Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
6aa8b732 AK |
2 | /* |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * This module enables machines with Intel VT-x extensions to run virtual | |
6 | * machines without emulation or binary translation. | |
7 | * | |
8 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 9 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
10 | * |
11 | * Authors: | |
12 | * Avi Kivity <avi@qumranet.com> | |
13 | * Yaniv Kamay <yaniv@qumranet.com> | |
6aa8b732 AK |
14 | */ |
15 | ||
af669ac6 | 16 | #include <kvm/iodev.h> |
6aa8b732 | 17 | |
edf88417 | 18 | #include <linux/kvm_host.h> |
6aa8b732 AK |
19 | #include <linux/kvm.h> |
20 | #include <linux/module.h> | |
21 | #include <linux/errno.h> | |
6aa8b732 | 22 | #include <linux/percpu.h> |
6aa8b732 AK |
23 | #include <linux/mm.h> |
24 | #include <linux/miscdevice.h> | |
25 | #include <linux/vmalloc.h> | |
6aa8b732 | 26 | #include <linux/reboot.h> |
6aa8b732 AK |
27 | #include <linux/debugfs.h> |
28 | #include <linux/highmem.h> | |
29 | #include <linux/file.h> | |
fb3600cc | 30 | #include <linux/syscore_ops.h> |
774c47f1 | 31 | #include <linux/cpu.h> |
174cd4b1 | 32 | #include <linux/sched/signal.h> |
6e84f315 | 33 | #include <linux/sched/mm.h> |
03441a34 | 34 | #include <linux/sched/stat.h> |
d9e368d6 AK |
35 | #include <linux/cpumask.h> |
36 | #include <linux/smp.h> | |
d6d28168 | 37 | #include <linux/anon_inodes.h> |
04d2cc77 | 38 | #include <linux/profile.h> |
7aa81cc0 | 39 | #include <linux/kvm_para.h> |
6fc138d2 | 40 | #include <linux/pagemap.h> |
8d4e1288 | 41 | #include <linux/mman.h> |
35149e21 | 42 | #include <linux/swap.h> |
e56d532f | 43 | #include <linux/bitops.h> |
547de29e | 44 | #include <linux/spinlock.h> |
6ff5894c | 45 | #include <linux/compat.h> |
bc6678a3 | 46 | #include <linux/srcu.h> |
8f0b1ab6 | 47 | #include <linux/hugetlb.h> |
5a0e3ad6 | 48 | #include <linux/slab.h> |
743eeb0b SL |
49 | #include <linux/sort.h> |
50 | #include <linux/bsearch.h> | |
c011d23b | 51 | #include <linux/io.h> |
2eb06c30 | 52 | #include <linux/lockdep.h> |
c57c8046 | 53 | #include <linux/kthread.h> |
2fdef3a2 | 54 | #include <linux/suspend.h> |
6aa8b732 | 55 | |
e495606d | 56 | #include <asm/processor.h> |
2ea75be3 | 57 | #include <asm/ioctl.h> |
7c0f6ba6 | 58 | #include <linux/uaccess.h> |
6aa8b732 | 59 | |
5f94c174 | 60 | #include "coalesced_mmio.h" |
af585b92 | 61 | #include "async_pf.h" |
982ed0de | 62 | #include "kvm_mm.h" |
3c3c29fd | 63 | #include "vfio.h" |
5f94c174 | 64 | |
4c8c3c7f VS |
65 | #include <trace/events/ipi.h> |
66 | ||
229456fc MT |
67 | #define CREATE_TRACE_POINTS |
68 | #include <trace/events/kvm.h> | |
69 | ||
fb04a1ed PX |
70 | #include <linux/kvm_dirty_ring.h> |
71 | ||
4c8c3c7f | 72 | |
536a6f88 JF |
73 | /* Worst case buffer size needed for holding an integer. */ |
74 | #define ITOA_MAX_LEN 12 | |
75 | ||
6aa8b732 AK |
76 | MODULE_AUTHOR("Qumranet"); |
77 | MODULE_LICENSE("GPL"); | |
78 | ||
920552b2 | 79 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 80 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
039c5d1b | 81 | module_param(halt_poll_ns, uint, 0644); |
ec76d819 | 82 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 83 | |
aca6ff29 | 84 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 85 | unsigned int halt_poll_ns_grow = 2; |
039c5d1b | 86 | module_param(halt_poll_ns_grow, uint, 0644); |
ec76d819 | 87 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 | 88 | |
49113d36 NW |
89 | /* The start value to grow halt_poll_ns from */ |
90 | unsigned int halt_poll_ns_grow_start = 10000; /* 10us */ | |
91 | module_param(halt_poll_ns_grow_start, uint, 0644); | |
92 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start); | |
93 | ||
aca6ff29 | 94 | /* Default resets per-vcpu halt_poll_ns . */ |
ec76d819 | 95 | unsigned int halt_poll_ns_shrink; |
039c5d1b | 96 | module_param(halt_poll_ns_shrink, uint, 0644); |
ec76d819 | 97 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 98 | |
fa40a821 MT |
99 | /* |
100 | * Ordering of locks: | |
101 | * | |
b7d409de | 102 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
103 | */ |
104 | ||
0d9ce162 | 105 | DEFINE_MUTEX(kvm_lock); |
e9b11c17 | 106 | LIST_HEAD(vm_list); |
133de902 | 107 | |
aaba298c | 108 | static struct kmem_cache *kvm_vcpu_cache; |
1165f5fe | 109 | |
15ad7146 | 110 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
7495e22b | 111 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu); |
15ad7146 | 112 | |
76f7c879 | 113 | struct dentry *kvm_debugfs_dir; |
e23a808b | 114 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 115 | |
09cbcef6 | 116 | static const struct file_operations stat_fops_per_vm; |
536a6f88 | 117 | |
5f6de5cb DM |
118 | static struct file_operations kvm_chardev_ops; |
119 | ||
bccf2150 AK |
120 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
121 | unsigned long arg); | |
de8e5d74 | 122 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
123 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
124 | unsigned long arg); | |
7ddfd3e0 MZ |
125 | #define KVM_COMPAT(c) .compat_ioctl = (c) |
126 | #else | |
9cb09e7c MZ |
127 | /* |
128 | * For architectures that don't implement a compat infrastructure, | |
129 | * adopt a double line of defense: | |
130 | * - Prevent a compat task from opening /dev/kvm | |
131 | * - If the open has been done by a 64bit task, and the KVM fd | |
132 | * passed to a compat task, let the ioctls fail. | |
133 | */ | |
7ddfd3e0 MZ |
134 | static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl, |
135 | unsigned long arg) { return -EINVAL; } | |
b9876e6d MZ |
136 | |
137 | static int kvm_no_compat_open(struct inode *inode, struct file *file) | |
138 | { | |
139 | return is_compat_task() ? -ENODEV : 0; | |
140 | } | |
141 | #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \ | |
142 | .open = kvm_no_compat_open | |
1dda606c | 143 | #endif |
10474ae8 AG |
144 | static int hardware_enable_all(void); |
145 | static void hardware_disable_all(void); | |
bccf2150 | 146 | |
e93f8a0f | 147 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 148 | |
286de8f6 CI |
149 | #define KVM_EVENT_CREATE_VM 0 |
150 | #define KVM_EVENT_DESTROY_VM 1 | |
151 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); | |
152 | static unsigned long long kvm_createvm_count; | |
153 | static unsigned long long kvm_active_vms; | |
154 | ||
baff59cc VK |
155 | static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask); |
156 | ||
e649b3f0 ET |
157 | __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
158 | unsigned long start, unsigned long end) | |
b1394e74 RK |
159 | { |
160 | } | |
161 | ||
683412cc MZ |
162 | __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm) |
163 | { | |
164 | } | |
165 | ||
284dc493 | 166 | bool kvm_is_zone_device_page(struct page *page) |
a78986aa SC |
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 | */ | |
284dc493 | 174 | if (WARN_ON_ONCE(!page_count(page))) |
a78986aa SC |
175 | return false; |
176 | ||
284dc493 | 177 | return is_zone_device_page(page); |
a78986aa SC |
178 | } |
179 | ||
b14b2690 SC |
180 | /* |
181 | * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted | |
182 | * page, NULL otherwise. Note, the list of refcounted PG_reserved page types | |
183 | * is likely incomplete, it has been compiled purely through people wanting to | |
184 | * back guest with a certain type of memory and encountering issues. | |
185 | */ | |
186 | struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn) | |
cbff90a7 | 187 | { |
b14b2690 SC |
188 | struct page *page; |
189 | ||
190 | if (!pfn_valid(pfn)) | |
191 | return NULL; | |
192 | ||
193 | page = pfn_to_page(pfn); | |
194 | if (!PageReserved(page)) | |
195 | return page; | |
196 | ||
197 | /* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */ | |
198 | if (is_zero_pfn(pfn)) | |
199 | return page; | |
200 | ||
a78986aa SC |
201 | /* |
202 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting | |
203 | * perspective they are "normal" pages, albeit with slightly different | |
204 | * usage rules. | |
205 | */ | |
b14b2690 SC |
206 | if (kvm_is_zone_device_page(page)) |
207 | return page; | |
cbff90a7 | 208 | |
b14b2690 | 209 | return NULL; |
cbff90a7 BAY |
210 | } |
211 | ||
bccf2150 AK |
212 | /* |
213 | * Switches to specified vcpu, until a matching vcpu_put() | |
214 | */ | |
ec7660cc | 215 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 216 | { |
ec7660cc | 217 | int cpu = get_cpu(); |
7495e22b PB |
218 | |
219 | __this_cpu_write(kvm_running_vcpu, vcpu); | |
15ad7146 | 220 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 221 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 222 | put_cpu(); |
6aa8b732 | 223 | } |
2f1fe811 | 224 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 225 | |
313a3dc7 | 226 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 227 | { |
15ad7146 | 228 | preempt_disable(); |
313a3dc7 | 229 | kvm_arch_vcpu_put(vcpu); |
15ad7146 | 230 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
7495e22b | 231 | __this_cpu_write(kvm_running_vcpu, NULL); |
15ad7146 | 232 | preempt_enable(); |
6aa8b732 | 233 | } |
2f1fe811 | 234 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 235 | |
7a97cec2 PB |
236 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
237 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
238 | { | |
239 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
240 | ||
241 | /* | |
242 | * We need to wait for the VCPU to reenable interrupts and get out of | |
243 | * READING_SHADOW_PAGE_TABLES mode. | |
244 | */ | |
245 | if (req & KVM_REQUEST_WAIT) | |
246 | return mode != OUTSIDE_GUEST_MODE; | |
247 | ||
248 | /* | |
249 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
250 | */ | |
251 | return mode == IN_GUEST_MODE; | |
252 | } | |
253 | ||
f24b44e4 | 254 | static void ack_kick(void *_completed) |
d9e368d6 | 255 | { |
d9e368d6 AK |
256 | } |
257 | ||
620b2438 | 258 | static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait) |
b49defe8 | 259 | { |
b49defe8 PB |
260 | if (cpumask_empty(cpus)) |
261 | return false; | |
262 | ||
f24b44e4 | 263 | smp_call_function_many(cpus, ack_kick, NULL, wait); |
b49defe8 PB |
264 | return true; |
265 | } | |
266 | ||
b56bd8e0 JL |
267 | static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req, |
268 | struct cpumask *tmp, int current_cpu) | |
ae0946cd VK |
269 | { |
270 | int cpu; | |
271 | ||
df06dae3 SC |
272 | if (likely(!(req & KVM_REQUEST_NO_ACTION))) |
273 | __kvm_make_request(req, vcpu); | |
ae0946cd VK |
274 | |
275 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) | |
276 | return; | |
277 | ||
ae0946cd VK |
278 | /* |
279 | * Note, the vCPU could get migrated to a different pCPU at any point | |
280 | * after kvm_request_needs_ipi(), which could result in sending an IPI | |
281 | * to the previous pCPU. But, that's OK because the purpose of the IPI | |
282 | * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is | |
283 | * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES | |
284 | * after this point is also OK, as the requirement is only that KVM wait | |
285 | * for vCPUs that were reading SPTEs _before_ any changes were | |
286 | * finalized. See kvm_vcpu_kick() for more details on handling requests. | |
287 | */ | |
288 | if (kvm_request_needs_ipi(vcpu, req)) { | |
289 | cpu = READ_ONCE(vcpu->cpu); | |
290 | if (cpu != -1 && cpu != current_cpu) | |
291 | __cpumask_set_cpu(cpu, tmp); | |
292 | } | |
293 | } | |
294 | ||
7053df4e | 295 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
620b2438 | 296 | unsigned long *vcpu_bitmap) |
d9e368d6 | 297 | { |
d9e368d6 | 298 | struct kvm_vcpu *vcpu; |
620b2438 | 299 | struct cpumask *cpus; |
ae0946cd | 300 | int i, me; |
7053df4e | 301 | bool called; |
6ef7a1bc | 302 | |
3cba4130 | 303 | me = get_cpu(); |
7053df4e | 304 | |
620b2438 VK |
305 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
306 | cpumask_clear(cpus); | |
307 | ||
ae0946cd VK |
308 | for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) { |
309 | vcpu = kvm_get_vcpu(kvm, i); | |
381cecc5 | 310 | if (!vcpu) |
7053df4e | 311 | continue; |
b56bd8e0 | 312 | kvm_make_vcpu_request(vcpu, req, cpus, me); |
49846896 | 313 | } |
7053df4e | 314 | |
620b2438 | 315 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); |
3cba4130 | 316 | put_cpu(); |
7053df4e VK |
317 | |
318 | return called; | |
319 | } | |
320 | ||
54163a34 SS |
321 | bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req, |
322 | struct kvm_vcpu *except) | |
7053df4e | 323 | { |
ae0946cd | 324 | struct kvm_vcpu *vcpu; |
baff59cc | 325 | struct cpumask *cpus; |
46808a4c | 326 | unsigned long i; |
7053df4e | 327 | bool called; |
46808a4c | 328 | int me; |
7053df4e | 329 | |
ae0946cd VK |
330 | me = get_cpu(); |
331 | ||
baff59cc VK |
332 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
333 | cpumask_clear(cpus); | |
334 | ||
ae0946cd VK |
335 | kvm_for_each_vcpu(i, vcpu, kvm) { |
336 | if (vcpu == except) | |
337 | continue; | |
b56bd8e0 | 338 | kvm_make_vcpu_request(vcpu, req, cpus, me); |
ae0946cd VK |
339 | } |
340 | ||
341 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); | |
342 | put_cpu(); | |
7053df4e | 343 | |
49846896 | 344 | return called; |
d9e368d6 AK |
345 | } |
346 | ||
54163a34 SS |
347 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
348 | { | |
349 | return kvm_make_all_cpus_request_except(kvm, req, NULL); | |
350 | } | |
a2486020 | 351 | EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request); |
54163a34 | 352 | |
a6d51016 | 353 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 354 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 355 | { |
3cc4e148 | 356 | ++kvm->stat.generic.remote_tlb_flush_requests; |
6bc6db00 | 357 | |
4ae3cb3a LT |
358 | /* |
359 | * We want to publish modifications to the page tables before reading | |
360 | * mode. Pairs with a memory barrier in arch-specific code. | |
361 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
362 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
363 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
364 | * | |
365 | * There is already an smp_mb__after_atomic() before | |
366 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
367 | * barrier here. | |
368 | */ | |
b08660e5 TL |
369 | if (!kvm_arch_flush_remote_tlb(kvm) |
370 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
0193cc90 | 371 | ++kvm->stat.generic.remote_tlb_flush; |
2e53d63a | 372 | } |
2ba9f0d8 | 373 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 374 | #endif |
2e53d63a | 375 | |
683412cc MZ |
376 | static void kvm_flush_shadow_all(struct kvm *kvm) |
377 | { | |
378 | kvm_arch_flush_shadow_all(kvm); | |
379 | kvm_arch_guest_memory_reclaimed(kvm); | |
380 | } | |
381 | ||
6926f95a SC |
382 | #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE |
383 | static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc, | |
384 | gfp_t gfp_flags) | |
385 | { | |
386 | gfp_flags |= mc->gfp_zero; | |
387 | ||
388 | if (mc->kmem_cache) | |
389 | return kmem_cache_alloc(mc->kmem_cache, gfp_flags); | |
390 | else | |
391 | return (void *)__get_free_page(gfp_flags); | |
392 | } | |
393 | ||
837f66c7 | 394 | int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min) |
6926f95a | 395 | { |
63f4b210 | 396 | gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT; |
6926f95a SC |
397 | void *obj; |
398 | ||
399 | if (mc->nobjs >= min) | |
400 | return 0; | |
837f66c7 DM |
401 | |
402 | if (unlikely(!mc->objects)) { | |
403 | if (WARN_ON_ONCE(!capacity)) | |
404 | return -EIO; | |
405 | ||
406 | mc->objects = kvmalloc_array(sizeof(void *), capacity, gfp); | |
407 | if (!mc->objects) | |
408 | return -ENOMEM; | |
409 | ||
410 | mc->capacity = capacity; | |
411 | } | |
412 | ||
413 | /* It is illegal to request a different capacity across topups. */ | |
414 | if (WARN_ON_ONCE(mc->capacity != capacity)) | |
415 | return -EIO; | |
416 | ||
417 | while (mc->nobjs < mc->capacity) { | |
418 | obj = mmu_memory_cache_alloc_obj(mc, gfp); | |
6926f95a SC |
419 | if (!obj) |
420 | return mc->nobjs >= min ? 0 : -ENOMEM; | |
421 | mc->objects[mc->nobjs++] = obj; | |
422 | } | |
423 | return 0; | |
424 | } | |
425 | ||
837f66c7 DM |
426 | int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min) |
427 | { | |
428 | return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min); | |
429 | } | |
430 | ||
6926f95a SC |
431 | int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc) |
432 | { | |
433 | return mc->nobjs; | |
434 | } | |
435 | ||
436 | void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
437 | { | |
438 | while (mc->nobjs) { | |
439 | if (mc->kmem_cache) | |
440 | kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]); | |
441 | else | |
442 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
443 | } | |
837f66c7 DM |
444 | |
445 | kvfree(mc->objects); | |
446 | ||
447 | mc->objects = NULL; | |
448 | mc->capacity = 0; | |
6926f95a SC |
449 | } |
450 | ||
451 | void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
452 | { | |
453 | void *p; | |
454 | ||
455 | if (WARN_ON(!mc->nobjs)) | |
456 | p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT); | |
457 | else | |
458 | p = mc->objects[--mc->nobjs]; | |
459 | BUG_ON(!p); | |
460 | return p; | |
461 | } | |
462 | #endif | |
463 | ||
8bd826d6 | 464 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 465 | { |
fb3f0f51 RR |
466 | mutex_init(&vcpu->mutex); |
467 | vcpu->cpu = -1; | |
fb3f0f51 RR |
468 | vcpu->kvm = kvm; |
469 | vcpu->vcpu_id = id; | |
34bb10b7 | 470 | vcpu->pid = NULL; |
510958e9 | 471 | #ifndef __KVM_HAVE_ARCH_WQP |
da4ad88c | 472 | rcuwait_init(&vcpu->wait); |
510958e9 | 473 | #endif |
af585b92 | 474 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 475 | |
4c088493 R |
476 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
477 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 478 | vcpu->preempted = false; |
d73eb57b | 479 | vcpu->ready = false; |
d5c48deb | 480 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
a54d8066 | 481 | vcpu->last_used_slot = NULL; |
58fc1166 OU |
482 | |
483 | /* Fill the stats id string for the vcpu */ | |
484 | snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d", | |
485 | task_pid_nr(current), id); | |
fb3f0f51 | 486 | } |
fb3f0f51 | 487 | |
27592ae8 | 488 | static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
4543bdc0 SC |
489 | { |
490 | kvm_arch_vcpu_destroy(vcpu); | |
5593473a | 491 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
e529ef66 | 492 | |
9941d224 SC |
493 | /* |
494 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
495 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
496 | * are already gone. | |
497 | */ | |
498 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
499 | ||
8bd826d6 | 500 | free_page((unsigned long)vcpu->run); |
e529ef66 | 501 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 | 502 | } |
27592ae8 MZ |
503 | |
504 | void kvm_destroy_vcpus(struct kvm *kvm) | |
505 | { | |
46808a4c | 506 | unsigned long i; |
27592ae8 MZ |
507 | struct kvm_vcpu *vcpu; |
508 | ||
509 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
510 | kvm_vcpu_destroy(vcpu); | |
c5b07754 | 511 | xa_erase(&kvm->vcpu_array, i); |
27592ae8 MZ |
512 | } |
513 | ||
514 | atomic_set(&kvm->online_vcpus, 0); | |
515 | } | |
516 | EXPORT_SYMBOL_GPL(kvm_destroy_vcpus); | |
4543bdc0 | 517 | |
e930bffe AA |
518 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
519 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
520 | { | |
521 | return container_of(mn, struct kvm, mmu_notifier); | |
522 | } | |
523 | ||
e649b3f0 ET |
524 | static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn, |
525 | struct mm_struct *mm, | |
526 | unsigned long start, unsigned long end) | |
527 | { | |
528 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
529 | int idx; | |
530 | ||
531 | idx = srcu_read_lock(&kvm->srcu); | |
532 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
533 | srcu_read_unlock(&kvm->srcu, idx); | |
534 | } | |
535 | ||
3039bcc7 SC |
536 | typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range); |
537 | ||
f922bd9b SC |
538 | typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start, |
539 | unsigned long end); | |
540 | ||
683412cc MZ |
541 | typedef void (*on_unlock_fn_t)(struct kvm *kvm); |
542 | ||
3039bcc7 SC |
543 | struct kvm_hva_range { |
544 | unsigned long start; | |
545 | unsigned long end; | |
546 | pte_t pte; | |
547 | hva_handler_t handler; | |
f922bd9b | 548 | on_lock_fn_t on_lock; |
683412cc | 549 | on_unlock_fn_t on_unlock; |
3039bcc7 SC |
550 | bool flush_on_ret; |
551 | bool may_block; | |
552 | }; | |
553 | ||
f922bd9b SC |
554 | /* |
555 | * Use a dedicated stub instead of NULL to indicate that there is no callback | |
556 | * function/handler. The compiler technically can't guarantee that a real | |
557 | * function will have a non-zero address, and so it will generate code to | |
558 | * check for !NULL, whereas comparing against a stub will be elided at compile | |
559 | * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9). | |
560 | */ | |
561 | static void kvm_null_fn(void) | |
562 | { | |
563 | ||
564 | } | |
565 | #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn) | |
566 | ||
ed922739 MS |
567 | /* Iterate over each memslot intersecting [start, last] (inclusive) range */ |
568 | #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \ | |
569 | for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \ | |
570 | node; \ | |
571 | node = interval_tree_iter_next(node, start, last)) \ | |
572 | ||
3039bcc7 SC |
573 | static __always_inline int __kvm_handle_hva_range(struct kvm *kvm, |
574 | const struct kvm_hva_range *range) | |
575 | { | |
8931a454 | 576 | bool ret = false, locked = false; |
f922bd9b | 577 | struct kvm_gfn_range gfn_range; |
3039bcc7 SC |
578 | struct kvm_memory_slot *slot; |
579 | struct kvm_memslots *slots; | |
3039bcc7 SC |
580 | int i, idx; |
581 | ||
ed922739 MS |
582 | if (WARN_ON_ONCE(range->end <= range->start)) |
583 | return 0; | |
584 | ||
f922bd9b SC |
585 | /* A null handler is allowed if and only if on_lock() is provided. */ |
586 | if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) && | |
587 | IS_KVM_NULL_FN(range->handler))) | |
588 | return 0; | |
589 | ||
3039bcc7 SC |
590 | idx = srcu_read_lock(&kvm->srcu); |
591 | ||
592 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
ed922739 MS |
593 | struct interval_tree_node *node; |
594 | ||
3039bcc7 | 595 | slots = __kvm_memslots(kvm, i); |
ed922739 MS |
596 | kvm_for_each_memslot_in_hva_range(node, slots, |
597 | range->start, range->end - 1) { | |
3039bcc7 SC |
598 | unsigned long hva_start, hva_end; |
599 | ||
a54d8066 | 600 | slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]); |
3039bcc7 SC |
601 | hva_start = max(range->start, slot->userspace_addr); |
602 | hva_end = min(range->end, slot->userspace_addr + | |
603 | (slot->npages << PAGE_SHIFT)); | |
3039bcc7 SC |
604 | |
605 | /* | |
606 | * To optimize for the likely case where the address | |
607 | * range is covered by zero or one memslots, don't | |
608 | * bother making these conditional (to avoid writes on | |
609 | * the second or later invocation of the handler). | |
610 | */ | |
611 | gfn_range.pte = range->pte; | |
612 | gfn_range.may_block = range->may_block; | |
613 | ||
614 | /* | |
615 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
616 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
617 | */ | |
618 | gfn_range.start = hva_to_gfn_memslot(hva_start, slot); | |
619 | gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot); | |
620 | gfn_range.slot = slot; | |
621 | ||
8931a454 SC |
622 | if (!locked) { |
623 | locked = true; | |
624 | KVM_MMU_LOCK(kvm); | |
071064f1 PB |
625 | if (!IS_KVM_NULL_FN(range->on_lock)) |
626 | range->on_lock(kvm, range->start, range->end); | |
627 | if (IS_KVM_NULL_FN(range->handler)) | |
628 | break; | |
8931a454 | 629 | } |
3039bcc7 SC |
630 | ret |= range->handler(kvm, &gfn_range); |
631 | } | |
632 | } | |
633 | ||
6bc6db00 | 634 | if (range->flush_on_ret && ret) |
3039bcc7 SC |
635 | kvm_flush_remote_tlbs(kvm); |
636 | ||
683412cc | 637 | if (locked) { |
8931a454 | 638 | KVM_MMU_UNLOCK(kvm); |
683412cc MZ |
639 | if (!IS_KVM_NULL_FN(range->on_unlock)) |
640 | range->on_unlock(kvm); | |
641 | } | |
f922bd9b | 642 | |
3039bcc7 SC |
643 | srcu_read_unlock(&kvm->srcu, idx); |
644 | ||
645 | /* The notifiers are averse to booleans. :-( */ | |
646 | return (int)ret; | |
647 | } | |
648 | ||
649 | static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn, | |
650 | unsigned long start, | |
651 | unsigned long end, | |
652 | pte_t pte, | |
653 | hva_handler_t handler) | |
654 | { | |
655 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
656 | const struct kvm_hva_range range = { | |
657 | .start = start, | |
658 | .end = end, | |
659 | .pte = pte, | |
660 | .handler = handler, | |
f922bd9b | 661 | .on_lock = (void *)kvm_null_fn, |
683412cc | 662 | .on_unlock = (void *)kvm_null_fn, |
3039bcc7 SC |
663 | .flush_on_ret = true, |
664 | .may_block = false, | |
665 | }; | |
3039bcc7 | 666 | |
f922bd9b | 667 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 SC |
668 | } |
669 | ||
670 | static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn, | |
671 | unsigned long start, | |
672 | unsigned long end, | |
673 | hva_handler_t handler) | |
674 | { | |
675 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
676 | const struct kvm_hva_range range = { | |
677 | .start = start, | |
678 | .end = end, | |
679 | .pte = __pte(0), | |
680 | .handler = handler, | |
f922bd9b | 681 | .on_lock = (void *)kvm_null_fn, |
683412cc | 682 | .on_unlock = (void *)kvm_null_fn, |
3039bcc7 SC |
683 | .flush_on_ret = false, |
684 | .may_block = false, | |
685 | }; | |
3039bcc7 | 686 | |
f922bd9b | 687 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 | 688 | } |
3da0dd43 IE |
689 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
690 | struct mm_struct *mm, | |
691 | unsigned long address, | |
692 | pte_t pte) | |
693 | { | |
694 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
695 | ||
501b9185 SC |
696 | trace_kvm_set_spte_hva(address); |
697 | ||
c13fda23 | 698 | /* |
52ac8b35 | 699 | * .change_pte() must be surrounded by .invalidate_range_{start,end}(). |
20ec3ebd CP |
700 | * If mmu_invalidate_in_progress is zero, then no in-progress |
701 | * invalidations, including this one, found a relevant memslot at | |
702 | * start(); rechecking memslots here is unnecessary. Note, a false | |
703 | * positive (count elevated by a different invalidation) is sub-optimal | |
704 | * but functionally ok. | |
c13fda23 | 705 | */ |
52ac8b35 | 706 | WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count)); |
20ec3ebd | 707 | if (!READ_ONCE(kvm->mmu_invalidate_in_progress)) |
071064f1 | 708 | return; |
c13fda23 | 709 | |
3039bcc7 | 710 | kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn); |
3da0dd43 IE |
711 | } |
712 | ||
20ec3ebd CP |
713 | void kvm_mmu_invalidate_begin(struct kvm *kvm, unsigned long start, |
714 | unsigned long end) | |
e930bffe | 715 | { |
e930bffe AA |
716 | /* |
717 | * The count increase must become visible at unlock time as no | |
718 | * spte can be established without taking the mmu_lock and | |
719 | * count is also read inside the mmu_lock critical section. | |
720 | */ | |
20ec3ebd CP |
721 | kvm->mmu_invalidate_in_progress++; |
722 | if (likely(kvm->mmu_invalidate_in_progress == 1)) { | |
723 | kvm->mmu_invalidate_range_start = start; | |
724 | kvm->mmu_invalidate_range_end = end; | |
4a42d848 DS |
725 | } else { |
726 | /* | |
a413a625 | 727 | * Fully tracking multiple concurrent ranges has diminishing |
4a42d848 DS |
728 | * returns. Keep things simple and just find the minimal range |
729 | * which includes the current and new ranges. As there won't be | |
730 | * enough information to subtract a range after its invalidate | |
731 | * completes, any ranges invalidated concurrently will | |
732 | * accumulate and persist until all outstanding invalidates | |
733 | * complete. | |
734 | */ | |
20ec3ebd CP |
735 | kvm->mmu_invalidate_range_start = |
736 | min(kvm->mmu_invalidate_range_start, start); | |
737 | kvm->mmu_invalidate_range_end = | |
738 | max(kvm->mmu_invalidate_range_end, end); | |
4a42d848 | 739 | } |
f922bd9b | 740 | } |
3039bcc7 | 741 | |
f922bd9b SC |
742 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
743 | const struct mmu_notifier_range *range) | |
744 | { | |
745 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
746 | const struct kvm_hva_range hva_range = { | |
747 | .start = range->start, | |
748 | .end = range->end, | |
749 | .pte = __pte(0), | |
750 | .handler = kvm_unmap_gfn_range, | |
20ec3ebd | 751 | .on_lock = kvm_mmu_invalidate_begin, |
683412cc | 752 | .on_unlock = kvm_arch_guest_memory_reclaimed, |
f922bd9b SC |
753 | .flush_on_ret = true, |
754 | .may_block = mmu_notifier_range_blockable(range), | |
755 | }; | |
565f3be2 | 756 | |
f922bd9b SC |
757 | trace_kvm_unmap_hva_range(range->start, range->end); |
758 | ||
52ac8b35 PB |
759 | /* |
760 | * Prevent memslot modification between range_start() and range_end() | |
761 | * so that conditionally locking provides the same result in both | |
20ec3ebd | 762 | * functions. Without that guarantee, the mmu_invalidate_in_progress |
52ac8b35 PB |
763 | * adjustments will be imbalanced. |
764 | * | |
765 | * Pairs with the decrement in range_end(). | |
766 | */ | |
767 | spin_lock(&kvm->mn_invalidate_lock); | |
768 | kvm->mn_active_invalidate_count++; | |
769 | spin_unlock(&kvm->mn_invalidate_lock); | |
770 | ||
58cd407c SC |
771 | /* |
772 | * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e. | |
773 | * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring | |
774 | * each cache's lock. There are relatively few caches in existence at | |
775 | * any given time, and the caches themselves can check for hva overlap, | |
776 | * i.e. don't need to rely on memslot overlap checks for performance. | |
777 | * Because this runs without holding mmu_lock, the pfn caches must use | |
20ec3ebd CP |
778 | * mn_active_invalidate_count (see above) instead of |
779 | * mmu_invalidate_in_progress. | |
58cd407c | 780 | */ |
982ed0de DW |
781 | gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end, |
782 | hva_range.may_block); | |
783 | ||
f922bd9b | 784 | __kvm_handle_hva_range(kvm, &hva_range); |
93065ac7 | 785 | |
e649b3f0 | 786 | return 0; |
e930bffe AA |
787 | } |
788 | ||
20ec3ebd CP |
789 | void kvm_mmu_invalidate_end(struct kvm *kvm, unsigned long start, |
790 | unsigned long end) | |
e930bffe | 791 | { |
e930bffe AA |
792 | /* |
793 | * This sequence increase will notify the kvm page fault that | |
794 | * the page that is going to be mapped in the spte could have | |
795 | * been freed. | |
796 | */ | |
20ec3ebd | 797 | kvm->mmu_invalidate_seq++; |
a355aa54 | 798 | smp_wmb(); |
e930bffe AA |
799 | /* |
800 | * The above sequence increase must be visible before the | |
a355aa54 | 801 | * below count decrease, which is ensured by the smp_wmb above |
20ec3ebd | 802 | * in conjunction with the smp_rmb in mmu_invalidate_retry(). |
e930bffe | 803 | */ |
20ec3ebd | 804 | kvm->mmu_invalidate_in_progress--; |
f922bd9b SC |
805 | } |
806 | ||
807 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
808 | const struct mmu_notifier_range *range) | |
809 | { | |
810 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
811 | const struct kvm_hva_range hva_range = { | |
812 | .start = range->start, | |
813 | .end = range->end, | |
814 | .pte = __pte(0), | |
815 | .handler = (void *)kvm_null_fn, | |
20ec3ebd | 816 | .on_lock = kvm_mmu_invalidate_end, |
683412cc | 817 | .on_unlock = (void *)kvm_null_fn, |
f922bd9b SC |
818 | .flush_on_ret = false, |
819 | .may_block = mmu_notifier_range_blockable(range), | |
820 | }; | |
52ac8b35 | 821 | bool wake; |
f922bd9b SC |
822 | |
823 | __kvm_handle_hva_range(kvm, &hva_range); | |
e930bffe | 824 | |
52ac8b35 PB |
825 | /* Pairs with the increment in range_start(). */ |
826 | spin_lock(&kvm->mn_invalidate_lock); | |
827 | wake = (--kvm->mn_active_invalidate_count == 0); | |
828 | spin_unlock(&kvm->mn_invalidate_lock); | |
829 | ||
830 | /* | |
831 | * There can only be one waiter, since the wait happens under | |
832 | * slots_lock. | |
833 | */ | |
834 | if (wake) | |
835 | rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait); | |
836 | ||
20ec3ebd | 837 | BUG_ON(kvm->mmu_invalidate_in_progress < 0); |
e930bffe AA |
838 | } |
839 | ||
840 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
841 | struct mm_struct *mm, | |
57128468 ALC |
842 | unsigned long start, |
843 | unsigned long end) | |
e930bffe | 844 | { |
501b9185 SC |
845 | trace_kvm_age_hva(start, end); |
846 | ||
3039bcc7 | 847 | return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn); |
e930bffe AA |
848 | } |
849 | ||
1d7715c6 VD |
850 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
851 | struct mm_struct *mm, | |
852 | unsigned long start, | |
853 | unsigned long end) | |
854 | { | |
501b9185 SC |
855 | trace_kvm_age_hva(start, end); |
856 | ||
1d7715c6 VD |
857 | /* |
858 | * Even though we do not flush TLB, this will still adversely | |
859 | * affect performance on pre-Haswell Intel EPT, where there is | |
860 | * no EPT Access Bit to clear so that we have to tear down EPT | |
861 | * tables instead. If we find this unacceptable, we can always | |
862 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
863 | * do anything on clear_young. | |
864 | * | |
865 | * Also note that currently we never issue secondary TLB flushes | |
866 | * from clear_young, leaving this job up to the regular system | |
867 | * cadence. If we find this inaccurate, we might come up with a | |
868 | * more sophisticated heuristic later. | |
869 | */ | |
3039bcc7 | 870 | return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn); |
1d7715c6 VD |
871 | } |
872 | ||
8ee53820 AA |
873 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
874 | struct mm_struct *mm, | |
875 | unsigned long address) | |
876 | { | |
501b9185 SC |
877 | trace_kvm_test_age_hva(address); |
878 | ||
3039bcc7 SC |
879 | return kvm_handle_hva_range_no_flush(mn, address, address + 1, |
880 | kvm_test_age_gfn); | |
8ee53820 AA |
881 | } |
882 | ||
85db06e5 MT |
883 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
884 | struct mm_struct *mm) | |
885 | { | |
886 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
887 | int idx; |
888 | ||
889 | idx = srcu_read_lock(&kvm->srcu); | |
683412cc | 890 | kvm_flush_shadow_all(kvm); |
eda2beda | 891 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
892 | } |
893 | ||
e930bffe | 894 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e649b3f0 | 895 | .invalidate_range = kvm_mmu_notifier_invalidate_range, |
e930bffe AA |
896 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
897 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
898 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 899 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 900 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 901 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 902 | .release = kvm_mmu_notifier_release, |
e930bffe | 903 | }; |
4c07b0a4 AK |
904 | |
905 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
906 | { | |
907 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
908 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
909 | } | |
910 | ||
911 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
912 | ||
913 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
914 | { | |
915 | return 0; | |
916 | } | |
917 | ||
e930bffe AA |
918 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
919 | ||
2fdef3a2 SS |
920 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
921 | static int kvm_pm_notifier_call(struct notifier_block *bl, | |
922 | unsigned long state, | |
923 | void *unused) | |
924 | { | |
925 | struct kvm *kvm = container_of(bl, struct kvm, pm_notifier); | |
926 | ||
927 | return kvm_arch_pm_notifier(kvm, state); | |
928 | } | |
929 | ||
930 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
931 | { | |
932 | kvm->pm_notifier.notifier_call = kvm_pm_notifier_call; | |
933 | /* Suspend KVM before we suspend ftrace, RCU, etc. */ | |
934 | kvm->pm_notifier.priority = INT_MAX; | |
935 | register_pm_notifier(&kvm->pm_notifier); | |
936 | } | |
937 | ||
938 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
939 | { | |
940 | unregister_pm_notifier(&kvm->pm_notifier); | |
941 | } | |
942 | #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
943 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
944 | { | |
945 | } | |
946 | ||
947 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
948 | { | |
949 | } | |
950 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
951 | ||
a47d2b07 PB |
952 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
953 | { | |
954 | if (!memslot->dirty_bitmap) | |
955 | return; | |
956 | ||
957 | kvfree(memslot->dirty_bitmap); | |
958 | memslot->dirty_bitmap = NULL; | |
959 | } | |
960 | ||
a54d8066 | 961 | /* This does not remove the slot from struct kvm_memslots data structures */ |
e96c81ee | 962 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 963 | { |
e96c81ee | 964 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 965 | |
e96c81ee | 966 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 967 | |
a54d8066 | 968 | kfree(slot); |
a47d2b07 PB |
969 | } |
970 | ||
971 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
972 | { | |
a54d8066 | 973 | struct hlist_node *idnode; |
a47d2b07 | 974 | struct kvm_memory_slot *memslot; |
a54d8066 | 975 | int bkt; |
a47d2b07 | 976 | |
a54d8066 MS |
977 | /* |
978 | * The same memslot objects live in both active and inactive sets, | |
979 | * arbitrarily free using index '1' so the second invocation of this | |
980 | * function isn't operating over a structure with dangling pointers | |
981 | * (even though this function isn't actually touching them). | |
982 | */ | |
983 | if (!slots->node_idx) | |
a47d2b07 PB |
984 | return; |
985 | ||
a54d8066 | 986 | hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1]) |
e96c81ee | 987 | kvm_free_memslot(kvm, memslot); |
bf3e05bc XG |
988 | } |
989 | ||
bc9e9e67 JZ |
990 | static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc) |
991 | { | |
992 | switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) { | |
993 | case KVM_STATS_TYPE_INSTANT: | |
994 | return 0444; | |
995 | case KVM_STATS_TYPE_CUMULATIVE: | |
996 | case KVM_STATS_TYPE_PEAK: | |
997 | default: | |
998 | return 0644; | |
999 | } | |
1000 | } | |
1001 | ||
1002 | ||
536a6f88 JF |
1003 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
1004 | { | |
1005 | int i; | |
bc9e9e67 JZ |
1006 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
1007 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 | 1008 | |
a44a4cc1 | 1009 | if (IS_ERR(kvm->debugfs_dentry)) |
536a6f88 JF |
1010 | return; |
1011 | ||
1012 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
1013 | ||
9d5a1dce LC |
1014 | if (kvm->debugfs_stat_data) { |
1015 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
1016 | kfree(kvm->debugfs_stat_data[i]); | |
1017 | kfree(kvm->debugfs_stat_data); | |
1018 | } | |
536a6f88 JF |
1019 | } |
1020 | ||
59f82aad | 1021 | static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname) |
536a6f88 | 1022 | { |
85cd39af PB |
1023 | static DEFINE_MUTEX(kvm_debugfs_lock); |
1024 | struct dentry *dent; | |
536a6f88 JF |
1025 | char dir_name[ITOA_MAX_LEN * 2]; |
1026 | struct kvm_stat_data *stat_data; | |
bc9e9e67 | 1027 | const struct _kvm_stats_desc *pdesc; |
b74ed7a6 | 1028 | int i, ret = -ENOMEM; |
bc9e9e67 JZ |
1029 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
1030 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
1031 | |
1032 | if (!debugfs_initialized()) | |
1033 | return 0; | |
1034 | ||
59f82aad | 1035 | snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname); |
85cd39af PB |
1036 | mutex_lock(&kvm_debugfs_lock); |
1037 | dent = debugfs_lookup(dir_name, kvm_debugfs_dir); | |
1038 | if (dent) { | |
1039 | pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name); | |
1040 | dput(dent); | |
1041 | mutex_unlock(&kvm_debugfs_lock); | |
1042 | return 0; | |
1043 | } | |
1044 | dent = debugfs_create_dir(dir_name, kvm_debugfs_dir); | |
1045 | mutex_unlock(&kvm_debugfs_lock); | |
1046 | if (IS_ERR(dent)) | |
1047 | return 0; | |
536a6f88 | 1048 | |
85cd39af | 1049 | kvm->debugfs_dentry = dent; |
536a6f88 JF |
1050 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
1051 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 1052 | GFP_KERNEL_ACCOUNT); |
536a6f88 | 1053 | if (!kvm->debugfs_stat_data) |
b74ed7a6 | 1054 | goto out_err; |
536a6f88 | 1055 | |
bc9e9e67 JZ |
1056 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
1057 | pdesc = &kvm_vm_stats_desc[i]; | |
b12ce36a | 1058 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 | 1059 | if (!stat_data) |
b74ed7a6 | 1060 | goto out_err; |
536a6f88 JF |
1061 | |
1062 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1063 | stat_data->desc = pdesc; |
1064 | stat_data->kind = KVM_STAT_VM; | |
1065 | kvm->debugfs_stat_data[i] = stat_data; | |
1066 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
1067 | kvm->debugfs_dentry, stat_data, | |
1068 | &stat_fops_per_vm); | |
1069 | } | |
1070 | ||
1071 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
1072 | pdesc = &kvm_vcpu_stats_desc[i]; | |
b12ce36a | 1073 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 | 1074 | if (!stat_data) |
b74ed7a6 | 1075 | goto out_err; |
536a6f88 JF |
1076 | |
1077 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1078 | stat_data->desc = pdesc; |
1079 | stat_data->kind = KVM_STAT_VCPU; | |
004d62eb | 1080 | kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data; |
bc9e9e67 | 1081 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), |
09cbcef6 MP |
1082 | kvm->debugfs_dentry, stat_data, |
1083 | &stat_fops_per_vm); | |
536a6f88 | 1084 | } |
3165af73 PX |
1085 | |
1086 | ret = kvm_arch_create_vm_debugfs(kvm); | |
b74ed7a6 OU |
1087 | if (ret) |
1088 | goto out_err; | |
3165af73 | 1089 | |
536a6f88 | 1090 | return 0; |
b74ed7a6 OU |
1091 | out_err: |
1092 | kvm_destroy_vm_debugfs(kvm); | |
1093 | return ret; | |
536a6f88 JF |
1094 | } |
1095 | ||
1aa9b957 JS |
1096 | /* |
1097 | * Called after the VM is otherwise initialized, but just before adding it to | |
1098 | * the vm_list. | |
1099 | */ | |
1100 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
1101 | { | |
1102 | return 0; | |
1103 | } | |
1104 | ||
1105 | /* | |
1106 | * Called just after removing the VM from the vm_list, but before doing any | |
1107 | * other destruction. | |
1108 | */ | |
1109 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
1110 | { | |
1111 | } | |
1112 | ||
3165af73 PX |
1113 | /* |
1114 | * Called after per-vm debugfs created. When called kvm->debugfs_dentry should | |
1115 | * be setup already, so we can create arch-specific debugfs entries under it. | |
1116 | * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so | |
1117 | * a per-arch destroy interface is not needed. | |
1118 | */ | |
1119 | int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm) | |
1120 | { | |
1121 | return 0; | |
1122 | } | |
1123 | ||
b74ed7a6 | 1124 | static struct kvm *kvm_create_vm(unsigned long type, const char *fdname) |
6aa8b732 | 1125 | { |
d89f5eff | 1126 | struct kvm *kvm = kvm_arch_alloc_vm(); |
a54d8066 | 1127 | struct kvm_memslots *slots; |
9121923c | 1128 | int r = -ENOMEM; |
a54d8066 | 1129 | int i, j; |
6aa8b732 | 1130 | |
d89f5eff JK |
1131 | if (!kvm) |
1132 | return ERR_PTR(-ENOMEM); | |
1133 | ||
405294f2 SC |
1134 | /* KVM is pinned via open("/dev/kvm"), the fd passed to this ioctl(). */ |
1135 | __module_get(kvm_chardev_ops.owner); | |
1136 | ||
531810ca | 1137 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 1138 | mmgrab(current->mm); |
e9ad4ec8 PB |
1139 | kvm->mm = current->mm; |
1140 | kvm_eventfd_init(kvm); | |
1141 | mutex_init(&kvm->lock); | |
1142 | mutex_init(&kvm->irq_lock); | |
1143 | mutex_init(&kvm->slots_lock); | |
b10a038e | 1144 | mutex_init(&kvm->slots_arch_lock); |
52ac8b35 PB |
1145 | spin_lock_init(&kvm->mn_invalidate_lock); |
1146 | rcuwait_init(&kvm->mn_memslots_update_rcuwait); | |
c5b07754 | 1147 | xa_init(&kvm->vcpu_array); |
52ac8b35 | 1148 | |
982ed0de DW |
1149 | INIT_LIST_HEAD(&kvm->gpc_list); |
1150 | spin_lock_init(&kvm->gpc_lock); | |
52ac8b35 | 1151 | |
e9ad4ec8 | 1152 | INIT_LIST_HEAD(&kvm->devices); |
f502cc56 | 1153 | kvm->max_vcpus = KVM_MAX_VCPUS; |
e9ad4ec8 | 1154 | |
1e702d9a AW |
1155 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
1156 | ||
5c697c36 SC |
1157 | /* |
1158 | * Force subsequent debugfs file creations to fail if the VM directory | |
1159 | * is not created (by kvm_create_vm_debugfs()). | |
1160 | */ | |
1161 | kvm->debugfs_dentry = ERR_PTR(-ENOENT); | |
1162 | ||
f2759c08 OU |
1163 | snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d", |
1164 | task_pid_nr(current)); | |
1165 | ||
8a44119a PB |
1166 | if (init_srcu_struct(&kvm->srcu)) |
1167 | goto out_err_no_srcu; | |
1168 | if (init_srcu_struct(&kvm->irq_srcu)) | |
1169 | goto out_err_no_irq_srcu; | |
1170 | ||
e2d3fcaf | 1171 | refcount_set(&kvm->users_count, 1); |
f481b069 | 1172 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
a54d8066 MS |
1173 | for (j = 0; j < 2; j++) { |
1174 | slots = &kvm->__memslots[i][j]; | |
9121923c | 1175 | |
a54d8066 MS |
1176 | atomic_long_set(&slots->last_used_slot, (unsigned long)NULL); |
1177 | slots->hva_tree = RB_ROOT_CACHED; | |
1178 | slots->gfn_tree = RB_ROOT; | |
1179 | hash_init(slots->id_hash); | |
1180 | slots->node_idx = j; | |
1181 | ||
1182 | /* Generations must be different for each address space. */ | |
1183 | slots->generation = i; | |
1184 | } | |
1185 | ||
1186 | rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]); | |
f481b069 | 1187 | } |
00f034a1 | 1188 | |
e93f8a0f | 1189 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 1190 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 1191 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 1192 | if (!kvm->buses[i]) |
a97b0e77 | 1193 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 1194 | } |
e930bffe | 1195 | |
e08b9637 | 1196 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 1197 | if (r) |
a97b0e77 | 1198 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
1199 | |
1200 | r = hardware_enable_all(); | |
1201 | if (r) | |
719d93cd | 1202 | goto out_err_no_disable; |
10474ae8 | 1203 | |
c77dcacb | 1204 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 1205 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 1206 | #endif |
6aa8b732 | 1207 | |
74b5c5bf | 1208 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
1209 | if (r) |
1210 | goto out_err_no_mmu_notifier; | |
1211 | ||
c2b82397 SC |
1212 | r = kvm_coalesced_mmio_init(kvm); |
1213 | if (r < 0) | |
1214 | goto out_no_coalesced_mmio; | |
1215 | ||
4ba4f419 SC |
1216 | r = kvm_create_vm_debugfs(kvm, fdname); |
1217 | if (r) | |
1218 | goto out_err_no_debugfs; | |
1219 | ||
1aa9b957 | 1220 | r = kvm_arch_post_init_vm(kvm); |
74b5c5bf | 1221 | if (r) |
4ba4f419 | 1222 | goto out_err; |
74b5c5bf | 1223 | |
0d9ce162 | 1224 | mutex_lock(&kvm_lock); |
5e58cfe4 | 1225 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 1226 | mutex_unlock(&kvm_lock); |
d89f5eff | 1227 | |
2ecd9d29 | 1228 | preempt_notifier_inc(); |
2fdef3a2 | 1229 | kvm_init_pm_notifier(kvm); |
2ecd9d29 | 1230 | |
f17abe9a | 1231 | return kvm; |
10474ae8 AG |
1232 | |
1233 | out_err: | |
4ba4f419 SC |
1234 | kvm_destroy_vm_debugfs(kvm); |
1235 | out_err_no_debugfs: | |
c2b82397 SC |
1236 | kvm_coalesced_mmio_free(kvm); |
1237 | out_no_coalesced_mmio: | |
1aa9b957 JS |
1238 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1239 | if (kvm->mmu_notifier.ops) | |
1240 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
1241 | #endif | |
1242 | out_err_no_mmu_notifier: | |
10474ae8 | 1243 | hardware_disable_all(); |
719d93cd | 1244 | out_err_no_disable: |
a97b0e77 | 1245 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 1246 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 1247 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 1248 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 1249 | kfree(kvm_get_bus(kvm, i)); |
8a44119a PB |
1250 | cleanup_srcu_struct(&kvm->irq_srcu); |
1251 | out_err_no_irq_srcu: | |
1252 | cleanup_srcu_struct(&kvm->srcu); | |
1253 | out_err_no_srcu: | |
d89f5eff | 1254 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 1255 | mmdrop(current->mm); |
405294f2 | 1256 | module_put(kvm_chardev_ops.owner); |
10474ae8 | 1257 | return ERR_PTR(r); |
f17abe9a AK |
1258 | } |
1259 | ||
07f0a7bd SW |
1260 | static void kvm_destroy_devices(struct kvm *kvm) |
1261 | { | |
e6e3b5a6 | 1262 | struct kvm_device *dev, *tmp; |
07f0a7bd | 1263 | |
a28ebea2 CD |
1264 | /* |
1265 | * We do not need to take the kvm->lock here, because nobody else | |
1266 | * has a reference to the struct kvm at this point and therefore | |
1267 | * cannot access the devices list anyhow. | |
1268 | */ | |
e6e3b5a6 GT |
1269 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
1270 | list_del(&dev->vm_node); | |
07f0a7bd SW |
1271 | dev->ops->destroy(dev); |
1272 | } | |
1273 | } | |
1274 | ||
f17abe9a AK |
1275 | static void kvm_destroy_vm(struct kvm *kvm) |
1276 | { | |
e93f8a0f | 1277 | int i; |
6d4e4c4f AK |
1278 | struct mm_struct *mm = kvm->mm; |
1279 | ||
2fdef3a2 | 1280 | kvm_destroy_pm_notifier(kvm); |
286de8f6 | 1281 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 1282 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 1283 | kvm_arch_sync_events(kvm); |
0d9ce162 | 1284 | mutex_lock(&kvm_lock); |
133de902 | 1285 | list_del(&kvm->vm_list); |
0d9ce162 | 1286 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
1287 | kvm_arch_pre_destroy_vm(kvm); |
1288 | ||
399ec807 | 1289 | kvm_free_irq_routing(kvm); |
df630b8c | 1290 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 1291 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 1292 | |
4a12f951 CB |
1293 | if (bus) |
1294 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
1295 | kvm->buses[i] = NULL; |
1296 | } | |
980da6ce | 1297 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
1298 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1299 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
52ac8b35 PB |
1300 | /* |
1301 | * At this point, pending calls to invalidate_range_start() | |
1302 | * have completed but no more MMU notifiers will run, so | |
1303 | * mn_active_invalidate_count may remain unbalanced. | |
b0d23708 | 1304 | * No threads can be waiting in kvm_swap_active_memslots() as the |
52ac8b35 PB |
1305 | * last reference on KVM has been dropped, but freeing |
1306 | * memslots would deadlock without this manual intervention. | |
1307 | */ | |
1308 | WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait)); | |
1309 | kvm->mn_active_invalidate_count = 0; | |
f00be0ca | 1310 | #else |
683412cc | 1311 | kvm_flush_shadow_all(kvm); |
5f94c174 | 1312 | #endif |
d19a9cd2 | 1313 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 1314 | kvm_destroy_devices(kvm); |
a54d8066 MS |
1315 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
1316 | kvm_free_memslots(kvm, &kvm->__memslots[i][0]); | |
1317 | kvm_free_memslots(kvm, &kvm->__memslots[i][1]); | |
1318 | } | |
820b3fcd | 1319 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
1320 | cleanup_srcu_struct(&kvm->srcu); |
1321 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 1322 | preempt_notifier_dec(); |
10474ae8 | 1323 | hardware_disable_all(); |
6d4e4c4f | 1324 | mmdrop(mm); |
5f6de5cb | 1325 | module_put(kvm_chardev_ops.owner); |
f17abe9a AK |
1326 | } |
1327 | ||
d39f13b0 IE |
1328 | void kvm_get_kvm(struct kvm *kvm) |
1329 | { | |
e3736c3e | 1330 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
1331 | } |
1332 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
1333 | ||
605c7130 PX |
1334 | /* |
1335 | * Make sure the vm is not during destruction, which is a safe version of | |
1336 | * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise. | |
1337 | */ | |
1338 | bool kvm_get_kvm_safe(struct kvm *kvm) | |
1339 | { | |
1340 | return refcount_inc_not_zero(&kvm->users_count); | |
1341 | } | |
1342 | EXPORT_SYMBOL_GPL(kvm_get_kvm_safe); | |
1343 | ||
d39f13b0 IE |
1344 | void kvm_put_kvm(struct kvm *kvm) |
1345 | { | |
e3736c3e | 1346 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
1347 | kvm_destroy_vm(kvm); |
1348 | } | |
1349 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
1350 | ||
149487bd SC |
1351 | /* |
1352 | * Used to put a reference that was taken on behalf of an object associated | |
1353 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
1354 | * of the new file descriptor fails and the reference cannot be transferred to | |
1355 | * its final owner. In such cases, the caller is still actively using @kvm and | |
1356 | * will fail miserably if the refcount unexpectedly hits zero. | |
1357 | */ | |
1358 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
1359 | { | |
1360 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
1361 | } | |
1362 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 1363 | |
f17abe9a AK |
1364 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
1365 | { | |
1366 | struct kvm *kvm = filp->private_data; | |
1367 | ||
721eecbf GH |
1368 | kvm_irqfd_release(kvm); |
1369 | ||
d39f13b0 | 1370 | kvm_put_kvm(kvm); |
6aa8b732 AK |
1371 | return 0; |
1372 | } | |
1373 | ||
515a0127 TY |
1374 | /* |
1375 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 1376 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 1377 | */ |
3c9bd400 | 1378 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 1379 | { |
37b2a651 | 1380 | unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 1381 | |
37b2a651 | 1382 | memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
1383 | if (!memslot->dirty_bitmap) |
1384 | return -ENOMEM; | |
1385 | ||
a36a57b1 TY |
1386 | return 0; |
1387 | } | |
1388 | ||
a54d8066 | 1389 | static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id) |
bf3e05bc | 1390 | { |
a54d8066 MS |
1391 | struct kvm_memslots *active = __kvm_memslots(kvm, as_id); |
1392 | int node_idx_inactive = active->node_idx ^ 1; | |
0e60b079 | 1393 | |
a54d8066 | 1394 | return &kvm->__memslots[as_id][node_idx_inactive]; |
0577d1ab SC |
1395 | } |
1396 | ||
1397 | /* | |
a54d8066 MS |
1398 | * Helper to get the address space ID when one of memslot pointers may be NULL. |
1399 | * This also serves as a sanity that at least one of the pointers is non-NULL, | |
1400 | * and that their address space IDs don't diverge. | |
0577d1ab | 1401 | */ |
a54d8066 MS |
1402 | static int kvm_memslots_get_as_id(struct kvm_memory_slot *a, |
1403 | struct kvm_memory_slot *b) | |
0577d1ab | 1404 | { |
a54d8066 MS |
1405 | if (WARN_ON_ONCE(!a && !b)) |
1406 | return 0; | |
0577d1ab | 1407 | |
a54d8066 MS |
1408 | if (!a) |
1409 | return b->as_id; | |
1410 | if (!b) | |
1411 | return a->as_id; | |
0577d1ab | 1412 | |
a54d8066 MS |
1413 | WARN_ON_ONCE(a->as_id != b->as_id); |
1414 | return a->as_id; | |
0577d1ab | 1415 | } |
efbeec70 | 1416 | |
a54d8066 MS |
1417 | static void kvm_insert_gfn_node(struct kvm_memslots *slots, |
1418 | struct kvm_memory_slot *slot) | |
0577d1ab | 1419 | { |
a54d8066 MS |
1420 | struct rb_root *gfn_tree = &slots->gfn_tree; |
1421 | struct rb_node **node, *parent; | |
1422 | int idx = slots->node_idx; | |
0577d1ab | 1423 | |
a54d8066 MS |
1424 | parent = NULL; |
1425 | for (node = &gfn_tree->rb_node; *node; ) { | |
1426 | struct kvm_memory_slot *tmp; | |
f85e2cb5 | 1427 | |
a54d8066 MS |
1428 | tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]); |
1429 | parent = *node; | |
1430 | if (slot->base_gfn < tmp->base_gfn) | |
1431 | node = &(*node)->rb_left; | |
1432 | else if (slot->base_gfn > tmp->base_gfn) | |
1433 | node = &(*node)->rb_right; | |
1434 | else | |
1435 | BUG(); | |
0577d1ab | 1436 | } |
a54d8066 MS |
1437 | |
1438 | rb_link_node(&slot->gfn_node[idx], parent, node); | |
1439 | rb_insert_color(&slot->gfn_node[idx], gfn_tree); | |
0577d1ab SC |
1440 | } |
1441 | ||
a54d8066 MS |
1442 | static void kvm_erase_gfn_node(struct kvm_memslots *slots, |
1443 | struct kvm_memory_slot *slot) | |
0577d1ab | 1444 | { |
a54d8066 MS |
1445 | rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree); |
1446 | } | |
0577d1ab | 1447 | |
a54d8066 MS |
1448 | static void kvm_replace_gfn_node(struct kvm_memslots *slots, |
1449 | struct kvm_memory_slot *old, | |
1450 | struct kvm_memory_slot *new) | |
1451 | { | |
1452 | int idx = slots->node_idx; | |
0577d1ab | 1453 | |
a54d8066 | 1454 | WARN_ON_ONCE(old->base_gfn != new->base_gfn); |
0577d1ab | 1455 | |
a54d8066 MS |
1456 | rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx], |
1457 | &slots->gfn_tree); | |
0577d1ab SC |
1458 | } |
1459 | ||
1460 | /* | |
a54d8066 | 1461 | * Replace @old with @new in the inactive memslots. |
0577d1ab | 1462 | * |
a54d8066 MS |
1463 | * With NULL @old this simply adds @new. |
1464 | * With NULL @new this simply removes @old. | |
0577d1ab | 1465 | * |
a54d8066 MS |
1466 | * If @new is non-NULL its hva_node[slots_idx] range has to be set |
1467 | * appropriately. | |
0577d1ab | 1468 | */ |
a54d8066 MS |
1469 | static void kvm_replace_memslot(struct kvm *kvm, |
1470 | struct kvm_memory_slot *old, | |
1471 | struct kvm_memory_slot *new) | |
0577d1ab | 1472 | { |
a54d8066 MS |
1473 | int as_id = kvm_memslots_get_as_id(old, new); |
1474 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); | |
1475 | int idx = slots->node_idx; | |
0577d1ab | 1476 | |
a54d8066 MS |
1477 | if (old) { |
1478 | hash_del(&old->id_node[idx]); | |
1479 | interval_tree_remove(&old->hva_node[idx], &slots->hva_tree); | |
0577d1ab | 1480 | |
a54d8066 MS |
1481 | if ((long)old == atomic_long_read(&slots->last_used_slot)) |
1482 | atomic_long_set(&slots->last_used_slot, (long)new); | |
0577d1ab | 1483 | |
a54d8066 MS |
1484 | if (!new) { |
1485 | kvm_erase_gfn_node(slots, old); | |
1e8617d3 | 1486 | return; |
a54d8066 MS |
1487 | } |
1488 | } | |
1e8617d3 | 1489 | |
a54d8066 MS |
1490 | /* |
1491 | * Initialize @new's hva range. Do this even when replacing an @old | |
1492 | * slot, kvm_copy_memslot() deliberately does not touch node data. | |
1493 | */ | |
1494 | new->hva_node[idx].start = new->userspace_addr; | |
1495 | new->hva_node[idx].last = new->userspace_addr + | |
1496 | (new->npages << PAGE_SHIFT) - 1; | |
1497 | ||
1498 | /* | |
1499 | * (Re)Add the new memslot. There is no O(1) interval_tree_replace(), | |
1500 | * hva_node needs to be swapped with remove+insert even though hva can't | |
1501 | * change when replacing an existing slot. | |
1502 | */ | |
1503 | hash_add(slots->id_hash, &new->id_node[idx], new->id); | |
1504 | interval_tree_insert(&new->hva_node[idx], &slots->hva_tree); | |
1505 | ||
1506 | /* | |
1507 | * If the memslot gfn is unchanged, rb_replace_node() can be used to | |
1508 | * switch the node in the gfn tree instead of removing the old and | |
1509 | * inserting the new as two separate operations. Replacement is a | |
1510 | * single O(1) operation versus two O(log(n)) operations for | |
1511 | * remove+insert. | |
1512 | */ | |
1513 | if (old && old->base_gfn == new->base_gfn) { | |
1514 | kvm_replace_gfn_node(slots, old, new); | |
1515 | } else { | |
1516 | if (old) | |
1517 | kvm_erase_gfn_node(slots, old); | |
1518 | kvm_insert_gfn_node(slots, new); | |
0577d1ab | 1519 | } |
bf3e05bc XG |
1520 | } |
1521 | ||
09170a49 | 1522 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 1523 | { |
4d8b81ab XG |
1524 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1525 | ||
0f8a4de3 | 1526 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
1527 | valid_flags |= KVM_MEM_READONLY; |
1528 | #endif | |
1529 | ||
1530 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1531 | return -EINVAL; |
1532 | ||
1533 | return 0; | |
1534 | } | |
1535 | ||
a54d8066 | 1536 | static void kvm_swap_active_memslots(struct kvm *kvm, int as_id) |
7ec4fb44 | 1537 | { |
a54d8066 MS |
1538 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); |
1539 | ||
1540 | /* Grab the generation from the activate memslots. */ | |
1541 | u64 gen = __kvm_memslots(kvm, as_id)->generation; | |
7ec4fb44 | 1542 | |
361209e0 SC |
1543 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1544 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1545 | |
52ac8b35 PB |
1546 | /* |
1547 | * Do not store the new memslots while there are invalidations in | |
071064f1 PB |
1548 | * progress, otherwise the locking in invalidate_range_start and |
1549 | * invalidate_range_end will be unbalanced. | |
52ac8b35 PB |
1550 | */ |
1551 | spin_lock(&kvm->mn_invalidate_lock); | |
1552 | prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
1553 | while (kvm->mn_active_invalidate_count) { | |
1554 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1555 | spin_unlock(&kvm->mn_invalidate_lock); | |
1556 | schedule(); | |
1557 | spin_lock(&kvm->mn_invalidate_lock); | |
1558 | } | |
1559 | finish_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
f481b069 | 1560 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
52ac8b35 | 1561 | spin_unlock(&kvm->mn_invalidate_lock); |
b10a038e BG |
1562 | |
1563 | /* | |
1564 | * Acquired in kvm_set_memslot. Must be released before synchronize | |
1565 | * SRCU below in order to avoid deadlock with another thread | |
1566 | * acquiring the slots_arch_lock in an srcu critical section. | |
1567 | */ | |
1568 | mutex_unlock(&kvm->slots_arch_lock); | |
1569 | ||
7ec4fb44 | 1570 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1571 | |
ee3d1570 | 1572 | /* |
361209e0 | 1573 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1574 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1575 | * the number of address spaces. This provides a unique and easily |
1576 | * identifiable generation number while the memslots are in flux. | |
1577 | */ | |
1578 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1579 | ||
1580 | /* | |
4bd518f1 PB |
1581 | * Generations must be unique even across address spaces. We do not need |
1582 | * a global counter for that, instead the generation space is evenly split | |
1583 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1584 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1585 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1586 | */ |
164bf7e5 | 1587 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 1588 | |
15248258 | 1589 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1590 | |
15248258 | 1591 | slots->generation = gen; |
7ec4fb44 GN |
1592 | } |
1593 | ||
07921665 SC |
1594 | static int kvm_prepare_memory_region(struct kvm *kvm, |
1595 | const struct kvm_memory_slot *old, | |
1596 | struct kvm_memory_slot *new, | |
1597 | enum kvm_mr_change change) | |
ddc12f2a | 1598 | { |
07921665 SC |
1599 | int r; |
1600 | ||
1601 | /* | |
1602 | * If dirty logging is disabled, nullify the bitmap; the old bitmap | |
1603 | * will be freed on "commit". If logging is enabled in both old and | |
1604 | * new, reuse the existing bitmap. If logging is enabled only in the | |
1605 | * new and KVM isn't using a ring buffer, allocate and initialize a | |
1606 | * new bitmap. | |
1607 | */ | |
244893fa SC |
1608 | if (change != KVM_MR_DELETE) { |
1609 | if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1610 | new->dirty_bitmap = NULL; | |
1611 | else if (old && old->dirty_bitmap) | |
1612 | new->dirty_bitmap = old->dirty_bitmap; | |
86bdf3eb | 1613 | else if (kvm_use_dirty_bitmap(kvm)) { |
244893fa SC |
1614 | r = kvm_alloc_dirty_bitmap(new); |
1615 | if (r) | |
1616 | return r; | |
1617 | ||
1618 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1619 | bitmap_set(new->dirty_bitmap, 0, new->npages); | |
1620 | } | |
07921665 SC |
1621 | } |
1622 | ||
1623 | r = kvm_arch_prepare_memory_region(kvm, old, new, change); | |
1624 | ||
1625 | /* Free the bitmap on failure if it was allocated above. */ | |
c87661f8 | 1626 | if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap)) |
07921665 SC |
1627 | kvm_destroy_dirty_bitmap(new); |
1628 | ||
1629 | return r; | |
ddc12f2a BG |
1630 | } |
1631 | ||
07921665 SC |
1632 | static void kvm_commit_memory_region(struct kvm *kvm, |
1633 | struct kvm_memory_slot *old, | |
1634 | const struct kvm_memory_slot *new, | |
1635 | enum kvm_mr_change change) | |
ddc12f2a | 1636 | { |
6c7b2202 PB |
1637 | int old_flags = old ? old->flags : 0; |
1638 | int new_flags = new ? new->flags : 0; | |
07921665 SC |
1639 | /* |
1640 | * Update the total number of memslot pages before calling the arch | |
1641 | * hook so that architectures can consume the result directly. | |
1642 | */ | |
1643 | if (change == KVM_MR_DELETE) | |
1644 | kvm->nr_memslot_pages -= old->npages; | |
1645 | else if (change == KVM_MR_CREATE) | |
1646 | kvm->nr_memslot_pages += new->npages; | |
1647 | ||
6c7b2202 PB |
1648 | if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES) { |
1649 | int change = (new_flags & KVM_MEM_LOG_DIRTY_PAGES) ? 1 : -1; | |
1650 | atomic_set(&kvm->nr_memslots_dirty_logging, | |
1651 | atomic_read(&kvm->nr_memslots_dirty_logging) + change); | |
1652 | } | |
1653 | ||
07921665 SC |
1654 | kvm_arch_commit_memory_region(kvm, old, new, change); |
1655 | ||
a54d8066 MS |
1656 | switch (change) { |
1657 | case KVM_MR_CREATE: | |
1658 | /* Nothing more to do. */ | |
1659 | break; | |
1660 | case KVM_MR_DELETE: | |
1661 | /* Free the old memslot and all its metadata. */ | |
1662 | kvm_free_memslot(kvm, old); | |
1663 | break; | |
1664 | case KVM_MR_MOVE: | |
1665 | case KVM_MR_FLAGS_ONLY: | |
1666 | /* | |
1667 | * Free the dirty bitmap as needed; the below check encompasses | |
1668 | * both the flags and whether a ring buffer is being used) | |
1669 | */ | |
1670 | if (old->dirty_bitmap && !new->dirty_bitmap) | |
1671 | kvm_destroy_dirty_bitmap(old); | |
1672 | ||
1673 | /* | |
1674 | * The final quirk. Free the detached, old slot, but only its | |
1675 | * memory, not any metadata. Metadata, including arch specific | |
1676 | * data, may be reused by @new. | |
1677 | */ | |
1678 | kfree(old); | |
1679 | break; | |
1680 | default: | |
1681 | BUG(); | |
1682 | } | |
ddc12f2a BG |
1683 | } |
1684 | ||
36947254 | 1685 | /* |
a54d8066 MS |
1686 | * Activate @new, which must be installed in the inactive slots by the caller, |
1687 | * by swapping the active slots and then propagating @new to @old once @old is | |
1688 | * unreachable and can be safely modified. | |
1689 | * | |
1690 | * With NULL @old this simply adds @new to @active (while swapping the sets). | |
1691 | * With NULL @new this simply removes @old from @active and frees it | |
1692 | * (while also swapping the sets). | |
36947254 | 1693 | */ |
a54d8066 MS |
1694 | static void kvm_activate_memslot(struct kvm *kvm, |
1695 | struct kvm_memory_slot *old, | |
1696 | struct kvm_memory_slot *new) | |
36947254 | 1697 | { |
a54d8066 | 1698 | int as_id = kvm_memslots_get_as_id(old, new); |
36947254 | 1699 | |
a54d8066 MS |
1700 | kvm_swap_active_memslots(kvm, as_id); |
1701 | ||
1702 | /* Propagate the new memslot to the now inactive memslots. */ | |
1703 | kvm_replace_memslot(kvm, old, new); | |
1704 | } | |
1705 | ||
1706 | static void kvm_copy_memslot(struct kvm_memory_slot *dest, | |
1707 | const struct kvm_memory_slot *src) | |
1708 | { | |
1709 | dest->base_gfn = src->base_gfn; | |
1710 | dest->npages = src->npages; | |
1711 | dest->dirty_bitmap = src->dirty_bitmap; | |
1712 | dest->arch = src->arch; | |
1713 | dest->userspace_addr = src->userspace_addr; | |
1714 | dest->flags = src->flags; | |
1715 | dest->id = src->id; | |
1716 | dest->as_id = src->as_id; | |
1717 | } | |
1718 | ||
1719 | static void kvm_invalidate_memslot(struct kvm *kvm, | |
1720 | struct kvm_memory_slot *old, | |
244893fa | 1721 | struct kvm_memory_slot *invalid_slot) |
a54d8066 | 1722 | { |
07921665 | 1723 | /* |
a54d8066 MS |
1724 | * Mark the current slot INVALID. As with all memslot modifications, |
1725 | * this must be done on an unreachable slot to avoid modifying the | |
1726 | * current slot in the active tree. | |
07921665 | 1727 | */ |
244893fa SC |
1728 | kvm_copy_memslot(invalid_slot, old); |
1729 | invalid_slot->flags |= KVM_MEMSLOT_INVALID; | |
1730 | kvm_replace_memslot(kvm, old, invalid_slot); | |
a54d8066 MS |
1731 | |
1732 | /* | |
1733 | * Activate the slot that is now marked INVALID, but don't propagate | |
1734 | * the slot to the now inactive slots. The slot is either going to be | |
1735 | * deleted or recreated as a new slot. | |
1736 | */ | |
1737 | kvm_swap_active_memslots(kvm, old->as_id); | |
1738 | ||
1739 | /* | |
1740 | * From this point no new shadow pages pointing to a deleted, or moved, | |
1741 | * memslot will be created. Validation of sp->gfn happens in: | |
1742 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1743 | * - kvm_is_visible_gfn (mmu_check_root) | |
1744 | */ | |
bcb63dcd | 1745 | kvm_arch_flush_shadow_memslot(kvm, old); |
683412cc | 1746 | kvm_arch_guest_memory_reclaimed(kvm); |
a54d8066 | 1747 | |
b0d23708 | 1748 | /* Was released by kvm_swap_active_memslots(), reacquire. */ |
a54d8066 MS |
1749 | mutex_lock(&kvm->slots_arch_lock); |
1750 | ||
1751 | /* | |
1752 | * Copy the arch-specific field of the newly-installed slot back to the | |
1753 | * old slot as the arch data could have changed between releasing | |
b0d23708 | 1754 | * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock |
a54d8066 MS |
1755 | * above. Writers are required to retrieve memslots *after* acquiring |
1756 | * slots_arch_lock, thus the active slot's data is guaranteed to be fresh. | |
1757 | */ | |
244893fa | 1758 | old->arch = invalid_slot->arch; |
a54d8066 MS |
1759 | } |
1760 | ||
1761 | static void kvm_create_memslot(struct kvm *kvm, | |
244893fa | 1762 | struct kvm_memory_slot *new) |
a54d8066 | 1763 | { |
244893fa SC |
1764 | /* Add the new memslot to the inactive set and activate. */ |
1765 | kvm_replace_memslot(kvm, NULL, new); | |
1766 | kvm_activate_memslot(kvm, NULL, new); | |
a54d8066 MS |
1767 | } |
1768 | ||
1769 | static void kvm_delete_memslot(struct kvm *kvm, | |
1770 | struct kvm_memory_slot *old, | |
1771 | struct kvm_memory_slot *invalid_slot) | |
1772 | { | |
1773 | /* | |
1774 | * Remove the old memslot (in the inactive memslots) by passing NULL as | |
244893fa | 1775 | * the "new" slot, and for the invalid version in the active slots. |
a54d8066 MS |
1776 | */ |
1777 | kvm_replace_memslot(kvm, old, NULL); | |
a54d8066 | 1778 | kvm_activate_memslot(kvm, invalid_slot, NULL); |
a54d8066 | 1779 | } |
36947254 | 1780 | |
244893fa SC |
1781 | static void kvm_move_memslot(struct kvm *kvm, |
1782 | struct kvm_memory_slot *old, | |
1783 | struct kvm_memory_slot *new, | |
1784 | struct kvm_memory_slot *invalid_slot) | |
a54d8066 | 1785 | { |
a54d8066 | 1786 | /* |
244893fa SC |
1787 | * Replace the old memslot in the inactive slots, and then swap slots |
1788 | * and replace the current INVALID with the new as well. | |
a54d8066 | 1789 | */ |
244893fa SC |
1790 | kvm_replace_memslot(kvm, old, new); |
1791 | kvm_activate_memslot(kvm, invalid_slot, new); | |
a54d8066 | 1792 | } |
36947254 | 1793 | |
a54d8066 MS |
1794 | static void kvm_update_flags_memslot(struct kvm *kvm, |
1795 | struct kvm_memory_slot *old, | |
244893fa | 1796 | struct kvm_memory_slot *new) |
a54d8066 MS |
1797 | { |
1798 | /* | |
1799 | * Similar to the MOVE case, but the slot doesn't need to be zapped as | |
1800 | * an intermediate step. Instead, the old memslot is simply replaced | |
1801 | * with a new, updated copy in both memslot sets. | |
1802 | */ | |
244893fa SC |
1803 | kvm_replace_memslot(kvm, old, new); |
1804 | kvm_activate_memslot(kvm, old, new); | |
36947254 SC |
1805 | } |
1806 | ||
cf47f50b | 1807 | static int kvm_set_memslot(struct kvm *kvm, |
a54d8066 | 1808 | struct kvm_memory_slot *old, |
ce5f0215 | 1809 | struct kvm_memory_slot *new, |
cf47f50b SC |
1810 | enum kvm_mr_change change) |
1811 | { | |
244893fa | 1812 | struct kvm_memory_slot *invalid_slot; |
cf47f50b SC |
1813 | int r; |
1814 | ||
b10a038e | 1815 | /* |
b0d23708 | 1816 | * Released in kvm_swap_active_memslots(). |
b10a038e | 1817 | * |
b0d23708 JM |
1818 | * Must be held from before the current memslots are copied until after |
1819 | * the new memslots are installed with rcu_assign_pointer, then | |
1820 | * released before the synchronize srcu in kvm_swap_active_memslots(). | |
b10a038e BG |
1821 | * |
1822 | * When modifying memslots outside of the slots_lock, must be held | |
1823 | * before reading the pointer to the current memslots until after all | |
1824 | * changes to those memslots are complete. | |
1825 | * | |
1826 | * These rules ensure that installing new memslots does not lose | |
1827 | * changes made to the previous memslots. | |
1828 | */ | |
1829 | mutex_lock(&kvm->slots_arch_lock); | |
1830 | ||
a54d8066 MS |
1831 | /* |
1832 | * Invalidate the old slot if it's being deleted or moved. This is | |
1833 | * done prior to actually deleting/moving the memslot to allow vCPUs to | |
1834 | * continue running by ensuring there are no mappings or shadow pages | |
1835 | * for the memslot when it is deleted/moved. Without pre-invalidation | |
1836 | * (and without a lock), a window would exist between effecting the | |
1837 | * delete/move and committing the changes in arch code where KVM or a | |
1838 | * guest could access a non-existent memslot. | |
244893fa SC |
1839 | * |
1840 | * Modifications are done on a temporary, unreachable slot. The old | |
1841 | * slot needs to be preserved in case a later step fails and the | |
1842 | * invalidation needs to be reverted. | |
a54d8066 | 1843 | */ |
cf47f50b | 1844 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
244893fa SC |
1845 | invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT); |
1846 | if (!invalid_slot) { | |
1847 | mutex_unlock(&kvm->slots_arch_lock); | |
1848 | return -ENOMEM; | |
1849 | } | |
1850 | kvm_invalidate_memslot(kvm, old, invalid_slot); | |
1851 | } | |
b10a038e | 1852 | |
a54d8066 MS |
1853 | r = kvm_prepare_memory_region(kvm, old, new, change); |
1854 | if (r) { | |
b10a038e | 1855 | /* |
a54d8066 MS |
1856 | * For DELETE/MOVE, revert the above INVALID change. No |
1857 | * modifications required since the original slot was preserved | |
1858 | * in the inactive slots. Changing the active memslots also | |
1859 | * release slots_arch_lock. | |
b10a038e | 1860 | */ |
244893fa SC |
1861 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
1862 | kvm_activate_memslot(kvm, invalid_slot, old); | |
1863 | kfree(invalid_slot); | |
1864 | } else { | |
a54d8066 | 1865 | mutex_unlock(&kvm->slots_arch_lock); |
244893fa | 1866 | } |
a54d8066 | 1867 | return r; |
cf47f50b SC |
1868 | } |
1869 | ||
bda44d84 | 1870 | /* |
a54d8066 MS |
1871 | * For DELETE and MOVE, the working slot is now active as the INVALID |
1872 | * version of the old slot. MOVE is particularly special as it reuses | |
1873 | * the old slot and returns a copy of the old slot (in working_slot). | |
1874 | * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the | |
1875 | * old slot is detached but otherwise preserved. | |
bda44d84 | 1876 | */ |
a54d8066 | 1877 | if (change == KVM_MR_CREATE) |
244893fa | 1878 | kvm_create_memslot(kvm, new); |
a54d8066 | 1879 | else if (change == KVM_MR_DELETE) |
244893fa | 1880 | kvm_delete_memslot(kvm, old, invalid_slot); |
a54d8066 | 1881 | else if (change == KVM_MR_MOVE) |
244893fa | 1882 | kvm_move_memslot(kvm, old, new, invalid_slot); |
a54d8066 | 1883 | else if (change == KVM_MR_FLAGS_ONLY) |
244893fa | 1884 | kvm_update_flags_memslot(kvm, old, new); |
a54d8066 MS |
1885 | else |
1886 | BUG(); | |
cf47f50b | 1887 | |
244893fa SC |
1888 | /* Free the temporary INVALID slot used for DELETE and MOVE. */ |
1889 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) | |
1890 | kfree(invalid_slot); | |
bda44d84 | 1891 | |
a54d8066 MS |
1892 | /* |
1893 | * No need to refresh new->arch, changes after dropping slots_arch_lock | |
a413a625 | 1894 | * will directly hit the final, active memslot. Architectures are |
a54d8066 MS |
1895 | * responsible for knowing that new->arch may be stale. |
1896 | */ | |
1897 | kvm_commit_memory_region(kvm, old, new, change); | |
cf47f50b | 1898 | |
cf47f50b | 1899 | return 0; |
cf47f50b SC |
1900 | } |
1901 | ||
44401a20 MS |
1902 | static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id, |
1903 | gfn_t start, gfn_t end) | |
5c0b4f3d | 1904 | { |
44401a20 | 1905 | struct kvm_memslot_iter iter; |
5c0b4f3d | 1906 | |
44401a20 MS |
1907 | kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) { |
1908 | if (iter.slot->id != id) | |
1909 | return true; | |
1910 | } | |
5c0b4f3d | 1911 | |
44401a20 | 1912 | return false; |
5c0b4f3d SC |
1913 | } |
1914 | ||
6aa8b732 AK |
1915 | /* |
1916 | * Allocate some memory and give it an address in the guest physical address | |
1917 | * space. | |
1918 | * | |
1919 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1920 | * |
02d5d55b | 1921 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1922 | */ |
f78e0e2e | 1923 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 1924 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 1925 | { |
244893fa | 1926 | struct kvm_memory_slot *old, *new; |
44401a20 | 1927 | struct kvm_memslots *slots; |
f64c0398 | 1928 | enum kvm_mr_change change; |
0f9bdef3 SC |
1929 | unsigned long npages; |
1930 | gfn_t base_gfn; | |
163da372 SC |
1931 | int as_id, id; |
1932 | int r; | |
6aa8b732 | 1933 | |
a50d64d6 XG |
1934 | r = check_memory_region_flags(mem); |
1935 | if (r) | |
71a4c30b | 1936 | return r; |
a50d64d6 | 1937 | |
f481b069 PB |
1938 | as_id = mem->slot >> 16; |
1939 | id = (u16)mem->slot; | |
1940 | ||
6aa8b732 | 1941 | /* General sanity checks */ |
6b285a55 SC |
1942 | if ((mem->memory_size & (PAGE_SIZE - 1)) || |
1943 | (mem->memory_size != (unsigned long)mem->memory_size)) | |
71a4c30b | 1944 | return -EINVAL; |
6aa8b732 | 1945 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 1946 | return -EINVAL; |
fa3d315a | 1947 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 1948 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 1949 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 1950 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 1951 | mem->memory_size)) |
71a4c30b | 1952 | return -EINVAL; |
f481b069 | 1953 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 1954 | return -EINVAL; |
6aa8b732 | 1955 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 1956 | return -EINVAL; |
0f9bdef3 SC |
1957 | if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES) |
1958 | return -EINVAL; | |
6aa8b732 | 1959 | |
44401a20 | 1960 | slots = __kvm_memslots(kvm, as_id); |
6aa8b732 | 1961 | |
5c0b4f3d | 1962 | /* |
7cd08553 SC |
1963 | * Note, the old memslot (and the pointer itself!) may be invalidated |
1964 | * and/or destroyed by kvm_set_memslot(). | |
5c0b4f3d | 1965 | */ |
44401a20 | 1966 | old = id_to_memslot(slots, id); |
163da372 | 1967 | |
47ea7d90 | 1968 | if (!mem->memory_size) { |
7cd08553 | 1969 | if (!old || !old->npages) |
47ea7d90 | 1970 | return -EINVAL; |
5c0b4f3d | 1971 | |
7cd08553 | 1972 | if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages)) |
47ea7d90 | 1973 | return -EIO; |
6aa8b732 | 1974 | |
244893fa | 1975 | return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE); |
47ea7d90 | 1976 | } |
5c0b4f3d | 1977 | |
0f9bdef3 SC |
1978 | base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT); |
1979 | npages = (mem->memory_size >> PAGE_SHIFT); | |
163da372 | 1980 | |
7cd08553 | 1981 | if (!old || !old->npages) { |
5c0b4f3d | 1982 | change = KVM_MR_CREATE; |
afa319a5 SC |
1983 | |
1984 | /* | |
1985 | * To simplify KVM internals, the total number of pages across | |
1986 | * all memslots must fit in an unsigned long. | |
1987 | */ | |
0f9bdef3 | 1988 | if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages) |
afa319a5 | 1989 | return -EINVAL; |
5c0b4f3d | 1990 | } else { /* Modify an existing slot. */ |
0f9bdef3 SC |
1991 | if ((mem->userspace_addr != old->userspace_addr) || |
1992 | (npages != old->npages) || | |
1993 | ((mem->flags ^ old->flags) & KVM_MEM_READONLY)) | |
71a4c30b | 1994 | return -EINVAL; |
09170a49 | 1995 | |
0f9bdef3 | 1996 | if (base_gfn != old->base_gfn) |
5c0b4f3d | 1997 | change = KVM_MR_MOVE; |
0f9bdef3 | 1998 | else if (mem->flags != old->flags) |
5c0b4f3d SC |
1999 | change = KVM_MR_FLAGS_ONLY; |
2000 | else /* Nothing to change. */ | |
2001 | return 0; | |
09170a49 | 2002 | } |
6aa8b732 | 2003 | |
44401a20 | 2004 | if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) && |
0f9bdef3 | 2005 | kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages)) |
44401a20 | 2006 | return -EEXIST; |
6aa8b732 | 2007 | |
244893fa SC |
2008 | /* Allocate a slot that will persist in the memslot. */ |
2009 | new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT); | |
2010 | if (!new) | |
2011 | return -ENOMEM; | |
3c9bd400 | 2012 | |
244893fa SC |
2013 | new->as_id = as_id; |
2014 | new->id = id; | |
2015 | new->base_gfn = base_gfn; | |
2016 | new->npages = npages; | |
2017 | new->flags = mem->flags; | |
2018 | new->userspace_addr = mem->userspace_addr; | |
6aa8b732 | 2019 | |
244893fa | 2020 | r = kvm_set_memslot(kvm, old, new, change); |
cf47f50b | 2021 | if (r) |
244893fa | 2022 | kfree(new); |
6aa8b732 | 2023 | return r; |
210c7c4d | 2024 | } |
f78e0e2e SY |
2025 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
2026 | ||
2027 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 2028 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
2029 | { |
2030 | int r; | |
2031 | ||
79fac95e | 2032 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 2033 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 2034 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
2035 | return r; |
2036 | } | |
210c7c4d IE |
2037 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
2038 | ||
7940876e SH |
2039 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
2040 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 2041 | { |
f481b069 | 2042 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 2043 | return -EINVAL; |
09170a49 | 2044 | |
47ae31e2 | 2045 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
2046 | } |
2047 | ||
0dff0846 | 2048 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
2049 | /** |
2050 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
2051 | * @kvm: pointer to kvm instance | |
2052 | * @log: slot id and address to which we copy the log | |
2053 | * @is_dirty: set to '1' if any dirty pages were found | |
2054 | * @memslot: set to the associated memslot, always valid on success | |
2055 | */ | |
2056 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
2057 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 2058 | { |
9f6b8029 | 2059 | struct kvm_memslots *slots; |
843574a3 | 2060 | int i, as_id, id; |
87bf6e7d | 2061 | unsigned long n; |
6aa8b732 AK |
2062 | unsigned long any = 0; |
2063 | ||
86bdf3eb GS |
2064 | /* Dirty ring tracking may be exclusive to dirty log tracking */ |
2065 | if (!kvm_use_dirty_bitmap(kvm)) | |
b2cc64c4 PX |
2066 | return -ENXIO; |
2067 | ||
2a49f61d SC |
2068 | *memslot = NULL; |
2069 | *is_dirty = 0; | |
2070 | ||
f481b069 PB |
2071 | as_id = log->slot >> 16; |
2072 | id = (u16)log->slot; | |
2073 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 2074 | return -EINVAL; |
6aa8b732 | 2075 | |
f481b069 | 2076 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 2077 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 2078 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 2079 | return -ENOENT; |
6aa8b732 | 2080 | |
2a49f61d SC |
2081 | kvm_arch_sync_dirty_log(kvm, *memslot); |
2082 | ||
2083 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 2084 | |
cd1a4a98 | 2085 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 2086 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 2087 | |
2a49f61d | 2088 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 2089 | return -EFAULT; |
6aa8b732 | 2090 | |
5bb064dc ZX |
2091 | if (any) |
2092 | *is_dirty = 1; | |
843574a3 | 2093 | return 0; |
6aa8b732 | 2094 | } |
2ba9f0d8 | 2095 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 2096 | |
0dff0846 | 2097 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 2098 | /** |
b8b00220 | 2099 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 2100 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
2101 | * @kvm: pointer to kvm instance |
2102 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
2103 | * |
2104 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
2105 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
2106 | * following order: | |
2107 | * | |
2108 | * 1. Take a snapshot of the bit and clear it if needed. | |
2109 | * 2. Write protect the corresponding page. | |
2110 | * 3. Copy the snapshot to the userspace. | |
2111 | * 4. Upon return caller flushes TLB's if needed. | |
2112 | * | |
2113 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
2114 | * entry. This is not a problem because the page is reported dirty using | |
2115 | * the snapshot taken before and step 4 ensures that writes done after | |
2116 | * exiting to userspace will be logged for the next call. | |
2117 | * | |
2118 | */ | |
0dff0846 | 2119 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 2120 | { |
9f6b8029 | 2121 | struct kvm_memslots *slots; |
ba0513b5 | 2122 | struct kvm_memory_slot *memslot; |
58d6db34 | 2123 | int i, as_id, id; |
ba0513b5 MS |
2124 | unsigned long n; |
2125 | unsigned long *dirty_bitmap; | |
2126 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2127 | bool flush; |
ba0513b5 | 2128 | |
86bdf3eb GS |
2129 | /* Dirty ring tracking may be exclusive to dirty log tracking */ |
2130 | if (!kvm_use_dirty_bitmap(kvm)) | |
b2cc64c4 PX |
2131 | return -ENXIO; |
2132 | ||
f481b069 PB |
2133 | as_id = log->slot >> 16; |
2134 | id = (u16)log->slot; | |
2135 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 2136 | return -EINVAL; |
ba0513b5 | 2137 | |
f481b069 PB |
2138 | slots = __kvm_memslots(kvm, as_id); |
2139 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2140 | if (!memslot || !memslot->dirty_bitmap) |
2141 | return -ENOENT; | |
ba0513b5 MS |
2142 | |
2143 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 2144 | |
0dff0846 SC |
2145 | kvm_arch_sync_dirty_log(kvm, memslot); |
2146 | ||
ba0513b5 | 2147 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 2148 | flush = false; |
2a31b9db PB |
2149 | if (kvm->manual_dirty_log_protect) { |
2150 | /* | |
2151 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
2152 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
2153 | * is some code duplication between this function and | |
2154 | * kvm_get_dirty_log, but hopefully all architecture | |
2155 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
2156 | * can be eliminated. | |
2157 | */ | |
2158 | dirty_bitmap_buffer = dirty_bitmap; | |
2159 | } else { | |
2160 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
2161 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 2162 | |
531810ca | 2163 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
2164 | for (i = 0; i < n / sizeof(long); i++) { |
2165 | unsigned long mask; | |
2166 | gfn_t offset; | |
ba0513b5 | 2167 | |
2a31b9db PB |
2168 | if (!dirty_bitmap[i]) |
2169 | continue; | |
2170 | ||
0dff0846 | 2171 | flush = true; |
2a31b9db PB |
2172 | mask = xchg(&dirty_bitmap[i], 0); |
2173 | dirty_bitmap_buffer[i] = mask; | |
2174 | ||
a67794ca LT |
2175 | offset = i * BITS_PER_LONG; |
2176 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
2177 | offset, mask); | |
2a31b9db | 2178 | } |
531810ca | 2179 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
2180 | } |
2181 | ||
0dff0846 SC |
2182 | if (flush) |
2183 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2184 | ||
2a31b9db PB |
2185 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
2186 | return -EFAULT; | |
2187 | return 0; | |
2188 | } | |
0dff0846 SC |
2189 | |
2190 | ||
2191 | /** | |
2192 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
2193 | * @kvm: kvm instance | |
2194 | * @log: slot id and address to which we copy the log | |
2195 | * | |
2196 | * Steps 1-4 below provide general overview of dirty page logging. See | |
2197 | * kvm_get_dirty_log_protect() function description for additional details. | |
2198 | * | |
2199 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
2200 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
2201 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
2202 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
2203 | * writes will be marked dirty for next log read. | |
2204 | * | |
2205 | * 1. Take a snapshot of the bit and clear it if needed. | |
2206 | * 2. Write protect the corresponding page. | |
2207 | * 3. Copy the snapshot to the userspace. | |
2208 | * 4. Flush TLB's if needed. | |
2209 | */ | |
2210 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
2211 | struct kvm_dirty_log *log) | |
2212 | { | |
2213 | int r; | |
2214 | ||
2215 | mutex_lock(&kvm->slots_lock); | |
2216 | ||
2217 | r = kvm_get_dirty_log_protect(kvm, log); | |
2218 | ||
2219 | mutex_unlock(&kvm->slots_lock); | |
2220 | return r; | |
2221 | } | |
2a31b9db PB |
2222 | |
2223 | /** | |
2224 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
2225 | * and reenable dirty page tracking for the corresponding pages. | |
2226 | * @kvm: pointer to kvm instance | |
2227 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
2228 | */ | |
0dff0846 SC |
2229 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
2230 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
2231 | { |
2232 | struct kvm_memslots *slots; | |
2233 | struct kvm_memory_slot *memslot; | |
98938aa8 | 2234 | int as_id, id; |
2a31b9db | 2235 | gfn_t offset; |
98938aa8 | 2236 | unsigned long i, n; |
2a31b9db PB |
2237 | unsigned long *dirty_bitmap; |
2238 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2239 | bool flush; |
2a31b9db | 2240 | |
86bdf3eb GS |
2241 | /* Dirty ring tracking may be exclusive to dirty log tracking */ |
2242 | if (!kvm_use_dirty_bitmap(kvm)) | |
b2cc64c4 PX |
2243 | return -ENXIO; |
2244 | ||
2a31b9db PB |
2245 | as_id = log->slot >> 16; |
2246 | id = (u16)log->slot; | |
2247 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
2248 | return -EINVAL; | |
2249 | ||
76d58e0f | 2250 | if (log->first_page & 63) |
2a31b9db PB |
2251 | return -EINVAL; |
2252 | ||
2253 | slots = __kvm_memslots(kvm, as_id); | |
2254 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2255 | if (!memslot || !memslot->dirty_bitmap) |
2256 | return -ENOENT; | |
2a31b9db PB |
2257 | |
2258 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 2259 | |
4ddc9204 | 2260 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
2261 | |
2262 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
2263 | log->num_pages > memslot->npages - log->first_page || |
2264 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
2265 | return -EINVAL; | |
98938aa8 | 2266 | |
0dff0846 SC |
2267 | kvm_arch_sync_dirty_log(kvm, memslot); |
2268 | ||
2269 | flush = false; | |
2a31b9db PB |
2270 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
2271 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
2272 | return -EFAULT; | |
ba0513b5 | 2273 | |
531810ca | 2274 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
2275 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
2276 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
2277 | i++, offset += BITS_PER_LONG) { |
2278 | unsigned long mask = *dirty_bitmap_buffer++; | |
2279 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
2280 | if (!mask) | |
ba0513b5 MS |
2281 | continue; |
2282 | ||
2a31b9db | 2283 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 2284 | |
2a31b9db PB |
2285 | /* |
2286 | * mask contains the bits that really have been cleared. This | |
2287 | * never includes any bits beyond the length of the memslot (if | |
2288 | * the length is not aligned to 64 pages), therefore it is not | |
2289 | * a problem if userspace sets them in log->dirty_bitmap. | |
2290 | */ | |
58d2930f | 2291 | if (mask) { |
0dff0846 | 2292 | flush = true; |
58d2930f TY |
2293 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
2294 | offset, mask); | |
2295 | } | |
ba0513b5 | 2296 | } |
531810ca | 2297 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 2298 | |
0dff0846 SC |
2299 | if (flush) |
2300 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2301 | ||
58d6db34 | 2302 | return 0; |
ba0513b5 | 2303 | } |
0dff0846 SC |
2304 | |
2305 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
2306 | struct kvm_clear_dirty_log *log) | |
2307 | { | |
2308 | int r; | |
2309 | ||
2310 | mutex_lock(&kvm->slots_lock); | |
2311 | ||
2312 | r = kvm_clear_dirty_log_protect(kvm, log); | |
2313 | ||
2314 | mutex_unlock(&kvm->slots_lock); | |
2315 | return r; | |
2316 | } | |
2317 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 2318 | |
49c7754c GN |
2319 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
2320 | { | |
2321 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
2322 | } | |
a1f4d395 | 2323 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 2324 | |
8e73485c PB |
2325 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
2326 | { | |
fe22ed82 | 2327 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
a54d8066 | 2328 | u64 gen = slots->generation; |
fe22ed82 | 2329 | struct kvm_memory_slot *slot; |
fe22ed82 | 2330 | |
a54d8066 MS |
2331 | /* |
2332 | * This also protects against using a memslot from a different address space, | |
2333 | * since different address spaces have different generation numbers. | |
2334 | */ | |
2335 | if (unlikely(gen != vcpu->last_used_slot_gen)) { | |
2336 | vcpu->last_used_slot = NULL; | |
2337 | vcpu->last_used_slot_gen = gen; | |
2338 | } | |
2339 | ||
2340 | slot = try_get_memslot(vcpu->last_used_slot, gfn); | |
fe22ed82 DM |
2341 | if (slot) |
2342 | return slot; | |
2343 | ||
2344 | /* | |
2345 | * Fall back to searching all memslots. We purposely use | |
2346 | * search_memslots() instead of __gfn_to_memslot() to avoid | |
a54d8066 | 2347 | * thrashing the VM-wide last_used_slot in kvm_memslots. |
fe22ed82 | 2348 | */ |
a54d8066 | 2349 | slot = search_memslots(slots, gfn, false); |
fe22ed82 | 2350 | if (slot) { |
a54d8066 | 2351 | vcpu->last_used_slot = slot; |
fe22ed82 DM |
2352 | return slot; |
2353 | } | |
2354 | ||
2355 | return NULL; | |
8e73485c PB |
2356 | } |
2357 | ||
33e94154 | 2358 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 2359 | { |
bf3e05bc | 2360 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 2361 | |
c36b7150 | 2362 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
2363 | } |
2364 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
2365 | ||
995decb6 VK |
2366 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
2367 | { | |
2368 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2369 | ||
2370 | return kvm_is_visible_memslot(memslot); | |
2371 | } | |
2372 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
2373 | ||
f9b84e19 | 2374 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
2375 | { |
2376 | struct vm_area_struct *vma; | |
2377 | unsigned long addr, size; | |
2378 | ||
2379 | size = PAGE_SIZE; | |
2380 | ||
42cde48b | 2381 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
2382 | if (kvm_is_error_hva(addr)) |
2383 | return PAGE_SIZE; | |
2384 | ||
d8ed45c5 | 2385 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
2386 | vma = find_vma(current->mm, addr); |
2387 | if (!vma) | |
2388 | goto out; | |
2389 | ||
2390 | size = vma_kernel_pagesize(vma); | |
2391 | ||
2392 | out: | |
d8ed45c5 | 2393 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
2394 | |
2395 | return size; | |
2396 | } | |
2397 | ||
8283e36a | 2398 | static bool memslot_is_readonly(const struct kvm_memory_slot *slot) |
4d8b81ab XG |
2399 | { |
2400 | return slot->flags & KVM_MEM_READONLY; | |
2401 | } | |
2402 | ||
8283e36a | 2403 | static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn, |
4d8b81ab | 2404 | gfn_t *nr_pages, bool write) |
539cb660 | 2405 | { |
bc6678a3 | 2406 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 2407 | return KVM_HVA_ERR_BAD; |
48987781 | 2408 | |
4d8b81ab XG |
2409 | if (memslot_is_readonly(slot) && write) |
2410 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
2411 | |
2412 | if (nr_pages) | |
2413 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
2414 | ||
4d8b81ab | 2415 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 2416 | } |
48987781 | 2417 | |
4d8b81ab XG |
2418 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2419 | gfn_t *nr_pages) | |
2420 | { | |
2421 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 2422 | } |
48987781 | 2423 | |
4d8b81ab | 2424 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 2425 | gfn_t gfn) |
4d8b81ab XG |
2426 | { |
2427 | return gfn_to_hva_many(slot, gfn, NULL); | |
2428 | } | |
2429 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
2430 | ||
48987781 XG |
2431 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
2432 | { | |
49c7754c | 2433 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 2434 | } |
0d150298 | 2435 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 2436 | |
8e73485c PB |
2437 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
2438 | { | |
2439 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
2440 | } | |
2441 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
2442 | ||
86ab8cff | 2443 | /* |
970c0d4b WY |
2444 | * Return the hva of a @gfn and the R/W attribute if possible. |
2445 | * | |
2446 | * @slot: the kvm_memory_slot which contains @gfn | |
2447 | * @gfn: the gfn to be translated | |
2448 | * @writable: used to return the read/write attribute of the @slot if the hva | |
2449 | * is valid and @writable is not NULL | |
86ab8cff | 2450 | */ |
64d83126 CD |
2451 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
2452 | gfn_t gfn, bool *writable) | |
86ab8cff | 2453 | { |
a2ac07fe GN |
2454 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
2455 | ||
2456 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
2457 | *writable = !memslot_is_readonly(slot); |
2458 | ||
a2ac07fe | 2459 | return hva; |
86ab8cff XG |
2460 | } |
2461 | ||
64d83126 CD |
2462 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
2463 | { | |
2464 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2465 | ||
2466 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2467 | } | |
2468 | ||
8e73485c PB |
2469 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
2470 | { | |
2471 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2472 | ||
2473 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2474 | } | |
2475 | ||
fafc3dba HY |
2476 | static inline int check_user_page_hwpoison(unsigned long addr) |
2477 | { | |
0d731759 | 2478 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 2479 | |
0d731759 | 2480 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
2481 | return rc == -EHWPOISON; |
2482 | } | |
2483 | ||
2fc84311 | 2484 | /* |
b9b33da2 PB |
2485 | * The fast path to get the writable pfn which will be stored in @pfn, |
2486 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 2487 | * only part that runs if we can in atomic context. |
2fc84311 | 2488 | */ |
b9b33da2 PB |
2489 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
2490 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 2491 | { |
8d4e1288 | 2492 | struct page *page[1]; |
954bbbc2 | 2493 | |
12ce13fe XG |
2494 | /* |
2495 | * Fast pin a writable pfn only if it is a write fault request | |
2496 | * or the caller allows to map a writable pfn for a read fault | |
2497 | * request. | |
2498 | */ | |
2499 | if (!(write_fault || writable)) | |
2500 | return false; | |
612819c3 | 2501 | |
dadbb612 | 2502 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 2503 | *pfn = page_to_pfn(page[0]); |
612819c3 | 2504 | |
2fc84311 XG |
2505 | if (writable) |
2506 | *writable = true; | |
2507 | return true; | |
2508 | } | |
af585b92 | 2509 | |
2fc84311 XG |
2510 | return false; |
2511 | } | |
612819c3 | 2512 | |
2fc84311 XG |
2513 | /* |
2514 | * The slow path to get the pfn of the specified host virtual address, | |
2515 | * 1 indicates success, -errno is returned if error is detected. | |
2516 | */ | |
2517 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
c8b88b33 | 2518 | bool interruptible, bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 2519 | { |
ce53053c AV |
2520 | unsigned int flags = FOLL_HWPOISON; |
2521 | struct page *page; | |
28249139 | 2522 | int npages; |
612819c3 | 2523 | |
2fc84311 XG |
2524 | might_sleep(); |
2525 | ||
2526 | if (writable) | |
2527 | *writable = write_fault; | |
2528 | ||
ce53053c AV |
2529 | if (write_fault) |
2530 | flags |= FOLL_WRITE; | |
2531 | if (async) | |
2532 | flags |= FOLL_NOWAIT; | |
c8b88b33 PX |
2533 | if (interruptible) |
2534 | flags |= FOLL_INTERRUPTIBLE; | |
d4944b0e | 2535 | |
ce53053c | 2536 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
2537 | if (npages != 1) |
2538 | return npages; | |
2539 | ||
2540 | /* map read fault as writable if possible */ | |
12ce13fe | 2541 | if (unlikely(!write_fault) && writable) { |
ce53053c | 2542 | struct page *wpage; |
2fc84311 | 2543 | |
dadbb612 | 2544 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 2545 | *writable = true; |
ce53053c AV |
2546 | put_page(page); |
2547 | page = wpage; | |
612819c3 | 2548 | } |
887c08ac | 2549 | } |
ce53053c | 2550 | *pfn = page_to_pfn(page); |
2fc84311 XG |
2551 | return npages; |
2552 | } | |
539cb660 | 2553 | |
4d8b81ab XG |
2554 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
2555 | { | |
2556 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
2557 | return false; | |
2e2e3738 | 2558 | |
4d8b81ab XG |
2559 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
2560 | return false; | |
887c08ac | 2561 | |
4d8b81ab XG |
2562 | return true; |
2563 | } | |
bf998156 | 2564 | |
f8be156b NP |
2565 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
2566 | { | |
b14b2690 SC |
2567 | struct page *page = kvm_pfn_to_refcounted_page(pfn); |
2568 | ||
2569 | if (!page) | |
f8be156b | 2570 | return 1; |
b14b2690 SC |
2571 | |
2572 | return get_page_unless_zero(page); | |
f8be156b NP |
2573 | } |
2574 | ||
92176a8e | 2575 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1625566e XT |
2576 | unsigned long addr, bool write_fault, |
2577 | bool *writable, kvm_pfn_t *p_pfn) | |
92176a8e | 2578 | { |
a9545779 | 2579 | kvm_pfn_t pfn; |
bd2fae8d PB |
2580 | pte_t *ptep; |
2581 | spinlock_t *ptl; | |
add6a0cd PB |
2582 | int r; |
2583 | ||
9fd6dad1 | 2584 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2585 | if (r) { |
2586 | /* | |
2587 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
2588 | * not call the fault handler, so do it here. | |
2589 | */ | |
2590 | bool unlocked = false; | |
64019a2e | 2591 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
2592 | (write_fault ? FAULT_FLAG_WRITE : 0), |
2593 | &unlocked); | |
a8387d0b PB |
2594 | if (unlocked) |
2595 | return -EAGAIN; | |
add6a0cd PB |
2596 | if (r) |
2597 | return r; | |
2598 | ||
9fd6dad1 | 2599 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2600 | if (r) |
2601 | return r; | |
bd2fae8d | 2602 | } |
add6a0cd | 2603 | |
bd2fae8d PB |
2604 | if (write_fault && !pte_write(*ptep)) { |
2605 | pfn = KVM_PFN_ERR_RO_FAULT; | |
2606 | goto out; | |
add6a0cd PB |
2607 | } |
2608 | ||
a340b3e2 | 2609 | if (writable) |
bd2fae8d PB |
2610 | *writable = pte_write(*ptep); |
2611 | pfn = pte_pfn(*ptep); | |
add6a0cd PB |
2612 | |
2613 | /* | |
2614 | * Get a reference here because callers of *hva_to_pfn* and | |
2615 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
2616 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
36c3ce6c | 2617 | * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will |
add6a0cd PB |
2618 | * simply do nothing for reserved pfns. |
2619 | * | |
2620 | * Whoever called remap_pfn_range is also going to call e.g. | |
2621 | * unmap_mapping_range before the underlying pages are freed, | |
2622 | * causing a call to our MMU notifier. | |
f8be156b NP |
2623 | * |
2624 | * Certain IO or PFNMAP mappings can be backed with valid | |
2625 | * struct pages, but be allocated without refcounting e.g., | |
2626 | * tail pages of non-compound higher order allocations, which | |
2627 | * would then underflow the refcount when the caller does the | |
2628 | * required put_page. Don't allow those pages here. | |
add6a0cd | 2629 | */ |
f8be156b NP |
2630 | if (!kvm_try_get_pfn(pfn)) |
2631 | r = -EFAULT; | |
add6a0cd | 2632 | |
bd2fae8d PB |
2633 | out: |
2634 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 2635 | *p_pfn = pfn; |
f8be156b NP |
2636 | |
2637 | return r; | |
92176a8e PB |
2638 | } |
2639 | ||
12ce13fe XG |
2640 | /* |
2641 | * Pin guest page in memory and return its pfn. | |
2642 | * @addr: host virtual address which maps memory to the guest | |
2643 | * @atomic: whether this function can sleep | |
c8b88b33 | 2644 | * @interruptible: whether the process can be interrupted by non-fatal signals |
12ce13fe XG |
2645 | * @async: whether this function need to wait IO complete if the |
2646 | * host page is not in the memory | |
2647 | * @write_fault: whether we should get a writable host page | |
2648 | * @writable: whether it allows to map a writable host page for !@write_fault | |
2649 | * | |
2650 | * The function will map a writable host page for these two cases: | |
2651 | * 1): @write_fault = true | |
2652 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
2653 | * whether the mapping is writable. | |
2654 | */ | |
c8b88b33 PX |
2655 | kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool interruptible, |
2656 | bool *async, bool write_fault, bool *writable) | |
2fc84311 XG |
2657 | { |
2658 | struct vm_area_struct *vma; | |
943dfea8 | 2659 | kvm_pfn_t pfn; |
92176a8e | 2660 | int npages, r; |
2e2e3738 | 2661 | |
2fc84311 XG |
2662 | /* we can do it either atomically or asynchronously, not both */ |
2663 | BUG_ON(atomic && async); | |
8d4e1288 | 2664 | |
b9b33da2 | 2665 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
2666 | return pfn; |
2667 | ||
2668 | if (atomic) | |
2669 | return KVM_PFN_ERR_FAULT; | |
2670 | ||
c8b88b33 PX |
2671 | npages = hva_to_pfn_slow(addr, async, write_fault, interruptible, |
2672 | writable, &pfn); | |
2fc84311 XG |
2673 | if (npages == 1) |
2674 | return pfn; | |
fe5ed56c PX |
2675 | if (npages == -EINTR) |
2676 | return KVM_PFN_ERR_SIGPENDING; | |
8d4e1288 | 2677 | |
d8ed45c5 | 2678 | mmap_read_lock(current->mm); |
2fc84311 XG |
2679 | if (npages == -EHWPOISON || |
2680 | (!async && check_user_page_hwpoison(addr))) { | |
2681 | pfn = KVM_PFN_ERR_HWPOISON; | |
2682 | goto exit; | |
2683 | } | |
2684 | ||
a8387d0b | 2685 | retry: |
fc98c03b | 2686 | vma = vma_lookup(current->mm, addr); |
2fc84311 XG |
2687 | |
2688 | if (vma == NULL) | |
2689 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 2690 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
1625566e | 2691 | r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn); |
a8387d0b PB |
2692 | if (r == -EAGAIN) |
2693 | goto retry; | |
92176a8e PB |
2694 | if (r < 0) |
2695 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 2696 | } else { |
4d8b81ab | 2697 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
2698 | *async = true; |
2699 | pfn = KVM_PFN_ERR_FAULT; | |
2700 | } | |
2701 | exit: | |
d8ed45c5 | 2702 | mmap_read_unlock(current->mm); |
2e2e3738 | 2703 | return pfn; |
35149e21 AL |
2704 | } |
2705 | ||
8283e36a | 2706 | kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn, |
c8b88b33 PX |
2707 | bool atomic, bool interruptible, bool *async, |
2708 | bool write_fault, bool *writable, hva_t *hva) | |
887c08ac | 2709 | { |
4d8b81ab XG |
2710 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
2711 | ||
4a42d848 DS |
2712 | if (hva) |
2713 | *hva = addr; | |
2714 | ||
b2740d35 PB |
2715 | if (addr == KVM_HVA_ERR_RO_BAD) { |
2716 | if (writable) | |
2717 | *writable = false; | |
4d8b81ab | 2718 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 2719 | } |
4d8b81ab | 2720 | |
b2740d35 PB |
2721 | if (kvm_is_error_hva(addr)) { |
2722 | if (writable) | |
2723 | *writable = false; | |
81c52c56 | 2724 | return KVM_PFN_NOSLOT; |
b2740d35 | 2725 | } |
4d8b81ab XG |
2726 | |
2727 | /* Do not map writable pfn in the readonly memslot. */ | |
2728 | if (writable && memslot_is_readonly(slot)) { | |
2729 | *writable = false; | |
2730 | writable = NULL; | |
2731 | } | |
2732 | ||
c8b88b33 | 2733 | return hva_to_pfn(addr, atomic, interruptible, async, write_fault, |
4d8b81ab | 2734 | writable); |
887c08ac | 2735 | } |
3520469d | 2736 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 2737 | |
ba049e93 | 2738 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
2739 | bool *writable) |
2740 | { | |
c8b88b33 PX |
2741 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, false, |
2742 | NULL, write_fault, writable, NULL); | |
612819c3 MT |
2743 | } |
2744 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
2745 | ||
8283e36a | 2746 | kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2747 | { |
c8b88b33 PX |
2748 | return __gfn_to_pfn_memslot(slot, gfn, false, false, NULL, true, |
2749 | NULL, NULL); | |
506f0d6f | 2750 | } |
e37afc6e | 2751 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 2752 | |
8283e36a | 2753 | kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2754 | { |
c8b88b33 PX |
2755 | return __gfn_to_pfn_memslot(slot, gfn, true, false, NULL, true, |
2756 | NULL, NULL); | |
506f0d6f | 2757 | } |
037d92dc | 2758 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 2759 | |
ba049e93 | 2760 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2761 | { |
2762 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2763 | } | |
2764 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
2765 | ||
ba049e93 | 2766 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
2767 | { |
2768 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
2769 | } | |
2770 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
2771 | ||
ba049e93 | 2772 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2773 | { |
2774 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2775 | } | |
2776 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
2777 | ||
d9ef13c2 PB |
2778 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2779 | struct page **pages, int nr_pages) | |
48987781 XG |
2780 | { |
2781 | unsigned long addr; | |
076b925d | 2782 | gfn_t entry = 0; |
48987781 | 2783 | |
d9ef13c2 | 2784 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
2785 | if (kvm_is_error_hva(addr)) |
2786 | return -1; | |
2787 | ||
2788 | if (entry < nr_pages) | |
2789 | return 0; | |
2790 | ||
dadbb612 | 2791 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
2792 | } |
2793 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
2794 | ||
b1624f99 SC |
2795 | /* |
2796 | * Do not use this helper unless you are absolutely certain the gfn _must_ be | |
2797 | * backed by 'struct page'. A valid example is if the backing memslot is | |
2798 | * controlled by KVM. Note, if the returned page is valid, it's refcount has | |
2799 | * been elevated by gfn_to_pfn(). | |
2800 | */ | |
35149e21 AL |
2801 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
2802 | { | |
b14b2690 | 2803 | struct page *page; |
ba049e93 | 2804 | kvm_pfn_t pfn; |
2e2e3738 AL |
2805 | |
2806 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 2807 | |
81c52c56 | 2808 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 2809 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2810 | |
b14b2690 SC |
2811 | page = kvm_pfn_to_refcounted_page(pfn); |
2812 | if (!page) | |
6cede2e6 | 2813 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2814 | |
b14b2690 | 2815 | return page; |
954bbbc2 AK |
2816 | } |
2817 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
2818 | ||
357a18ad | 2819 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty) |
91724814 | 2820 | { |
91724814 BO |
2821 | if (dirty) |
2822 | kvm_release_pfn_dirty(pfn); | |
2823 | else | |
2824 | kvm_release_pfn_clean(pfn); | |
2825 | } | |
2826 | ||
357a18ad | 2827 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
e45adf66 KA |
2828 | { |
2829 | kvm_pfn_t pfn; | |
2830 | void *hva = NULL; | |
2831 | struct page *page = KVM_UNMAPPED_PAGE; | |
2832 | ||
2833 | if (!map) | |
2834 | return -EINVAL; | |
2835 | ||
357a18ad | 2836 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
e45adf66 KA |
2837 | if (is_error_noslot_pfn(pfn)) |
2838 | return -EINVAL; | |
2839 | ||
2840 | if (pfn_valid(pfn)) { | |
2841 | page = pfn_to_page(pfn); | |
357a18ad | 2842 | hva = kmap(page); |
d30b214d | 2843 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2844 | } else { |
357a18ad | 2845 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
d30b214d | 2846 | #endif |
e45adf66 KA |
2847 | } |
2848 | ||
2849 | if (!hva) | |
2850 | return -EFAULT; | |
2851 | ||
2852 | map->page = page; | |
2853 | map->hva = hva; | |
2854 | map->pfn = pfn; | |
2855 | map->gfn = gfn; | |
2856 | ||
2857 | return 0; | |
2858 | } | |
e45adf66 KA |
2859 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); |
2860 | ||
357a18ad | 2861 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) |
e45adf66 KA |
2862 | { |
2863 | if (!map) | |
2864 | return; | |
2865 | ||
2866 | if (!map->hva) | |
2867 | return; | |
2868 | ||
357a18ad DW |
2869 | if (map->page != KVM_UNMAPPED_PAGE) |
2870 | kunmap(map->page); | |
eb1f2f38 | 2871 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2872 | else |
357a18ad | 2873 | memunmap(map->hva); |
eb1f2f38 | 2874 | #endif |
e45adf66 | 2875 | |
91724814 | 2876 | if (dirty) |
357a18ad | 2877 | kvm_vcpu_mark_page_dirty(vcpu, map->gfn); |
91724814 | 2878 | |
357a18ad | 2879 | kvm_release_pfn(map->pfn, dirty); |
e45adf66 KA |
2880 | |
2881 | map->hva = NULL; | |
2882 | map->page = NULL; | |
2883 | } | |
2884 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); | |
2885 | ||
8e1c6914 | 2886 | static bool kvm_is_ad_tracked_page(struct page *page) |
8e73485c | 2887 | { |
8e1c6914 SC |
2888 | /* |
2889 | * Per page-flags.h, pages tagged PG_reserved "should in general not be | |
2890 | * touched (e.g. set dirty) except by its owner". | |
2891 | */ | |
2892 | return !PageReserved(page); | |
2893 | } | |
8e73485c | 2894 | |
8e1c6914 SC |
2895 | static void kvm_set_page_dirty(struct page *page) |
2896 | { | |
2897 | if (kvm_is_ad_tracked_page(page)) | |
2898 | SetPageDirty(page); | |
2899 | } | |
8e73485c | 2900 | |
8e1c6914 SC |
2901 | static void kvm_set_page_accessed(struct page *page) |
2902 | { | |
2903 | if (kvm_is_ad_tracked_page(page)) | |
2904 | mark_page_accessed(page); | |
8e73485c | 2905 | } |
8e73485c | 2906 | |
b4231d61 IE |
2907 | void kvm_release_page_clean(struct page *page) |
2908 | { | |
32cad84f XG |
2909 | WARN_ON(is_error_page(page)); |
2910 | ||
8e1c6914 SC |
2911 | kvm_set_page_accessed(page); |
2912 | put_page(page); | |
b4231d61 IE |
2913 | } |
2914 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
2915 | ||
ba049e93 | 2916 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 2917 | { |
b14b2690 SC |
2918 | struct page *page; |
2919 | ||
2920 | if (is_error_noslot_pfn(pfn)) | |
2921 | return; | |
2922 | ||
2923 | page = kvm_pfn_to_refcounted_page(pfn); | |
2924 | if (!page) | |
2925 | return; | |
2926 | ||
2927 | kvm_release_page_clean(page); | |
35149e21 AL |
2928 | } |
2929 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
2930 | ||
b4231d61 | 2931 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 2932 | { |
a2766325 XG |
2933 | WARN_ON(is_error_page(page)); |
2934 | ||
8e1c6914 SC |
2935 | kvm_set_page_dirty(page); |
2936 | kvm_release_page_clean(page); | |
35149e21 AL |
2937 | } |
2938 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
2939 | ||
f7a6509f | 2940 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2941 | { |
b14b2690 SC |
2942 | struct page *page; |
2943 | ||
2944 | if (is_error_noslot_pfn(pfn)) | |
2945 | return; | |
2946 | ||
2947 | page = kvm_pfn_to_refcounted_page(pfn); | |
2948 | if (!page) | |
2949 | return; | |
2950 | ||
2951 | kvm_release_page_dirty(page); | |
35149e21 | 2952 | } |
f7a6509f | 2953 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 2954 | |
8e1c6914 SC |
2955 | /* |
2956 | * Note, checking for an error/noslot pfn is the caller's responsibility when | |
2957 | * directly marking a page dirty/accessed. Unlike the "release" helpers, the | |
2958 | * "set" helpers are not to be used when the pfn might point at garbage. | |
2959 | */ | |
ba049e93 | 2960 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2961 | { |
8e1c6914 SC |
2962 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
2963 | return; | |
2964 | ||
2965 | if (pfn_valid(pfn)) | |
2966 | kvm_set_page_dirty(pfn_to_page(pfn)); | |
8a7ae055 | 2967 | } |
35149e21 AL |
2968 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
2969 | ||
ba049e93 | 2970 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 2971 | { |
8e1c6914 SC |
2972 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
2973 | return; | |
2974 | ||
2975 | if (pfn_valid(pfn)) | |
2976 | kvm_set_page_accessed(pfn_to_page(pfn)); | |
35149e21 AL |
2977 | } |
2978 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
2979 | ||
195aefde IE |
2980 | static int next_segment(unsigned long len, int offset) |
2981 | { | |
2982 | if (len > PAGE_SIZE - offset) | |
2983 | return PAGE_SIZE - offset; | |
2984 | else | |
2985 | return len; | |
2986 | } | |
2987 | ||
8e73485c PB |
2988 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
2989 | void *data, int offset, int len) | |
195aefde | 2990 | { |
e0506bcb IE |
2991 | int r; |
2992 | unsigned long addr; | |
195aefde | 2993 | |
8e73485c | 2994 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
2995 | if (kvm_is_error_hva(addr)) |
2996 | return -EFAULT; | |
3180a7fc | 2997 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 2998 | if (r) |
195aefde | 2999 | return -EFAULT; |
195aefde IE |
3000 | return 0; |
3001 | } | |
8e73485c PB |
3002 | |
3003 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
3004 | int len) | |
3005 | { | |
3006 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
3007 | ||
3008 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
3009 | } | |
195aefde IE |
3010 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
3011 | ||
8e73485c PB |
3012 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
3013 | int offset, int len) | |
3014 | { | |
3015 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3016 | ||
3017 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
3018 | } | |
3019 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
3020 | ||
195aefde IE |
3021 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
3022 | { | |
3023 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3024 | int seg; | |
3025 | int offset = offset_in_page(gpa); | |
3026 | int ret; | |
3027 | ||
3028 | while ((seg = next_segment(len, offset)) != 0) { | |
3029 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
3030 | if (ret < 0) | |
3031 | return ret; | |
3032 | offset = 0; | |
3033 | len -= seg; | |
3034 | data += seg; | |
3035 | ++gfn; | |
3036 | } | |
3037 | return 0; | |
3038 | } | |
3039 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
3040 | ||
8e73485c | 3041 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 3042 | { |
7ec54588 | 3043 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 3044 | int seg; |
7ec54588 | 3045 | int offset = offset_in_page(gpa); |
8e73485c PB |
3046 | int ret; |
3047 | ||
3048 | while ((seg = next_segment(len, offset)) != 0) { | |
3049 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
3050 | if (ret < 0) | |
3051 | return ret; | |
3052 | offset = 0; | |
3053 | len -= seg; | |
3054 | data += seg; | |
3055 | ++gfn; | |
3056 | } | |
3057 | return 0; | |
3058 | } | |
3059 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 3060 | |
8e73485c PB |
3061 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
3062 | void *data, int offset, unsigned long len) | |
3063 | { | |
3064 | int r; | |
3065 | unsigned long addr; | |
3066 | ||
3067 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
3068 | if (kvm_is_error_hva(addr)) |
3069 | return -EFAULT; | |
0aac03f0 | 3070 | pagefault_disable(); |
3180a7fc | 3071 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 3072 | pagefault_enable(); |
7ec54588 MT |
3073 | if (r) |
3074 | return -EFAULT; | |
3075 | return 0; | |
3076 | } | |
7ec54588 | 3077 | |
8e73485c PB |
3078 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
3079 | void *data, unsigned long len) | |
3080 | { | |
3081 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3082 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3083 | int offset = offset_in_page(gpa); | |
3084 | ||
3085 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
3086 | } | |
3087 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
3088 | ||
28bd726a PX |
3089 | static int __kvm_write_guest_page(struct kvm *kvm, |
3090 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 3091 | const void *data, int offset, int len) |
195aefde | 3092 | { |
e0506bcb IE |
3093 | int r; |
3094 | unsigned long addr; | |
195aefde | 3095 | |
251eb841 | 3096 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
3097 | if (kvm_is_error_hva(addr)) |
3098 | return -EFAULT; | |
8b0cedff | 3099 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 3100 | if (r) |
195aefde | 3101 | return -EFAULT; |
28bd726a | 3102 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
3103 | return 0; |
3104 | } | |
8e73485c PB |
3105 | |
3106 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
3107 | const void *data, int offset, int len) | |
3108 | { | |
3109 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
3110 | ||
28bd726a | 3111 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 3112 | } |
195aefde IE |
3113 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
3114 | ||
8e73485c PB |
3115 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3116 | const void *data, int offset, int len) | |
3117 | { | |
3118 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3119 | ||
28bd726a | 3120 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
3121 | } |
3122 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
3123 | ||
195aefde IE |
3124 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
3125 | unsigned long len) | |
3126 | { | |
3127 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3128 | int seg; | |
3129 | int offset = offset_in_page(gpa); | |
3130 | int ret; | |
3131 | ||
3132 | while ((seg = next_segment(len, offset)) != 0) { | |
3133 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
3134 | if (ret < 0) | |
3135 | return ret; | |
3136 | offset = 0; | |
3137 | len -= seg; | |
3138 | data += seg; | |
3139 | ++gfn; | |
3140 | } | |
3141 | return 0; | |
3142 | } | |
ff651cb6 | 3143 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 3144 | |
8e73485c PB |
3145 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
3146 | unsigned long len) | |
3147 | { | |
3148 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3149 | int seg; | |
3150 | int offset = offset_in_page(gpa); | |
3151 | int ret; | |
3152 | ||
3153 | while ((seg = next_segment(len, offset)) != 0) { | |
3154 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
3155 | if (ret < 0) | |
3156 | return ret; | |
3157 | offset = 0; | |
3158 | len -= seg; | |
3159 | data += seg; | |
3160 | ++gfn; | |
3161 | } | |
3162 | return 0; | |
3163 | } | |
3164 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
3165 | ||
5a2d4365 PB |
3166 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
3167 | struct gfn_to_hva_cache *ghc, | |
3168 | gpa_t gpa, unsigned long len) | |
49c7754c | 3169 | { |
49c7754c | 3170 | int offset = offset_in_page(gpa); |
8f964525 AH |
3171 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
3172 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
3173 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
3174 | gfn_t nr_pages_avail; | |
49c7754c | 3175 | |
6ad1e29f | 3176 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 3177 | ghc->generation = slots->generation; |
6ad1e29f SC |
3178 | |
3179 | if (start_gfn > end_gfn) { | |
3180 | ghc->hva = KVM_HVA_ERR_BAD; | |
3181 | return -EINVAL; | |
3182 | } | |
f1b9dd5e JM |
3183 | |
3184 | /* | |
3185 | * If the requested region crosses two memslots, we still | |
3186 | * verify that the entire region is valid here. | |
3187 | */ | |
6ad1e29f | 3188 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
3189 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
3190 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
3191 | &nr_pages_avail); | |
3192 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 3193 | return -EFAULT; |
f1b9dd5e JM |
3194 | } |
3195 | ||
3196 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 3197 | if (nr_pages_needed == 1) |
49c7754c | 3198 | ghc->hva += offset; |
f1b9dd5e | 3199 | else |
8f964525 | 3200 | ghc->memslot = NULL; |
f1b9dd5e | 3201 | |
6ad1e29f SC |
3202 | ghc->gpa = gpa; |
3203 | ghc->len = len; | |
3204 | return 0; | |
49c7754c | 3205 | } |
5a2d4365 | 3206 | |
4e335d9e | 3207 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
3208 | gpa_t gpa, unsigned long len) |
3209 | { | |
4e335d9e | 3210 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
3211 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
3212 | } | |
4e335d9e | 3213 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 3214 | |
4e335d9e | 3215 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
3216 | void *data, unsigned int offset, |
3217 | unsigned long len) | |
49c7754c | 3218 | { |
4e335d9e | 3219 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 3220 | int r; |
4ec6e863 | 3221 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 3222 | |
5f25e71e PB |
3223 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3224 | return -EINVAL; | |
8f964525 | 3225 | |
dc9ce71e SC |
3226 | if (slots->generation != ghc->generation) { |
3227 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3228 | return -EFAULT; | |
3229 | } | |
8f964525 | 3230 | |
49c7754c GN |
3231 | if (kvm_is_error_hva(ghc->hva)) |
3232 | return -EFAULT; | |
3233 | ||
fcfbc617 SC |
3234 | if (unlikely(!ghc->memslot)) |
3235 | return kvm_write_guest(kvm, gpa, data, len); | |
3236 | ||
4ec6e863 | 3237 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
3238 | if (r) |
3239 | return -EFAULT; | |
28bd726a | 3240 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
3241 | |
3242 | return 0; | |
3243 | } | |
4e335d9e | 3244 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 3245 | |
4e335d9e PB |
3246 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3247 | void *data, unsigned long len) | |
4ec6e863 | 3248 | { |
4e335d9e | 3249 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 3250 | } |
4e335d9e | 3251 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 3252 | |
0958f0ce VK |
3253 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3254 | void *data, unsigned int offset, | |
3255 | unsigned long len) | |
e03b644f | 3256 | { |
4e335d9e | 3257 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 3258 | int r; |
0958f0ce | 3259 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 3260 | |
5f25e71e PB |
3261 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3262 | return -EINVAL; | |
8f964525 | 3263 | |
dc9ce71e SC |
3264 | if (slots->generation != ghc->generation) { |
3265 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3266 | return -EFAULT; | |
3267 | } | |
8f964525 | 3268 | |
e03b644f GN |
3269 | if (kvm_is_error_hva(ghc->hva)) |
3270 | return -EFAULT; | |
3271 | ||
fcfbc617 | 3272 | if (unlikely(!ghc->memslot)) |
0958f0ce | 3273 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 3274 | |
0958f0ce | 3275 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
3276 | if (r) |
3277 | return -EFAULT; | |
3278 | ||
3279 | return 0; | |
3280 | } | |
0958f0ce VK |
3281 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
3282 | ||
3283 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
3284 | void *data, unsigned long len) | |
3285 | { | |
3286 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
3287 | } | |
4e335d9e | 3288 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 3289 | |
195aefde IE |
3290 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
3291 | { | |
2f541442 | 3292 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
3293 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3294 | int seg; | |
3295 | int offset = offset_in_page(gpa); | |
3296 | int ret; | |
3297 | ||
bfda0e84 | 3298 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 3299 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
3300 | if (ret < 0) |
3301 | return ret; | |
3302 | offset = 0; | |
3303 | len -= seg; | |
3304 | ++gfn; | |
3305 | } | |
3306 | return 0; | |
3307 | } | |
3308 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
3309 | ||
28bd726a | 3310 | void mark_page_dirty_in_slot(struct kvm *kvm, |
8283e36a | 3311 | const struct kvm_memory_slot *memslot, |
28bd726a | 3312 | gfn_t gfn) |
6aa8b732 | 3313 | { |
2efd61a6 DW |
3314 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); |
3315 | ||
e09fccb5 | 3316 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
86bdf3eb | 3317 | if (WARN_ON_ONCE(vcpu && vcpu->kvm != kvm)) |
2efd61a6 | 3318 | return; |
86bdf3eb | 3319 | |
c57351a7 | 3320 | WARN_ON_ONCE(!vcpu && !kvm_arch_allow_write_without_running_vcpu(kvm)); |
e09fccb5 | 3321 | #endif |
2efd61a6 | 3322 | |
044c59c4 | 3323 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 3324 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 3325 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 3326 | |
86bdf3eb | 3327 | if (kvm->dirty_ring_size && vcpu) |
cf87ac73 | 3328 | kvm_dirty_ring_push(vcpu, slot, rel_gfn); |
c57351a7 | 3329 | else if (memslot->dirty_bitmap) |
fb04a1ed | 3330 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
3331 | } |
3332 | } | |
a6a0b05d | 3333 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 3334 | |
49c7754c GN |
3335 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
3336 | { | |
3337 | struct kvm_memory_slot *memslot; | |
3338 | ||
3339 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 3340 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 3341 | } |
2ba9f0d8 | 3342 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 3343 | |
8e73485c PB |
3344 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
3345 | { | |
3346 | struct kvm_memory_slot *memslot; | |
3347 | ||
3348 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 3349 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
3350 | } |
3351 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
3352 | ||
20b7035c JS |
3353 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
3354 | { | |
3355 | if (!vcpu->sigset_active) | |
3356 | return; | |
3357 | ||
3358 | /* | |
3359 | * This does a lockless modification of ->real_blocked, which is fine | |
3360 | * because, only current can change ->real_blocked and all readers of | |
3361 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
3362 | * of ->blocked. | |
3363 | */ | |
3364 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
3365 | } | |
3366 | ||
3367 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
3368 | { | |
3369 | if (!vcpu->sigset_active) | |
3370 | return; | |
3371 | ||
3372 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
3373 | sigemptyset(¤t->real_blocked); | |
3374 | } | |
3375 | ||
aca6ff29 WL |
3376 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
3377 | { | |
dee339b5 | 3378 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 3379 | |
2cbd7824 | 3380 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 3381 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3382 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
3383 | if (!grow) |
3384 | goto out; | |
3385 | ||
dee339b5 NW |
3386 | val *= grow; |
3387 | if (val < grow_start) | |
3388 | val = grow_start; | |
aca6ff29 WL |
3389 | |
3390 | vcpu->halt_poll_ns = val; | |
7fa08e71 | 3391 | out: |
2cbd7824 | 3392 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3393 | } |
3394 | ||
3395 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
3396 | { | |
ae232ea4 | 3397 | unsigned int old, val, shrink, grow_start; |
aca6ff29 | 3398 | |
2cbd7824 | 3399 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 3400 | shrink = READ_ONCE(halt_poll_ns_shrink); |
ae232ea4 | 3401 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3402 | if (shrink == 0) |
aca6ff29 WL |
3403 | val = 0; |
3404 | else | |
6b6de68c | 3405 | val /= shrink; |
aca6ff29 | 3406 | |
ae232ea4 SS |
3407 | if (val < grow_start) |
3408 | val = 0; | |
3409 | ||
aca6ff29 | 3410 | vcpu->halt_poll_ns = val; |
2cbd7824 | 3411 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3412 | } |
3413 | ||
f7819512 PB |
3414 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
3415 | { | |
50c28f21 JS |
3416 | int ret = -EINTR; |
3417 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
3418 | ||
c59fb127 | 3419 | if (kvm_arch_vcpu_runnable(vcpu)) |
50c28f21 | 3420 | goto out; |
f7819512 | 3421 | if (kvm_cpu_has_pending_timer(vcpu)) |
50c28f21 | 3422 | goto out; |
f7819512 | 3423 | if (signal_pending(current)) |
50c28f21 | 3424 | goto out; |
084071d5 MT |
3425 | if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu)) |
3426 | goto out; | |
f7819512 | 3427 | |
50c28f21 JS |
3428 | ret = 0; |
3429 | out: | |
3430 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
3431 | return ret; | |
f7819512 PB |
3432 | } |
3433 | ||
fac42688 SC |
3434 | /* |
3435 | * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is | |
3436 | * pending. This is mostly used when halting a vCPU, but may also be used | |
3437 | * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI. | |
3438 | */ | |
3439 | bool kvm_vcpu_block(struct kvm_vcpu *vcpu) | |
cb953129 | 3440 | { |
fac42688 SC |
3441 | struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); |
3442 | bool waited = false; | |
3443 | ||
c3858335 JZ |
3444 | vcpu->stat.generic.blocking = 1; |
3445 | ||
18869f26 | 3446 | preempt_disable(); |
fac42688 | 3447 | kvm_arch_vcpu_blocking(vcpu); |
fac42688 | 3448 | prepare_to_rcuwait(wait); |
18869f26 ML |
3449 | preempt_enable(); |
3450 | ||
fac42688 SC |
3451 | for (;;) { |
3452 | set_current_state(TASK_INTERRUPTIBLE); | |
3453 | ||
3454 | if (kvm_vcpu_check_block(vcpu) < 0) | |
3455 | break; | |
3456 | ||
3457 | waited = true; | |
3458 | schedule(); | |
3459 | } | |
fac42688 | 3460 | |
18869f26 ML |
3461 | preempt_disable(); |
3462 | finish_rcuwait(wait); | |
fac42688 | 3463 | kvm_arch_vcpu_unblocking(vcpu); |
18869f26 | 3464 | preempt_enable(); |
fac42688 | 3465 | |
c3858335 JZ |
3466 | vcpu->stat.generic.blocking = 0; |
3467 | ||
fac42688 SC |
3468 | return waited; |
3469 | } | |
3470 | ||
29e72893 SC |
3471 | static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start, |
3472 | ktime_t end, bool success) | |
cb953129 | 3473 | { |
30c94347 | 3474 | struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic; |
29e72893 SC |
3475 | u64 poll_ns = ktime_to_ns(ktime_sub(end, start)); |
3476 | ||
30c94347 SC |
3477 | ++vcpu->stat.generic.halt_attempted_poll; |
3478 | ||
3479 | if (success) { | |
3480 | ++vcpu->stat.generic.halt_successful_poll; | |
3481 | ||
3482 | if (!vcpu_valid_wakeup(vcpu)) | |
3483 | ++vcpu->stat.generic.halt_poll_invalid; | |
3484 | ||
3485 | stats->halt_poll_success_ns += poll_ns; | |
3486 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns); | |
3487 | } else { | |
3488 | stats->halt_poll_fail_ns += poll_ns; | |
3489 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns); | |
3490 | } | |
cb953129 DM |
3491 | } |
3492 | ||
175d5dc7 DM |
3493 | static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu *vcpu) |
3494 | { | |
9eb8ca04 DM |
3495 | struct kvm *kvm = vcpu->kvm; |
3496 | ||
3497 | if (kvm->override_halt_poll_ns) { | |
3498 | /* | |
3499 | * Ensure kvm->max_halt_poll_ns is not read before | |
3500 | * kvm->override_halt_poll_ns. | |
3501 | * | |
3502 | * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL. | |
3503 | */ | |
3504 | smp_rmb(); | |
3505 | return READ_ONCE(kvm->max_halt_poll_ns); | |
3506 | } | |
3507 | ||
3508 | return READ_ONCE(halt_poll_ns); | |
175d5dc7 DM |
3509 | } |
3510 | ||
b6958ce4 | 3511 | /* |
fac42688 SC |
3512 | * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt |
3513 | * polling is enabled, busy wait for a short time before blocking to avoid the | |
3514 | * expensive block+unblock sequence if a wake event arrives soon after the vCPU | |
3515 | * is halted. | |
b6958ce4 | 3516 | */ |
91b99ea7 | 3517 | void kvm_vcpu_halt(struct kvm_vcpu *vcpu) |
d3bef15f | 3518 | { |
175d5dc7 | 3519 | unsigned int max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu); |
6f390916 | 3520 | bool halt_poll_allowed = !kvm_arch_no_poll(vcpu); |
cb953129 | 3521 | ktime_t start, cur, poll_end; |
f7819512 | 3522 | bool waited = false; |
97b6847a | 3523 | bool do_halt_poll; |
91b99ea7 | 3524 | u64 halt_ns; |
07ab0f8d | 3525 | |
175d5dc7 DM |
3526 | if (vcpu->halt_poll_ns > max_halt_poll_ns) |
3527 | vcpu->halt_poll_ns = max_halt_poll_ns; | |
97b6847a DM |
3528 | |
3529 | do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns; | |
3530 | ||
cb953129 | 3531 | start = cur = poll_end = ktime_get(); |
8df6a61c | 3532 | if (do_halt_poll) { |
109a9826 | 3533 | ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns); |
f95ef0cd | 3534 | |
f7819512 | 3535 | do { |
30c94347 | 3536 | if (kvm_vcpu_check_block(vcpu) < 0) |
f7819512 | 3537 | goto out; |
74775654 | 3538 | cpu_relax(); |
cb953129 | 3539 | poll_end = cur = ktime_get(); |
6bd5b743 | 3540 | } while (kvm_vcpu_can_poll(cur, stop)); |
f7819512 | 3541 | } |
e5c239cf | 3542 | |
fac42688 | 3543 | waited = kvm_vcpu_block(vcpu); |
8ccba534 | 3544 | |
f7819512 | 3545 | cur = ktime_get(); |
87bcc5fa JZ |
3546 | if (waited) { |
3547 | vcpu->stat.generic.halt_wait_ns += | |
3548 | ktime_to_ns(cur) - ktime_to_ns(poll_end); | |
8ccba534 JZ |
3549 | KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist, |
3550 | ktime_to_ns(cur) - ktime_to_ns(poll_end)); | |
87bcc5fa | 3551 | } |
f7819512 | 3552 | out: |
91b99ea7 SC |
3553 | /* The total time the vCPU was "halted", including polling time. */ |
3554 | halt_ns = ktime_to_ns(cur) - ktime_to_ns(start); | |
aca6ff29 | 3555 | |
29e72893 SC |
3556 | /* |
3557 | * Note, halt-polling is considered successful so long as the vCPU was | |
3558 | * never actually scheduled out, i.e. even if the wake event arrived | |
3559 | * after of the halt-polling loop itself, but before the full wait. | |
3560 | */ | |
8df6a61c | 3561 | if (do_halt_poll) |
29e72893 | 3562 | update_halt_poll_stats(vcpu, start, poll_end, !waited); |
cb953129 | 3563 | |
6f390916 | 3564 | if (halt_poll_allowed) { |
175d5dc7 DM |
3565 | /* Recompute the max halt poll time in case it changed. */ |
3566 | max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu); | |
3567 | ||
44551b2f | 3568 | if (!vcpu_valid_wakeup(vcpu)) { |
aca6ff29 | 3569 | shrink_halt_poll_ns(vcpu); |
175d5dc7 | 3570 | } else if (max_halt_poll_ns) { |
91b99ea7 | 3571 | if (halt_ns <= vcpu->halt_poll_ns) |
44551b2f WL |
3572 | ; |
3573 | /* we had a long block, shrink polling */ | |
acd05785 | 3574 | else if (vcpu->halt_poll_ns && |
175d5dc7 | 3575 | halt_ns > max_halt_poll_ns) |
44551b2f WL |
3576 | shrink_halt_poll_ns(vcpu); |
3577 | /* we had a short halt and our poll time is too small */ | |
175d5dc7 DM |
3578 | else if (vcpu->halt_poll_ns < max_halt_poll_ns && |
3579 | halt_ns < max_halt_poll_ns) | |
44551b2f WL |
3580 | grow_halt_poll_ns(vcpu); |
3581 | } else { | |
3582 | vcpu->halt_poll_ns = 0; | |
3583 | } | |
3584 | } | |
aca6ff29 | 3585 | |
91b99ea7 | 3586 | trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu)); |
b6958ce4 | 3587 | } |
91b99ea7 | 3588 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
b6958ce4 | 3589 | |
178f02ff | 3590 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 3591 | { |
d92a5d1c | 3592 | if (__kvm_vcpu_wake_up(vcpu)) { |
d73eb57b | 3593 | WRITE_ONCE(vcpu->ready, true); |
0193cc90 | 3594 | ++vcpu->stat.generic.halt_wakeup; |
178f02ff | 3595 | return true; |
b6d33834 CD |
3596 | } |
3597 | ||
178f02ff | 3598 | return false; |
dd1a4cc1 RK |
3599 | } |
3600 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
3601 | ||
0266c894 | 3602 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
3603 | /* |
3604 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
3605 | */ | |
3606 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
3607 | { | |
85b64045 | 3608 | int me, cpu; |
dd1a4cc1 | 3609 | |
178f02ff RK |
3610 | if (kvm_vcpu_wake_up(vcpu)) |
3611 | return; | |
3612 | ||
aefdc2ed PB |
3613 | me = get_cpu(); |
3614 | /* | |
3615 | * The only state change done outside the vcpu mutex is IN_GUEST_MODE | |
3616 | * to EXITING_GUEST_MODE. Therefore the moderately expensive "should | |
3617 | * kick" check does not need atomic operations if kvm_vcpu_kick is used | |
3618 | * within the vCPU thread itself. | |
3619 | */ | |
3620 | if (vcpu == __this_cpu_read(kvm_running_vcpu)) { | |
3621 | if (vcpu->mode == IN_GUEST_MODE) | |
3622 | WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE); | |
3623 | goto out; | |
3624 | } | |
3625 | ||
85b64045 SC |
3626 | /* |
3627 | * Note, the vCPU could get migrated to a different pCPU at any point | |
3628 | * after kvm_arch_vcpu_should_kick(), which could result in sending an | |
3629 | * IPI to the previous pCPU. But, that's ok because the purpose of the | |
3630 | * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the | |
3631 | * vCPU also requires it to leave IN_GUEST_MODE. | |
3632 | */ | |
85b64045 SC |
3633 | if (kvm_arch_vcpu_should_kick(vcpu)) { |
3634 | cpu = READ_ONCE(vcpu->cpu); | |
3635 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
b6d33834 | 3636 | smp_send_reschedule(cpu); |
85b64045 | 3637 | } |
aefdc2ed | 3638 | out: |
b6d33834 CD |
3639 | put_cpu(); |
3640 | } | |
a20ed54d | 3641 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 3642 | #endif /* !CONFIG_S390 */ |
b6d33834 | 3643 | |
fa93384f | 3644 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
3645 | { |
3646 | struct pid *pid; | |
3647 | struct task_struct *task = NULL; | |
fa93384f | 3648 | int ret = 0; |
41628d33 KW |
3649 | |
3650 | rcu_read_lock(); | |
3651 | pid = rcu_dereference(target->pid); | |
3652 | if (pid) | |
27fbe64b | 3653 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
3654 | rcu_read_unlock(); |
3655 | if (!task) | |
c45c528e | 3656 | return ret; |
c45c528e | 3657 | ret = yield_to(task, 1); |
41628d33 | 3658 | put_task_struct(task); |
c45c528e R |
3659 | |
3660 | return ret; | |
41628d33 KW |
3661 | } |
3662 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
3663 | ||
06e48c51 R |
3664 | /* |
3665 | * Helper that checks whether a VCPU is eligible for directed yield. | |
3666 | * Most eligible candidate to yield is decided by following heuristics: | |
3667 | * | |
3668 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
3669 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 3670 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
3671 | * |
3672 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
3673 | * chance last time (mostly it has become eligible now since we have probably | |
3674 | * yielded to lockholder in last iteration. This is done by toggling | |
3675 | * @dy_eligible each time a VCPU checked for eligibility.) | |
3676 | * | |
3677 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
3678 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
3679 | * burning. Giving priority for a potential lock-holder increases lock | |
3680 | * progress. | |
3681 | * | |
3682 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
3683 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
3684 | * and continue with next VCPU and so on. | |
3685 | */ | |
7940876e | 3686 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 3687 | { |
4a55dd72 | 3688 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
3689 | bool eligible; |
3690 | ||
3691 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 3692 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
3693 | |
3694 | if (vcpu->spin_loop.in_spin_loop) | |
3695 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
3696 | ||
3697 | return eligible; | |
4a55dd72 SW |
3698 | #else |
3699 | return true; | |
06e48c51 | 3700 | #endif |
4a55dd72 | 3701 | } |
c45c528e | 3702 | |
17e433b5 WL |
3703 | /* |
3704 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
3705 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
3706 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
3707 | */ | |
3708 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
3709 | { | |
3710 | return kvm_arch_vcpu_runnable(vcpu); | |
3711 | } | |
3712 | ||
3713 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
3714 | { | |
3715 | if (kvm_arch_dy_runnable(vcpu)) | |
3716 | return true; | |
3717 | ||
3718 | #ifdef CONFIG_KVM_ASYNC_PF | |
3719 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
3720 | return true; | |
3721 | #endif | |
3722 | ||
3723 | return false; | |
3724 | } | |
3725 | ||
52acd22f WL |
3726 | bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
3727 | { | |
3728 | return false; | |
3729 | } | |
3730 | ||
199b5763 | 3731 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 3732 | { |
217ece61 RR |
3733 | struct kvm *kvm = me->kvm; |
3734 | struct kvm_vcpu *vcpu; | |
3735 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
46808a4c | 3736 | unsigned long i; |
217ece61 | 3737 | int yielded = 0; |
c45c528e | 3738 | int try = 3; |
217ece61 | 3739 | int pass; |
d255f4f2 | 3740 | |
4c088493 | 3741 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
3742 | /* |
3743 | * We boost the priority of a VCPU that is runnable but not | |
3744 | * currently running, because it got preempted by something | |
3745 | * else and called schedule in __vcpu_run. Hopefully that | |
3746 | * VCPU is holding the lock that we need and will release it. | |
3747 | * We approximate round-robin by starting at the last boosted VCPU. | |
3748 | */ | |
c45c528e | 3749 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 3750 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 3751 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
3752 | i = last_boosted_vcpu; |
3753 | continue; | |
3754 | } else if (pass && i > last_boosted_vcpu) | |
3755 | break; | |
d73eb57b | 3756 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 3757 | continue; |
217ece61 RR |
3758 | if (vcpu == me) |
3759 | continue; | |
d92a5d1c | 3760 | if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu)) |
217ece61 | 3761 | continue; |
046ddeed | 3762 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
52acd22f WL |
3763 | !kvm_arch_dy_has_pending_interrupt(vcpu) && |
3764 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 3765 | continue; |
06e48c51 R |
3766 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
3767 | continue; | |
c45c528e R |
3768 | |
3769 | yielded = kvm_vcpu_yield_to(vcpu); | |
3770 | if (yielded > 0) { | |
217ece61 | 3771 | kvm->last_boosted_vcpu = i; |
217ece61 | 3772 | break; |
c45c528e R |
3773 | } else if (yielded < 0) { |
3774 | try--; | |
3775 | if (!try) | |
3776 | break; | |
217ece61 | 3777 | } |
217ece61 RR |
3778 | } |
3779 | } | |
4c088493 | 3780 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
3781 | |
3782 | /* Ensure vcpu is not eligible during next spinloop */ | |
3783 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
3784 | } |
3785 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
3786 | ||
fb04a1ed PX |
3787 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
3788 | { | |
dc70ec21 | 3789 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
fb04a1ed PX |
3790 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && |
3791 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
3792 | kvm->dirty_ring_size / PAGE_SIZE); | |
3793 | #else | |
3794 | return false; | |
3795 | #endif | |
3796 | } | |
3797 | ||
1499fa80 | 3798 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 3799 | { |
11bac800 | 3800 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
3801 | struct page *page; |
3802 | ||
e4a533a4 | 3803 | if (vmf->pgoff == 0) |
039576c0 | 3804 | page = virt_to_page(vcpu->run); |
09566765 | 3805 | #ifdef CONFIG_X86 |
e4a533a4 | 3806 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 3807 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 3808 | #endif |
4b4357e0 | 3809 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3810 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
3811 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 3812 | #endif |
fb04a1ed PX |
3813 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
3814 | page = kvm_dirty_ring_get_page( | |
3815 | &vcpu->dirty_ring, | |
3816 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 3817 | else |
5b1c1493 | 3818 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 3819 | get_page(page); |
e4a533a4 | 3820 | vmf->page = page; |
3821 | return 0; | |
9a2bb7f4 AK |
3822 | } |
3823 | ||
f0f37e2f | 3824 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 3825 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
3826 | }; |
3827 | ||
3828 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
3829 | { | |
fb04a1ed | 3830 | struct kvm_vcpu *vcpu = file->private_data; |
11476d27 | 3831 | unsigned long pages = vma_pages(vma); |
fb04a1ed PX |
3832 | |
3833 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
3834 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
3835 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
3836 | return -EINVAL; | |
3837 | ||
9a2bb7f4 AK |
3838 | vma->vm_ops = &kvm_vcpu_vm_ops; |
3839 | return 0; | |
3840 | } | |
3841 | ||
bccf2150 AK |
3842 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
3843 | { | |
3844 | struct kvm_vcpu *vcpu = filp->private_data; | |
3845 | ||
66c0b394 | 3846 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
3847 | return 0; |
3848 | } | |
3849 | ||
70375c2d | 3850 | static const struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
3851 | .release = kvm_vcpu_release, |
3852 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 3853 | .mmap = kvm_vcpu_mmap, |
6038f373 | 3854 | .llseek = noop_llseek, |
7ddfd3e0 | 3855 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
3856 | }; |
3857 | ||
3858 | /* | |
3859 | * Allocates an inode for the vcpu. | |
3860 | */ | |
3861 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
3862 | { | |
e46b4692 MY |
3863 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
3864 | ||
3865 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
3866 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
3867 | } |
3868 | ||
e36de87d VP |
3869 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
3870 | static int vcpu_get_pid(void *data, u64 *val) | |
3871 | { | |
14aa40a1 | 3872 | struct kvm_vcpu *vcpu = data; |
e36de87d VP |
3873 | *val = pid_nr(rcu_access_pointer(vcpu->pid)); |
3874 | return 0; | |
3875 | } | |
3876 | ||
3877 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n"); | |
3878 | ||
3e7093d0 | 3879 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 3880 | { |
d56f5136 | 3881 | struct dentry *debugfs_dentry; |
45b5939e | 3882 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 3883 | |
45b5939e | 3884 | if (!debugfs_initialized()) |
3e7093d0 | 3885 | return; |
45b5939e LC |
3886 | |
3887 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
3888 | debugfs_dentry = debugfs_create_dir(dir_name, |
3889 | vcpu->kvm->debugfs_dentry); | |
e36de87d VP |
3890 | debugfs_create_file("pid", 0444, debugfs_dentry, vcpu, |
3891 | &vcpu_get_pid_fops); | |
45b5939e | 3892 | |
d56f5136 | 3893 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
45b5939e | 3894 | } |
e36de87d | 3895 | #endif |
45b5939e | 3896 | |
c5ea7660 AK |
3897 | /* |
3898 | * Creates some virtual cpus. Good luck creating more than one. | |
3899 | */ | |
73880c80 | 3900 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
3901 | { |
3902 | int r; | |
e09fefde | 3903 | struct kvm_vcpu *vcpu; |
8bd826d6 | 3904 | struct page *page; |
c5ea7660 | 3905 | |
a1c42dde | 3906 | if (id >= KVM_MAX_VCPU_IDS) |
338c7dba AH |
3907 | return -EINVAL; |
3908 | ||
6c7caebc | 3909 | mutex_lock(&kvm->lock); |
f502cc56 | 3910 | if (kvm->created_vcpus >= kvm->max_vcpus) { |
6c7caebc PB |
3911 | mutex_unlock(&kvm->lock); |
3912 | return -EINVAL; | |
3913 | } | |
3914 | ||
1d5e740d ZG |
3915 | r = kvm_arch_vcpu_precreate(kvm, id); |
3916 | if (r) { | |
3917 | mutex_unlock(&kvm->lock); | |
3918 | return r; | |
3919 | } | |
3920 | ||
6c7caebc PB |
3921 | kvm->created_vcpus++; |
3922 | mutex_unlock(&kvm->lock); | |
3923 | ||
85f47930 | 3924 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
e529ef66 SC |
3925 | if (!vcpu) { |
3926 | r = -ENOMEM; | |
6c7caebc PB |
3927 | goto vcpu_decrement; |
3928 | } | |
c5ea7660 | 3929 | |
fcd97ad5 | 3930 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 3931 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
3932 | if (!page) { |
3933 | r = -ENOMEM; | |
e529ef66 | 3934 | goto vcpu_free; |
8bd826d6 SC |
3935 | } |
3936 | vcpu->run = page_address(page); | |
3937 | ||
3938 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
3939 | |
3940 | r = kvm_arch_vcpu_create(vcpu); | |
3941 | if (r) | |
8bd826d6 | 3942 | goto vcpu_free_run_page; |
e529ef66 | 3943 | |
fb04a1ed PX |
3944 | if (kvm->dirty_ring_size) { |
3945 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
3946 | id, kvm->dirty_ring_size); | |
3947 | if (r) | |
3948 | goto arch_vcpu_destroy; | |
3949 | } | |
3950 | ||
11ec2804 | 3951 | mutex_lock(&kvm->lock); |
42a90008 DW |
3952 | |
3953 | #ifdef CONFIG_LOCKDEP | |
3954 | /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */ | |
3955 | mutex_lock(&vcpu->mutex); | |
3956 | mutex_unlock(&vcpu->mutex); | |
3957 | #endif | |
3958 | ||
e09fefde DH |
3959 | if (kvm_get_vcpu_by_id(kvm, id)) { |
3960 | r = -EEXIST; | |
3961 | goto unlock_vcpu_destroy; | |
3962 | } | |
73880c80 | 3963 | |
8750e72a | 3964 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
c5b07754 MZ |
3965 | r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT); |
3966 | BUG_ON(r == -EBUSY); | |
3967 | if (r) | |
3968 | goto unlock_vcpu_destroy; | |
c5ea7660 | 3969 | |
fb3f0f51 | 3970 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 3971 | kvm_get_kvm(kvm); |
bccf2150 | 3972 | r = create_vcpu_fd(vcpu); |
73880c80 | 3973 | if (r < 0) { |
c5b07754 | 3974 | xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx); |
149487bd | 3975 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 3976 | goto unlock_vcpu_destroy; |
73880c80 GN |
3977 | } |
3978 | ||
dd489240 | 3979 | /* |
c5b07754 MZ |
3980 | * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu |
3981 | * pointer before kvm->online_vcpu's incremented value. | |
dd489240 | 3982 | */ |
73880c80 GN |
3983 | smp_wmb(); |
3984 | atomic_inc(&kvm->online_vcpus); | |
3985 | ||
73880c80 | 3986 | mutex_unlock(&kvm->lock); |
42897d86 | 3987 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 3988 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 3989 | return r; |
39c3b86e | 3990 | |
d780592b | 3991 | unlock_vcpu_destroy: |
7d8fece6 | 3992 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
3993 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
3994 | arch_vcpu_destroy: | |
d40ccc62 | 3995 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
3996 | vcpu_free_run_page: |
3997 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
3998 | vcpu_free: |
3999 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
4000 | vcpu_decrement: |
4001 | mutex_lock(&kvm->lock); | |
4002 | kvm->created_vcpus--; | |
4003 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
4004 | return r; |
4005 | } | |
4006 | ||
1961d276 AK |
4007 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
4008 | { | |
4009 | if (sigset) { | |
4010 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
4011 | vcpu->sigset_active = 1; | |
4012 | vcpu->sigset = *sigset; | |
4013 | } else | |
4014 | vcpu->sigset_active = 0; | |
4015 | return 0; | |
4016 | } | |
4017 | ||
ce55c049 JZ |
4018 | static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer, |
4019 | size_t size, loff_t *offset) | |
4020 | { | |
4021 | struct kvm_vcpu *vcpu = file->private_data; | |
4022 | ||
4023 | return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header, | |
4024 | &kvm_vcpu_stats_desc[0], &vcpu->stat, | |
4025 | sizeof(vcpu->stat), user_buffer, size, offset); | |
4026 | } | |
4027 | ||
4028 | static const struct file_operations kvm_vcpu_stats_fops = { | |
4029 | .read = kvm_vcpu_stats_read, | |
4030 | .llseek = noop_llseek, | |
4031 | }; | |
4032 | ||
4033 | static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu) | |
4034 | { | |
4035 | int fd; | |
4036 | struct file *file; | |
4037 | char name[15 + ITOA_MAX_LEN + 1]; | |
4038 | ||
4039 | snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id); | |
4040 | ||
4041 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4042 | if (fd < 0) | |
4043 | return fd; | |
4044 | ||
4045 | file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY); | |
4046 | if (IS_ERR(file)) { | |
4047 | put_unused_fd(fd); | |
4048 | return PTR_ERR(file); | |
4049 | } | |
4050 | file->f_mode |= FMODE_PREAD; | |
4051 | fd_install(fd, file); | |
4052 | ||
4053 | return fd; | |
4054 | } | |
4055 | ||
bccf2150 AK |
4056 | static long kvm_vcpu_ioctl(struct file *filp, |
4057 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 4058 | { |
bccf2150 | 4059 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 4060 | void __user *argp = (void __user *)arg; |
313a3dc7 | 4061 | int r; |
fa3795a7 DH |
4062 | struct kvm_fpu *fpu = NULL; |
4063 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 4064 | |
f4d31653 | 4065 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
6d4e4c4f | 4066 | return -EIO; |
2122ff5e | 4067 | |
2ea75be3 DM |
4068 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
4069 | return -EINVAL; | |
4070 | ||
2122ff5e | 4071 | /* |
5cb0944c PB |
4072 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
4073 | * execution; mutex_lock() would break them. | |
2122ff5e | 4074 | */ |
5cb0944c PB |
4075 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
4076 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 4077 | return r; |
2122ff5e | 4078 | |
ec7660cc CD |
4079 | if (mutex_lock_killable(&vcpu->mutex)) |
4080 | return -EINTR; | |
6aa8b732 | 4081 | switch (ioctl) { |
0e4524a5 CB |
4082 | case KVM_RUN: { |
4083 | struct pid *oldpid; | |
f0fe5108 AK |
4084 | r = -EINVAL; |
4085 | if (arg) | |
4086 | goto out; | |
0e4524a5 | 4087 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 4088 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 4089 | /* The thread running this VCPU changed. */ |
bd2a6394 | 4090 | struct pid *newpid; |
f95ef0cd | 4091 | |
bd2a6394 CD |
4092 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
4093 | if (r) | |
4094 | break; | |
4095 | ||
4096 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
4097 | rcu_assign_pointer(vcpu->pid, newpid); |
4098 | if (oldpid) | |
4099 | synchronize_rcu(); | |
4100 | put_pid(oldpid); | |
4101 | } | |
1b94f6f8 | 4102 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 4103 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 4104 | break; |
0e4524a5 | 4105 | } |
6aa8b732 | 4106 | case KVM_GET_REGS: { |
3e4bb3ac | 4107 | struct kvm_regs *kvm_regs; |
6aa8b732 | 4108 | |
3e4bb3ac | 4109 | r = -ENOMEM; |
b12ce36a | 4110 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 4111 | if (!kvm_regs) |
6aa8b732 | 4112 | goto out; |
3e4bb3ac XZ |
4113 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
4114 | if (r) | |
4115 | goto out_free1; | |
6aa8b732 | 4116 | r = -EFAULT; |
3e4bb3ac XZ |
4117 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
4118 | goto out_free1; | |
6aa8b732 | 4119 | r = 0; |
3e4bb3ac XZ |
4120 | out_free1: |
4121 | kfree(kvm_regs); | |
6aa8b732 AK |
4122 | break; |
4123 | } | |
4124 | case KVM_SET_REGS: { | |
3e4bb3ac | 4125 | struct kvm_regs *kvm_regs; |
6aa8b732 | 4126 | |
ff5c2c03 SL |
4127 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
4128 | if (IS_ERR(kvm_regs)) { | |
4129 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 4130 | goto out; |
ff5c2c03 | 4131 | } |
3e4bb3ac | 4132 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 4133 | kfree(kvm_regs); |
6aa8b732 AK |
4134 | break; |
4135 | } | |
4136 | case KVM_GET_SREGS: { | |
b12ce36a BG |
4137 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
4138 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
4139 | r = -ENOMEM; |
4140 | if (!kvm_sregs) | |
4141 | goto out; | |
4142 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
4143 | if (r) |
4144 | goto out; | |
4145 | r = -EFAULT; | |
fa3795a7 | 4146 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
4147 | goto out; |
4148 | r = 0; | |
4149 | break; | |
4150 | } | |
4151 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
4152 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
4153 | if (IS_ERR(kvm_sregs)) { | |
4154 | r = PTR_ERR(kvm_sregs); | |
18595411 | 4155 | kvm_sregs = NULL; |
6aa8b732 | 4156 | goto out; |
ff5c2c03 | 4157 | } |
fa3795a7 | 4158 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
4159 | break; |
4160 | } | |
62d9f0db MT |
4161 | case KVM_GET_MP_STATE: { |
4162 | struct kvm_mp_state mp_state; | |
4163 | ||
4164 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
4165 | if (r) | |
4166 | goto out; | |
4167 | r = -EFAULT; | |
893bdbf1 | 4168 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
4169 | goto out; |
4170 | r = 0; | |
4171 | break; | |
4172 | } | |
4173 | case KVM_SET_MP_STATE: { | |
4174 | struct kvm_mp_state mp_state; | |
4175 | ||
4176 | r = -EFAULT; | |
893bdbf1 | 4177 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
4178 | goto out; |
4179 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
4180 | break; |
4181 | } | |
6aa8b732 AK |
4182 | case KVM_TRANSLATE: { |
4183 | struct kvm_translation tr; | |
4184 | ||
4185 | r = -EFAULT; | |
893bdbf1 | 4186 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 4187 | goto out; |
8b006791 | 4188 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
4189 | if (r) |
4190 | goto out; | |
4191 | r = -EFAULT; | |
893bdbf1 | 4192 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
4193 | goto out; |
4194 | r = 0; | |
4195 | break; | |
4196 | } | |
d0bfb940 JK |
4197 | case KVM_SET_GUEST_DEBUG: { |
4198 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
4199 | |
4200 | r = -EFAULT; | |
893bdbf1 | 4201 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 4202 | goto out; |
d0bfb940 | 4203 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
4204 | break; |
4205 | } | |
1961d276 AK |
4206 | case KVM_SET_SIGNAL_MASK: { |
4207 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4208 | struct kvm_signal_mask kvm_sigmask; | |
4209 | sigset_t sigset, *p; | |
4210 | ||
4211 | p = NULL; | |
4212 | if (argp) { | |
4213 | r = -EFAULT; | |
4214 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4215 | sizeof(kvm_sigmask))) |
1961d276 AK |
4216 | goto out; |
4217 | r = -EINVAL; | |
893bdbf1 | 4218 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
4219 | goto out; |
4220 | r = -EFAULT; | |
4221 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 4222 | sizeof(sigset))) |
1961d276 AK |
4223 | goto out; |
4224 | p = &sigset; | |
4225 | } | |
376d41ff | 4226 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
4227 | break; |
4228 | } | |
b8836737 | 4229 | case KVM_GET_FPU: { |
b12ce36a | 4230 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
4231 | r = -ENOMEM; |
4232 | if (!fpu) | |
4233 | goto out; | |
4234 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
4235 | if (r) |
4236 | goto out; | |
4237 | r = -EFAULT; | |
fa3795a7 | 4238 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
4239 | goto out; |
4240 | r = 0; | |
4241 | break; | |
4242 | } | |
4243 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
4244 | fpu = memdup_user(argp, sizeof(*fpu)); |
4245 | if (IS_ERR(fpu)) { | |
4246 | r = PTR_ERR(fpu); | |
18595411 | 4247 | fpu = NULL; |
b8836737 | 4248 | goto out; |
ff5c2c03 | 4249 | } |
fa3795a7 | 4250 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
4251 | break; |
4252 | } | |
ce55c049 JZ |
4253 | case KVM_GET_STATS_FD: { |
4254 | r = kvm_vcpu_ioctl_get_stats_fd(vcpu); | |
4255 | break; | |
4256 | } | |
bccf2150 | 4257 | default: |
313a3dc7 | 4258 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
4259 | } |
4260 | out: | |
ec7660cc | 4261 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
4262 | kfree(fpu); |
4263 | kfree(kvm_sregs); | |
bccf2150 AK |
4264 | return r; |
4265 | } | |
4266 | ||
de8e5d74 | 4267 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
4268 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
4269 | unsigned int ioctl, unsigned long arg) | |
4270 | { | |
4271 | struct kvm_vcpu *vcpu = filp->private_data; | |
4272 | void __user *argp = compat_ptr(arg); | |
4273 | int r; | |
4274 | ||
f4d31653 | 4275 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
1dda606c AG |
4276 | return -EIO; |
4277 | ||
4278 | switch (ioctl) { | |
4279 | case KVM_SET_SIGNAL_MASK: { | |
4280 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4281 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
4282 | sigset_t sigset; |
4283 | ||
4284 | if (argp) { | |
4285 | r = -EFAULT; | |
4286 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4287 | sizeof(kvm_sigmask))) |
1dda606c AG |
4288 | goto out; |
4289 | r = -EINVAL; | |
3968cf62 | 4290 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
4291 | goto out; |
4292 | r = -EFAULT; | |
1393b4aa PB |
4293 | if (get_compat_sigset(&sigset, |
4294 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 4295 | goto out; |
760a9a30 AC |
4296 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
4297 | } else | |
4298 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
4299 | break; |
4300 | } | |
4301 | default: | |
4302 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
4303 | } | |
4304 | ||
4305 | out: | |
4306 | return r; | |
4307 | } | |
4308 | #endif | |
4309 | ||
a1cd3f08 CLG |
4310 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
4311 | { | |
4312 | struct kvm_device *dev = filp->private_data; | |
4313 | ||
4314 | if (dev->ops->mmap) | |
4315 | return dev->ops->mmap(dev, vma); | |
4316 | ||
4317 | return -ENODEV; | |
4318 | } | |
4319 | ||
852b6d57 SW |
4320 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
4321 | int (*accessor)(struct kvm_device *dev, | |
4322 | struct kvm_device_attr *attr), | |
4323 | unsigned long arg) | |
4324 | { | |
4325 | struct kvm_device_attr attr; | |
4326 | ||
4327 | if (!accessor) | |
4328 | return -EPERM; | |
4329 | ||
4330 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
4331 | return -EFAULT; | |
4332 | ||
4333 | return accessor(dev, &attr); | |
4334 | } | |
4335 | ||
4336 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
4337 | unsigned long arg) | |
4338 | { | |
4339 | struct kvm_device *dev = filp->private_data; | |
4340 | ||
f4d31653 | 4341 | if (dev->kvm->mm != current->mm || dev->kvm->vm_dead) |
ddba9180 SC |
4342 | return -EIO; |
4343 | ||
852b6d57 SW |
4344 | switch (ioctl) { |
4345 | case KVM_SET_DEVICE_ATTR: | |
4346 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
4347 | case KVM_GET_DEVICE_ATTR: | |
4348 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
4349 | case KVM_HAS_DEVICE_ATTR: | |
4350 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
4351 | default: | |
4352 | if (dev->ops->ioctl) | |
4353 | return dev->ops->ioctl(dev, ioctl, arg); | |
4354 | ||
4355 | return -ENOTTY; | |
4356 | } | |
4357 | } | |
4358 | ||
852b6d57 SW |
4359 | static int kvm_device_release(struct inode *inode, struct file *filp) |
4360 | { | |
4361 | struct kvm_device *dev = filp->private_data; | |
4362 | struct kvm *kvm = dev->kvm; | |
4363 | ||
2bde9b3e CLG |
4364 | if (dev->ops->release) { |
4365 | mutex_lock(&kvm->lock); | |
4366 | list_del(&dev->vm_node); | |
4367 | dev->ops->release(dev); | |
4368 | mutex_unlock(&kvm->lock); | |
4369 | } | |
4370 | ||
852b6d57 SW |
4371 | kvm_put_kvm(kvm); |
4372 | return 0; | |
4373 | } | |
4374 | ||
4375 | static const struct file_operations kvm_device_fops = { | |
4376 | .unlocked_ioctl = kvm_device_ioctl, | |
4377 | .release = kvm_device_release, | |
7ddfd3e0 | 4378 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 4379 | .mmap = kvm_device_mmap, |
852b6d57 SW |
4380 | }; |
4381 | ||
4382 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
4383 | { | |
4384 | if (filp->f_op != &kvm_device_fops) | |
4385 | return NULL; | |
4386 | ||
4387 | return filp->private_data; | |
4388 | } | |
4389 | ||
8538cb22 | 4390 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 4391 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
4392 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
4393 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 4394 | #endif |
d60eacb0 WD |
4395 | }; |
4396 | ||
8538cb22 | 4397 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
4398 | { |
4399 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
4400 | return -ENOSPC; | |
4401 | ||
4402 | if (kvm_device_ops_table[type] != NULL) | |
4403 | return -EEXIST; | |
4404 | ||
4405 | kvm_device_ops_table[type] = ops; | |
4406 | return 0; | |
4407 | } | |
4408 | ||
571ee1b6 WL |
4409 | void kvm_unregister_device_ops(u32 type) |
4410 | { | |
4411 | if (kvm_device_ops_table[type] != NULL) | |
4412 | kvm_device_ops_table[type] = NULL; | |
4413 | } | |
4414 | ||
852b6d57 SW |
4415 | static int kvm_ioctl_create_device(struct kvm *kvm, |
4416 | struct kvm_create_device *cd) | |
4417 | { | |
eceb6e1d | 4418 | const struct kvm_device_ops *ops; |
852b6d57 SW |
4419 | struct kvm_device *dev; |
4420 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 4421 | int type; |
852b6d57 SW |
4422 | int ret; |
4423 | ||
d60eacb0 WD |
4424 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
4425 | return -ENODEV; | |
4426 | ||
1d487e9b PB |
4427 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
4428 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 4429 | if (ops == NULL) |
852b6d57 | 4430 | return -ENODEV; |
852b6d57 SW |
4431 | |
4432 | if (test) | |
4433 | return 0; | |
4434 | ||
b12ce36a | 4435 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
4436 | if (!dev) |
4437 | return -ENOMEM; | |
4438 | ||
4439 | dev->ops = ops; | |
4440 | dev->kvm = kvm; | |
852b6d57 | 4441 | |
a28ebea2 | 4442 | mutex_lock(&kvm->lock); |
1d487e9b | 4443 | ret = ops->create(dev, type); |
852b6d57 | 4444 | if (ret < 0) { |
a28ebea2 | 4445 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
4446 | kfree(dev); |
4447 | return ret; | |
4448 | } | |
a28ebea2 CD |
4449 | list_add(&dev->vm_node, &kvm->devices); |
4450 | mutex_unlock(&kvm->lock); | |
852b6d57 | 4451 | |
023e9fdd CD |
4452 | if (ops->init) |
4453 | ops->init(dev); | |
4454 | ||
cfa39381 | 4455 | kvm_get_kvm(kvm); |
24009b05 | 4456 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 4457 | if (ret < 0) { |
149487bd | 4458 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
4459 | mutex_lock(&kvm->lock); |
4460 | list_del(&dev->vm_node); | |
e8bc2427 AK |
4461 | if (ops->release) |
4462 | ops->release(dev); | |
a28ebea2 | 4463 | mutex_unlock(&kvm->lock); |
e8bc2427 AK |
4464 | if (ops->destroy) |
4465 | ops->destroy(dev); | |
852b6d57 SW |
4466 | return ret; |
4467 | } | |
4468 | ||
852b6d57 SW |
4469 | cd->fd = ret; |
4470 | return 0; | |
4471 | } | |
4472 | ||
f15ba52b | 4473 | static int kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
92b591a4 AG |
4474 | { |
4475 | switch (arg) { | |
4476 | case KVM_CAP_USER_MEMORY: | |
4477 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
4478 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
4479 | case KVM_CAP_INTERNAL_ERROR_DATA: |
4480 | #ifdef CONFIG_HAVE_KVM_MSI | |
4481 | case KVM_CAP_SIGNAL_MSI: | |
4482 | #endif | |
297e2105 | 4483 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 4484 | case KVM_CAP_IRQFD: |
92b591a4 | 4485 | #endif |
e9ea5069 | 4486 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 4487 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 4488 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 4489 | case KVM_CAP_HALT_POLL: |
92b591a4 | 4490 | return 1; |
4b4357e0 | 4491 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
4492 | case KVM_CAP_COALESCED_MMIO: |
4493 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
4494 | case KVM_CAP_COALESCED_PIO: |
4495 | return 1; | |
30422558 | 4496 | #endif |
3c9bd400 JZ |
4497 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4498 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
4499 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
4500 | #endif | |
92b591a4 AG |
4501 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4502 | case KVM_CAP_IRQ_ROUTING: | |
4503 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
4504 | #endif |
4505 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
4506 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
4507 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 4508 | #endif |
c110ae57 PB |
4509 | case KVM_CAP_NR_MEMSLOTS: |
4510 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed | 4511 | case KVM_CAP_DIRTY_LOG_RING: |
17601bfe MZ |
4512 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO |
4513 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); | |
4514 | #else | |
4515 | return 0; | |
4516 | #endif | |
4517 | case KVM_CAP_DIRTY_LOG_RING_ACQ_REL: | |
4518 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL | |
fb04a1ed PX |
4519 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); |
4520 | #else | |
4521 | return 0; | |
86bdf3eb GS |
4522 | #endif |
4523 | #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP | |
4524 | case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: | |
fb04a1ed | 4525 | #endif |
ce55c049 | 4526 | case KVM_CAP_BINARY_STATS_FD: |
d495f942 | 4527 | case KVM_CAP_SYSTEM_EVENT_DATA: |
ce55c049 | 4528 | return 1; |
92b591a4 AG |
4529 | default: |
4530 | break; | |
4531 | } | |
4532 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
4533 | } | |
4534 | ||
fb04a1ed PX |
4535 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
4536 | { | |
4537 | int r; | |
4538 | ||
4539 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
4540 | return -EINVAL; | |
4541 | ||
4542 | /* the size should be power of 2 */ | |
4543 | if (!size || (size & (size - 1))) | |
4544 | return -EINVAL; | |
4545 | ||
4546 | /* Should be bigger to keep the reserved entries, or a page */ | |
4547 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
4548 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
4549 | return -EINVAL; | |
4550 | ||
4551 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
4552 | sizeof(struct kvm_dirty_gfn)) | |
4553 | return -E2BIG; | |
4554 | ||
4555 | /* We only allow it to set once */ | |
4556 | if (kvm->dirty_ring_size) | |
4557 | return -EINVAL; | |
4558 | ||
4559 | mutex_lock(&kvm->lock); | |
4560 | ||
4561 | if (kvm->created_vcpus) { | |
4562 | /* We don't allow to change this value after vcpu created */ | |
4563 | r = -EINVAL; | |
4564 | } else { | |
4565 | kvm->dirty_ring_size = size; | |
4566 | r = 0; | |
4567 | } | |
4568 | ||
4569 | mutex_unlock(&kvm->lock); | |
4570 | return r; | |
4571 | } | |
4572 | ||
4573 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
4574 | { | |
46808a4c | 4575 | unsigned long i; |
fb04a1ed PX |
4576 | struct kvm_vcpu *vcpu; |
4577 | int cleared = 0; | |
4578 | ||
4579 | if (!kvm->dirty_ring_size) | |
4580 | return -EINVAL; | |
4581 | ||
4582 | mutex_lock(&kvm->slots_lock); | |
4583 | ||
4584 | kvm_for_each_vcpu(i, vcpu, kvm) | |
4585 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
4586 | ||
4587 | mutex_unlock(&kvm->slots_lock); | |
4588 | ||
4589 | if (cleared) | |
4590 | kvm_flush_remote_tlbs(kvm); | |
4591 | ||
4592 | return cleared; | |
4593 | } | |
4594 | ||
e5d83c74 PB |
4595 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4596 | struct kvm_enable_cap *cap) | |
4597 | { | |
4598 | return -EINVAL; | |
4599 | } | |
4600 | ||
86bdf3eb GS |
4601 | static bool kvm_are_all_memslots_empty(struct kvm *kvm) |
4602 | { | |
4603 | int i; | |
4604 | ||
4605 | lockdep_assert_held(&kvm->slots_lock); | |
4606 | ||
4607 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
4608 | if (!kvm_memslots_empty(__kvm_memslots(kvm, i))) | |
4609 | return false; | |
4610 | } | |
4611 | ||
4612 | return true; | |
4613 | } | |
4614 | ||
e5d83c74 PB |
4615 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, |
4616 | struct kvm_enable_cap *cap) | |
4617 | { | |
4618 | switch (cap->cap) { | |
2a31b9db | 4619 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
4620 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
4621 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
4622 | ||
4623 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
4624 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
4625 | ||
4626 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
4627 | return -EINVAL; |
4628 | kvm->manual_dirty_log_protect = cap->args[0]; | |
4629 | return 0; | |
3c9bd400 | 4630 | } |
2a31b9db | 4631 | #endif |
acd05785 DM |
4632 | case KVM_CAP_HALT_POLL: { |
4633 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
4634 | return -EINVAL; | |
4635 | ||
4636 | kvm->max_halt_poll_ns = cap->args[0]; | |
9eb8ca04 DM |
4637 | |
4638 | /* | |
4639 | * Ensure kvm->override_halt_poll_ns does not become visible | |
4640 | * before kvm->max_halt_poll_ns. | |
4641 | * | |
4642 | * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns(). | |
4643 | */ | |
4644 | smp_wmb(); | |
4645 | kvm->override_halt_poll_ns = true; | |
4646 | ||
acd05785 DM |
4647 | return 0; |
4648 | } | |
fb04a1ed | 4649 | case KVM_CAP_DIRTY_LOG_RING: |
17601bfe | 4650 | case KVM_CAP_DIRTY_LOG_RING_ACQ_REL: |
7a2726ec GS |
4651 | if (!kvm_vm_ioctl_check_extension_generic(kvm, cap->cap)) |
4652 | return -EINVAL; | |
4653 | ||
fb04a1ed | 4654 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); |
86bdf3eb GS |
4655 | case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: { |
4656 | int r = -EINVAL; | |
4657 | ||
4658 | if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP) || | |
4659 | !kvm->dirty_ring_size || cap->flags) | |
4660 | return r; | |
4661 | ||
4662 | mutex_lock(&kvm->slots_lock); | |
4663 | ||
4664 | /* | |
4665 | * For simplicity, allow enabling ring+bitmap if and only if | |
4666 | * there are no memslots, e.g. to ensure all memslots allocate | |
4667 | * a bitmap after the capability is enabled. | |
4668 | */ | |
4669 | if (kvm_are_all_memslots_empty(kvm)) { | |
4670 | kvm->dirty_ring_with_bitmap = true; | |
4671 | r = 0; | |
4672 | } | |
4673 | ||
4674 | mutex_unlock(&kvm->slots_lock); | |
4675 | ||
4676 | return r; | |
4677 | } | |
e5d83c74 PB |
4678 | default: |
4679 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
4680 | } | |
4681 | } | |
4682 | ||
fcfe1bae JZ |
4683 | static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer, |
4684 | size_t size, loff_t *offset) | |
4685 | { | |
4686 | struct kvm *kvm = file->private_data; | |
4687 | ||
4688 | return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header, | |
4689 | &kvm_vm_stats_desc[0], &kvm->stat, | |
4690 | sizeof(kvm->stat), user_buffer, size, offset); | |
4691 | } | |
4692 | ||
4693 | static const struct file_operations kvm_vm_stats_fops = { | |
4694 | .read = kvm_vm_stats_read, | |
4695 | .llseek = noop_llseek, | |
4696 | }; | |
4697 | ||
4698 | static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm) | |
4699 | { | |
4700 | int fd; | |
4701 | struct file *file; | |
4702 | ||
4703 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4704 | if (fd < 0) | |
4705 | return fd; | |
4706 | ||
4707 | file = anon_inode_getfile("kvm-vm-stats", | |
4708 | &kvm_vm_stats_fops, kvm, O_RDONLY); | |
4709 | if (IS_ERR(file)) { | |
4710 | put_unused_fd(fd); | |
4711 | return PTR_ERR(file); | |
4712 | } | |
4713 | file->f_mode |= FMODE_PREAD; | |
4714 | fd_install(fd, file); | |
4715 | ||
4716 | return fd; | |
4717 | } | |
4718 | ||
bccf2150 AK |
4719 | static long kvm_vm_ioctl(struct file *filp, |
4720 | unsigned int ioctl, unsigned long arg) | |
4721 | { | |
4722 | struct kvm *kvm = filp->private_data; | |
4723 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 4724 | int r; |
bccf2150 | 4725 | |
f4d31653 | 4726 | if (kvm->mm != current->mm || kvm->vm_dead) |
6d4e4c4f | 4727 | return -EIO; |
bccf2150 AK |
4728 | switch (ioctl) { |
4729 | case KVM_CREATE_VCPU: | |
4730 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 4731 | break; |
e5d83c74 PB |
4732 | case KVM_ENABLE_CAP: { |
4733 | struct kvm_enable_cap cap; | |
4734 | ||
4735 | r = -EFAULT; | |
4736 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4737 | goto out; | |
4738 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
4739 | break; | |
4740 | } | |
6fc138d2 IE |
4741 | case KVM_SET_USER_MEMORY_REGION: { |
4742 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
4743 | ||
4744 | r = -EFAULT; | |
4745 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 4746 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
4747 | goto out; |
4748 | ||
47ae31e2 | 4749 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
4750 | break; |
4751 | } | |
4752 | case KVM_GET_DIRTY_LOG: { | |
4753 | struct kvm_dirty_log log; | |
4754 | ||
4755 | r = -EFAULT; | |
893bdbf1 | 4756 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 4757 | goto out; |
2c6f5df9 | 4758 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
4759 | break; |
4760 | } | |
2a31b9db PB |
4761 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4762 | case KVM_CLEAR_DIRTY_LOG: { | |
4763 | struct kvm_clear_dirty_log log; | |
4764 | ||
4765 | r = -EFAULT; | |
4766 | if (copy_from_user(&log, argp, sizeof(log))) | |
4767 | goto out; | |
4768 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4769 | break; | |
4770 | } | |
4771 | #endif | |
4b4357e0 | 4772 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
4773 | case KVM_REGISTER_COALESCED_MMIO: { |
4774 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4775 | |
5f94c174 | 4776 | r = -EFAULT; |
893bdbf1 | 4777 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4778 | goto out; |
5f94c174 | 4779 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4780 | break; |
4781 | } | |
4782 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
4783 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4784 | |
5f94c174 | 4785 | r = -EFAULT; |
893bdbf1 | 4786 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4787 | goto out; |
5f94c174 | 4788 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4789 | break; |
4790 | } | |
4791 | #endif | |
721eecbf GH |
4792 | case KVM_IRQFD: { |
4793 | struct kvm_irqfd data; | |
4794 | ||
4795 | r = -EFAULT; | |
893bdbf1 | 4796 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 4797 | goto out; |
d4db2935 | 4798 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
4799 | break; |
4800 | } | |
d34e6b17 GH |
4801 | case KVM_IOEVENTFD: { |
4802 | struct kvm_ioeventfd data; | |
4803 | ||
4804 | r = -EFAULT; | |
893bdbf1 | 4805 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
4806 | goto out; |
4807 | r = kvm_ioeventfd(kvm, &data); | |
4808 | break; | |
4809 | } | |
07975ad3 JK |
4810 | #ifdef CONFIG_HAVE_KVM_MSI |
4811 | case KVM_SIGNAL_MSI: { | |
4812 | struct kvm_msi msi; | |
4813 | ||
4814 | r = -EFAULT; | |
893bdbf1 | 4815 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
4816 | goto out; |
4817 | r = kvm_send_userspace_msi(kvm, &msi); | |
4818 | break; | |
4819 | } | |
23d43cf9 CD |
4820 | #endif |
4821 | #ifdef __KVM_HAVE_IRQ_LINE | |
4822 | case KVM_IRQ_LINE_STATUS: | |
4823 | case KVM_IRQ_LINE: { | |
4824 | struct kvm_irq_level irq_event; | |
4825 | ||
4826 | r = -EFAULT; | |
893bdbf1 | 4827 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
4828 | goto out; |
4829 | ||
aa2fbe6d YZ |
4830 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
4831 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
4832 | if (r) |
4833 | goto out; | |
4834 | ||
4835 | r = -EFAULT; | |
4836 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 4837 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
4838 | goto out; |
4839 | } | |
4840 | ||
4841 | r = 0; | |
4842 | break; | |
4843 | } | |
73880c80 | 4844 | #endif |
aa8d5944 AG |
4845 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4846 | case KVM_SET_GSI_ROUTING: { | |
4847 | struct kvm_irq_routing routing; | |
4848 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 4849 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
4850 | |
4851 | r = -EFAULT; | |
4852 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
4853 | goto out; | |
4854 | r = -EINVAL; | |
5c0aea0e DH |
4855 | if (!kvm_arch_can_set_irq_routing(kvm)) |
4856 | goto out; | |
caf1ff26 | 4857 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
4858 | goto out; |
4859 | if (routing.flags) | |
4860 | goto out; | |
f8c1b85b | 4861 | if (routing.nr) { |
f8c1b85b | 4862 | urouting = argp; |
7ec28e26 DE |
4863 | entries = vmemdup_user(urouting->entries, |
4864 | array_size(sizeof(*entries), | |
4865 | routing.nr)); | |
4866 | if (IS_ERR(entries)) { | |
4867 | r = PTR_ERR(entries); | |
4868 | goto out; | |
4869 | } | |
f8c1b85b | 4870 | } |
aa8d5944 AG |
4871 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
4872 | routing.flags); | |
7ec28e26 | 4873 | kvfree(entries); |
aa8d5944 AG |
4874 | break; |
4875 | } | |
4876 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
4877 | case KVM_CREATE_DEVICE: { |
4878 | struct kvm_create_device cd; | |
4879 | ||
4880 | r = -EFAULT; | |
4881 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
4882 | goto out; | |
4883 | ||
4884 | r = kvm_ioctl_create_device(kvm, &cd); | |
4885 | if (r) | |
4886 | goto out; | |
4887 | ||
4888 | r = -EFAULT; | |
4889 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
4890 | goto out; | |
4891 | ||
4892 | r = 0; | |
4893 | break; | |
4894 | } | |
92b591a4 AG |
4895 | case KVM_CHECK_EXTENSION: |
4896 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
4897 | break; | |
fb04a1ed PX |
4898 | case KVM_RESET_DIRTY_RINGS: |
4899 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
4900 | break; | |
fcfe1bae JZ |
4901 | case KVM_GET_STATS_FD: |
4902 | r = kvm_vm_ioctl_get_stats_fd(kvm); | |
4903 | break; | |
f17abe9a | 4904 | default: |
1fe779f8 | 4905 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
4906 | } |
4907 | out: | |
4908 | return r; | |
4909 | } | |
4910 | ||
de8e5d74 | 4911 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
4912 | struct compat_kvm_dirty_log { |
4913 | __u32 slot; | |
4914 | __u32 padding1; | |
4915 | union { | |
4916 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4917 | __u64 padding2; | |
4918 | }; | |
4919 | }; | |
4920 | ||
8750f9bb PB |
4921 | struct compat_kvm_clear_dirty_log { |
4922 | __u32 slot; | |
4923 | __u32 num_pages; | |
4924 | __u64 first_page; | |
4925 | union { | |
4926 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4927 | __u64 padding2; | |
4928 | }; | |
4929 | }; | |
4930 | ||
ed51862f AG |
4931 | long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl, |
4932 | unsigned long arg) | |
4933 | { | |
4934 | return -ENOTTY; | |
4935 | } | |
4936 | ||
6ff5894c AB |
4937 | static long kvm_vm_compat_ioctl(struct file *filp, |
4938 | unsigned int ioctl, unsigned long arg) | |
4939 | { | |
4940 | struct kvm *kvm = filp->private_data; | |
4941 | int r; | |
4942 | ||
f4d31653 | 4943 | if (kvm->mm != current->mm || kvm->vm_dead) |
6ff5894c | 4944 | return -EIO; |
ed51862f AG |
4945 | |
4946 | r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg); | |
4947 | if (r != -ENOTTY) | |
4948 | return r; | |
4949 | ||
6ff5894c | 4950 | switch (ioctl) { |
8750f9bb PB |
4951 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4952 | case KVM_CLEAR_DIRTY_LOG: { | |
4953 | struct compat_kvm_clear_dirty_log compat_log; | |
4954 | struct kvm_clear_dirty_log log; | |
4955 | ||
4956 | if (copy_from_user(&compat_log, (void __user *)arg, | |
4957 | sizeof(compat_log))) | |
4958 | return -EFAULT; | |
4959 | log.slot = compat_log.slot; | |
4960 | log.num_pages = compat_log.num_pages; | |
4961 | log.first_page = compat_log.first_page; | |
4962 | log.padding2 = compat_log.padding2; | |
4963 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4964 | ||
4965 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4966 | break; | |
4967 | } | |
4968 | #endif | |
6ff5894c AB |
4969 | case KVM_GET_DIRTY_LOG: { |
4970 | struct compat_kvm_dirty_log compat_log; | |
4971 | struct kvm_dirty_log log; | |
4972 | ||
6ff5894c AB |
4973 | if (copy_from_user(&compat_log, (void __user *)arg, |
4974 | sizeof(compat_log))) | |
f6a3b168 | 4975 | return -EFAULT; |
6ff5894c AB |
4976 | log.slot = compat_log.slot; |
4977 | log.padding1 = compat_log.padding1; | |
4978 | log.padding2 = compat_log.padding2; | |
4979 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4980 | ||
4981 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
4982 | break; |
4983 | } | |
4984 | default: | |
4985 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
4986 | } | |
6ff5894c AB |
4987 | return r; |
4988 | } | |
4989 | #endif | |
4990 | ||
70375c2d | 4991 | static const struct file_operations kvm_vm_fops = { |
f17abe9a AK |
4992 | .release = kvm_vm_release, |
4993 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 4994 | .llseek = noop_llseek, |
7ddfd3e0 | 4995 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
4996 | }; |
4997 | ||
54526d1f NT |
4998 | bool file_is_kvm(struct file *file) |
4999 | { | |
5000 | return file && file->f_op == &kvm_vm_fops; | |
5001 | } | |
5002 | EXPORT_SYMBOL_GPL(file_is_kvm); | |
5003 | ||
e08b9637 | 5004 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 5005 | { |
59f82aad | 5006 | char fdname[ITOA_MAX_LEN + 1]; |
20020f4c | 5007 | int r, fd; |
f17abe9a | 5008 | struct kvm *kvm; |
506cfba9 | 5009 | struct file *file; |
f17abe9a | 5010 | |
20020f4c OU |
5011 | fd = get_unused_fd_flags(O_CLOEXEC); |
5012 | if (fd < 0) | |
5013 | return fd; | |
5014 | ||
59f82aad OU |
5015 | snprintf(fdname, sizeof(fdname), "%d", fd); |
5016 | ||
b74ed7a6 | 5017 | kvm = kvm_create_vm(type, fdname); |
20020f4c OU |
5018 | if (IS_ERR(kvm)) { |
5019 | r = PTR_ERR(kvm); | |
5020 | goto put_fd; | |
5021 | } | |
5022 | ||
506cfba9 AV |
5023 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
5024 | if (IS_ERR(file)) { | |
78588335 ME |
5025 | r = PTR_ERR(file); |
5026 | goto put_kvm; | |
506cfba9 | 5027 | } |
536a6f88 | 5028 | |
525df861 PB |
5029 | /* |
5030 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
5031 | * already set, with ->release() being kvm_vm_release(). In error | |
5032 | * cases it will be called by the final fput(file) and will take | |
5033 | * care of doing kvm_put_kvm(kvm). | |
5034 | */ | |
286de8f6 | 5035 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 5036 | |
20020f4c OU |
5037 | fd_install(fd, file); |
5038 | return fd; | |
78588335 ME |
5039 | |
5040 | put_kvm: | |
5041 | kvm_put_kvm(kvm); | |
20020f4c OU |
5042 | put_fd: |
5043 | put_unused_fd(fd); | |
78588335 | 5044 | return r; |
f17abe9a AK |
5045 | } |
5046 | ||
5047 | static long kvm_dev_ioctl(struct file *filp, | |
5048 | unsigned int ioctl, unsigned long arg) | |
5049 | { | |
f15ba52b | 5050 | int r = -EINVAL; |
f17abe9a AK |
5051 | |
5052 | switch (ioctl) { | |
5053 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
5054 | if (arg) |
5055 | goto out; | |
f17abe9a AK |
5056 | r = KVM_API_VERSION; |
5057 | break; | |
5058 | case KVM_CREATE_VM: | |
e08b9637 | 5059 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 5060 | break; |
018d00d2 | 5061 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 5062 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 5063 | break; |
07c45a36 | 5064 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
5065 | if (arg) |
5066 | goto out; | |
adb1ff46 AK |
5067 | r = PAGE_SIZE; /* struct kvm_run */ |
5068 | #ifdef CONFIG_X86 | |
5069 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 5070 | #endif |
4b4357e0 | 5071 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 5072 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 5073 | #endif |
07c45a36 | 5074 | break; |
d4c9ff2d FEL |
5075 | case KVM_TRACE_ENABLE: |
5076 | case KVM_TRACE_PAUSE: | |
5077 | case KVM_TRACE_DISABLE: | |
2023a29c | 5078 | r = -EOPNOTSUPP; |
d4c9ff2d | 5079 | break; |
6aa8b732 | 5080 | default: |
043405e1 | 5081 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
5082 | } |
5083 | out: | |
5084 | return r; | |
5085 | } | |
5086 | ||
6aa8b732 | 5087 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 5088 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 5089 | .llseek = noop_llseek, |
7ddfd3e0 | 5090 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
5091 | }; |
5092 | ||
5093 | static struct miscdevice kvm_dev = { | |
bbe4432e | 5094 | KVM_MINOR, |
6aa8b732 AK |
5095 | "kvm", |
5096 | &kvm_chardev_ops, | |
5097 | }; | |
5098 | ||
441f7bfa SC |
5099 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
5100 | __visible bool kvm_rebooting; | |
5101 | EXPORT_SYMBOL_GPL(kvm_rebooting); | |
5102 | ||
5103 | static DEFINE_PER_CPU(bool, hardware_enabled); | |
5104 | static int kvm_usage_count; | |
5105 | ||
e6fb7d6e | 5106 | static int __hardware_enable_nolock(void) |
1b6c0168 | 5107 | { |
37d25881 | 5108 | if (__this_cpu_read(hardware_enabled)) |
e6fb7d6e | 5109 | return 0; |
10474ae8 | 5110 | |
37d25881 | 5111 | if (kvm_arch_hardware_enable()) { |
37d25881 SC |
5112 | pr_info("kvm: enabling virtualization on CPU%d failed\n", |
5113 | raw_smp_processor_id()); | |
e6fb7d6e | 5114 | return -EIO; |
10474ae8 | 5115 | } |
37d25881 SC |
5116 | |
5117 | __this_cpu_write(hardware_enabled, true); | |
e6fb7d6e IY |
5118 | return 0; |
5119 | } | |
5120 | ||
5121 | static void hardware_enable_nolock(void *failed) | |
5122 | { | |
5123 | if (__hardware_enable_nolock()) | |
5124 | atomic_inc(failed); | |
1b6c0168 AK |
5125 | } |
5126 | ||
aaf12a7b | 5127 | static int kvm_online_cpu(unsigned int cpu) |
75b7127c | 5128 | { |
aaf12a7b CG |
5129 | int ret = 0; |
5130 | ||
5131 | /* | |
5132 | * Abort the CPU online process if hardware virtualization cannot | |
5133 | * be enabled. Otherwise running VMs would encounter unrecoverable | |
5134 | * errors when scheduled to this CPU. | |
5135 | */ | |
0bf50497 | 5136 | mutex_lock(&kvm_lock); |
e6fb7d6e IY |
5137 | if (kvm_usage_count) |
5138 | ret = __hardware_enable_nolock(); | |
0bf50497 | 5139 | mutex_unlock(&kvm_lock); |
aaf12a7b | 5140 | return ret; |
75b7127c TY |
5141 | } |
5142 | ||
5143 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 | 5144 | { |
37d25881 SC |
5145 | /* |
5146 | * Note, hardware_disable_all_nolock() tells all online CPUs to disable | |
5147 | * hardware, not just CPUs that successfully enabled hardware! | |
5148 | */ | |
5149 | if (!__this_cpu_read(hardware_enabled)) | |
1b6c0168 | 5150 | return; |
37d25881 | 5151 | |
13a34e06 | 5152 | kvm_arch_hardware_disable(); |
37d25881 SC |
5153 | |
5154 | __this_cpu_write(hardware_enabled, false); | |
1b6c0168 AK |
5155 | } |
5156 | ||
aaf12a7b | 5157 | static int kvm_offline_cpu(unsigned int cpu) |
75b7127c | 5158 | { |
0bf50497 | 5159 | mutex_lock(&kvm_lock); |
4fa92fb2 PB |
5160 | if (kvm_usage_count) |
5161 | hardware_disable_nolock(NULL); | |
0bf50497 | 5162 | mutex_unlock(&kvm_lock); |
8c18b2d2 | 5163 | return 0; |
75b7127c TY |
5164 | } |
5165 | ||
10474ae8 AG |
5166 | static void hardware_disable_all_nolock(void) |
5167 | { | |
5168 | BUG_ON(!kvm_usage_count); | |
5169 | ||
5170 | kvm_usage_count--; | |
5171 | if (!kvm_usage_count) | |
75b7127c | 5172 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
5173 | } |
5174 | ||
5175 | static void hardware_disable_all(void) | |
5176 | { | |
e4aa7f88 | 5177 | cpus_read_lock(); |
0bf50497 | 5178 | mutex_lock(&kvm_lock); |
10474ae8 | 5179 | hardware_disable_all_nolock(); |
0bf50497 | 5180 | mutex_unlock(&kvm_lock); |
e4aa7f88 | 5181 | cpus_read_unlock(); |
10474ae8 AG |
5182 | } |
5183 | ||
5184 | static int hardware_enable_all(void) | |
5185 | { | |
e6fb7d6e | 5186 | atomic_t failed = ATOMIC_INIT(0); |
10474ae8 AG |
5187 | int r = 0; |
5188 | ||
e4aa7f88 CG |
5189 | /* |
5190 | * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu() | |
5191 | * is called, and so on_each_cpu() between them includes the CPU that | |
5192 | * is being onlined. As a result, hardware_enable_nolock() may get | |
5193 | * invoked before kvm_online_cpu(), which also enables hardware if the | |
5194 | * usage count is non-zero. Disable CPU hotplug to avoid attempting to | |
5195 | * enable hardware multiple times. | |
5196 | */ | |
5197 | cpus_read_lock(); | |
0bf50497 | 5198 | mutex_lock(&kvm_lock); |
10474ae8 AG |
5199 | |
5200 | kvm_usage_count++; | |
5201 | if (kvm_usage_count == 1) { | |
e6fb7d6e | 5202 | on_each_cpu(hardware_enable_nolock, &failed, 1); |
10474ae8 | 5203 | |
e6fb7d6e | 5204 | if (atomic_read(&failed)) { |
10474ae8 AG |
5205 | hardware_disable_all_nolock(); |
5206 | r = -EBUSY; | |
5207 | } | |
5208 | } | |
5209 | ||
0bf50497 | 5210 | mutex_unlock(&kvm_lock); |
e4aa7f88 | 5211 | cpus_read_unlock(); |
10474ae8 AG |
5212 | |
5213 | return r; | |
5214 | } | |
5215 | ||
9a2b85c6 | 5216 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 5217 | void *v) |
9a2b85c6 | 5218 | { |
8e1c1815 SY |
5219 | /* |
5220 | * Some (well, at least mine) BIOSes hang on reboot if | |
5221 | * in vmx root mode. | |
5222 | * | |
5223 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
5224 | */ | |
1170adc6 | 5225 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 5226 | kvm_rebooting = true; |
75b7127c | 5227 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
5228 | return NOTIFY_OK; |
5229 | } | |
5230 | ||
5231 | static struct notifier_block kvm_reboot_notifier = { | |
5232 | .notifier_call = kvm_reboot, | |
5233 | .priority = 0, | |
5234 | }; | |
5235 | ||
35774a9f SC |
5236 | static int kvm_suspend(void) |
5237 | { | |
5238 | /* | |
5239 | * Secondary CPUs and CPU hotplug are disabled across the suspend/resume | |
5240 | * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count | |
5241 | * is stable. Assert that kvm_lock is not held to ensure the system | |
5242 | * isn't suspended while KVM is enabling hardware. Hardware enabling | |
5243 | * can be preempted, but the task cannot be frozen until it has dropped | |
5244 | * all locks (userspace tasks are frozen via a fake signal). | |
5245 | */ | |
5246 | lockdep_assert_not_held(&kvm_lock); | |
5247 | lockdep_assert_irqs_disabled(); | |
5248 | ||
5249 | if (kvm_usage_count) | |
5250 | hardware_disable_nolock(NULL); | |
5251 | return 0; | |
5252 | } | |
5253 | ||
5254 | static void kvm_resume(void) | |
5255 | { | |
5256 | lockdep_assert_not_held(&kvm_lock); | |
5257 | lockdep_assert_irqs_disabled(); | |
5258 | ||
5259 | if (kvm_usage_count) | |
5260 | WARN_ON_ONCE(__hardware_enable_nolock()); | |
5261 | } | |
5262 | ||
5263 | static struct syscore_ops kvm_syscore_ops = { | |
5264 | .suspend = kvm_suspend, | |
5265 | .resume = kvm_resume, | |
5266 | }; | |
441f7bfa SC |
5267 | #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */ |
5268 | static int hardware_enable_all(void) | |
5269 | { | |
5270 | return 0; | |
5271 | } | |
5272 | ||
5273 | static void hardware_disable_all(void) | |
5274 | { | |
5275 | ||
5276 | } | |
5277 | #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */ | |
35774a9f | 5278 | |
e93f8a0f | 5279 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
5280 | { |
5281 | int i; | |
5282 | ||
5283 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 5284 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
5285 | |
5286 | kvm_iodevice_destructor(pos); | |
5287 | } | |
e93f8a0f | 5288 | kfree(bus); |
2eeb2e94 GH |
5289 | } |
5290 | ||
c21fbff1 | 5291 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 5292 | const struct kvm_io_range *r2) |
743eeb0b | 5293 | { |
8f4216c7 JW |
5294 | gpa_t addr1 = r1->addr; |
5295 | gpa_t addr2 = r2->addr; | |
5296 | ||
5297 | if (addr1 < addr2) | |
743eeb0b | 5298 | return -1; |
8f4216c7 JW |
5299 | |
5300 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
5301 | * accept any overlapping write. Any order is acceptable for | |
5302 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
5303 | * we process all of them. | |
5304 | */ | |
5305 | if (r2->len) { | |
5306 | addr1 += r1->len; | |
5307 | addr2 += r2->len; | |
5308 | } | |
5309 | ||
5310 | if (addr1 > addr2) | |
743eeb0b | 5311 | return 1; |
8f4216c7 | 5312 | |
743eeb0b SL |
5313 | return 0; |
5314 | } | |
5315 | ||
a343c9b7 PB |
5316 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
5317 | { | |
c21fbff1 | 5318 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
5319 | } |
5320 | ||
39369f7a | 5321 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
5322 | gpa_t addr, int len) |
5323 | { | |
5324 | struct kvm_io_range *range, key; | |
5325 | int off; | |
5326 | ||
5327 | key = (struct kvm_io_range) { | |
5328 | .addr = addr, | |
5329 | .len = len, | |
5330 | }; | |
5331 | ||
5332 | range = bsearch(&key, bus->range, bus->dev_count, | |
5333 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
5334 | if (range == NULL) | |
5335 | return -ENOENT; | |
5336 | ||
5337 | off = range - bus->range; | |
5338 | ||
c21fbff1 | 5339 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
5340 | off--; |
5341 | ||
5342 | return off; | |
5343 | } | |
5344 | ||
e32edf4f | 5345 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
5346 | struct kvm_io_range *range, const void *val) |
5347 | { | |
5348 | int idx; | |
5349 | ||
5350 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
5351 | if (idx < 0) | |
5352 | return -EOPNOTSUPP; | |
5353 | ||
5354 | while (idx < bus->dev_count && | |
c21fbff1 | 5355 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5356 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5357 | range->len, val)) |
5358 | return idx; | |
5359 | idx++; | |
5360 | } | |
5361 | ||
5362 | return -EOPNOTSUPP; | |
5363 | } | |
5364 | ||
bda9020e | 5365 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 5366 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 5367 | int len, const void *val) |
2eeb2e94 | 5368 | { |
90d83dc3 | 5369 | struct kvm_io_bus *bus; |
743eeb0b | 5370 | struct kvm_io_range range; |
126a5af5 | 5371 | int r; |
743eeb0b SL |
5372 | |
5373 | range = (struct kvm_io_range) { | |
5374 | .addr = addr, | |
5375 | .len = len, | |
5376 | }; | |
90d83dc3 | 5377 | |
e32edf4f | 5378 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5379 | if (!bus) |
5380 | return -ENOMEM; | |
e32edf4f | 5381 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5382 | return r < 0 ? r : 0; |
5383 | } | |
a2420107 | 5384 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
5385 | |
5386 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
5387 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
5388 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
5389 | { |
5390 | struct kvm_io_bus *bus; | |
5391 | struct kvm_io_range range; | |
5392 | ||
5393 | range = (struct kvm_io_range) { | |
5394 | .addr = addr, | |
5395 | .len = len, | |
5396 | }; | |
5397 | ||
e32edf4f | 5398 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5399 | if (!bus) |
5400 | return -ENOMEM; | |
126a5af5 CH |
5401 | |
5402 | /* First try the device referenced by cookie. */ | |
5403 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 5404 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 5405 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
5406 | val)) |
5407 | return cookie; | |
5408 | ||
5409 | /* | |
5410 | * cookie contained garbage; fall back to search and return the | |
5411 | * correct cookie value. | |
5412 | */ | |
e32edf4f | 5413 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5414 | } |
5415 | ||
e32edf4f NN |
5416 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
5417 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
5418 | { |
5419 | int idx; | |
5420 | ||
5421 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
5422 | if (idx < 0) |
5423 | return -EOPNOTSUPP; | |
5424 | ||
5425 | while (idx < bus->dev_count && | |
c21fbff1 | 5426 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5427 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5428 | range->len, val)) |
5429 | return idx; | |
743eeb0b SL |
5430 | idx++; |
5431 | } | |
5432 | ||
bda9020e MT |
5433 | return -EOPNOTSUPP; |
5434 | } | |
2eeb2e94 | 5435 | |
bda9020e | 5436 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 5437 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 5438 | int len, void *val) |
bda9020e | 5439 | { |
90d83dc3 | 5440 | struct kvm_io_bus *bus; |
743eeb0b | 5441 | struct kvm_io_range range; |
126a5af5 | 5442 | int r; |
743eeb0b SL |
5443 | |
5444 | range = (struct kvm_io_range) { | |
5445 | .addr = addr, | |
5446 | .len = len, | |
5447 | }; | |
e93f8a0f | 5448 | |
e32edf4f | 5449 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5450 | if (!bus) |
5451 | return -ENOMEM; | |
e32edf4f | 5452 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
5453 | return r < 0 ? r : 0; |
5454 | } | |
743eeb0b | 5455 | |
79fac95e | 5456 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
5457 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
5458 | int len, struct kvm_io_device *dev) | |
6c474694 | 5459 | { |
d4c67a7a | 5460 | int i; |
e93f8a0f | 5461 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 5462 | struct kvm_io_range range; |
090b7aff | 5463 | |
4a12f951 | 5464 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
5465 | if (!bus) |
5466 | return -ENOMEM; | |
5467 | ||
6ea34c9b AK |
5468 | /* exclude ioeventfd which is limited by maximum fd */ |
5469 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 5470 | return -ENOSPC; |
2eeb2e94 | 5471 | |
90952cd3 | 5472 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 5473 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
5474 | if (!new_bus) |
5475 | return -ENOMEM; | |
d4c67a7a GH |
5476 | |
5477 | range = (struct kvm_io_range) { | |
5478 | .addr = addr, | |
5479 | .len = len, | |
5480 | .dev = dev, | |
5481 | }; | |
5482 | ||
5483 | for (i = 0; i < bus->dev_count; i++) | |
5484 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
5485 | break; | |
5486 | ||
5487 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
5488 | new_bus->dev_count++; | |
5489 | new_bus->range[i] = range; | |
5490 | memcpy(new_bus->range + i + 1, bus->range + i, | |
5491 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
5492 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
5493 | synchronize_srcu_expedited(&kvm->srcu); | |
5494 | kfree(bus); | |
090b7aff GH |
5495 | |
5496 | return 0; | |
5497 | } | |
5498 | ||
5d3c4c79 SC |
5499 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5500 | struct kvm_io_device *dev) | |
090b7aff | 5501 | { |
f6588660 | 5502 | int i, j; |
e93f8a0f | 5503 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 5504 | |
7c896d37 SC |
5505 | lockdep_assert_held(&kvm->slots_lock); |
5506 | ||
4a12f951 | 5507 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 5508 | if (!bus) |
5d3c4c79 | 5509 | return 0; |
df630b8c | 5510 | |
7c896d37 | 5511 | for (i = 0; i < bus->dev_count; i++) { |
a1300716 | 5512 | if (bus->range[i].dev == dev) { |
090b7aff GH |
5513 | break; |
5514 | } | |
7c896d37 | 5515 | } |
e93f8a0f | 5516 | |
90db1043 | 5517 | if (i == bus->dev_count) |
5d3c4c79 | 5518 | return 0; |
a1300716 | 5519 | |
90952cd3 | 5520 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 5521 | GFP_KERNEL_ACCOUNT); |
f6588660 | 5522 | if (new_bus) { |
871c433b | 5523 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
5524 | new_bus->dev_count--; |
5525 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 5526 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
2ee37574 SC |
5527 | } |
5528 | ||
5529 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
5530 | synchronize_srcu_expedited(&kvm->srcu); | |
5531 | ||
5532 | /* Destroy the old bus _after_ installing the (null) bus. */ | |
5533 | if (!new_bus) { | |
90db1043 | 5534 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
f6588660 RK |
5535 | for (j = 0; j < bus->dev_count; j++) { |
5536 | if (j == i) | |
5537 | continue; | |
5538 | kvm_iodevice_destructor(bus->range[j].dev); | |
5539 | } | |
90db1043 | 5540 | } |
a1300716 | 5541 | |
e93f8a0f | 5542 | kfree(bus); |
5d3c4c79 | 5543 | return new_bus ? 0 : -ENOMEM; |
2eeb2e94 GH |
5544 | } |
5545 | ||
8a39d006 AP |
5546 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5547 | gpa_t addr) | |
5548 | { | |
5549 | struct kvm_io_bus *bus; | |
5550 | int dev_idx, srcu_idx; | |
5551 | struct kvm_io_device *iodev = NULL; | |
5552 | ||
5553 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
5554 | ||
5555 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
5556 | if (!bus) |
5557 | goto out_unlock; | |
8a39d006 AP |
5558 | |
5559 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
5560 | if (dev_idx < 0) | |
5561 | goto out_unlock; | |
5562 | ||
5563 | iodev = bus->range[dev_idx].dev; | |
5564 | ||
5565 | out_unlock: | |
5566 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
5567 | ||
5568 | return iodev; | |
5569 | } | |
5570 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
5571 | ||
536a6f88 JF |
5572 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
5573 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
5574 | const char *fmt) | |
5575 | { | |
180418e2 | 5576 | int ret; |
14aa40a1 | 5577 | struct kvm_stat_data *stat_data = inode->i_private; |
536a6f88 | 5578 | |
605c7130 PX |
5579 | /* |
5580 | * The debugfs files are a reference to the kvm struct which | |
5581 | * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe | |
5582 | * avoids the race between open and the removal of the debugfs directory. | |
536a6f88 | 5583 | */ |
605c7130 | 5584 | if (!kvm_get_kvm_safe(stat_data->kvm)) |
536a6f88 JF |
5585 | return -ENOENT; |
5586 | ||
180418e2 HW |
5587 | ret = simple_attr_open(inode, file, get, |
5588 | kvm_stats_debugfs_mode(stat_data->desc) & 0222 | |
5589 | ? set : NULL, fmt); | |
5590 | if (ret) | |
536a6f88 | 5591 | kvm_put_kvm(stat_data->kvm); |
536a6f88 | 5592 | |
180418e2 | 5593 | return ret; |
536a6f88 JF |
5594 | } |
5595 | ||
5596 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
5597 | { | |
14aa40a1 | 5598 | struct kvm_stat_data *stat_data = inode->i_private; |
536a6f88 JF |
5599 | |
5600 | simple_attr_release(inode, file); | |
5601 | kvm_put_kvm(stat_data->kvm); | |
5602 | ||
5603 | return 0; | |
5604 | } | |
5605 | ||
09cbcef6 | 5606 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 5607 | { |
bc9e9e67 | 5608 | *val = *(u64 *)((void *)(&kvm->stat) + offset); |
536a6f88 | 5609 | |
09cbcef6 MP |
5610 | return 0; |
5611 | } | |
5612 | ||
5613 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
5614 | { | |
bc9e9e67 | 5615 | *(u64 *)((void *)(&kvm->stat) + offset) = 0; |
536a6f88 JF |
5616 | |
5617 | return 0; | |
5618 | } | |
5619 | ||
09cbcef6 | 5620 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 5621 | { |
46808a4c | 5622 | unsigned long i; |
09cbcef6 | 5623 | struct kvm_vcpu *vcpu; |
ce35ef27 | 5624 | |
09cbcef6 | 5625 | *val = 0; |
ce35ef27 | 5626 | |
09cbcef6 | 5627 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5628 | *val += *(u64 *)((void *)(&vcpu->stat) + offset); |
ce35ef27 SJS |
5629 | |
5630 | return 0; | |
5631 | } | |
5632 | ||
09cbcef6 | 5633 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 5634 | { |
46808a4c | 5635 | unsigned long i; |
09cbcef6 | 5636 | struct kvm_vcpu *vcpu; |
536a6f88 | 5637 | |
09cbcef6 | 5638 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5639 | *(u64 *)((void *)(&vcpu->stat) + offset) = 0; |
09cbcef6 MP |
5640 | |
5641 | return 0; | |
5642 | } | |
536a6f88 | 5643 | |
09cbcef6 | 5644 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 5645 | { |
09cbcef6 | 5646 | int r = -EFAULT; |
14aa40a1 | 5647 | struct kvm_stat_data *stat_data = data; |
536a6f88 | 5648 | |
bc9e9e67 | 5649 | switch (stat_data->kind) { |
09cbcef6 MP |
5650 | case KVM_STAT_VM: |
5651 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5652 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5653 | break; |
5654 | case KVM_STAT_VCPU: | |
5655 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5656 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5657 | break; |
5658 | } | |
536a6f88 | 5659 | |
09cbcef6 | 5660 | return r; |
536a6f88 JF |
5661 | } |
5662 | ||
09cbcef6 | 5663 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 5664 | { |
09cbcef6 | 5665 | int r = -EFAULT; |
14aa40a1 | 5666 | struct kvm_stat_data *stat_data = data; |
ce35ef27 SJS |
5667 | |
5668 | if (val) | |
5669 | return -EINVAL; | |
5670 | ||
bc9e9e67 | 5671 | switch (stat_data->kind) { |
09cbcef6 MP |
5672 | case KVM_STAT_VM: |
5673 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5674 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5675 | break; |
5676 | case KVM_STAT_VCPU: | |
5677 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5678 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5679 | break; |
5680 | } | |
ce35ef27 | 5681 | |
09cbcef6 | 5682 | return r; |
ce35ef27 SJS |
5683 | } |
5684 | ||
09cbcef6 | 5685 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
5686 | { |
5687 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
5688 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
5689 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
5690 | } |
5691 | ||
09cbcef6 MP |
5692 | static const struct file_operations stat_fops_per_vm = { |
5693 | .owner = THIS_MODULE, | |
5694 | .open = kvm_stat_data_open, | |
536a6f88 | 5695 | .release = kvm_debugfs_release, |
09cbcef6 MP |
5696 | .read = simple_attr_read, |
5697 | .write = simple_attr_write, | |
5698 | .llseek = no_llseek, | |
536a6f88 JF |
5699 | }; |
5700 | ||
8b88b099 | 5701 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
5702 | { |
5703 | unsigned offset = (long)_offset; | |
ba1389b7 | 5704 | struct kvm *kvm; |
536a6f88 | 5705 | u64 tmp_val; |
ba1389b7 | 5706 | |
8b88b099 | 5707 | *val = 0; |
0d9ce162 | 5708 | mutex_lock(&kvm_lock); |
536a6f88 | 5709 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5710 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
5711 | *val += tmp_val; |
5712 | } | |
0d9ce162 | 5713 | mutex_unlock(&kvm_lock); |
8b88b099 | 5714 | return 0; |
ba1389b7 AK |
5715 | } |
5716 | ||
ce35ef27 SJS |
5717 | static int vm_stat_clear(void *_offset, u64 val) |
5718 | { | |
5719 | unsigned offset = (long)_offset; | |
5720 | struct kvm *kvm; | |
ce35ef27 SJS |
5721 | |
5722 | if (val) | |
5723 | return -EINVAL; | |
5724 | ||
0d9ce162 | 5725 | mutex_lock(&kvm_lock); |
ce35ef27 | 5726 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5727 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 5728 | } |
0d9ce162 | 5729 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5730 | |
5731 | return 0; | |
5732 | } | |
5733 | ||
5734 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
bc9e9e67 | 5735 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n"); |
ba1389b7 | 5736 | |
8b88b099 | 5737 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
5738 | { |
5739 | unsigned offset = (long)_offset; | |
1165f5fe | 5740 | struct kvm *kvm; |
536a6f88 | 5741 | u64 tmp_val; |
1165f5fe | 5742 | |
8b88b099 | 5743 | *val = 0; |
0d9ce162 | 5744 | mutex_lock(&kvm_lock); |
536a6f88 | 5745 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5746 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
5747 | *val += tmp_val; |
5748 | } | |
0d9ce162 | 5749 | mutex_unlock(&kvm_lock); |
8b88b099 | 5750 | return 0; |
1165f5fe AK |
5751 | } |
5752 | ||
ce35ef27 SJS |
5753 | static int vcpu_stat_clear(void *_offset, u64 val) |
5754 | { | |
5755 | unsigned offset = (long)_offset; | |
5756 | struct kvm *kvm; | |
ce35ef27 SJS |
5757 | |
5758 | if (val) | |
5759 | return -EINVAL; | |
5760 | ||
0d9ce162 | 5761 | mutex_lock(&kvm_lock); |
ce35ef27 | 5762 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5763 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 5764 | } |
0d9ce162 | 5765 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5766 | |
5767 | return 0; | |
5768 | } | |
5769 | ||
5770 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
5771 | "%llu\n"); | |
bc9e9e67 | 5772 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n"); |
1165f5fe | 5773 | |
286de8f6 CI |
5774 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
5775 | { | |
5776 | struct kobj_uevent_env *env; | |
286de8f6 CI |
5777 | unsigned long long created, active; |
5778 | ||
5779 | if (!kvm_dev.this_device || !kvm) | |
5780 | return; | |
5781 | ||
0d9ce162 | 5782 | mutex_lock(&kvm_lock); |
286de8f6 CI |
5783 | if (type == KVM_EVENT_CREATE_VM) { |
5784 | kvm_createvm_count++; | |
5785 | kvm_active_vms++; | |
5786 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
5787 | kvm_active_vms--; | |
5788 | } | |
5789 | created = kvm_createvm_count; | |
5790 | active = kvm_active_vms; | |
0d9ce162 | 5791 | mutex_unlock(&kvm_lock); |
286de8f6 | 5792 | |
b12ce36a | 5793 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
5794 | if (!env) |
5795 | return; | |
5796 | ||
5797 | add_uevent_var(env, "CREATED=%llu", created); | |
5798 | add_uevent_var(env, "COUNT=%llu", active); | |
5799 | ||
fdeaf7e3 | 5800 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 5801 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
5802 | kvm->userspace_pid = task_pid_nr(current); |
5803 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 5804 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
5805 | } |
5806 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 5807 | |
a44a4cc1 | 5808 | if (!IS_ERR(kvm->debugfs_dentry)) { |
b12ce36a | 5809 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
5810 | |
5811 | if (p) { | |
5812 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
5813 | if (!IS_ERR(tmp)) | |
5814 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
5815 | kfree(p); | |
286de8f6 CI |
5816 | } |
5817 | } | |
5818 | /* no need for checks, since we are adding at most only 5 keys */ | |
5819 | env->envp[env->envp_idx++] = NULL; | |
5820 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
5821 | kfree(env); | |
286de8f6 CI |
5822 | } |
5823 | ||
929f45e3 | 5824 | static void kvm_init_debug(void) |
6aa8b732 | 5825 | { |
bc9e9e67 JZ |
5826 | const struct file_operations *fops; |
5827 | const struct _kvm_stats_desc *pdesc; | |
5828 | int i; | |
6aa8b732 | 5829 | |
76f7c879 | 5830 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 5831 | |
bc9e9e67 JZ |
5832 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
5833 | pdesc = &kvm_vm_stats_desc[i]; | |
5834 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5835 | fops = &vm_stat_fops; | |
5836 | else | |
5837 | fops = &vm_stat_readonly_fops; | |
5838 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5839 | kvm_debugfs_dir, | |
5840 | (void *)(long)pdesc->desc.offset, fops); | |
5841 | } | |
5842 | ||
5843 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
5844 | pdesc = &kvm_vcpu_stats_desc[i]; | |
5845 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5846 | fops = &vcpu_stat_fops; | |
5847 | else | |
5848 | fops = &vcpu_stat_readonly_fops; | |
5849 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5850 | kvm_debugfs_dir, | |
5851 | (void *)(long)pdesc->desc.offset, fops); | |
4f69b680 | 5852 | } |
6aa8b732 AK |
5853 | } |
5854 | ||
15ad7146 AK |
5855 | static inline |
5856 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
5857 | { | |
5858 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
5859 | } | |
5860 | ||
5861 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
5862 | { | |
5863 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 5864 | |
046ddeed | 5865 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 5866 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 5867 | |
7495e22b | 5868 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 5869 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 5870 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
5871 | } |
5872 | ||
5873 | static void kvm_sched_out(struct preempt_notifier *pn, | |
5874 | struct task_struct *next) | |
5875 | { | |
5876 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
5877 | ||
3ba9f93b | 5878 | if (current->on_rq) { |
046ddeed | 5879 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
5880 | WRITE_ONCE(vcpu->ready, true); |
5881 | } | |
e9b11c17 | 5882 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
5883 | __this_cpu_write(kvm_running_vcpu, NULL); |
5884 | } | |
5885 | ||
5886 | /** | |
5887 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
5888 | * |
5889 | * We can disable preemption locally around accessing the per-CPU variable, | |
5890 | * and use the resolved vcpu pointer after enabling preemption again, | |
5891 | * because even if the current thread is migrated to another CPU, reading | |
5892 | * the per-CPU value later will give us the same value as we update the | |
5893 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
5894 | */ |
5895 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
5896 | { | |
1f03b2bc MZ |
5897 | struct kvm_vcpu *vcpu; |
5898 | ||
5899 | preempt_disable(); | |
5900 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
5901 | preempt_enable(); | |
5902 | ||
5903 | return vcpu; | |
7495e22b | 5904 | } |
379a3c8e | 5905 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
5906 | |
5907 | /** | |
5908 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
5909 | */ | |
5910 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
5911 | { | |
5912 | return &kvm_running_vcpu; | |
15ad7146 AK |
5913 | } |
5914 | ||
e1bfc245 SC |
5915 | #ifdef CONFIG_GUEST_PERF_EVENTS |
5916 | static unsigned int kvm_guest_state(void) | |
5917 | { | |
5918 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
5919 | unsigned int state; | |
5920 | ||
5921 | if (!kvm_arch_pmi_in_guest(vcpu)) | |
5922 | return 0; | |
5923 | ||
5924 | state = PERF_GUEST_ACTIVE; | |
5925 | if (!kvm_arch_vcpu_in_kernel(vcpu)) | |
5926 | state |= PERF_GUEST_USER; | |
5927 | ||
5928 | return state; | |
5929 | } | |
5930 | ||
5931 | static unsigned long kvm_guest_get_ip(void) | |
5932 | { | |
5933 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
5934 | ||
5935 | /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */ | |
5936 | if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu))) | |
5937 | return 0; | |
5938 | ||
5939 | return kvm_arch_vcpu_get_ip(vcpu); | |
5940 | } | |
5941 | ||
5942 | static struct perf_guest_info_callbacks kvm_guest_cbs = { | |
5943 | .state = kvm_guest_state, | |
5944 | .get_ip = kvm_guest_get_ip, | |
5945 | .handle_intel_pt_intr = NULL, | |
5946 | }; | |
5947 | ||
5948 | void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void)) | |
5949 | { | |
5950 | kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler; | |
5951 | perf_register_guest_info_callbacks(&kvm_guest_cbs); | |
5952 | } | |
5953 | void kvm_unregister_perf_callbacks(void) | |
5954 | { | |
5955 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); | |
5956 | } | |
5957 | #endif | |
5958 | ||
81a1cf9f | 5959 | int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module) |
f257d6dc | 5960 | { |
6aa8b732 | 5961 | int r; |
002c7f7c | 5962 | int cpu; |
6aa8b732 | 5963 | |
441f7bfa | 5964 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
aaf12a7b CG |
5965 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE, "kvm/cpu:online", |
5966 | kvm_online_cpu, kvm_offline_cpu); | |
774c47f1 | 5967 | if (r) |
37d25881 SC |
5968 | return r; |
5969 | ||
6aa8b732 | 5970 | register_reboot_notifier(&kvm_reboot_notifier); |
35774a9f | 5971 | register_syscore_ops(&kvm_syscore_ops); |
441f7bfa | 5972 | #endif |
6aa8b732 | 5973 | |
c16f862d | 5974 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
5975 | if (!vcpu_align) |
5976 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
5977 | kvm_vcpu_cache = |
5978 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
5979 | SLAB_ACCOUNT, | |
5980 | offsetof(struct kvm_vcpu, arch), | |
ce55c049 JZ |
5981 | offsetofend(struct kvm_vcpu, stats_id) |
5982 | - offsetof(struct kvm_vcpu, arch), | |
46515736 | 5983 | NULL); |
c16f862d RR |
5984 | if (!kvm_vcpu_cache) { |
5985 | r = -ENOMEM; | |
9f1a4c00 | 5986 | goto err_vcpu_cache; |
c16f862d RR |
5987 | } |
5988 | ||
baff59cc VK |
5989 | for_each_possible_cpu(cpu) { |
5990 | if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu), | |
5991 | GFP_KERNEL, cpu_to_node(cpu))) { | |
5992 | r = -ENOMEM; | |
9f1a4c00 | 5993 | goto err_cpu_kick_mask; |
baff59cc VK |
5994 | } |
5995 | } | |
5996 | ||
5910ccf0 SC |
5997 | r = kvm_irqfd_init(); |
5998 | if (r) | |
5999 | goto err_irqfd; | |
6000 | ||
af585b92 GN |
6001 | r = kvm_async_pf_init(); |
6002 | if (r) | |
5910ccf0 | 6003 | goto err_async_pf; |
af585b92 | 6004 | |
6aa8b732 AK |
6005 | kvm_chardev_ops.owner = module; |
6006 | ||
15ad7146 AK |
6007 | kvm_preempt_ops.sched_in = kvm_sched_in; |
6008 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
6009 | ||
929f45e3 | 6010 | kvm_init_debug(); |
0ea4ed8e | 6011 | |
3c3c29fd | 6012 | r = kvm_vfio_ops_init(); |
2b012812 SC |
6013 | if (WARN_ON_ONCE(r)) |
6014 | goto err_vfio; | |
6015 | ||
6016 | /* | |
6017 | * Registration _must_ be the very last thing done, as this exposes | |
6018 | * /dev/kvm to userspace, i.e. all infrastructure must be setup! | |
6019 | */ | |
6020 | r = misc_register(&kvm_dev); | |
6021 | if (r) { | |
6022 | pr_err("kvm: misc device register failed\n"); | |
6023 | goto err_register; | |
6024 | } | |
3c3c29fd | 6025 | |
c7addb90 | 6026 | return 0; |
6aa8b732 | 6027 | |
2b012812 SC |
6028 | err_register: |
6029 | kvm_vfio_ops_exit(); | |
6030 | err_vfio: | |
af585b92 | 6031 | kvm_async_pf_deinit(); |
5910ccf0 SC |
6032 | err_async_pf: |
6033 | kvm_irqfd_exit(); | |
6034 | err_irqfd: | |
9f1a4c00 | 6035 | err_cpu_kick_mask: |
baff59cc VK |
6036 | for_each_possible_cpu(cpu) |
6037 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
c16f862d | 6038 | kmem_cache_destroy(kvm_vcpu_cache); |
9f1a4c00 | 6039 | err_vcpu_cache: |
441f7bfa | 6040 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
35774a9f | 6041 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 6042 | unregister_reboot_notifier(&kvm_reboot_notifier); |
aaf12a7b | 6043 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE); |
441f7bfa | 6044 | #endif |
6aa8b732 AK |
6045 | return r; |
6046 | } | |
cb498ea2 | 6047 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 6048 | |
cb498ea2 | 6049 | void kvm_exit(void) |
6aa8b732 | 6050 | { |
baff59cc VK |
6051 | int cpu; |
6052 | ||
2b012812 SC |
6053 | /* |
6054 | * Note, unregistering /dev/kvm doesn't strictly need to come first, | |
6055 | * fops_get(), a.k.a. try_module_get(), prevents acquiring references | |
6056 | * to KVM while the module is being stopped. | |
6057 | */ | |
6aa8b732 | 6058 | misc_deregister(&kvm_dev); |
2b012812 SC |
6059 | |
6060 | debugfs_remove_recursive(kvm_debugfs_dir); | |
baff59cc VK |
6061 | for_each_possible_cpu(cpu) |
6062 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
c16f862d | 6063 | kmem_cache_destroy(kvm_vcpu_cache); |
73b8dc04 | 6064 | kvm_vfio_ops_exit(); |
af585b92 | 6065 | kvm_async_pf_deinit(); |
441f7bfa | 6066 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
fb3600cc | 6067 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 6068 | unregister_reboot_notifier(&kvm_reboot_notifier); |
aaf12a7b | 6069 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE); |
441f7bfa | 6070 | #endif |
5910ccf0 | 6071 | kvm_irqfd_exit(); |
6aa8b732 | 6072 | } |
cb498ea2 | 6073 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
6074 | |
6075 | struct kvm_vm_worker_thread_context { | |
6076 | struct kvm *kvm; | |
6077 | struct task_struct *parent; | |
6078 | struct completion init_done; | |
6079 | kvm_vm_thread_fn_t thread_fn; | |
6080 | uintptr_t data; | |
6081 | int err; | |
6082 | }; | |
6083 | ||
6084 | static int kvm_vm_worker_thread(void *context) | |
6085 | { | |
6086 | /* | |
6087 | * The init_context is allocated on the stack of the parent thread, so | |
6088 | * we have to locally copy anything that is needed beyond initialization | |
6089 | */ | |
6090 | struct kvm_vm_worker_thread_context *init_context = context; | |
e45cce30 | 6091 | struct task_struct *parent; |
c57c8046 JS |
6092 | struct kvm *kvm = init_context->kvm; |
6093 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
6094 | uintptr_t data = init_context->data; | |
6095 | int err; | |
6096 | ||
6097 | err = kthread_park(current); | |
6098 | /* kthread_park(current) is never supposed to return an error */ | |
6099 | WARN_ON(err != 0); | |
6100 | if (err) | |
6101 | goto init_complete; | |
6102 | ||
6103 | err = cgroup_attach_task_all(init_context->parent, current); | |
6104 | if (err) { | |
6105 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
6106 | __func__, err); | |
6107 | goto init_complete; | |
6108 | } | |
6109 | ||
6110 | set_user_nice(current, task_nice(init_context->parent)); | |
6111 | ||
6112 | init_complete: | |
6113 | init_context->err = err; | |
6114 | complete(&init_context->init_done); | |
6115 | init_context = NULL; | |
6116 | ||
6117 | if (err) | |
e45cce30 | 6118 | goto out; |
c57c8046 JS |
6119 | |
6120 | /* Wait to be woken up by the spawner before proceeding. */ | |
6121 | kthread_parkme(); | |
6122 | ||
6123 | if (!kthread_should_stop()) | |
6124 | err = thread_fn(kvm, data); | |
6125 | ||
e45cce30 VS |
6126 | out: |
6127 | /* | |
6128 | * Move kthread back to its original cgroup to prevent it lingering in | |
6129 | * the cgroup of the VM process, after the latter finishes its | |
6130 | * execution. | |
6131 | * | |
6132 | * kthread_stop() waits on the 'exited' completion condition which is | |
6133 | * set in exit_mm(), via mm_release(), in do_exit(). However, the | |
6134 | * kthread is removed from the cgroup in the cgroup_exit() which is | |
6135 | * called after the exit_mm(). This causes the kthread_stop() to return | |
6136 | * before the kthread actually quits the cgroup. | |
6137 | */ | |
6138 | rcu_read_lock(); | |
6139 | parent = rcu_dereference(current->real_parent); | |
6140 | get_task_struct(parent); | |
6141 | rcu_read_unlock(); | |
6142 | cgroup_attach_task_all(parent, current); | |
6143 | put_task_struct(parent); | |
6144 | ||
c57c8046 JS |
6145 | return err; |
6146 | } | |
6147 | ||
6148 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
6149 | uintptr_t data, const char *name, | |
6150 | struct task_struct **thread_ptr) | |
6151 | { | |
6152 | struct kvm_vm_worker_thread_context init_context = {}; | |
6153 | struct task_struct *thread; | |
6154 | ||
6155 | *thread_ptr = NULL; | |
6156 | init_context.kvm = kvm; | |
6157 | init_context.parent = current; | |
6158 | init_context.thread_fn = thread_fn; | |
6159 | init_context.data = data; | |
6160 | init_completion(&init_context.init_done); | |
6161 | ||
6162 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
6163 | "%s-%d", name, task_pid_nr(current)); | |
6164 | if (IS_ERR(thread)) | |
6165 | return PTR_ERR(thread); | |
6166 | ||
6167 | /* kthread_run is never supposed to return NULL */ | |
6168 | WARN_ON(thread == NULL); | |
6169 | ||
6170 | wait_for_completion(&init_context.init_done); | |
6171 | ||
6172 | if (!init_context.err) | |
6173 | *thread_ptr = thread; | |
6174 | ||
6175 | return init_context.err; | |
6176 | } |