 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 2499 | #endif /* !CONFIG_S390 */ |
b6d33834 | 2500 | |
fa93384f | 2501 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
2502 | { |
2503 | struct pid *pid; | |
2504 | struct task_struct *task = NULL; | |
fa93384f | 2505 | int ret = 0; |
41628d33 KW |
2506 | |
2507 | rcu_read_lock(); | |
2508 | pid = rcu_dereference(target->pid); | |
2509 | if (pid) | |
27fbe64b | 2510 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
2511 | rcu_read_unlock(); |
2512 | if (!task) | |
c45c528e | 2513 | return ret; |
c45c528e | 2514 | ret = yield_to(task, 1); |
41628d33 | 2515 | put_task_struct(task); |
c45c528e R |
2516 | |
2517 | return ret; | |
41628d33 KW |
2518 | } |
2519 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
2520 | ||
06e48c51 R |
2521 | /* |
2522 | * Helper that checks whether a VCPU is eligible for directed yield. | |
2523 | * Most eligible candidate to yield is decided by following heuristics: | |
2524 | * | |
2525 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
2526 | * (preempted lock holder), indicated by @in_spin_loop. | |
2527 | * Set at the beiginning and cleared at the end of interception/PLE handler. | |
2528 | * | |
2529 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
2530 | * chance last time (mostly it has become eligible now since we have probably | |
2531 | * yielded to lockholder in last iteration. This is done by toggling | |
2532 | * @dy_eligible each time a VCPU checked for eligibility.) | |
2533 | * | |
2534 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
2535 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
2536 | * burning. Giving priority for a potential lock-holder increases lock | |
2537 | * progress. | |
2538 | * | |
2539 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
2540 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
2541 | * and continue with next VCPU and so on. | |
2542 | */ | |
7940876e | 2543 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 2544 | { |
4a55dd72 | 2545 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
2546 | bool eligible; |
2547 | ||
2548 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 2549 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
2550 | |
2551 | if (vcpu->spin_loop.in_spin_loop) | |
2552 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
2553 | ||
2554 | return eligible; | |
4a55dd72 SW |
2555 | #else |
2556 | return true; | |
06e48c51 | 2557 | #endif |
4a55dd72 | 2558 | } |
c45c528e | 2559 | |
17e433b5 WL |
2560 | /* |
2561 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
2562 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
2563 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
2564 | */ | |
2565 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
2566 | { | |
2567 | return kvm_arch_vcpu_runnable(vcpu); | |
2568 | } | |
2569 | ||
2570 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
2571 | { | |
2572 | if (kvm_arch_dy_runnable(vcpu)) | |
2573 | return true; | |
2574 | ||
2575 | #ifdef CONFIG_KVM_ASYNC_PF | |
2576 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
2577 | return true; | |
2578 | #endif | |
2579 | ||
2580 | return false; | |
2581 | } | |
2582 | ||
199b5763 | 2583 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 2584 | { |
217ece61 RR |
2585 | struct kvm *kvm = me->kvm; |
2586 | struct kvm_vcpu *vcpu; | |
2587 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
2588 | int yielded = 0; | |
c45c528e | 2589 | int try = 3; |
217ece61 RR |
2590 | int pass; |
2591 | int i; | |
d255f4f2 | 2592 | |
4c088493 | 2593 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
2594 | /* |
2595 | * We boost the priority of a VCPU that is runnable but not | |
2596 | * currently running, because it got preempted by something | |
2597 | * else and called schedule in __vcpu_run. Hopefully that | |
2598 | * VCPU is holding the lock that we need and will release it. | |
2599 | * We approximate round-robin by starting at the last boosted VCPU. | |
2600 | */ | |
c45c528e | 2601 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 2602 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 2603 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
2604 | i = last_boosted_vcpu; |
2605 | continue; | |
2606 | } else if (pass && i > last_boosted_vcpu) | |
2607 | break; | |
d73eb57b | 2608 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 2609 | continue; |
217ece61 RR |
2610 | if (vcpu == me) |
2611 | continue; | |
17e433b5 | 2612 | if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu)) |
217ece61 | 2613 | continue; |
046ddeed WL |
2614 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
2615 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 2616 | continue; |
06e48c51 R |
2617 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
2618 | continue; | |
c45c528e R |
2619 | |
2620 | yielded = kvm_vcpu_yield_to(vcpu); | |
2621 | if (yielded > 0) { | |
217ece61 | 2622 | kvm->last_boosted_vcpu = i; |
217ece61 | 2623 | break; |
c45c528e R |
2624 | } else if (yielded < 0) { |
2625 | try--; | |
2626 | if (!try) | |
2627 | break; | |
217ece61 | 2628 | } |
217ece61 RR |
2629 | } |
2630 | } | |
4c088493 | 2631 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
2632 | |
2633 | /* Ensure vcpu is not eligible during next spinloop */ | |
2634 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
2635 | } |
2636 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
2637 | ||
1499fa80 | 2638 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 2639 | { |
11bac800 | 2640 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
2641 | struct page *page; |
2642 | ||
e4a533a4 | 2643 | if (vmf->pgoff == 0) |
039576c0 | 2644 | page = virt_to_page(vcpu->run); |
09566765 | 2645 | #ifdef CONFIG_X86 |
e4a533a4 | 2646 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 2647 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 2648 | #endif |
4b4357e0 | 2649 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
2650 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
2651 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 2652 | #endif |
039576c0 | 2653 | else |
5b1c1493 | 2654 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 2655 | get_page(page); |
e4a533a4 NP |
2656 | vmf->page = page; |
2657 | return 0; | |
9a2bb7f4 AK |
2658 | } |
2659 | ||
f0f37e2f | 2660 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 2661 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
2662 | }; |
2663 | ||
2664 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
2665 | { | |
2666 | vma->vm_ops = &kvm_vcpu_vm_ops; | |
2667 | return 0; | |
2668 | } | |
2669 | ||
bccf2150 AK |
2670 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
2671 | { | |
2672 | struct kvm_vcpu *vcpu = filp->private_data; | |
2673 | ||
45b5939e | 2674 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
66c0b394 | 2675 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
2676 | return 0; |
2677 | } | |
2678 | ||
3d3aab1b | 2679 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
2680 | .release = kvm_vcpu_release, |
2681 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 2682 | .mmap = kvm_vcpu_mmap, |
6038f373 | 2683 | .llseek = noop_llseek, |
7ddfd3e0 | 2684 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
2685 | }; |
2686 | ||
2687 | /* | |
2688 | * Allocates an inode for the vcpu. | |
2689 | */ | |
2690 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
2691 | { | |
e46b4692 MY |
2692 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
2693 | ||
2694 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
2695 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
2696 | } |
2697 | ||
3e7093d0 | 2698 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 2699 | { |
741cbbae | 2700 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
45b5939e | 2701 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 2702 | |
45b5939e | 2703 | if (!debugfs_initialized()) |
3e7093d0 | 2704 | return; |
45b5939e LC |
2705 | |
2706 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
2707 | vcpu->debugfs_dentry = debugfs_create_dir(dir_name, | |
3e7093d0 | 2708 | vcpu->kvm->debugfs_dentry); |
45b5939e | 2709 | |
3e7093d0 | 2710 | kvm_arch_create_vcpu_debugfs(vcpu); |
741cbbae | 2711 | #endif |
45b5939e LC |
2712 | } |
2713 | ||
c5ea7660 AK |
2714 | /* |
2715 | * Creates some virtual cpus. Good luck creating more than one. | |
2716 | */ | |
73880c80 | 2717 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
2718 | { |
2719 | int r; | |
e09fefde | 2720 | struct kvm_vcpu *vcpu; |
c5ea7660 | 2721 | |
0b1b1dfd | 2722 | if (id >= KVM_MAX_VCPU_ID) |
338c7dba AH |
2723 | return -EINVAL; |
2724 | ||
6c7caebc PB |
2725 | mutex_lock(&kvm->lock); |
2726 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
2727 | mutex_unlock(&kvm->lock); | |
2728 | return -EINVAL; | |
2729 | } | |
2730 | ||
2731 | kvm->created_vcpus++; | |
2732 | mutex_unlock(&kvm->lock); | |
2733 | ||
73880c80 | 2734 | vcpu = kvm_arch_vcpu_create(kvm, id); |
6c7caebc PB |
2735 | if (IS_ERR(vcpu)) { |
2736 | r = PTR_ERR(vcpu); | |
2737 | goto vcpu_decrement; | |
2738 | } | |
c5ea7660 | 2739 | |
15ad7146 AK |
2740 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
2741 | ||
26e5215f AK |
2742 | r = kvm_arch_vcpu_setup(vcpu); |
2743 | if (r) | |
d780592b | 2744 | goto vcpu_destroy; |
26e5215f | 2745 | |
3e7093d0 | 2746 | kvm_create_vcpu_debugfs(vcpu); |
45b5939e | 2747 | |
11ec2804 | 2748 | mutex_lock(&kvm->lock); |
e09fefde DH |
2749 | if (kvm_get_vcpu_by_id(kvm, id)) { |
2750 | r = -EEXIST; | |
2751 | goto unlock_vcpu_destroy; | |
2752 | } | |
73880c80 | 2753 | |
8750e72a RK |
2754 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
2755 | BUG_ON(kvm->vcpus[vcpu->vcpu_idx]); | |
c5ea7660 | 2756 | |
fb3f0f51 | 2757 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 2758 | kvm_get_kvm(kvm); |
bccf2150 | 2759 | r = create_vcpu_fd(vcpu); |
73880c80 | 2760 | if (r < 0) { |
149487bd | 2761 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 2762 | goto unlock_vcpu_destroy; |
73880c80 GN |
2763 | } |
2764 | ||
8750e72a | 2765 | kvm->vcpus[vcpu->vcpu_idx] = vcpu; |
dd489240 PB |
2766 | |
2767 | /* | |
2768 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
2769 | * before kvm->online_vcpu's incremented value. | |
2770 | */ | |
73880c80 GN |
2771 | smp_wmb(); |
2772 | atomic_inc(&kvm->online_vcpus); | |
2773 | ||
73880c80 | 2774 | mutex_unlock(&kvm->lock); |
42897d86 | 2775 | kvm_arch_vcpu_postcreate(vcpu); |
fb3f0f51 | 2776 | return r; |
39c3b86e | 2777 | |
d780592b | 2778 | unlock_vcpu_destroy: |
7d8fece6 | 2779 | mutex_unlock(&kvm->lock); |
45b5939e | 2780 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
d780592b | 2781 | vcpu_destroy: |
d40ccc62 | 2782 | kvm_arch_vcpu_destroy(vcpu); |
6c7caebc PB |
2783 | vcpu_decrement: |
2784 | mutex_lock(&kvm->lock); | |
2785 | kvm->created_vcpus--; | |
2786 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
2787 | return r; |
2788 | } | |
2789 | ||
1961d276 AK |
2790 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
2791 | { | |
2792 | if (sigset) { | |
2793 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
2794 | vcpu->sigset_active = 1; | |
2795 | vcpu->sigset = *sigset; | |
2796 | } else | |
2797 | vcpu->sigset_active = 0; | |
2798 | return 0; | |
2799 | } | |
2800 | ||
bccf2150 AK |
2801 | static long kvm_vcpu_ioctl(struct file *filp, |
2802 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 2803 | { |
bccf2150 | 2804 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 2805 | void __user *argp = (void __user *)arg; |
313a3dc7 | 2806 | int r; |
fa3795a7 DH |
2807 | struct kvm_fpu *fpu = NULL; |
2808 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 2809 | |
6d4e4c4f AK |
2810 | if (vcpu->kvm->mm != current->mm) |
2811 | return -EIO; | |
2122ff5e | 2812 | |
2ea75be3 DM |
2813 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
2814 | return -EINVAL; | |
2815 | ||
2122ff5e | 2816 | /* |
5cb0944c PB |
2817 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
2818 | * execution; mutex_lock() would break them. | |
2122ff5e | 2819 | */ |
5cb0944c PB |
2820 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
2821 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 2822 | return r; |
2122ff5e | 2823 | |
ec7660cc CD |
2824 | if (mutex_lock_killable(&vcpu->mutex)) |
2825 | return -EINTR; | |
6aa8b732 | 2826 | switch (ioctl) { |
0e4524a5 CB |
2827 | case KVM_RUN: { |
2828 | struct pid *oldpid; | |
f0fe5108 AK |
2829 | r = -EINVAL; |
2830 | if (arg) | |
2831 | goto out; | |
0e4524a5 | 2832 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 2833 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 2834 | /* The thread running this VCPU changed. */ |
bd2a6394 | 2835 | struct pid *newpid; |
f95ef0cd | 2836 | |
bd2a6394 CD |
2837 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
2838 | if (r) | |
2839 | break; | |
2840 | ||
2841 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
2842 | rcu_assign_pointer(vcpu->pid, newpid); |
2843 | if (oldpid) | |
2844 | synchronize_rcu(); | |
2845 | put_pid(oldpid); | |
2846 | } | |
b6c7a5dc | 2847 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
64be5007 | 2848 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 2849 | break; |
0e4524a5 | 2850 | } |
6aa8b732 | 2851 | case KVM_GET_REGS: { |
3e4bb3ac | 2852 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2853 | |
3e4bb3ac | 2854 | r = -ENOMEM; |
b12ce36a | 2855 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 2856 | if (!kvm_regs) |
6aa8b732 | 2857 | goto out; |
3e4bb3ac XZ |
2858 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
2859 | if (r) | |
2860 | goto out_free1; | |
6aa8b732 | 2861 | r = -EFAULT; |
3e4bb3ac XZ |
2862 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
2863 | goto out_free1; | |
6aa8b732 | 2864 | r = 0; |
3e4bb3ac XZ |
2865 | out_free1: |
2866 | kfree(kvm_regs); | |
6aa8b732 AK |
2867 | break; |
2868 | } | |
2869 | case KVM_SET_REGS: { | |
3e4bb3ac | 2870 | struct kvm_regs *kvm_regs; |
6aa8b732 | 2871 | |
3e4bb3ac | 2872 | r = -ENOMEM; |
ff5c2c03 SL |
2873 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
2874 | if (IS_ERR(kvm_regs)) { | |
2875 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 2876 | goto out; |
ff5c2c03 | 2877 | } |
3e4bb3ac | 2878 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 2879 | kfree(kvm_regs); |
6aa8b732 AK |
2880 | break; |
2881 | } | |
2882 | case KVM_GET_SREGS: { | |
b12ce36a BG |
2883 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
2884 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
2885 | r = -ENOMEM; |
2886 | if (!kvm_sregs) | |
2887 | goto out; | |
2888 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
2889 | if (r) |
2890 | goto out; | |
2891 | r = -EFAULT; | |
fa3795a7 | 2892 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
2893 | goto out; |
2894 | r = 0; | |
2895 | break; | |
2896 | } | |
2897 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
2898 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
2899 | if (IS_ERR(kvm_sregs)) { | |
2900 | r = PTR_ERR(kvm_sregs); | |
18595411 | 2901 | kvm_sregs = NULL; |
6aa8b732 | 2902 | goto out; |
ff5c2c03 | 2903 | } |
fa3795a7 | 2904 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
2905 | break; |
2906 | } | |
62d9f0db MT |
2907 | case KVM_GET_MP_STATE: { |
2908 | struct kvm_mp_state mp_state; | |
2909 | ||
2910 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
2911 | if (r) | |
2912 | goto out; | |
2913 | r = -EFAULT; | |
893bdbf1 | 2914 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
2915 | goto out; |
2916 | r = 0; | |
2917 | break; | |
2918 | } | |
2919 | case KVM_SET_MP_STATE: { | |
2920 | struct kvm_mp_state mp_state; | |
2921 | ||
2922 | r = -EFAULT; | |
893bdbf1 | 2923 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
2924 | goto out; |
2925 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
2926 | break; |
2927 | } | |
6aa8b732 AK |
2928 | case KVM_TRANSLATE: { |
2929 | struct kvm_translation tr; | |
2930 | ||
2931 | r = -EFAULT; | |
893bdbf1 | 2932 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 2933 | goto out; |
8b006791 | 2934 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
2935 | if (r) |
2936 | goto out; | |
2937 | r = -EFAULT; | |
893bdbf1 | 2938 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
2939 | goto out; |
2940 | r = 0; | |
2941 | break; | |
2942 | } | |
d0bfb940 JK |
2943 | case KVM_SET_GUEST_DEBUG: { |
2944 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
2945 | |
2946 | r = -EFAULT; | |
893bdbf1 | 2947 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 2948 | goto out; |
d0bfb940 | 2949 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
2950 | break; |
2951 | } | |
1961d276 AK |
2952 | case KVM_SET_SIGNAL_MASK: { |
2953 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
2954 | struct kvm_signal_mask kvm_sigmask; | |
2955 | sigset_t sigset, *p; | |
2956 | ||
2957 | p = NULL; | |
2958 | if (argp) { | |
2959 | r = -EFAULT; | |
2960 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 2961 | sizeof(kvm_sigmask))) |
1961d276 AK |
2962 | goto out; |
2963 | r = -EINVAL; | |
893bdbf1 | 2964 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
2965 | goto out; |
2966 | r = -EFAULT; | |
2967 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 2968 | sizeof(sigset))) |
1961d276 AK |
2969 | goto out; |
2970 | p = &sigset; | |
2971 | } | |
376d41ff | 2972 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
2973 | break; |
2974 | } | |
b8836737 | 2975 | case KVM_GET_FPU: { |
b12ce36a | 2976 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
2977 | r = -ENOMEM; |
2978 | if (!fpu) | |
2979 | goto out; | |
2980 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
2981 | if (r) |
2982 | goto out; | |
2983 | r = -EFAULT; | |
fa3795a7 | 2984 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
2985 | goto out; |
2986 | r = 0; | |
2987 | break; | |
2988 | } | |
2989 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
2990 | fpu = memdup_user(argp, sizeof(*fpu)); |
2991 | if (IS_ERR(fpu)) { | |
2992 | r = PTR_ERR(fpu); | |
18595411 | 2993 | fpu = NULL; |
b8836737 | 2994 | goto out; |
ff5c2c03 | 2995 | } |
fa3795a7 | 2996 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
2997 | break; |
2998 | } | |
bccf2150 | 2999 | default: |
313a3dc7 | 3000 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
3001 | } |
3002 | out: | |
ec7660cc | 3003 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
3004 | kfree(fpu); |
3005 | kfree(kvm_sregs); | |
bccf2150 AK |
3006 | return r; |
3007 | } | |
3008 | ||
de8e5d74 | 3009 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
3010 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
3011 | unsigned int ioctl, unsigned long arg) | |
3012 | { | |
3013 | struct kvm_vcpu *vcpu = filp->private_data; | |
3014 | void __user *argp = compat_ptr(arg); | |
3015 | int r; | |
3016 | ||
3017 | if (vcpu->kvm->mm != current->mm) | |
3018 | return -EIO; | |
3019 | ||
3020 | switch (ioctl) { | |
3021 | case KVM_SET_SIGNAL_MASK: { | |
3022 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3023 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
3024 | sigset_t sigset; |
3025 | ||
3026 | if (argp) { | |
3027 | r = -EFAULT; | |
3028 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3029 | sizeof(kvm_sigmask))) |
1dda606c AG |
3030 | goto out; |
3031 | r = -EINVAL; | |
3968cf62 | 3032 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
3033 | goto out; |
3034 | r = -EFAULT; | |
3968cf62 | 3035 | if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset)) |
1dda606c | 3036 | goto out; |
760a9a30 AC |
3037 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
3038 | } else | |
3039 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
3040 | break; |
3041 | } | |
3042 | default: | |
3043 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
3044 | } | |
3045 | ||
3046 | out: | |
3047 | return r; | |
3048 | } | |
3049 | #endif | |
3050 | ||
a1cd3f08 CLG |
3051 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
3052 | { | |
3053 | struct kvm_device *dev = filp->private_data; | |
3054 | ||
3055 | if (dev->ops->mmap) | |
3056 | return dev->ops->mmap(dev, vma); | |
3057 | ||
3058 | return -ENODEV; | |
3059 | } | |
3060 | ||
852b6d57 SW |
3061 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
3062 | int (*accessor)(struct kvm_device *dev, | |
3063 | struct kvm_device_attr *attr), | |
3064 | unsigned long arg) | |
3065 | { | |
3066 | struct kvm_device_attr attr; | |
3067 | ||
3068 | if (!accessor) | |
3069 | return -EPERM; | |
3070 | ||
3071 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
3072 | return -EFAULT; | |
3073 | ||
3074 | return accessor(dev, &attr); | |
3075 | } | |
3076 | ||
3077 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
3078 | unsigned long arg) | |
3079 | { | |
3080 | struct kvm_device *dev = filp->private_data; | |
3081 | ||
ddba9180 SC |
3082 | if (dev->kvm->mm != current->mm) |
3083 | return -EIO; | |
3084 | ||
852b6d57 SW |
3085 | switch (ioctl) { |
3086 | case KVM_SET_DEVICE_ATTR: | |
3087 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
3088 | case KVM_GET_DEVICE_ATTR: | |
3089 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
3090 | case KVM_HAS_DEVICE_ATTR: | |
3091 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
3092 | default: | |
3093 | if (dev->ops->ioctl) | |
3094 | return dev->ops->ioctl(dev, ioctl, arg); | |
3095 | ||
3096 | return -ENOTTY; | |
3097 | } | |
3098 | } | |
3099 | ||
852b6d57 SW |
3100 | static int kvm_device_release(struct inode *inode, struct file *filp) |
3101 | { | |
3102 | struct kvm_device *dev = filp->private_data; | |
3103 | struct kvm *kvm = dev->kvm; | |
3104 | ||
2bde9b3e CLG |
3105 | if (dev->ops->release) { |
3106 | mutex_lock(&kvm->lock); | |
3107 | list_del(&dev->vm_node); | |
3108 | dev->ops->release(dev); | |
3109 | mutex_unlock(&kvm->lock); | |
3110 | } | |
3111 | ||
852b6d57 SW |
3112 | kvm_put_kvm(kvm); |
3113 | return 0; | |
3114 | } | |
3115 | ||
3116 | static const struct file_operations kvm_device_fops = { | |
3117 | .unlocked_ioctl = kvm_device_ioctl, | |
3118 | .release = kvm_device_release, | |
7ddfd3e0 | 3119 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 3120 | .mmap = kvm_device_mmap, |
852b6d57 SW |
3121 | }; |
3122 | ||
3123 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
3124 | { | |
3125 | if (filp->f_op != &kvm_device_fops) | |
3126 | return NULL; | |
3127 | ||
3128 | return filp->private_data; | |
3129 | } | |
3130 | ||
8538cb22 | 3131 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 3132 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
3133 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
3134 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 3135 | #endif |
d60eacb0 WD |
3136 | }; |
3137 | ||
8538cb22 | 3138 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
3139 | { |
3140 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
3141 | return -ENOSPC; | |
3142 | ||
3143 | if (kvm_device_ops_table[type] != NULL) | |
3144 | return -EEXIST; | |
3145 | ||
3146 | kvm_device_ops_table[type] = ops; | |
3147 | return 0; | |
3148 | } | |
3149 | ||
571ee1b6 WL |
3150 | void kvm_unregister_device_ops(u32 type) |
3151 | { | |
3152 | if (kvm_device_ops_table[type] != NULL) | |
3153 | kvm_device_ops_table[type] = NULL; | |
3154 | } | |
3155 | ||
852b6d57 SW |
3156 | static int kvm_ioctl_create_device(struct kvm *kvm, |
3157 | struct kvm_create_device *cd) | |
3158 | { | |
8538cb22 | 3159 | const struct kvm_device_ops *ops = NULL; |
852b6d57 SW |
3160 | struct kvm_device *dev; |
3161 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 3162 | int type; |
852b6d57 SW |
3163 | int ret; |
3164 | ||
d60eacb0 WD |
3165 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
3166 | return -ENODEV; | |
3167 | ||
1d487e9b PB |
3168 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
3169 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 3170 | if (ops == NULL) |
852b6d57 | 3171 | return -ENODEV; |
852b6d57 SW |
3172 | |
3173 | if (test) | |
3174 | return 0; | |
3175 | ||
b12ce36a | 3176 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
3177 | if (!dev) |
3178 | return -ENOMEM; | |
3179 | ||
3180 | dev->ops = ops; | |
3181 | dev->kvm = kvm; | |
852b6d57 | 3182 | |
a28ebea2 | 3183 | mutex_lock(&kvm->lock); |
1d487e9b | 3184 | ret = ops->create(dev, type); |
852b6d57 | 3185 | if (ret < 0) { |
a28ebea2 | 3186 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
3187 | kfree(dev); |
3188 | return ret; | |
3189 | } | |
a28ebea2 CD |
3190 | list_add(&dev->vm_node, &kvm->devices); |
3191 | mutex_unlock(&kvm->lock); | |
852b6d57 | 3192 | |
023e9fdd CD |
3193 | if (ops->init) |
3194 | ops->init(dev); | |
3195 | ||
cfa39381 | 3196 | kvm_get_kvm(kvm); |
24009b05 | 3197 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 3198 | if (ret < 0) { |
149487bd | 3199 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
3200 | mutex_lock(&kvm->lock); |
3201 | list_del(&dev->vm_node); | |
3202 | mutex_unlock(&kvm->lock); | |
a0f1d21c | 3203 | ops->destroy(dev); |
852b6d57 SW |
3204 | return ret; |
3205 | } | |
3206 | ||
852b6d57 SW |
3207 | cd->fd = ret; |
3208 | return 0; | |
3209 | } | |
3210 | ||
92b591a4 AG |
3211 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
3212 | { | |
3213 | switch (arg) { | |
3214 | case KVM_CAP_USER_MEMORY: | |
3215 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
3216 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
3217 | case KVM_CAP_INTERNAL_ERROR_DATA: |
3218 | #ifdef CONFIG_HAVE_KVM_MSI | |
3219 | case KVM_CAP_SIGNAL_MSI: | |
3220 | #endif | |
297e2105 | 3221 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 3222 | case KVM_CAP_IRQFD: |
92b591a4 AG |
3223 | case KVM_CAP_IRQFD_RESAMPLE: |
3224 | #endif | |
e9ea5069 | 3225 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 3226 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 3227 | case KVM_CAP_ENABLE_CAP_VM: |
2a31b9db | 3228 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
d7547c55 | 3229 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: |
2a31b9db | 3230 | #endif |
92b591a4 | 3231 | return 1; |
4b4357e0 | 3232 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
3233 | case KVM_CAP_COALESCED_MMIO: |
3234 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
3235 | case KVM_CAP_COALESCED_PIO: |
3236 | return 1; | |
30422558 | 3237 | #endif |
92b591a4 AG |
3238 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
3239 | case KVM_CAP_IRQ_ROUTING: | |
3240 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
3241 | #endif |
3242 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
3243 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
3244 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 3245 | #endif |
c110ae57 PB |
3246 | case KVM_CAP_NR_MEMSLOTS: |
3247 | return KVM_USER_MEM_SLOTS; | |
92b591a4 AG |
3248 | default: |
3249 | break; | |
3250 | } | |
3251 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
3252 | } | |
3253 | ||
e5d83c74 PB |
3254 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
3255 | struct kvm_enable_cap *cap) | |
3256 | { | |
3257 | return -EINVAL; | |
3258 | } | |
3259 | ||
3260 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
3261 | struct kvm_enable_cap *cap) | |
3262 | { | |
3263 | switch (cap->cap) { | |
2a31b9db | 3264 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
d7547c55 | 3265 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: |
2a31b9db PB |
3266 | if (cap->flags || (cap->args[0] & ~1)) |
3267 | return -EINVAL; | |
3268 | kvm->manual_dirty_log_protect = cap->args[0]; | |
3269 | return 0; | |
3270 | #endif | |
e5d83c74 PB |
3271 | default: |
3272 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
3273 | } | |
3274 | } | |
3275 | ||
bccf2150 AK |
3276 | static long kvm_vm_ioctl(struct file *filp, |
3277 | unsigned int ioctl, unsigned long arg) | |
3278 | { | |
3279 | struct kvm *kvm = filp->private_data; | |
3280 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 3281 | int r; |
bccf2150 | 3282 | |
6d4e4c4f AK |
3283 | if (kvm->mm != current->mm) |
3284 | return -EIO; | |
bccf2150 AK |
3285 | switch (ioctl) { |
3286 | case KVM_CREATE_VCPU: | |
3287 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 3288 | break; |
e5d83c74 PB |
3289 | case KVM_ENABLE_CAP: { |
3290 | struct kvm_enable_cap cap; | |
3291 | ||
3292 | r = -EFAULT; | |
3293 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
3294 | goto out; | |
3295 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
3296 | break; | |
3297 | } | |
6fc138d2 IE |
3298 | case KVM_SET_USER_MEMORY_REGION: { |
3299 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
3300 | ||
3301 | r = -EFAULT; | |
3302 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 3303 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
3304 | goto out; |
3305 | ||
47ae31e2 | 3306 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
3307 | break; |
3308 | } | |
3309 | case KVM_GET_DIRTY_LOG: { | |
3310 | struct kvm_dirty_log log; | |
3311 | ||
3312 | r = -EFAULT; | |
893bdbf1 | 3313 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 3314 | goto out; |
2c6f5df9 | 3315 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
3316 | break; |
3317 | } | |
2a31b9db PB |
3318 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3319 | case KVM_CLEAR_DIRTY_LOG: { | |
3320 | struct kvm_clear_dirty_log log; | |
3321 | ||
3322 | r = -EFAULT; | |
3323 | if (copy_from_user(&log, argp, sizeof(log))) | |
3324 | goto out; | |
3325 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
3326 | break; | |
3327 | } | |
3328 | #endif | |
4b4357e0 | 3329 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3330 | case KVM_REGISTER_COALESCED_MMIO: { |
3331 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 3332 | |
5f94c174 | 3333 | r = -EFAULT; |
893bdbf1 | 3334 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 3335 | goto out; |
5f94c174 | 3336 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
3337 | break; |
3338 | } | |
3339 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
3340 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 3341 | |
5f94c174 | 3342 | r = -EFAULT; |
893bdbf1 | 3343 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 3344 | goto out; |
5f94c174 | 3345 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
3346 | break; |
3347 | } | |
3348 | #endif | |
721eecbf GH |
3349 | case KVM_IRQFD: { |
3350 | struct kvm_irqfd data; | |
3351 | ||
3352 | r = -EFAULT; | |
893bdbf1 | 3353 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 3354 | goto out; |
d4db2935 | 3355 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
3356 | break; |
3357 | } | |
d34e6b17 GH |
3358 | case KVM_IOEVENTFD: { |
3359 | struct kvm_ioeventfd data; | |
3360 | ||
3361 | r = -EFAULT; | |
893bdbf1 | 3362 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
3363 | goto out; |
3364 | r = kvm_ioeventfd(kvm, &data); | |
3365 | break; | |
3366 | } | |
07975ad3 JK |
3367 | #ifdef CONFIG_HAVE_KVM_MSI |
3368 | case KVM_SIGNAL_MSI: { | |
3369 | struct kvm_msi msi; | |
3370 | ||
3371 | r = -EFAULT; | |
893bdbf1 | 3372 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
3373 | goto out; |
3374 | r = kvm_send_userspace_msi(kvm, &msi); | |
3375 | break; | |
3376 | } | |
23d43cf9 CD |
3377 | #endif |
3378 | #ifdef __KVM_HAVE_IRQ_LINE | |
3379 | case KVM_IRQ_LINE_STATUS: | |
3380 | case KVM_IRQ_LINE: { | |
3381 | struct kvm_irq_level irq_event; | |
3382 | ||
3383 | r = -EFAULT; | |
893bdbf1 | 3384 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
3385 | goto out; |
3386 | ||
aa2fbe6d YZ |
3387 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
3388 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
3389 | if (r) |
3390 | goto out; | |
3391 | ||
3392 | r = -EFAULT; | |
3393 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 3394 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
3395 | goto out; |
3396 | } | |
3397 | ||
3398 | r = 0; | |
3399 | break; | |
3400 | } | |
73880c80 | 3401 | #endif |
aa8d5944 AG |
3402 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
3403 | case KVM_SET_GSI_ROUTING: { | |
3404 | struct kvm_irq_routing routing; | |
3405 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 3406 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
3407 | |
3408 | r = -EFAULT; | |
3409 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
3410 | goto out; | |
3411 | r = -EINVAL; | |
5c0aea0e DH |
3412 | if (!kvm_arch_can_set_irq_routing(kvm)) |
3413 | goto out; | |
caf1ff26 | 3414 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
3415 | goto out; |
3416 | if (routing.flags) | |
3417 | goto out; | |
f8c1b85b PB |
3418 | if (routing.nr) { |
3419 | r = -ENOMEM; | |
42bc47b3 KC |
3420 | entries = vmalloc(array_size(sizeof(*entries), |
3421 | routing.nr)); | |
f8c1b85b PB |
3422 | if (!entries) |
3423 | goto out; | |
3424 | r = -EFAULT; | |
3425 | urouting = argp; | |
3426 | if (copy_from_user(entries, urouting->entries, | |
3427 | routing.nr * sizeof(*entries))) | |
3428 | goto out_free_irq_routing; | |
3429 | } | |
aa8d5944 AG |
3430 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
3431 | routing.flags); | |
a642a175 | 3432 | out_free_irq_routing: |
aa8d5944 AG |
3433 | vfree(entries); |
3434 | break; | |
3435 | } | |
3436 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
3437 | case KVM_CREATE_DEVICE: { |
3438 | struct kvm_create_device cd; | |
3439 | ||
3440 | r = -EFAULT; | |
3441 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
3442 | goto out; | |
3443 | ||
3444 | r = kvm_ioctl_create_device(kvm, &cd); | |
3445 | if (r) | |
3446 | goto out; | |
3447 | ||
3448 | r = -EFAULT; | |
3449 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
3450 | goto out; | |
3451 | ||
3452 | r = 0; | |
3453 | break; | |
3454 | } | |
92b591a4 AG |
3455 | case KVM_CHECK_EXTENSION: |
3456 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
3457 | break; | |
f17abe9a | 3458 | default: |
1fe779f8 | 3459 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
3460 | } |
3461 | out: | |
3462 | return r; | |
3463 | } | |
3464 | ||
de8e5d74 | 3465 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
3466 | struct compat_kvm_dirty_log { |
3467 | __u32 slot; | |
3468 | __u32 padding1; | |
3469 | union { | |
3470 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
3471 | __u64 padding2; | |
3472 | }; | |
3473 | }; | |
3474 | ||
3475 | static long kvm_vm_compat_ioctl(struct file *filp, | |
3476 | unsigned int ioctl, unsigned long arg) | |
3477 | { | |
3478 | struct kvm *kvm = filp->private_data; | |
3479 | int r; | |
3480 | ||
3481 | if (kvm->mm != current->mm) | |
3482 | return -EIO; | |
3483 | switch (ioctl) { | |
3484 | case KVM_GET_DIRTY_LOG: { | |
3485 | struct compat_kvm_dirty_log compat_log; | |
3486 | struct kvm_dirty_log log; | |
3487 | ||
6ff5894c AB |
3488 | if (copy_from_user(&compat_log, (void __user *)arg, |
3489 | sizeof(compat_log))) | |
f6a3b168 | 3490 | return -EFAULT; |
6ff5894c AB |
3491 | log.slot = compat_log.slot; |
3492 | log.padding1 = compat_log.padding1; | |
3493 | log.padding2 = compat_log.padding2; | |
3494 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
3495 | ||
3496 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
3497 | break; |
3498 | } | |
3499 | default: | |
3500 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
3501 | } | |
6ff5894c AB |
3502 | return r; |
3503 | } | |
3504 | #endif | |
3505 | ||
3d3aab1b | 3506 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
3507 | .release = kvm_vm_release, |
3508 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 3509 | .llseek = noop_llseek, |
7ddfd3e0 | 3510 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
3511 | }; |
3512 | ||
e08b9637 | 3513 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 3514 | { |
aac87636 | 3515 | int r; |
f17abe9a | 3516 | struct kvm *kvm; |
506cfba9 | 3517 | struct file *file; |
f17abe9a | 3518 | |
e08b9637 | 3519 | kvm = kvm_create_vm(type); |
d6d28168 AK |
3520 | if (IS_ERR(kvm)) |
3521 | return PTR_ERR(kvm); | |
4b4357e0 | 3522 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 3523 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
3524 | if (r < 0) |
3525 | goto put_kvm; | |
6ce5a090 | 3526 | #endif |
506cfba9 | 3527 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
3528 | if (r < 0) |
3529 | goto put_kvm; | |
3530 | ||
506cfba9 AV |
3531 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
3532 | if (IS_ERR(file)) { | |
3533 | put_unused_fd(r); | |
78588335 ME |
3534 | r = PTR_ERR(file); |
3535 | goto put_kvm; | |
506cfba9 | 3536 | } |
536a6f88 | 3537 | |
525df861 PB |
3538 | /* |
3539 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
3540 | * already set, with ->release() being kvm_vm_release(). In error | |
3541 | * cases it will be called by the final fput(file) and will take | |
3542 | * care of doing kvm_put_kvm(kvm). | |
3543 | */ | |
536a6f88 | 3544 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
3545 | put_unused_fd(r); |
3546 | fput(file); | |
536a6f88 JF |
3547 | return -ENOMEM; |
3548 | } | |
286de8f6 | 3549 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 3550 | |
506cfba9 | 3551 | fd_install(r, file); |
aac87636 | 3552 | return r; |
78588335 ME |
3553 | |
3554 | put_kvm: | |
3555 | kvm_put_kvm(kvm); | |
3556 | return r; | |
f17abe9a AK |
3557 | } |
3558 | ||
3559 | static long kvm_dev_ioctl(struct file *filp, | |
3560 | unsigned int ioctl, unsigned long arg) | |
3561 | { | |
07c45a36 | 3562 | long r = -EINVAL; |
f17abe9a AK |
3563 | |
3564 | switch (ioctl) { | |
3565 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
3566 | if (arg) |
3567 | goto out; | |
f17abe9a AK |
3568 | r = KVM_API_VERSION; |
3569 | break; | |
3570 | case KVM_CREATE_VM: | |
e08b9637 | 3571 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 3572 | break; |
018d00d2 | 3573 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 3574 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 3575 | break; |
07c45a36 | 3576 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
3577 | if (arg) |
3578 | goto out; | |
adb1ff46 AK |
3579 | r = PAGE_SIZE; /* struct kvm_run */ |
3580 | #ifdef CONFIG_X86 | |
3581 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 3582 | #endif |
4b4357e0 | 3583 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 3584 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 3585 | #endif |
07c45a36 | 3586 | break; |
d4c9ff2d FEL |
3587 | case KVM_TRACE_ENABLE: |
3588 | case KVM_TRACE_PAUSE: | |
3589 | case KVM_TRACE_DISABLE: | |
2023a29c | 3590 | r = -EOPNOTSUPP; |
d4c9ff2d | 3591 | break; |
6aa8b732 | 3592 | default: |
043405e1 | 3593 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
3594 | } |
3595 | out: | |
3596 | return r; | |
3597 | } | |
3598 | ||
6aa8b732 | 3599 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 3600 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 3601 | .llseek = noop_llseek, |
7ddfd3e0 | 3602 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
3603 | }; |
3604 | ||
3605 | static struct miscdevice kvm_dev = { | |
bbe4432e | 3606 | KVM_MINOR, |
6aa8b732 AK |
3607 | "kvm", |
3608 | &kvm_chardev_ops, | |
3609 | }; | |
3610 | ||
75b7127c | 3611 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
3612 | { |
3613 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 3614 | int r; |
1b6c0168 | 3615 | |
7f59f492 | 3616 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3617 | return; |
10474ae8 | 3618 | |
7f59f492 | 3619 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 3620 | |
13a34e06 | 3621 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
3622 | |
3623 | if (r) { | |
3624 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
3625 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 3626 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 3627 | } |
1b6c0168 AK |
3628 | } |
3629 | ||
8c18b2d2 | 3630 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 3631 | { |
4a937f96 | 3632 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3633 | if (kvm_usage_count) |
3634 | hardware_enable_nolock(NULL); | |
4a937f96 | 3635 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3636 | return 0; |
75b7127c TY |
3637 | } |
3638 | ||
3639 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
3640 | { |
3641 | int cpu = raw_smp_processor_id(); | |
3642 | ||
7f59f492 | 3643 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 3644 | return; |
7f59f492 | 3645 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 3646 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
3647 | } |
3648 | ||
8c18b2d2 | 3649 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 3650 | { |
4a937f96 | 3651 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
3652 | if (kvm_usage_count) |
3653 | hardware_disable_nolock(NULL); | |
4a937f96 | 3654 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 3655 | return 0; |
75b7127c TY |
3656 | } |
3657 | ||
10474ae8 AG |
3658 | static void hardware_disable_all_nolock(void) |
3659 | { | |
3660 | BUG_ON(!kvm_usage_count); | |
3661 | ||
3662 | kvm_usage_count--; | |
3663 | if (!kvm_usage_count) | |
75b7127c | 3664 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
3665 | } |
3666 | ||
3667 | static void hardware_disable_all(void) | |
3668 | { | |
4a937f96 | 3669 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 3670 | hardware_disable_all_nolock(); |
4a937f96 | 3671 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3672 | } |
3673 | ||
3674 | static int hardware_enable_all(void) | |
3675 | { | |
3676 | int r = 0; | |
3677 | ||
4a937f96 | 3678 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
3679 | |
3680 | kvm_usage_count++; | |
3681 | if (kvm_usage_count == 1) { | |
3682 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 3683 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
3684 | |
3685 | if (atomic_read(&hardware_enable_failed)) { | |
3686 | hardware_disable_all_nolock(); | |
3687 | r = -EBUSY; | |
3688 | } | |
3689 | } | |
3690 | ||
4a937f96 | 3691 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
3692 | |
3693 | return r; | |
3694 | } | |
3695 | ||
9a2b85c6 | 3696 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 3697 | void *v) |
9a2b85c6 | 3698 | { |
8e1c1815 SY |
3699 | /* |
3700 | * Some (well, at least mine) BIOSes hang on reboot if | |
3701 | * in vmx root mode. | |
3702 | * | |
3703 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
3704 | */ | |
1170adc6 | 3705 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 3706 | kvm_rebooting = true; |
75b7127c | 3707 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
3708 | return NOTIFY_OK; |
3709 | } | |
3710 | ||
3711 | static struct notifier_block kvm_reboot_notifier = { | |
3712 | .notifier_call = kvm_reboot, | |
3713 | .priority = 0, | |
3714 | }; | |
3715 | ||
e93f8a0f | 3716 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
3717 | { |
3718 | int i; | |
3719 | ||
3720 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 3721 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
3722 | |
3723 | kvm_iodevice_destructor(pos); | |
3724 | } | |
e93f8a0f | 3725 | kfree(bus); |
2eeb2e94 GH |
3726 | } |
3727 | ||
c21fbff1 | 3728 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 3729 | const struct kvm_io_range *r2) |
743eeb0b | 3730 | { |
8f4216c7 JW |
3731 | gpa_t addr1 = r1->addr; |
3732 | gpa_t addr2 = r2->addr; | |
3733 | ||
3734 | if (addr1 < addr2) | |
743eeb0b | 3735 | return -1; |
8f4216c7 JW |
3736 | |
3737 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
3738 | * accept any overlapping write. Any order is acceptable for | |
3739 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
3740 | * we process all of them. | |
3741 | */ | |
3742 | if (r2->len) { | |
3743 | addr1 += r1->len; | |
3744 | addr2 += r2->len; | |
3745 | } | |
3746 | ||
3747 | if (addr1 > addr2) | |
743eeb0b | 3748 | return 1; |
8f4216c7 | 3749 | |
743eeb0b SL |
3750 | return 0; |
3751 | } | |
3752 | ||
a343c9b7 PB |
3753 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
3754 | { | |
c21fbff1 | 3755 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
3756 | } |
3757 | ||
39369f7a | 3758 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
3759 | gpa_t addr, int len) |
3760 | { | |
3761 | struct kvm_io_range *range, key; | |
3762 | int off; | |
3763 | ||
3764 | key = (struct kvm_io_range) { | |
3765 | .addr = addr, | |
3766 | .len = len, | |
3767 | }; | |
3768 | ||
3769 | range = bsearch(&key, bus->range, bus->dev_count, | |
3770 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
3771 | if (range == NULL) | |
3772 | return -ENOENT; | |
3773 | ||
3774 | off = range - bus->range; | |
3775 | ||
c21fbff1 | 3776 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
3777 | off--; |
3778 | ||
3779 | return off; | |
3780 | } | |
3781 | ||
e32edf4f | 3782 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
3783 | struct kvm_io_range *range, const void *val) |
3784 | { | |
3785 | int idx; | |
3786 | ||
3787 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
3788 | if (idx < 0) | |
3789 | return -EOPNOTSUPP; | |
3790 | ||
3791 | while (idx < bus->dev_count && | |
c21fbff1 | 3792 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3793 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3794 | range->len, val)) |
3795 | return idx; | |
3796 | idx++; | |
3797 | } | |
3798 | ||
3799 | return -EOPNOTSUPP; | |
3800 | } | |
3801 | ||
bda9020e | 3802 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 3803 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 3804 | int len, const void *val) |
2eeb2e94 | 3805 | { |
90d83dc3 | 3806 | struct kvm_io_bus *bus; |
743eeb0b | 3807 | struct kvm_io_range range; |
126a5af5 | 3808 | int r; |
743eeb0b SL |
3809 | |
3810 | range = (struct kvm_io_range) { | |
3811 | .addr = addr, | |
3812 | .len = len, | |
3813 | }; | |
90d83dc3 | 3814 | |
e32edf4f | 3815 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
3816 | if (!bus) |
3817 | return -ENOMEM; | |
e32edf4f | 3818 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3819 | return r < 0 ? r : 0; |
3820 | } | |
a2420107 | 3821 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
3822 | |
3823 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
3824 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
3825 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
3826 | { |
3827 | struct kvm_io_bus *bus; | |
3828 | struct kvm_io_range range; | |
3829 | ||
3830 | range = (struct kvm_io_range) { | |
3831 | .addr = addr, | |
3832 | .len = len, | |
3833 | }; | |
3834 | ||
e32edf4f | 3835 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
3836 | if (!bus) |
3837 | return -ENOMEM; | |
126a5af5 CH |
3838 | |
3839 | /* First try the device referenced by cookie. */ | |
3840 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 3841 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 3842 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
3843 | val)) |
3844 | return cookie; | |
3845 | ||
3846 | /* | |
3847 | * cookie contained garbage; fall back to search and return the | |
3848 | * correct cookie value. | |
3849 | */ | |
e32edf4f | 3850 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
3851 | } |
3852 | ||
e32edf4f NN |
3853 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
3854 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
3855 | { |
3856 | int idx; | |
3857 | ||
3858 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
3859 | if (idx < 0) |
3860 | return -EOPNOTSUPP; | |
3861 | ||
3862 | while (idx < bus->dev_count && | |
c21fbff1 | 3863 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 3864 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
3865 | range->len, val)) |
3866 | return idx; | |
743eeb0b SL |
3867 | idx++; |
3868 | } | |
3869 | ||
bda9020e MT |
3870 | return -EOPNOTSUPP; |
3871 | } | |
2eeb2e94 | 3872 | |
bda9020e | 3873 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 3874 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 3875 | int len, void *val) |
bda9020e | 3876 | { |
90d83dc3 | 3877 | struct kvm_io_bus *bus; |
743eeb0b | 3878 | struct kvm_io_range range; |
126a5af5 | 3879 | int r; |
743eeb0b SL |
3880 | |
3881 | range = (struct kvm_io_range) { | |
3882 | .addr = addr, | |
3883 | .len = len, | |
3884 | }; | |
e93f8a0f | 3885 | |
e32edf4f | 3886 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
3887 | if (!bus) |
3888 | return -ENOMEM; | |
e32edf4f | 3889 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
3890 | return r < 0 ? r : 0; |
3891 | } | |
743eeb0b | 3892 | |
79fac95e | 3893 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
3894 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
3895 | int len, struct kvm_io_device *dev) | |
6c474694 | 3896 | { |
d4c67a7a | 3897 | int i; |
e93f8a0f | 3898 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 3899 | struct kvm_io_range range; |
090b7aff | 3900 | |
4a12f951 | 3901 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
3902 | if (!bus) |
3903 | return -ENOMEM; | |
3904 | ||
6ea34c9b AK |
3905 | /* exclude ioeventfd which is limited by maximum fd */ |
3906 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 3907 | return -ENOSPC; |
2eeb2e94 | 3908 | |
90952cd3 | 3909 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 3910 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
3911 | if (!new_bus) |
3912 | return -ENOMEM; | |
d4c67a7a GH |
3913 | |
3914 | range = (struct kvm_io_range) { | |
3915 | .addr = addr, | |
3916 | .len = len, | |
3917 | .dev = dev, | |
3918 | }; | |
3919 | ||
3920 | for (i = 0; i < bus->dev_count; i++) | |
3921 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
3922 | break; | |
3923 | ||
3924 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3925 | new_bus->dev_count++; | |
3926 | new_bus->range[i] = range; | |
3927 | memcpy(new_bus->range + i + 1, bus->range + i, | |
3928 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
3929 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3930 | synchronize_srcu_expedited(&kvm->srcu); | |
3931 | kfree(bus); | |
090b7aff GH |
3932 | |
3933 | return 0; | |
3934 | } | |
3935 | ||
79fac95e | 3936 | /* Caller must hold slots_lock. */ |
90db1043 DH |
3937 | void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3938 | struct kvm_io_device *dev) | |
090b7aff | 3939 | { |
90db1043 | 3940 | int i; |
e93f8a0f | 3941 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 3942 | |
4a12f951 | 3943 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 3944 | if (!bus) |
90db1043 | 3945 | return; |
df630b8c | 3946 | |
a1300716 AK |
3947 | for (i = 0; i < bus->dev_count; i++) |
3948 | if (bus->range[i].dev == dev) { | |
090b7aff GH |
3949 | break; |
3950 | } | |
e93f8a0f | 3951 | |
90db1043 DH |
3952 | if (i == bus->dev_count) |
3953 | return; | |
a1300716 | 3954 | |
90952cd3 | 3955 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 3956 | GFP_KERNEL_ACCOUNT); |
90db1043 DH |
3957 | if (!new_bus) { |
3958 | pr_err("kvm: failed to shrink bus, removing it completely\n"); | |
3959 | goto broken; | |
3960 | } | |
a1300716 AK |
3961 | |
3962 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
3963 | new_bus->dev_count--; | |
3964 | memcpy(new_bus->range + i, bus->range + i + 1, | |
3965 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f | 3966 | |
90db1043 | 3967 | broken: |
e93f8a0f MT |
3968 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
3969 | synchronize_srcu_expedited(&kvm->srcu); | |
3970 | kfree(bus); | |
90db1043 | 3971 | return; |
2eeb2e94 GH |
3972 | } |
3973 | ||
8a39d006 AP |
3974 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
3975 | gpa_t addr) | |
3976 | { | |
3977 | struct kvm_io_bus *bus; | |
3978 | int dev_idx, srcu_idx; | |
3979 | struct kvm_io_device *iodev = NULL; | |
3980 | ||
3981 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
3982 | ||
3983 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
3984 | if (!bus) |
3985 | goto out_unlock; | |
8a39d006 AP |
3986 | |
3987 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
3988 | if (dev_idx < 0) | |
3989 | goto out_unlock; | |
3990 | ||
3991 | iodev = bus->range[dev_idx].dev; | |
3992 | ||
3993 | out_unlock: | |
3994 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
3995 | ||
3996 | return iodev; | |
3997 | } | |
3998 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
3999 | ||
536a6f88 JF |
4000 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
4001 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
4002 | const char *fmt) | |
4003 | { | |
4004 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
4005 | inode->i_private; | |
4006 | ||
4007 | /* The debugfs files are a reference to the kvm struct which | |
4008 | * is still valid when kvm_destroy_vm is called. | |
4009 | * To avoid the race between open and the removal of the debugfs | |
4010 | * directory we test against the users count. | |
4011 | */ | |
e3736c3e | 4012 | if (!refcount_inc_not_zero(&stat_data->kvm->users_count)) |
536a6f88 JF |
4013 | return -ENOENT; |
4014 | ||
833b45de PB |
4015 | if (simple_attr_open(inode, file, get, |
4016 | stat_data->mode & S_IWUGO ? set : NULL, | |
4017 | fmt)) { | |
536a6f88 JF |
4018 | kvm_put_kvm(stat_data->kvm); |
4019 | return -ENOMEM; | |
4020 | } | |
4021 | ||
4022 | return 0; | |
4023 | } | |
4024 | ||
4025 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
4026 | { | |
4027 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
4028 | inode->i_private; | |
4029 | ||
4030 | simple_attr_release(inode, file); | |
4031 | kvm_put_kvm(stat_data->kvm); | |
4032 | ||
4033 | return 0; | |
4034 | } | |
4035 | ||
4036 | static int vm_stat_get_per_vm(void *data, u64 *val) | |
4037 | { | |
4038 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
4039 | ||
8a7e75d4 | 4040 | *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset); |
536a6f88 JF |
4041 | |
4042 | return 0; | |
4043 | } | |
4044 | ||
ce35ef27 SJS |
4045 | static int vm_stat_clear_per_vm(void *data, u64 val) |
4046 | { | |
4047 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
4048 | ||
4049 | if (val) | |
4050 | return -EINVAL; | |
4051 | ||
4052 | *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0; | |
4053 | ||
4054 | return 0; | |
4055 | } | |
4056 | ||
536a6f88 JF |
4057 | static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file) |
4058 | { | |
4059 | __simple_attr_check_format("%llu\n", 0ull); | |
4060 | return kvm_debugfs_open(inode, file, vm_stat_get_per_vm, | |
ce35ef27 | 4061 | vm_stat_clear_per_vm, "%llu\n"); |
536a6f88 JF |
4062 | } |
4063 | ||
4064 | static const struct file_operations vm_stat_get_per_vm_fops = { | |
4065 | .owner = THIS_MODULE, | |
4066 | .open = vm_stat_get_per_vm_open, | |
4067 | .release = kvm_debugfs_release, | |
4068 | .read = simple_attr_read, | |
4069 | .write = simple_attr_write, | |
3bed8888 | 4070 | .llseek = no_llseek, |
536a6f88 JF |
4071 | }; |
4072 | ||
4073 | static int vcpu_stat_get_per_vm(void *data, u64 *val) | |
4074 | { | |
4075 | int i; | |
4076 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
4077 | struct kvm_vcpu *vcpu; | |
4078 | ||
4079 | *val = 0; | |
4080 | ||
4081 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) | |
8a7e75d4 | 4082 | *val += *(u64 *)((void *)vcpu + stat_data->offset); |
536a6f88 JF |
4083 | |
4084 | return 0; | |
4085 | } | |
4086 | ||
ce35ef27 SJS |
4087 | static int vcpu_stat_clear_per_vm(void *data, u64 val) |
4088 | { | |
4089 | int i; | |
4090 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; | |
4091 | struct kvm_vcpu *vcpu; | |
4092 | ||
4093 | if (val) | |
4094 | return -EINVAL; | |
4095 | ||
4096 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) | |
4097 | *(u64 *)((void *)vcpu + stat_data->offset) = 0; | |
4098 | ||
4099 | return 0; | |
4100 | } | |
4101 | ||
536a6f88 JF |
4102 | static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file) |
4103 | { | |
4104 | __simple_attr_check_format("%llu\n", 0ull); | |
4105 | return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm, | |
ce35ef27 | 4106 | vcpu_stat_clear_per_vm, "%llu\n"); |
536a6f88 JF |
4107 | } |
4108 | ||
4109 | static const struct file_operations vcpu_stat_get_per_vm_fops = { | |
4110 | .owner = THIS_MODULE, | |
4111 | .open = vcpu_stat_get_per_vm_open, | |
4112 | .release = kvm_debugfs_release, | |
4113 | .read = simple_attr_read, | |
4114 | .write = simple_attr_write, | |
3bed8888 | 4115 | .llseek = no_llseek, |
536a6f88 JF |
4116 | }; |
4117 | ||
4118 | static const struct file_operations *stat_fops_per_vm[] = { | |
4119 | [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops, | |
4120 | [KVM_STAT_VM] = &vm_stat_get_per_vm_fops, | |
4121 | }; | |
4122 | ||
8b88b099 | 4123 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
4124 | { |
4125 | unsigned offset = (long)_offset; | |
ba1389b7 | 4126 | struct kvm *kvm; |
536a6f88 JF |
4127 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
4128 | u64 tmp_val; | |
ba1389b7 | 4129 | |
8b88b099 | 4130 | *val = 0; |
0d9ce162 | 4131 | mutex_lock(&kvm_lock); |
536a6f88 JF |
4132 | list_for_each_entry(kvm, &vm_list, vm_list) { |
4133 | stat_tmp.kvm = kvm; | |
4134 | vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val); | |
4135 | *val += tmp_val; | |
4136 | } | |
0d9ce162 | 4137 | mutex_unlock(&kvm_lock); |
8b88b099 | 4138 | return 0; |
ba1389b7 AK |
4139 | } |
4140 | ||
ce35ef27 SJS |
4141 | static int vm_stat_clear(void *_offset, u64 val) |
4142 | { | |
4143 | unsigned offset = (long)_offset; | |
4144 | struct kvm *kvm; | |
4145 | struct kvm_stat_data stat_tmp = {.offset = offset}; | |
4146 | ||
4147 | if (val) | |
4148 | return -EINVAL; | |
4149 | ||
0d9ce162 | 4150 | mutex_lock(&kvm_lock); |
ce35ef27 SJS |
4151 | list_for_each_entry(kvm, &vm_list, vm_list) { |
4152 | stat_tmp.kvm = kvm; | |
4153 | vm_stat_clear_per_vm((void *)&stat_tmp, 0); | |
4154 | } | |
0d9ce162 | 4155 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
4156 | |
4157 | return 0; | |
4158 | } | |
4159 | ||
4160 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
ba1389b7 | 4161 | |
8b88b099 | 4162 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
4163 | { |
4164 | unsigned offset = (long)_offset; | |
1165f5fe | 4165 | struct kvm *kvm; |
536a6f88 JF |
4166 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
4167 | u64 tmp_val; | |
1165f5fe | 4168 | |
8b88b099 | 4169 | *val = 0; |
0d9ce162 | 4170 | mutex_lock(&kvm_lock); |
536a6f88 JF |
4171 | list_for_each_entry(kvm, &vm_list, vm_list) { |
4172 | stat_tmp.kvm = kvm; | |
4173 | vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val); | |
4174 | *val += tmp_val; | |
4175 | } | |
0d9ce162 | 4176 | mutex_unlock(&kvm_lock); |
8b88b099 | 4177 | return 0; |
1165f5fe AK |
4178 | } |
4179 | ||
ce35ef27 SJS |
4180 | static int vcpu_stat_clear(void *_offset, u64 val) |
4181 | { | |
4182 | unsigned offset = (long)_offset; | |
4183 | struct kvm *kvm; | |
4184 | struct kvm_stat_data stat_tmp = {.offset = offset}; | |
4185 | ||
4186 | if (val) | |
4187 | return -EINVAL; | |
4188 | ||
0d9ce162 | 4189 | mutex_lock(&kvm_lock); |
ce35ef27 SJS |
4190 | list_for_each_entry(kvm, &vm_list, vm_list) { |
4191 | stat_tmp.kvm = kvm; | |
4192 | vcpu_stat_clear_per_vm((void *)&stat_tmp, 0); | |
4193 | } | |
0d9ce162 | 4194 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
4195 | |
4196 | return 0; | |
4197 | } | |
4198 | ||
4199 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
4200 | "%llu\n"); | |
ba1389b7 | 4201 | |
828c0950 | 4202 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
4203 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
4204 | [KVM_STAT_VM] = &vm_stat_fops, | |
4205 | }; | |
1165f5fe | 4206 | |
286de8f6 CI |
4207 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
4208 | { | |
4209 | struct kobj_uevent_env *env; | |
286de8f6 CI |
4210 | unsigned long long created, active; |
4211 | ||
4212 | if (!kvm_dev.this_device || !kvm) | |
4213 | return; | |
4214 | ||
0d9ce162 | 4215 | mutex_lock(&kvm_lock); |
286de8f6 CI |
4216 | if (type == KVM_EVENT_CREATE_VM) { |
4217 | kvm_createvm_count++; | |
4218 | kvm_active_vms++; | |
4219 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
4220 | kvm_active_vms--; | |
4221 | } | |
4222 | created = kvm_createvm_count; | |
4223 | active = kvm_active_vms; | |
0d9ce162 | 4224 | mutex_unlock(&kvm_lock); |
286de8f6 | 4225 | |
b12ce36a | 4226 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
4227 | if (!env) |
4228 | return; | |
4229 | ||
4230 | add_uevent_var(env, "CREATED=%llu", created); | |
4231 | add_uevent_var(env, "COUNT=%llu", active); | |
4232 | ||
fdeaf7e3 | 4233 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 4234 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
4235 | kvm->userspace_pid = task_pid_nr(current); |
4236 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 4237 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
4238 | } |
4239 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 4240 | |
8ed0579c | 4241 | if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) { |
b12ce36a | 4242 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
4243 | |
4244 | if (p) { | |
4245 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
4246 | if (!IS_ERR(tmp)) | |
4247 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
4248 | kfree(p); | |
286de8f6 CI |
4249 | } |
4250 | } | |
4251 | /* no need for checks, since we are adding at most only 5 keys */ | |
4252 | env->envp[env->envp_idx++] = NULL; | |
4253 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
4254 | kfree(env); | |
286de8f6 CI |
4255 | } |
4256 | ||
929f45e3 | 4257 | static void kvm_init_debug(void) |
6aa8b732 AK |
4258 | { |
4259 | struct kvm_stats_debugfs_item *p; | |
4260 | ||
76f7c879 | 4261 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 4262 | |
536a6f88 JF |
4263 | kvm_debugfs_num_entries = 0; |
4264 | for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) { | |
833b45de PB |
4265 | int mode = p->mode ? p->mode : 0644; |
4266 | debugfs_create_file(p->name, mode, kvm_debugfs_dir, | |
929f45e3 GKH |
4267 | (void *)(long)p->offset, |
4268 | stat_fops[p->kind]); | |
4f69b680 | 4269 | } |
6aa8b732 AK |
4270 | } |
4271 | ||
fb3600cc | 4272 | static int kvm_suspend(void) |
59ae6c6b | 4273 | { |
10474ae8 | 4274 | if (kvm_usage_count) |
75b7127c | 4275 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
4276 | return 0; |
4277 | } | |
4278 | ||
fb3600cc | 4279 | static void kvm_resume(void) |
59ae6c6b | 4280 | { |
ca84d1a2 | 4281 | if (kvm_usage_count) { |
2eb06c30 WL |
4282 | #ifdef CONFIG_LOCKDEP |
4283 | WARN_ON(lockdep_is_held(&kvm_count_lock)); | |
4284 | #endif | |
75b7127c | 4285 | hardware_enable_nolock(NULL); |
ca84d1a2 | 4286 | } |
59ae6c6b AK |
4287 | } |
4288 | ||
fb3600cc | 4289 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
4290 | .suspend = kvm_suspend, |
4291 | .resume = kvm_resume, | |
4292 | }; | |
4293 | ||
15ad7146 AK |
4294 | static inline |
4295 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
4296 | { | |
4297 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
4298 | } | |
4299 | ||
4300 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
4301 | { | |
4302 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 4303 | |
046ddeed | 4304 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 4305 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 4306 | |
e790d9ef RK |
4307 | kvm_arch_sched_in(vcpu, cpu); |
4308 | ||
e9b11c17 | 4309 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
4310 | } |
4311 | ||
4312 | static void kvm_sched_out(struct preempt_notifier *pn, | |
4313 | struct task_struct *next) | |
4314 | { | |
4315 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
4316 | ||
d73eb57b | 4317 | if (current->state == TASK_RUNNING) { |
046ddeed | 4318 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
4319 | WRITE_ONCE(vcpu->ready, true); |
4320 | } | |
e9b11c17 | 4321 | kvm_arch_vcpu_put(vcpu); |
15ad7146 AK |
4322 | } |
4323 | ||
f257d6dc SC |
4324 | static void check_processor_compat(void *rtn) |
4325 | { | |
4326 | *(int *)rtn = kvm_arch_check_processor_compat(); | |
4327 | } | |
4328 | ||
0ee75bea | 4329 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 4330 | struct module *module) |
6aa8b732 AK |
4331 | { |
4332 | int r; | |
002c7f7c | 4333 | int cpu; |
6aa8b732 | 4334 | |
f8c16bba ZX |
4335 | r = kvm_arch_init(opaque); |
4336 | if (r) | |
d2308784 | 4337 | goto out_fail; |
cb498ea2 | 4338 | |
7dac16c3 AH |
4339 | /* |
4340 | * kvm_arch_init makes sure there's at most one caller | |
4341 | * for architectures that support multiple implementations, | |
4342 | * like intel and amd on x86. | |
36343f6e PB |
4343 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
4344 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 4345 | */ |
36343f6e PB |
4346 | r = kvm_irqfd_init(); |
4347 | if (r) | |
4348 | goto out_irqfd; | |
7dac16c3 | 4349 | |
8437a617 | 4350 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
4351 | r = -ENOMEM; |
4352 | goto out_free_0; | |
4353 | } | |
4354 | ||
e9b11c17 | 4355 | r = kvm_arch_hardware_setup(); |
6aa8b732 | 4356 | if (r < 0) |
faf0be22 | 4357 | goto out_free_1; |
6aa8b732 | 4358 | |
002c7f7c | 4359 | for_each_online_cpu(cpu) { |
f257d6dc | 4360 | smp_call_function_single(cpu, check_processor_compat, &r, 1); |
002c7f7c | 4361 | if (r < 0) |
faf0be22 | 4362 | goto out_free_2; |
002c7f7c YS |
4363 | } |
4364 | ||
73c1b41e | 4365 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 4366 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 4367 | if (r) |
d2308784 | 4368 | goto out_free_2; |
6aa8b732 AK |
4369 | register_reboot_notifier(&kvm_reboot_notifier); |
4370 | ||
c16f862d | 4371 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
4372 | if (!vcpu_align) |
4373 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
4374 | kvm_vcpu_cache = |
4375 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
4376 | SLAB_ACCOUNT, | |
4377 | offsetof(struct kvm_vcpu, arch), | |
4378 | sizeof_field(struct kvm_vcpu, arch), | |
4379 | NULL); | |
c16f862d RR |
4380 | if (!kvm_vcpu_cache) { |
4381 | r = -ENOMEM; | |
fb3600cc | 4382 | goto out_free_3; |
c16f862d RR |
4383 | } |
4384 | ||
af585b92 GN |
4385 | r = kvm_async_pf_init(); |
4386 | if (r) | |
4387 | goto out_free; | |
4388 | ||
6aa8b732 | 4389 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
4390 | kvm_vm_fops.owner = module; |
4391 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
4392 | |
4393 | r = misc_register(&kvm_dev); | |
4394 | if (r) { | |
1170adc6 | 4395 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 4396 | goto out_unreg; |
6aa8b732 AK |
4397 | } |
4398 | ||
fb3600cc RW |
4399 | register_syscore_ops(&kvm_syscore_ops); |
4400 | ||
15ad7146 AK |
4401 | kvm_preempt_ops.sched_in = kvm_sched_in; |
4402 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
4403 | ||
929f45e3 | 4404 | kvm_init_debug(); |
0ea4ed8e | 4405 | |
3c3c29fd PB |
4406 | r = kvm_vfio_ops_init(); |
4407 | WARN_ON(r); | |
4408 | ||
c7addb90 | 4409 | return 0; |
6aa8b732 | 4410 | |
af585b92 GN |
4411 | out_unreg: |
4412 | kvm_async_pf_deinit(); | |
6aa8b732 | 4413 | out_free: |
c16f862d | 4414 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 4415 | out_free_3: |
6aa8b732 | 4416 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4417 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 4418 | out_free_2: |
e9b11c17 | 4419 | kvm_arch_hardware_unsetup(); |
faf0be22 | 4420 | out_free_1: |
7f59f492 | 4421 | free_cpumask_var(cpus_hardware_enabled); |
d2308784 | 4422 | out_free_0: |
a0f155e9 | 4423 | kvm_irqfd_exit(); |
36343f6e | 4424 | out_irqfd: |
7dac16c3 AH |
4425 | kvm_arch_exit(); |
4426 | out_fail: | |
6aa8b732 AK |
4427 | return r; |
4428 | } | |
cb498ea2 | 4429 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 4430 | |
cb498ea2 | 4431 | void kvm_exit(void) |
6aa8b732 | 4432 | { |
4bd33b56 | 4433 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 4434 | misc_deregister(&kvm_dev); |
c16f862d | 4435 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 4436 | kvm_async_pf_deinit(); |
fb3600cc | 4437 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 4438 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4439 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 4440 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 4441 | kvm_arch_hardware_unsetup(); |
f8c16bba | 4442 | kvm_arch_exit(); |
a0f155e9 | 4443 | kvm_irqfd_exit(); |
7f59f492 | 4444 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 4445 | kvm_vfio_ops_exit(); |
6aa8b732 | 4446 | } |
cb498ea2 | 4447 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
4448 | |
4449 | struct kvm_vm_worker_thread_context { | |
4450 | struct kvm *kvm; | |
4451 | struct task_struct *parent; | |
4452 | struct completion init_done; | |
4453 | kvm_vm_thread_fn_t thread_fn; | |
4454 | uintptr_t data; | |
4455 | int err; | |
4456 | }; | |
4457 | ||
4458 | static int kvm_vm_worker_thread(void *context) | |
4459 | { | |
4460 | /* | |
4461 | * The init_context is allocated on the stack of the parent thread, so | |
4462 | * we have to locally copy anything that is needed beyond initialization | |
4463 | */ | |
4464 | struct kvm_vm_worker_thread_context *init_context = context; | |
4465 | struct kvm *kvm = init_context->kvm; | |
4466 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
4467 | uintptr_t data = init_context->data; | |
4468 | int err; | |
4469 | ||
4470 | err = kthread_park(current); | |
4471 | /* kthread_park(current) is never supposed to return an error */ | |
4472 | WARN_ON(err != 0); | |
4473 | if (err) | |
4474 | goto init_complete; | |
4475 | ||
4476 | err = cgroup_attach_task_all(init_context->parent, current); | |
4477 | if (err) { | |
4478 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
4479 | __func__, err); | |
4480 | goto init_complete; | |
4481 | } | |
4482 | ||
4483 | set_user_nice(current, task_nice(init_context->parent)); | |
4484 | ||
4485 | init_complete: | |
4486 | init_context->err = err; | |
4487 | complete(&init_context->init_done); | |
4488 | init_context = NULL; | |
4489 | ||
4490 | if (err) | |
4491 | return err; | |
4492 | ||
4493 | /* Wait to be woken up by the spawner before proceeding. */ | |
4494 | kthread_parkme(); | |
4495 | ||
4496 | if (!kthread_should_stop()) | |
4497 | err = thread_fn(kvm, data); | |
4498 | ||
4499 | return err; | |
4500 | } | |
4501 | ||
4502 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
4503 | uintptr_t data, const char *name, | |
4504 | struct task_struct **thread_ptr) | |
4505 | { | |
4506 | struct kvm_vm_worker_thread_context init_context = {}; | |
4507 | struct task_struct *thread; | |
4508 | ||
4509 | *thread_ptr = NULL; | |
4510 | init_context.kvm = kvm; | |
4511 | init_context.parent = current; | |
4512 | init_context.thread_fn = thread_fn; | |
4513 | init_context.data = data; | |
4514 | init_completion(&init_context.init_done); | |
4515 | ||
4516 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
4517 | "%s-%d", name, task_pid_nr(current)); | |
4518 | if (IS_ERR(thread)) | |
4519 | return PTR_ERR(thread); | |
4520 | ||
4521 | /* kthread_run is never supposed to return NULL */ | |
4522 | WARN_ON(thread == NULL); | |
4523 | ||
4524 | wait_for_completion(&init_context.init_done); | |
4525 | ||
4526 | if (!init_context.err) | |
4527 | *thread_ptr = thread; | |
4528 | ||
4529 | return init_context.err; | |
4530 | } |