]>
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 | |
229456fc MT |
65 | #define CREATE_TRACE_POINTS |
66 | #include <trace/events/kvm.h> | |
67 | ||
fb04a1ed PX |
68 | #include <linux/kvm_dirty_ring.h> |
69 | ||
536a6f88 JF |
70 | /* Worst case buffer size needed for holding an integer. */ |
71 | #define ITOA_MAX_LEN 12 | |
72 | ||
6aa8b732 AK |
73 | MODULE_AUTHOR("Qumranet"); |
74 | MODULE_LICENSE("GPL"); | |
75 | ||
920552b2 | 76 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 77 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
039c5d1b | 78 | module_param(halt_poll_ns, uint, 0644); |
ec76d819 | 79 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 80 | |
aca6ff29 | 81 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 82 | unsigned int halt_poll_ns_grow = 2; |
039c5d1b | 83 | module_param(halt_poll_ns_grow, uint, 0644); |
ec76d819 | 84 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 | 85 | |
49113d36 NW |
86 | /* The start value to grow halt_poll_ns from */ |
87 | unsigned int halt_poll_ns_grow_start = 10000; /* 10us */ | |
88 | module_param(halt_poll_ns_grow_start, uint, 0644); | |
89 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start); | |
90 | ||
aca6ff29 | 91 | /* Default resets per-vcpu halt_poll_ns . */ |
ec76d819 | 92 | unsigned int halt_poll_ns_shrink; |
039c5d1b | 93 | module_param(halt_poll_ns_shrink, uint, 0644); |
ec76d819 | 94 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 95 | |
fa40a821 MT |
96 | /* |
97 | * Ordering of locks: | |
98 | * | |
b7d409de | 99 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
100 | */ |
101 | ||
0d9ce162 | 102 | DEFINE_MUTEX(kvm_lock); |
4a937f96 | 103 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 104 | LIST_HEAD(vm_list); |
133de902 | 105 | |
7f59f492 | 106 | static cpumask_var_t cpus_hardware_enabled; |
f4fee932 | 107 | static int kvm_usage_count; |
10474ae8 | 108 | static atomic_t hardware_enable_failed; |
1b6c0168 | 109 | |
aaba298c | 110 | static struct kmem_cache *kvm_vcpu_cache; |
1165f5fe | 111 | |
15ad7146 | 112 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
7495e22b | 113 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu); |
15ad7146 | 114 | |
76f7c879 | 115 | struct dentry *kvm_debugfs_dir; |
e23a808b | 116 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 117 | |
09cbcef6 | 118 | static const struct file_operations stat_fops_per_vm; |
536a6f88 | 119 | |
5f6de5cb DM |
120 | static struct file_operations kvm_chardev_ops; |
121 | ||
bccf2150 AK |
122 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
123 | unsigned long arg); | |
de8e5d74 | 124 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
125 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
126 | unsigned long arg); | |
7ddfd3e0 MZ |
127 | #define KVM_COMPAT(c) .compat_ioctl = (c) |
128 | #else | |
9cb09e7c MZ |
129 | /* |
130 | * For architectures that don't implement a compat infrastructure, | |
131 | * adopt a double line of defense: | |
132 | * - Prevent a compat task from opening /dev/kvm | |
133 | * - If the open has been done by a 64bit task, and the KVM fd | |
134 | * passed to a compat task, let the ioctls fail. | |
135 | */ | |
7ddfd3e0 MZ |
136 | static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl, |
137 | unsigned long arg) { return -EINVAL; } | |
b9876e6d MZ |
138 | |
139 | static int kvm_no_compat_open(struct inode *inode, struct file *file) | |
140 | { | |
141 | return is_compat_task() ? -ENODEV : 0; | |
142 | } | |
143 | #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \ | |
144 | .open = kvm_no_compat_open | |
1dda606c | 145 | #endif |
10474ae8 AG |
146 | static int hardware_enable_all(void); |
147 | static void hardware_disable_all(void); | |
bccf2150 | 148 | |
e93f8a0f | 149 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 150 | |
52480137 | 151 | __visible bool kvm_rebooting; |
b7c4145b | 152 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 153 | |
286de8f6 CI |
154 | #define KVM_EVENT_CREATE_VM 0 |
155 | #define KVM_EVENT_DESTROY_VM 1 | |
156 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); | |
157 | static unsigned long long kvm_createvm_count; | |
158 | static unsigned long long kvm_active_vms; | |
159 | ||
baff59cc VK |
160 | static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask); |
161 | ||
e649b3f0 ET |
162 | __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
163 | unsigned long start, unsigned long end) | |
b1394e74 RK |
164 | { |
165 | } | |
166 | ||
683412cc MZ |
167 | __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm) |
168 | { | |
169 | } | |
170 | ||
284dc493 | 171 | bool kvm_is_zone_device_page(struct page *page) |
a78986aa SC |
172 | { |
173 | /* | |
174 | * The metadata used by is_zone_device_page() to determine whether or | |
175 | * not a page is ZONE_DEVICE is guaranteed to be valid if and only if | |
176 | * the device has been pinned, e.g. by get_user_pages(). WARN if the | |
177 | * page_count() is zero to help detect bad usage of this helper. | |
178 | */ | |
284dc493 | 179 | if (WARN_ON_ONCE(!page_count(page))) |
a78986aa SC |
180 | return false; |
181 | ||
284dc493 | 182 | return is_zone_device_page(page); |
a78986aa SC |
183 | } |
184 | ||
b14b2690 SC |
185 | /* |
186 | * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted | |
187 | * page, NULL otherwise. Note, the list of refcounted PG_reserved page types | |
188 | * is likely incomplete, it has been compiled purely through people wanting to | |
189 | * back guest with a certain type of memory and encountering issues. | |
190 | */ | |
191 | struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn) | |
cbff90a7 | 192 | { |
b14b2690 SC |
193 | struct page *page; |
194 | ||
195 | if (!pfn_valid(pfn)) | |
196 | return NULL; | |
197 | ||
198 | page = pfn_to_page(pfn); | |
199 | if (!PageReserved(page)) | |
200 | return page; | |
201 | ||
202 | /* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */ | |
203 | if (is_zero_pfn(pfn)) | |
204 | return page; | |
205 | ||
a78986aa SC |
206 | /* |
207 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting | |
208 | * perspective they are "normal" pages, albeit with slightly different | |
209 | * usage rules. | |
210 | */ | |
b14b2690 SC |
211 | if (kvm_is_zone_device_page(page)) |
212 | return page; | |
cbff90a7 | 213 | |
b14b2690 | 214 | return NULL; |
cbff90a7 BAY |
215 | } |
216 | ||
bccf2150 AK |
217 | /* |
218 | * Switches to specified vcpu, until a matching vcpu_put() | |
219 | */ | |
ec7660cc | 220 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 221 | { |
ec7660cc | 222 | int cpu = get_cpu(); |
7495e22b PB |
223 | |
224 | __this_cpu_write(kvm_running_vcpu, vcpu); | |
15ad7146 | 225 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 226 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 227 | put_cpu(); |
6aa8b732 | 228 | } |
2f1fe811 | 229 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 230 | |
313a3dc7 | 231 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 232 | { |
15ad7146 | 233 | preempt_disable(); |
313a3dc7 | 234 | kvm_arch_vcpu_put(vcpu); |
15ad7146 | 235 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
7495e22b | 236 | __this_cpu_write(kvm_running_vcpu, NULL); |
15ad7146 | 237 | preempt_enable(); |
6aa8b732 | 238 | } |
2f1fe811 | 239 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 240 | |
7a97cec2 PB |
241 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
242 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
243 | { | |
244 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
245 | ||
246 | /* | |
247 | * We need to wait for the VCPU to reenable interrupts and get out of | |
248 | * READING_SHADOW_PAGE_TABLES mode. | |
249 | */ | |
250 | if (req & KVM_REQUEST_WAIT) | |
251 | return mode != OUTSIDE_GUEST_MODE; | |
252 | ||
253 | /* | |
254 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
255 | */ | |
256 | return mode == IN_GUEST_MODE; | |
257 | } | |
258 | ||
f24b44e4 | 259 | static void ack_kick(void *_completed) |
d9e368d6 | 260 | { |
d9e368d6 AK |
261 | } |
262 | ||
620b2438 | 263 | static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait) |
b49defe8 | 264 | { |
b49defe8 PB |
265 | if (cpumask_empty(cpus)) |
266 | return false; | |
267 | ||
f24b44e4 | 268 | smp_call_function_many(cpus, ack_kick, NULL, wait); |
b49defe8 PB |
269 | return true; |
270 | } | |
271 | ||
b56bd8e0 JL |
272 | static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req, |
273 | struct cpumask *tmp, int current_cpu) | |
ae0946cd VK |
274 | { |
275 | int cpu; | |
276 | ||
df06dae3 SC |
277 | if (likely(!(req & KVM_REQUEST_NO_ACTION))) |
278 | __kvm_make_request(req, vcpu); | |
ae0946cd VK |
279 | |
280 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) | |
281 | return; | |
282 | ||
ae0946cd VK |
283 | /* |
284 | * Note, the vCPU could get migrated to a different pCPU at any point | |
285 | * after kvm_request_needs_ipi(), which could result in sending an IPI | |
286 | * to the previous pCPU. But, that's OK because the purpose of the IPI | |
287 | * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is | |
288 | * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES | |
289 | * after this point is also OK, as the requirement is only that KVM wait | |
290 | * for vCPUs that were reading SPTEs _before_ any changes were | |
291 | * finalized. See kvm_vcpu_kick() for more details on handling requests. | |
292 | */ | |
293 | if (kvm_request_needs_ipi(vcpu, req)) { | |
294 | cpu = READ_ONCE(vcpu->cpu); | |
295 | if (cpu != -1 && cpu != current_cpu) | |
296 | __cpumask_set_cpu(cpu, tmp); | |
297 | } | |
298 | } | |
299 | ||
7053df4e | 300 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
620b2438 | 301 | unsigned long *vcpu_bitmap) |
d9e368d6 | 302 | { |
d9e368d6 | 303 | struct kvm_vcpu *vcpu; |
620b2438 | 304 | struct cpumask *cpus; |
ae0946cd | 305 | int i, me; |
7053df4e | 306 | bool called; |
6ef7a1bc | 307 | |
3cba4130 | 308 | me = get_cpu(); |
7053df4e | 309 | |
620b2438 VK |
310 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
311 | cpumask_clear(cpus); | |
312 | ||
ae0946cd VK |
313 | for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) { |
314 | vcpu = kvm_get_vcpu(kvm, i); | |
381cecc5 | 315 | if (!vcpu) |
7053df4e | 316 | continue; |
b56bd8e0 | 317 | kvm_make_vcpu_request(vcpu, req, cpus, me); |
49846896 | 318 | } |
7053df4e | 319 | |
620b2438 | 320 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); |
3cba4130 | 321 | put_cpu(); |
7053df4e VK |
322 | |
323 | return called; | |
324 | } | |
325 | ||
54163a34 SS |
326 | bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req, |
327 | struct kvm_vcpu *except) | |
7053df4e | 328 | { |
ae0946cd | 329 | struct kvm_vcpu *vcpu; |
baff59cc | 330 | struct cpumask *cpus; |
46808a4c | 331 | unsigned long i; |
7053df4e | 332 | bool called; |
46808a4c | 333 | int me; |
7053df4e | 334 | |
ae0946cd VK |
335 | me = get_cpu(); |
336 | ||
baff59cc VK |
337 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
338 | cpumask_clear(cpus); | |
339 | ||
ae0946cd VK |
340 | kvm_for_each_vcpu(i, vcpu, kvm) { |
341 | if (vcpu == except) | |
342 | continue; | |
b56bd8e0 | 343 | kvm_make_vcpu_request(vcpu, req, cpus, me); |
ae0946cd VK |
344 | } |
345 | ||
346 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); | |
347 | put_cpu(); | |
7053df4e | 348 | |
49846896 | 349 | return called; |
d9e368d6 AK |
350 | } |
351 | ||
54163a34 SS |
352 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
353 | { | |
354 | return kvm_make_all_cpus_request_except(kvm, req, NULL); | |
355 | } | |
a2486020 | 356 | EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request); |
54163a34 | 357 | |
a6d51016 | 358 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 359 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 360 | { |
3cc4e148 | 361 | ++kvm->stat.generic.remote_tlb_flush_requests; |
6bc6db00 | 362 | |
4ae3cb3a LT |
363 | /* |
364 | * We want to publish modifications to the page tables before reading | |
365 | * mode. Pairs with a memory barrier in arch-specific code. | |
366 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
367 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
368 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
369 | * | |
370 | * There is already an smp_mb__after_atomic() before | |
371 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
372 | * barrier here. | |
373 | */ | |
b08660e5 TL |
374 | if (!kvm_arch_flush_remote_tlb(kvm) |
375 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
0193cc90 | 376 | ++kvm->stat.generic.remote_tlb_flush; |
2e53d63a | 377 | } |
2ba9f0d8 | 378 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 379 | #endif |
2e53d63a | 380 | |
683412cc MZ |
381 | static void kvm_flush_shadow_all(struct kvm *kvm) |
382 | { | |
383 | kvm_arch_flush_shadow_all(kvm); | |
384 | kvm_arch_guest_memory_reclaimed(kvm); | |
385 | } | |
386 | ||
6926f95a SC |
387 | #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE |
388 | static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc, | |
389 | gfp_t gfp_flags) | |
390 | { | |
391 | gfp_flags |= mc->gfp_zero; | |
392 | ||
393 | if (mc->kmem_cache) | |
394 | return kmem_cache_alloc(mc->kmem_cache, gfp_flags); | |
395 | else | |
396 | return (void *)__get_free_page(gfp_flags); | |
397 | } | |
398 | ||
837f66c7 | 399 | int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min) |
6926f95a | 400 | { |
837f66c7 | 401 | gfp_t gfp = GFP_KERNEL_ACCOUNT; |
6926f95a SC |
402 | void *obj; |
403 | ||
404 | if (mc->nobjs >= min) | |
405 | return 0; | |
837f66c7 DM |
406 | |
407 | if (unlikely(!mc->objects)) { | |
408 | if (WARN_ON_ONCE(!capacity)) | |
409 | return -EIO; | |
410 | ||
411 | mc->objects = kvmalloc_array(sizeof(void *), capacity, gfp); | |
412 | if (!mc->objects) | |
413 | return -ENOMEM; | |
414 | ||
415 | mc->capacity = capacity; | |
416 | } | |
417 | ||
418 | /* It is illegal to request a different capacity across topups. */ | |
419 | if (WARN_ON_ONCE(mc->capacity != capacity)) | |
420 | return -EIO; | |
421 | ||
422 | while (mc->nobjs < mc->capacity) { | |
423 | obj = mmu_memory_cache_alloc_obj(mc, gfp); | |
6926f95a SC |
424 | if (!obj) |
425 | return mc->nobjs >= min ? 0 : -ENOMEM; | |
426 | mc->objects[mc->nobjs++] = obj; | |
427 | } | |
428 | return 0; | |
429 | } | |
430 | ||
837f66c7 DM |
431 | int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min) |
432 | { | |
433 | return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min); | |
434 | } | |
435 | ||
6926f95a SC |
436 | int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc) |
437 | { | |
438 | return mc->nobjs; | |
439 | } | |
440 | ||
441 | void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
442 | { | |
443 | while (mc->nobjs) { | |
444 | if (mc->kmem_cache) | |
445 | kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]); | |
446 | else | |
447 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
448 | } | |
837f66c7 DM |
449 | |
450 | kvfree(mc->objects); | |
451 | ||
452 | mc->objects = NULL; | |
453 | mc->capacity = 0; | |
6926f95a SC |
454 | } |
455 | ||
456 | void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
457 | { | |
458 | void *p; | |
459 | ||
460 | if (WARN_ON(!mc->nobjs)) | |
461 | p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT); | |
462 | else | |
463 | p = mc->objects[--mc->nobjs]; | |
464 | BUG_ON(!p); | |
465 | return p; | |
466 | } | |
467 | #endif | |
468 | ||
8bd826d6 | 469 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 470 | { |
fb3f0f51 RR |
471 | mutex_init(&vcpu->mutex); |
472 | vcpu->cpu = -1; | |
fb3f0f51 RR |
473 | vcpu->kvm = kvm; |
474 | vcpu->vcpu_id = id; | |
34bb10b7 | 475 | vcpu->pid = NULL; |
510958e9 | 476 | #ifndef __KVM_HAVE_ARCH_WQP |
da4ad88c | 477 | rcuwait_init(&vcpu->wait); |
510958e9 | 478 | #endif |
af585b92 | 479 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 480 | |
4c088493 R |
481 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
482 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 483 | vcpu->preempted = false; |
d73eb57b | 484 | vcpu->ready = false; |
d5c48deb | 485 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
a54d8066 | 486 | vcpu->last_used_slot = NULL; |
fb3f0f51 | 487 | } |
fb3f0f51 | 488 | |
27592ae8 | 489 | static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
4543bdc0 SC |
490 | { |
491 | kvm_arch_vcpu_destroy(vcpu); | |
5593473a | 492 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
e529ef66 | 493 | |
9941d224 SC |
494 | /* |
495 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
496 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
497 | * are already gone. | |
498 | */ | |
499 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
500 | ||
8bd826d6 | 501 | free_page((unsigned long)vcpu->run); |
e529ef66 | 502 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 | 503 | } |
27592ae8 MZ |
504 | |
505 | void kvm_destroy_vcpus(struct kvm *kvm) | |
506 | { | |
46808a4c | 507 | unsigned long i; |
27592ae8 MZ |
508 | struct kvm_vcpu *vcpu; |
509 | ||
510 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
511 | kvm_vcpu_destroy(vcpu); | |
c5b07754 | 512 | xa_erase(&kvm->vcpu_array, i); |
27592ae8 MZ |
513 | } |
514 | ||
515 | atomic_set(&kvm->online_vcpus, 0); | |
516 | } | |
517 | EXPORT_SYMBOL_GPL(kvm_destroy_vcpus); | |
4543bdc0 | 518 | |
e930bffe AA |
519 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
520 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
521 | { | |
522 | return container_of(mn, struct kvm, mmu_notifier); | |
523 | } | |
524 | ||
e649b3f0 ET |
525 | static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn, |
526 | struct mm_struct *mm, | |
527 | unsigned long start, unsigned long end) | |
528 | { | |
529 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
530 | int idx; | |
531 | ||
532 | idx = srcu_read_lock(&kvm->srcu); | |
533 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
534 | srcu_read_unlock(&kvm->srcu, idx); | |
535 | } | |
536 | ||
3039bcc7 SC |
537 | typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range); |
538 | ||
f922bd9b SC |
539 | typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start, |
540 | unsigned long end); | |
541 | ||
683412cc MZ |
542 | typedef void (*on_unlock_fn_t)(struct kvm *kvm); |
543 | ||
3039bcc7 SC |
544 | struct kvm_hva_range { |
545 | unsigned long start; | |
546 | unsigned long end; | |
547 | pte_t pte; | |
548 | hva_handler_t handler; | |
f922bd9b | 549 | on_lock_fn_t on_lock; |
683412cc | 550 | on_unlock_fn_t on_unlock; |
3039bcc7 SC |
551 | bool flush_on_ret; |
552 | bool may_block; | |
553 | }; | |
554 | ||
f922bd9b SC |
555 | /* |
556 | * Use a dedicated stub instead of NULL to indicate that there is no callback | |
557 | * function/handler. The compiler technically can't guarantee that a real | |
558 | * function will have a non-zero address, and so it will generate code to | |
559 | * check for !NULL, whereas comparing against a stub will be elided at compile | |
560 | * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9). | |
561 | */ | |
562 | static void kvm_null_fn(void) | |
563 | { | |
564 | ||
565 | } | |
566 | #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn) | |
567 | ||
ed922739 MS |
568 | /* Iterate over each memslot intersecting [start, last] (inclusive) range */ |
569 | #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \ | |
570 | for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \ | |
571 | node; \ | |
572 | node = interval_tree_iter_next(node, start, last)) \ | |
573 | ||
3039bcc7 SC |
574 | static __always_inline int __kvm_handle_hva_range(struct kvm *kvm, |
575 | const struct kvm_hva_range *range) | |
576 | { | |
8931a454 | 577 | bool ret = false, locked = false; |
f922bd9b | 578 | struct kvm_gfn_range gfn_range; |
3039bcc7 SC |
579 | struct kvm_memory_slot *slot; |
580 | struct kvm_memslots *slots; | |
3039bcc7 SC |
581 | int i, idx; |
582 | ||
ed922739 MS |
583 | if (WARN_ON_ONCE(range->end <= range->start)) |
584 | return 0; | |
585 | ||
f922bd9b SC |
586 | /* A null handler is allowed if and only if on_lock() is provided. */ |
587 | if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) && | |
588 | IS_KVM_NULL_FN(range->handler))) | |
589 | return 0; | |
590 | ||
3039bcc7 SC |
591 | idx = srcu_read_lock(&kvm->srcu); |
592 | ||
593 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
ed922739 MS |
594 | struct interval_tree_node *node; |
595 | ||
3039bcc7 | 596 | slots = __kvm_memslots(kvm, i); |
ed922739 MS |
597 | kvm_for_each_memslot_in_hva_range(node, slots, |
598 | range->start, range->end - 1) { | |
3039bcc7 SC |
599 | unsigned long hva_start, hva_end; |
600 | ||
a54d8066 | 601 | slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]); |
3039bcc7 SC |
602 | hva_start = max(range->start, slot->userspace_addr); |
603 | hva_end = min(range->end, slot->userspace_addr + | |
604 | (slot->npages << PAGE_SHIFT)); | |
3039bcc7 SC |
605 | |
606 | /* | |
607 | * To optimize for the likely case where the address | |
608 | * range is covered by zero or one memslots, don't | |
609 | * bother making these conditional (to avoid writes on | |
610 | * the second or later invocation of the handler). | |
611 | */ | |
612 | gfn_range.pte = range->pte; | |
613 | gfn_range.may_block = range->may_block; | |
614 | ||
615 | /* | |
616 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
617 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
618 | */ | |
619 | gfn_range.start = hva_to_gfn_memslot(hva_start, slot); | |
620 | gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot); | |
621 | gfn_range.slot = slot; | |
622 | ||
8931a454 SC |
623 | if (!locked) { |
624 | locked = true; | |
625 | KVM_MMU_LOCK(kvm); | |
071064f1 PB |
626 | if (!IS_KVM_NULL_FN(range->on_lock)) |
627 | range->on_lock(kvm, range->start, range->end); | |
628 | if (IS_KVM_NULL_FN(range->handler)) | |
629 | break; | |
8931a454 | 630 | } |
3039bcc7 SC |
631 | ret |= range->handler(kvm, &gfn_range); |
632 | } | |
633 | } | |
634 | ||
6bc6db00 | 635 | if (range->flush_on_ret && ret) |
3039bcc7 SC |
636 | kvm_flush_remote_tlbs(kvm); |
637 | ||
683412cc | 638 | if (locked) { |
8931a454 | 639 | KVM_MMU_UNLOCK(kvm); |
683412cc MZ |
640 | if (!IS_KVM_NULL_FN(range->on_unlock)) |
641 | range->on_unlock(kvm); | |
642 | } | |
f922bd9b | 643 | |
3039bcc7 SC |
644 | srcu_read_unlock(&kvm->srcu, idx); |
645 | ||
646 | /* The notifiers are averse to booleans. :-( */ | |
647 | return (int)ret; | |
648 | } | |
649 | ||
650 | static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn, | |
651 | unsigned long start, | |
652 | unsigned long end, | |
653 | pte_t pte, | |
654 | hva_handler_t handler) | |
655 | { | |
656 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
657 | const struct kvm_hva_range range = { | |
658 | .start = start, | |
659 | .end = end, | |
660 | .pte = pte, | |
661 | .handler = handler, | |
f922bd9b | 662 | .on_lock = (void *)kvm_null_fn, |
683412cc | 663 | .on_unlock = (void *)kvm_null_fn, |
3039bcc7 SC |
664 | .flush_on_ret = true, |
665 | .may_block = false, | |
666 | }; | |
3039bcc7 | 667 | |
f922bd9b | 668 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 SC |
669 | } |
670 | ||
671 | static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn, | |
672 | unsigned long start, | |
673 | unsigned long end, | |
674 | hva_handler_t handler) | |
675 | { | |
676 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
677 | const struct kvm_hva_range range = { | |
678 | .start = start, | |
679 | .end = end, | |
680 | .pte = __pte(0), | |
681 | .handler = handler, | |
f922bd9b | 682 | .on_lock = (void *)kvm_null_fn, |
683412cc | 683 | .on_unlock = (void *)kvm_null_fn, |
3039bcc7 SC |
684 | .flush_on_ret = false, |
685 | .may_block = false, | |
686 | }; | |
3039bcc7 | 687 | |
f922bd9b | 688 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 | 689 | } |
3da0dd43 IE |
690 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
691 | struct mm_struct *mm, | |
692 | unsigned long address, | |
693 | pte_t pte) | |
694 | { | |
695 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
696 | ||
501b9185 SC |
697 | trace_kvm_set_spte_hva(address); |
698 | ||
c13fda23 | 699 | /* |
52ac8b35 | 700 | * .change_pte() must be surrounded by .invalidate_range_{start,end}(). |
071064f1 PB |
701 | * If mmu_notifier_count is zero, then no in-progress invalidations, |
702 | * including this one, found a relevant memslot at start(); rechecking | |
703 | * memslots here is unnecessary. Note, a false positive (count elevated | |
704 | * by a different invalidation) is sub-optimal but functionally ok. | |
c13fda23 | 705 | */ |
52ac8b35 | 706 | WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count)); |
071064f1 PB |
707 | if (!READ_ONCE(kvm->mmu_notifier_count)) |
708 | return; | |
c13fda23 | 709 | |
3039bcc7 | 710 | kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn); |
3da0dd43 IE |
711 | } |
712 | ||
edb298c6 | 713 | void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 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 | */ | |
721 | kvm->mmu_notifier_count++; | |
4a42d848 | 722 | if (likely(kvm->mmu_notifier_count == 1)) { |
f922bd9b SC |
723 | kvm->mmu_notifier_range_start = start; |
724 | kvm->mmu_notifier_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 | */ | |
735 | kvm->mmu_notifier_range_start = | |
f922bd9b | 736 | min(kvm->mmu_notifier_range_start, start); |
4a42d848 | 737 | kvm->mmu_notifier_range_end = |
f922bd9b | 738 | max(kvm->mmu_notifier_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, | |
751 | .on_lock = kvm_inc_notifier_count, | |
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 | |
762 | * functions. Without that guarantee, the mmu_notifier_count | |
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 | |
778 | * mn_active_invalidate_count (see above) instead of mmu_notifier_count. | |
779 | */ | |
982ed0de DW |
780 | gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end, |
781 | hva_range.may_block); | |
782 | ||
f922bd9b | 783 | __kvm_handle_hva_range(kvm, &hva_range); |
93065ac7 | 784 | |
e649b3f0 | 785 | return 0; |
e930bffe AA |
786 | } |
787 | ||
edb298c6 | 788 | void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 789 | unsigned long end) |
e930bffe | 790 | { |
e930bffe AA |
791 | /* |
792 | * This sequence increase will notify the kvm page fault that | |
793 | * the page that is going to be mapped in the spte could have | |
794 | * been freed. | |
795 | */ | |
796 | kvm->mmu_notifier_seq++; | |
a355aa54 | 797 | smp_wmb(); |
e930bffe AA |
798 | /* |
799 | * The above sequence increase must be visible before the | |
a355aa54 PM |
800 | * below count decrease, which is ensured by the smp_wmb above |
801 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
802 | */ |
803 | kvm->mmu_notifier_count--; | |
f922bd9b SC |
804 | } |
805 | ||
806 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
807 | const struct mmu_notifier_range *range) | |
808 | { | |
809 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
810 | const struct kvm_hva_range hva_range = { | |
811 | .start = range->start, | |
812 | .end = range->end, | |
813 | .pte = __pte(0), | |
814 | .handler = (void *)kvm_null_fn, | |
815 | .on_lock = kvm_dec_notifier_count, | |
683412cc | 816 | .on_unlock = (void *)kvm_null_fn, |
f922bd9b SC |
817 | .flush_on_ret = false, |
818 | .may_block = mmu_notifier_range_blockable(range), | |
819 | }; | |
52ac8b35 | 820 | bool wake; |
f922bd9b SC |
821 | |
822 | __kvm_handle_hva_range(kvm, &hva_range); | |
e930bffe | 823 | |
52ac8b35 PB |
824 | /* Pairs with the increment in range_start(). */ |
825 | spin_lock(&kvm->mn_invalidate_lock); | |
826 | wake = (--kvm->mn_active_invalidate_count == 0); | |
827 | spin_unlock(&kvm->mn_invalidate_lock); | |
828 | ||
829 | /* | |
830 | * There can only be one waiter, since the wait happens under | |
831 | * slots_lock. | |
832 | */ | |
833 | if (wake) | |
834 | rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait); | |
835 | ||
e930bffe AA |
836 | BUG_ON(kvm->mmu_notifier_count < 0); |
837 | } | |
838 | ||
839 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
840 | struct mm_struct *mm, | |
57128468 ALC |
841 | unsigned long start, |
842 | unsigned long end) | |
e930bffe | 843 | { |
501b9185 SC |
844 | trace_kvm_age_hva(start, end); |
845 | ||
3039bcc7 | 846 | return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn); |
e930bffe AA |
847 | } |
848 | ||
1d7715c6 VD |
849 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
850 | struct mm_struct *mm, | |
851 | unsigned long start, | |
852 | unsigned long end) | |
853 | { | |
501b9185 SC |
854 | trace_kvm_age_hva(start, end); |
855 | ||
1d7715c6 VD |
856 | /* |
857 | * Even though we do not flush TLB, this will still adversely | |
858 | * affect performance on pre-Haswell Intel EPT, where there is | |
859 | * no EPT Access Bit to clear so that we have to tear down EPT | |
860 | * tables instead. If we find this unacceptable, we can always | |
861 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
862 | * do anything on clear_young. | |
863 | * | |
864 | * Also note that currently we never issue secondary TLB flushes | |
865 | * from clear_young, leaving this job up to the regular system | |
866 | * cadence. If we find this inaccurate, we might come up with a | |
867 | * more sophisticated heuristic later. | |
868 | */ | |
3039bcc7 | 869 | return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn); |
1d7715c6 VD |
870 | } |
871 | ||
8ee53820 AA |
872 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
873 | struct mm_struct *mm, | |
874 | unsigned long address) | |
875 | { | |
501b9185 SC |
876 | trace_kvm_test_age_hva(address); |
877 | ||
3039bcc7 SC |
878 | return kvm_handle_hva_range_no_flush(mn, address, address + 1, |
879 | kvm_test_age_gfn); | |
8ee53820 AA |
880 | } |
881 | ||
85db06e5 MT |
882 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
883 | struct mm_struct *mm) | |
884 | { | |
885 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
886 | int idx; |
887 | ||
888 | idx = srcu_read_lock(&kvm->srcu); | |
683412cc | 889 | kvm_flush_shadow_all(kvm); |
eda2beda | 890 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
891 | } |
892 | ||
e930bffe | 893 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e649b3f0 | 894 | .invalidate_range = kvm_mmu_notifier_invalidate_range, |
e930bffe AA |
895 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
896 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
897 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 898 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 899 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 900 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 901 | .release = kvm_mmu_notifier_release, |
e930bffe | 902 | }; |
4c07b0a4 AK |
903 | |
904 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
905 | { | |
906 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
907 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
908 | } | |
909 | ||
910 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
911 | ||
912 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
913 | { | |
914 | return 0; | |
915 | } | |
916 | ||
e930bffe AA |
917 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
918 | ||
2fdef3a2 SS |
919 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
920 | static int kvm_pm_notifier_call(struct notifier_block *bl, | |
921 | unsigned long state, | |
922 | void *unused) | |
923 | { | |
924 | struct kvm *kvm = container_of(bl, struct kvm, pm_notifier); | |
925 | ||
926 | return kvm_arch_pm_notifier(kvm, state); | |
927 | } | |
928 | ||
929 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
930 | { | |
931 | kvm->pm_notifier.notifier_call = kvm_pm_notifier_call; | |
932 | /* Suspend KVM before we suspend ftrace, RCU, etc. */ | |
933 | kvm->pm_notifier.priority = INT_MAX; | |
934 | register_pm_notifier(&kvm->pm_notifier); | |
935 | } | |
936 | ||
937 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
938 | { | |
939 | unregister_pm_notifier(&kvm->pm_notifier); | |
940 | } | |
941 | #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
942 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
943 | { | |
944 | } | |
945 | ||
946 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
947 | { | |
948 | } | |
949 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
950 | ||
a47d2b07 PB |
951 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
952 | { | |
953 | if (!memslot->dirty_bitmap) | |
954 | return; | |
955 | ||
956 | kvfree(memslot->dirty_bitmap); | |
957 | memslot->dirty_bitmap = NULL; | |
958 | } | |
959 | ||
a54d8066 | 960 | /* This does not remove the slot from struct kvm_memslots data structures */ |
e96c81ee | 961 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 962 | { |
e96c81ee | 963 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 964 | |
e96c81ee | 965 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 966 | |
a54d8066 | 967 | kfree(slot); |
a47d2b07 PB |
968 | } |
969 | ||
970 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
971 | { | |
a54d8066 | 972 | struct hlist_node *idnode; |
a47d2b07 | 973 | struct kvm_memory_slot *memslot; |
a54d8066 | 974 | int bkt; |
a47d2b07 | 975 | |
a54d8066 MS |
976 | /* |
977 | * The same memslot objects live in both active and inactive sets, | |
978 | * arbitrarily free using index '1' so the second invocation of this | |
979 | * function isn't operating over a structure with dangling pointers | |
980 | * (even though this function isn't actually touching them). | |
981 | */ | |
982 | if (!slots->node_idx) | |
a47d2b07 PB |
983 | return; |
984 | ||
a54d8066 | 985 | hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1]) |
e96c81ee | 986 | kvm_free_memslot(kvm, memslot); |
bf3e05bc XG |
987 | } |
988 | ||
bc9e9e67 JZ |
989 | static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc) |
990 | { | |
991 | switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) { | |
992 | case KVM_STATS_TYPE_INSTANT: | |
993 | return 0444; | |
994 | case KVM_STATS_TYPE_CUMULATIVE: | |
995 | case KVM_STATS_TYPE_PEAK: | |
996 | default: | |
997 | return 0644; | |
998 | } | |
999 | } | |
1000 | ||
1001 | ||
536a6f88 JF |
1002 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
1003 | { | |
1004 | int i; | |
bc9e9e67 JZ |
1005 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
1006 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 | 1007 | |
a44a4cc1 | 1008 | if (IS_ERR(kvm->debugfs_dentry)) |
536a6f88 JF |
1009 | return; |
1010 | ||
1011 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
1012 | ||
9d5a1dce LC |
1013 | if (kvm->debugfs_stat_data) { |
1014 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
1015 | kfree(kvm->debugfs_stat_data[i]); | |
1016 | kfree(kvm->debugfs_stat_data); | |
1017 | } | |
536a6f88 JF |
1018 | } |
1019 | ||
1020 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
1021 | { | |
85cd39af PB |
1022 | static DEFINE_MUTEX(kvm_debugfs_lock); |
1023 | struct dentry *dent; | |
536a6f88 JF |
1024 | char dir_name[ITOA_MAX_LEN * 2]; |
1025 | struct kvm_stat_data *stat_data; | |
bc9e9e67 | 1026 | const struct _kvm_stats_desc *pdesc; |
3165af73 | 1027 | int i, ret; |
bc9e9e67 JZ |
1028 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
1029 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
1030 | |
1031 | if (!debugfs_initialized()) | |
1032 | return 0; | |
1033 | ||
1034 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
85cd39af PB |
1035 | mutex_lock(&kvm_debugfs_lock); |
1036 | dent = debugfs_lookup(dir_name, kvm_debugfs_dir); | |
1037 | if (dent) { | |
1038 | pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name); | |
1039 | dput(dent); | |
1040 | mutex_unlock(&kvm_debugfs_lock); | |
1041 | return 0; | |
1042 | } | |
1043 | dent = debugfs_create_dir(dir_name, kvm_debugfs_dir); | |
1044 | mutex_unlock(&kvm_debugfs_lock); | |
1045 | if (IS_ERR(dent)) | |
1046 | return 0; | |
536a6f88 | 1047 | |
85cd39af | 1048 | kvm->debugfs_dentry = dent; |
536a6f88 JF |
1049 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
1050 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 1051 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
1052 | if (!kvm->debugfs_stat_data) |
1053 | return -ENOMEM; | |
1054 | ||
bc9e9e67 JZ |
1055 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
1056 | pdesc = &kvm_vm_stats_desc[i]; | |
b12ce36a | 1057 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
1058 | if (!stat_data) |
1059 | return -ENOMEM; | |
1060 | ||
1061 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1062 | stat_data->desc = pdesc; |
1063 | stat_data->kind = KVM_STAT_VM; | |
1064 | kvm->debugfs_stat_data[i] = stat_data; | |
1065 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
1066 | kvm->debugfs_dentry, stat_data, | |
1067 | &stat_fops_per_vm); | |
1068 | } | |
1069 | ||
1070 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
1071 | pdesc = &kvm_vcpu_stats_desc[i]; | |
b12ce36a | 1072 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
1073 | if (!stat_data) |
1074 | return -ENOMEM; | |
1075 | ||
1076 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1077 | stat_data->desc = pdesc; |
1078 | stat_data->kind = KVM_STAT_VCPU; | |
004d62eb | 1079 | kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data; |
bc9e9e67 | 1080 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), |
09cbcef6 MP |
1081 | kvm->debugfs_dentry, stat_data, |
1082 | &stat_fops_per_vm); | |
536a6f88 | 1083 | } |
3165af73 PX |
1084 | |
1085 | ret = kvm_arch_create_vm_debugfs(kvm); | |
1086 | if (ret) { | |
1087 | kvm_destroy_vm_debugfs(kvm); | |
1088 | return i; | |
1089 | } | |
1090 | ||
536a6f88 JF |
1091 | return 0; |
1092 | } | |
1093 | ||
1aa9b957 JS |
1094 | /* |
1095 | * Called after the VM is otherwise initialized, but just before adding it to | |
1096 | * the vm_list. | |
1097 | */ | |
1098 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
1099 | { | |
1100 | return 0; | |
1101 | } | |
1102 | ||
1103 | /* | |
1104 | * Called just after removing the VM from the vm_list, but before doing any | |
1105 | * other destruction. | |
1106 | */ | |
1107 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
1108 | { | |
1109 | } | |
1110 | ||
3165af73 PX |
1111 | /* |
1112 | * Called after per-vm debugfs created. When called kvm->debugfs_dentry should | |
1113 | * be setup already, so we can create arch-specific debugfs entries under it. | |
1114 | * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so | |
1115 | * a per-arch destroy interface is not needed. | |
1116 | */ | |
1117 | int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm) | |
1118 | { | |
1119 | return 0; | |
1120 | } | |
1121 | ||
e08b9637 | 1122 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 1123 | { |
d89f5eff | 1124 | struct kvm *kvm = kvm_arch_alloc_vm(); |
a54d8066 | 1125 | struct kvm_memslots *slots; |
9121923c | 1126 | int r = -ENOMEM; |
a54d8066 | 1127 | int i, j; |
6aa8b732 | 1128 | |
d89f5eff JK |
1129 | if (!kvm) |
1130 | return ERR_PTR(-ENOMEM); | |
1131 | ||
531810ca | 1132 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 1133 | mmgrab(current->mm); |
e9ad4ec8 PB |
1134 | kvm->mm = current->mm; |
1135 | kvm_eventfd_init(kvm); | |
1136 | mutex_init(&kvm->lock); | |
1137 | mutex_init(&kvm->irq_lock); | |
1138 | mutex_init(&kvm->slots_lock); | |
b10a038e | 1139 | mutex_init(&kvm->slots_arch_lock); |
52ac8b35 PB |
1140 | spin_lock_init(&kvm->mn_invalidate_lock); |
1141 | rcuwait_init(&kvm->mn_memslots_update_rcuwait); | |
c5b07754 | 1142 | xa_init(&kvm->vcpu_array); |
52ac8b35 | 1143 | |
982ed0de DW |
1144 | INIT_LIST_HEAD(&kvm->gpc_list); |
1145 | spin_lock_init(&kvm->gpc_lock); | |
52ac8b35 | 1146 | |
e9ad4ec8 | 1147 | INIT_LIST_HEAD(&kvm->devices); |
f502cc56 | 1148 | kvm->max_vcpus = KVM_MAX_VCPUS; |
e9ad4ec8 | 1149 | |
1e702d9a AW |
1150 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
1151 | ||
5c697c36 SC |
1152 | /* |
1153 | * Force subsequent debugfs file creations to fail if the VM directory | |
1154 | * is not created (by kvm_create_vm_debugfs()). | |
1155 | */ | |
1156 | kvm->debugfs_dentry = ERR_PTR(-ENOENT); | |
1157 | ||
8a44119a PB |
1158 | if (init_srcu_struct(&kvm->srcu)) |
1159 | goto out_err_no_srcu; | |
1160 | if (init_srcu_struct(&kvm->irq_srcu)) | |
1161 | goto out_err_no_irq_srcu; | |
1162 | ||
e2d3fcaf | 1163 | refcount_set(&kvm->users_count, 1); |
f481b069 | 1164 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
a54d8066 MS |
1165 | for (j = 0; j < 2; j++) { |
1166 | slots = &kvm->__memslots[i][j]; | |
9121923c | 1167 | |
a54d8066 MS |
1168 | atomic_long_set(&slots->last_used_slot, (unsigned long)NULL); |
1169 | slots->hva_tree = RB_ROOT_CACHED; | |
1170 | slots->gfn_tree = RB_ROOT; | |
1171 | hash_init(slots->id_hash); | |
1172 | slots->node_idx = j; | |
1173 | ||
1174 | /* Generations must be different for each address space. */ | |
1175 | slots->generation = i; | |
1176 | } | |
1177 | ||
1178 | rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]); | |
f481b069 | 1179 | } |
00f034a1 | 1180 | |
e93f8a0f | 1181 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 1182 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 1183 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 1184 | if (!kvm->buses[i]) |
a97b0e77 | 1185 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 1186 | } |
e930bffe | 1187 | |
acd05785 DM |
1188 | kvm->max_halt_poll_ns = halt_poll_ns; |
1189 | ||
e08b9637 | 1190 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 1191 | if (r) |
a97b0e77 | 1192 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
1193 | |
1194 | r = hardware_enable_all(); | |
1195 | if (r) | |
719d93cd | 1196 | goto out_err_no_disable; |
10474ae8 | 1197 | |
c77dcacb | 1198 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 1199 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 1200 | #endif |
6aa8b732 | 1201 | |
74b5c5bf | 1202 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
1203 | if (r) |
1204 | goto out_err_no_mmu_notifier; | |
1205 | ||
1206 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
1207 | if (r) |
1208 | goto out_err; | |
1209 | ||
0d9ce162 | 1210 | mutex_lock(&kvm_lock); |
5e58cfe4 | 1211 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 1212 | mutex_unlock(&kvm_lock); |
d89f5eff | 1213 | |
2ecd9d29 | 1214 | preempt_notifier_inc(); |
2fdef3a2 | 1215 | kvm_init_pm_notifier(kvm); |
2ecd9d29 | 1216 | |
5f6de5cb DM |
1217 | /* |
1218 | * When the fd passed to this ioctl() is opened it pins the module, | |
1219 | * but try_module_get() also prevents getting a reference if the module | |
1220 | * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait"). | |
1221 | */ | |
1222 | if (!try_module_get(kvm_chardev_ops.owner)) { | |
1223 | r = -ENODEV; | |
1224 | goto out_err; | |
1225 | } | |
1226 | ||
f17abe9a | 1227 | return kvm; |
10474ae8 AG |
1228 | |
1229 | out_err: | |
1aa9b957 JS |
1230 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1231 | if (kvm->mmu_notifier.ops) | |
1232 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
1233 | #endif | |
1234 | out_err_no_mmu_notifier: | |
10474ae8 | 1235 | hardware_disable_all(); |
719d93cd | 1236 | out_err_no_disable: |
a97b0e77 | 1237 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 1238 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 1239 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 1240 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 1241 | kfree(kvm_get_bus(kvm, i)); |
8a44119a PB |
1242 | cleanup_srcu_struct(&kvm->irq_srcu); |
1243 | out_err_no_irq_srcu: | |
1244 | cleanup_srcu_struct(&kvm->srcu); | |
1245 | out_err_no_srcu: | |
d89f5eff | 1246 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 1247 | mmdrop(current->mm); |
10474ae8 | 1248 | return ERR_PTR(r); |
f17abe9a AK |
1249 | } |
1250 | ||
07f0a7bd SW |
1251 | static void kvm_destroy_devices(struct kvm *kvm) |
1252 | { | |
e6e3b5a6 | 1253 | struct kvm_device *dev, *tmp; |
07f0a7bd | 1254 | |
a28ebea2 CD |
1255 | /* |
1256 | * We do not need to take the kvm->lock here, because nobody else | |
1257 | * has a reference to the struct kvm at this point and therefore | |
1258 | * cannot access the devices list anyhow. | |
1259 | */ | |
e6e3b5a6 GT |
1260 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
1261 | list_del(&dev->vm_node); | |
07f0a7bd SW |
1262 | dev->ops->destroy(dev); |
1263 | } | |
1264 | } | |
1265 | ||
f17abe9a AK |
1266 | static void kvm_destroy_vm(struct kvm *kvm) |
1267 | { | |
e93f8a0f | 1268 | int i; |
6d4e4c4f AK |
1269 | struct mm_struct *mm = kvm->mm; |
1270 | ||
2fdef3a2 | 1271 | kvm_destroy_pm_notifier(kvm); |
286de8f6 | 1272 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 1273 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 1274 | kvm_arch_sync_events(kvm); |
0d9ce162 | 1275 | mutex_lock(&kvm_lock); |
133de902 | 1276 | list_del(&kvm->vm_list); |
0d9ce162 | 1277 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
1278 | kvm_arch_pre_destroy_vm(kvm); |
1279 | ||
399ec807 | 1280 | kvm_free_irq_routing(kvm); |
df630b8c | 1281 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 1282 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 1283 | |
4a12f951 CB |
1284 | if (bus) |
1285 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
1286 | kvm->buses[i] = NULL; |
1287 | } | |
980da6ce | 1288 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
1289 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1290 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
52ac8b35 PB |
1291 | /* |
1292 | * At this point, pending calls to invalidate_range_start() | |
1293 | * have completed but no more MMU notifiers will run, so | |
1294 | * mn_active_invalidate_count may remain unbalanced. | |
1295 | * No threads can be waiting in install_new_memslots as the | |
1296 | * last reference on KVM has been dropped, but freeing | |
1297 | * memslots would deadlock without this manual intervention. | |
1298 | */ | |
1299 | WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait)); | |
1300 | kvm->mn_active_invalidate_count = 0; | |
f00be0ca | 1301 | #else |
683412cc | 1302 | kvm_flush_shadow_all(kvm); |
5f94c174 | 1303 | #endif |
d19a9cd2 | 1304 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 1305 | kvm_destroy_devices(kvm); |
a54d8066 MS |
1306 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
1307 | kvm_free_memslots(kvm, &kvm->__memslots[i][0]); | |
1308 | kvm_free_memslots(kvm, &kvm->__memslots[i][1]); | |
1309 | } | |
820b3fcd | 1310 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
1311 | cleanup_srcu_struct(&kvm->srcu); |
1312 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 1313 | preempt_notifier_dec(); |
10474ae8 | 1314 | hardware_disable_all(); |
6d4e4c4f | 1315 | mmdrop(mm); |
5f6de5cb | 1316 | module_put(kvm_chardev_ops.owner); |
f17abe9a AK |
1317 | } |
1318 | ||
d39f13b0 IE |
1319 | void kvm_get_kvm(struct kvm *kvm) |
1320 | { | |
e3736c3e | 1321 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
1322 | } |
1323 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
1324 | ||
605c7130 PX |
1325 | /* |
1326 | * Make sure the vm is not during destruction, which is a safe version of | |
1327 | * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise. | |
1328 | */ | |
1329 | bool kvm_get_kvm_safe(struct kvm *kvm) | |
1330 | { | |
1331 | return refcount_inc_not_zero(&kvm->users_count); | |
1332 | } | |
1333 | EXPORT_SYMBOL_GPL(kvm_get_kvm_safe); | |
1334 | ||
d39f13b0 IE |
1335 | void kvm_put_kvm(struct kvm *kvm) |
1336 | { | |
e3736c3e | 1337 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
1338 | kvm_destroy_vm(kvm); |
1339 | } | |
1340 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
1341 | ||
149487bd SC |
1342 | /* |
1343 | * Used to put a reference that was taken on behalf of an object associated | |
1344 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
1345 | * of the new file descriptor fails and the reference cannot be transferred to | |
1346 | * its final owner. In such cases, the caller is still actively using @kvm and | |
1347 | * will fail miserably if the refcount unexpectedly hits zero. | |
1348 | */ | |
1349 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
1350 | { | |
1351 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
1352 | } | |
1353 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 1354 | |
f17abe9a AK |
1355 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
1356 | { | |
1357 | struct kvm *kvm = filp->private_data; | |
1358 | ||
721eecbf GH |
1359 | kvm_irqfd_release(kvm); |
1360 | ||
d39f13b0 | 1361 | kvm_put_kvm(kvm); |
6aa8b732 AK |
1362 | return 0; |
1363 | } | |
1364 | ||
515a0127 TY |
1365 | /* |
1366 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 1367 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 1368 | */ |
3c9bd400 | 1369 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 1370 | { |
37b2a651 | 1371 | unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 1372 | |
37b2a651 | 1373 | memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
1374 | if (!memslot->dirty_bitmap) |
1375 | return -ENOMEM; | |
1376 | ||
a36a57b1 TY |
1377 | return 0; |
1378 | } | |
1379 | ||
a54d8066 | 1380 | static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id) |
bf3e05bc | 1381 | { |
a54d8066 MS |
1382 | struct kvm_memslots *active = __kvm_memslots(kvm, as_id); |
1383 | int node_idx_inactive = active->node_idx ^ 1; | |
0e60b079 | 1384 | |
a54d8066 | 1385 | return &kvm->__memslots[as_id][node_idx_inactive]; |
0577d1ab SC |
1386 | } |
1387 | ||
1388 | /* | |
a54d8066 MS |
1389 | * Helper to get the address space ID when one of memslot pointers may be NULL. |
1390 | * This also serves as a sanity that at least one of the pointers is non-NULL, | |
1391 | * and that their address space IDs don't diverge. | |
0577d1ab | 1392 | */ |
a54d8066 MS |
1393 | static int kvm_memslots_get_as_id(struct kvm_memory_slot *a, |
1394 | struct kvm_memory_slot *b) | |
0577d1ab | 1395 | { |
a54d8066 MS |
1396 | if (WARN_ON_ONCE(!a && !b)) |
1397 | return 0; | |
0577d1ab | 1398 | |
a54d8066 MS |
1399 | if (!a) |
1400 | return b->as_id; | |
1401 | if (!b) | |
1402 | return a->as_id; | |
0577d1ab | 1403 | |
a54d8066 MS |
1404 | WARN_ON_ONCE(a->as_id != b->as_id); |
1405 | return a->as_id; | |
0577d1ab | 1406 | } |
efbeec70 | 1407 | |
a54d8066 MS |
1408 | static void kvm_insert_gfn_node(struct kvm_memslots *slots, |
1409 | struct kvm_memory_slot *slot) | |
0577d1ab | 1410 | { |
a54d8066 MS |
1411 | struct rb_root *gfn_tree = &slots->gfn_tree; |
1412 | struct rb_node **node, *parent; | |
1413 | int idx = slots->node_idx; | |
0577d1ab | 1414 | |
a54d8066 MS |
1415 | parent = NULL; |
1416 | for (node = &gfn_tree->rb_node; *node; ) { | |
1417 | struct kvm_memory_slot *tmp; | |
f85e2cb5 | 1418 | |
a54d8066 MS |
1419 | tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]); |
1420 | parent = *node; | |
1421 | if (slot->base_gfn < tmp->base_gfn) | |
1422 | node = &(*node)->rb_left; | |
1423 | else if (slot->base_gfn > tmp->base_gfn) | |
1424 | node = &(*node)->rb_right; | |
1425 | else | |
1426 | BUG(); | |
0577d1ab | 1427 | } |
a54d8066 MS |
1428 | |
1429 | rb_link_node(&slot->gfn_node[idx], parent, node); | |
1430 | rb_insert_color(&slot->gfn_node[idx], gfn_tree); | |
0577d1ab SC |
1431 | } |
1432 | ||
a54d8066 MS |
1433 | static void kvm_erase_gfn_node(struct kvm_memslots *slots, |
1434 | struct kvm_memory_slot *slot) | |
0577d1ab | 1435 | { |
a54d8066 MS |
1436 | rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree); |
1437 | } | |
0577d1ab | 1438 | |
a54d8066 MS |
1439 | static void kvm_replace_gfn_node(struct kvm_memslots *slots, |
1440 | struct kvm_memory_slot *old, | |
1441 | struct kvm_memory_slot *new) | |
1442 | { | |
1443 | int idx = slots->node_idx; | |
0577d1ab | 1444 | |
a54d8066 | 1445 | WARN_ON_ONCE(old->base_gfn != new->base_gfn); |
0577d1ab | 1446 | |
a54d8066 MS |
1447 | rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx], |
1448 | &slots->gfn_tree); | |
0577d1ab SC |
1449 | } |
1450 | ||
1451 | /* | |
a54d8066 | 1452 | * Replace @old with @new in the inactive memslots. |
0577d1ab | 1453 | * |
a54d8066 MS |
1454 | * With NULL @old this simply adds @new. |
1455 | * With NULL @new this simply removes @old. | |
0577d1ab | 1456 | * |
a54d8066 MS |
1457 | * If @new is non-NULL its hva_node[slots_idx] range has to be set |
1458 | * appropriately. | |
0577d1ab | 1459 | */ |
a54d8066 MS |
1460 | static void kvm_replace_memslot(struct kvm *kvm, |
1461 | struct kvm_memory_slot *old, | |
1462 | struct kvm_memory_slot *new) | |
0577d1ab | 1463 | { |
a54d8066 MS |
1464 | int as_id = kvm_memslots_get_as_id(old, new); |
1465 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); | |
1466 | int idx = slots->node_idx; | |
0577d1ab | 1467 | |
a54d8066 MS |
1468 | if (old) { |
1469 | hash_del(&old->id_node[idx]); | |
1470 | interval_tree_remove(&old->hva_node[idx], &slots->hva_tree); | |
0577d1ab | 1471 | |
a54d8066 MS |
1472 | if ((long)old == atomic_long_read(&slots->last_used_slot)) |
1473 | atomic_long_set(&slots->last_used_slot, (long)new); | |
0577d1ab | 1474 | |
a54d8066 MS |
1475 | if (!new) { |
1476 | kvm_erase_gfn_node(slots, old); | |
1e8617d3 | 1477 | return; |
a54d8066 MS |
1478 | } |
1479 | } | |
1e8617d3 | 1480 | |
a54d8066 MS |
1481 | /* |
1482 | * Initialize @new's hva range. Do this even when replacing an @old | |
1483 | * slot, kvm_copy_memslot() deliberately does not touch node data. | |
1484 | */ | |
1485 | new->hva_node[idx].start = new->userspace_addr; | |
1486 | new->hva_node[idx].last = new->userspace_addr + | |
1487 | (new->npages << PAGE_SHIFT) - 1; | |
1488 | ||
1489 | /* | |
1490 | * (Re)Add the new memslot. There is no O(1) interval_tree_replace(), | |
1491 | * hva_node needs to be swapped with remove+insert even though hva can't | |
1492 | * change when replacing an existing slot. | |
1493 | */ | |
1494 | hash_add(slots->id_hash, &new->id_node[idx], new->id); | |
1495 | interval_tree_insert(&new->hva_node[idx], &slots->hva_tree); | |
1496 | ||
1497 | /* | |
1498 | * If the memslot gfn is unchanged, rb_replace_node() can be used to | |
1499 | * switch the node in the gfn tree instead of removing the old and | |
1500 | * inserting the new as two separate operations. Replacement is a | |
1501 | * single O(1) operation versus two O(log(n)) operations for | |
1502 | * remove+insert. | |
1503 | */ | |
1504 | if (old && old->base_gfn == new->base_gfn) { | |
1505 | kvm_replace_gfn_node(slots, old, new); | |
1506 | } else { | |
1507 | if (old) | |
1508 | kvm_erase_gfn_node(slots, old); | |
1509 | kvm_insert_gfn_node(slots, new); | |
0577d1ab | 1510 | } |
bf3e05bc XG |
1511 | } |
1512 | ||
09170a49 | 1513 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 1514 | { |
4d8b81ab XG |
1515 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1516 | ||
0f8a4de3 | 1517 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
1518 | valid_flags |= KVM_MEM_READONLY; |
1519 | #endif | |
1520 | ||
1521 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1522 | return -EINVAL; |
1523 | ||
1524 | return 0; | |
1525 | } | |
1526 | ||
a54d8066 | 1527 | static void kvm_swap_active_memslots(struct kvm *kvm, int as_id) |
7ec4fb44 | 1528 | { |
a54d8066 MS |
1529 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); |
1530 | ||
1531 | /* Grab the generation from the activate memslots. */ | |
1532 | u64 gen = __kvm_memslots(kvm, as_id)->generation; | |
7ec4fb44 | 1533 | |
361209e0 SC |
1534 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1535 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1536 | |
52ac8b35 PB |
1537 | /* |
1538 | * Do not store the new memslots while there are invalidations in | |
071064f1 PB |
1539 | * progress, otherwise the locking in invalidate_range_start and |
1540 | * invalidate_range_end will be unbalanced. | |
52ac8b35 PB |
1541 | */ |
1542 | spin_lock(&kvm->mn_invalidate_lock); | |
1543 | prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
1544 | while (kvm->mn_active_invalidate_count) { | |
1545 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1546 | spin_unlock(&kvm->mn_invalidate_lock); | |
1547 | schedule(); | |
1548 | spin_lock(&kvm->mn_invalidate_lock); | |
1549 | } | |
1550 | finish_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
f481b069 | 1551 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
52ac8b35 | 1552 | spin_unlock(&kvm->mn_invalidate_lock); |
b10a038e BG |
1553 | |
1554 | /* | |
1555 | * Acquired in kvm_set_memslot. Must be released before synchronize | |
1556 | * SRCU below in order to avoid deadlock with another thread | |
1557 | * acquiring the slots_arch_lock in an srcu critical section. | |
1558 | */ | |
1559 | mutex_unlock(&kvm->slots_arch_lock); | |
1560 | ||
7ec4fb44 | 1561 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1562 | |
ee3d1570 | 1563 | /* |
361209e0 | 1564 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1565 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1566 | * the number of address spaces. This provides a unique and easily |
1567 | * identifiable generation number while the memslots are in flux. | |
1568 | */ | |
1569 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1570 | ||
1571 | /* | |
4bd518f1 PB |
1572 | * Generations must be unique even across address spaces. We do not need |
1573 | * a global counter for that, instead the generation space is evenly split | |
1574 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1575 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1576 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1577 | */ |
164bf7e5 | 1578 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 1579 | |
15248258 | 1580 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1581 | |
15248258 | 1582 | slots->generation = gen; |
7ec4fb44 GN |
1583 | } |
1584 | ||
07921665 SC |
1585 | static int kvm_prepare_memory_region(struct kvm *kvm, |
1586 | const struct kvm_memory_slot *old, | |
1587 | struct kvm_memory_slot *new, | |
1588 | enum kvm_mr_change change) | |
ddc12f2a | 1589 | { |
07921665 SC |
1590 | int r; |
1591 | ||
1592 | /* | |
1593 | * If dirty logging is disabled, nullify the bitmap; the old bitmap | |
1594 | * will be freed on "commit". If logging is enabled in both old and | |
1595 | * new, reuse the existing bitmap. If logging is enabled only in the | |
1596 | * new and KVM isn't using a ring buffer, allocate and initialize a | |
1597 | * new bitmap. | |
1598 | */ | |
244893fa SC |
1599 | if (change != KVM_MR_DELETE) { |
1600 | if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1601 | new->dirty_bitmap = NULL; | |
1602 | else if (old && old->dirty_bitmap) | |
1603 | new->dirty_bitmap = old->dirty_bitmap; | |
1604 | else if (!kvm->dirty_ring_size) { | |
1605 | r = kvm_alloc_dirty_bitmap(new); | |
1606 | if (r) | |
1607 | return r; | |
1608 | ||
1609 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1610 | bitmap_set(new->dirty_bitmap, 0, new->npages); | |
1611 | } | |
07921665 SC |
1612 | } |
1613 | ||
1614 | r = kvm_arch_prepare_memory_region(kvm, old, new, change); | |
1615 | ||
1616 | /* Free the bitmap on failure if it was allocated above. */ | |
c87661f8 | 1617 | if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap)) |
07921665 SC |
1618 | kvm_destroy_dirty_bitmap(new); |
1619 | ||
1620 | return r; | |
ddc12f2a BG |
1621 | } |
1622 | ||
07921665 SC |
1623 | static void kvm_commit_memory_region(struct kvm *kvm, |
1624 | struct kvm_memory_slot *old, | |
1625 | const struct kvm_memory_slot *new, | |
1626 | enum kvm_mr_change change) | |
ddc12f2a | 1627 | { |
07921665 SC |
1628 | /* |
1629 | * Update the total number of memslot pages before calling the arch | |
1630 | * hook so that architectures can consume the result directly. | |
1631 | */ | |
1632 | if (change == KVM_MR_DELETE) | |
1633 | kvm->nr_memslot_pages -= old->npages; | |
1634 | else if (change == KVM_MR_CREATE) | |
1635 | kvm->nr_memslot_pages += new->npages; | |
1636 | ||
1637 | kvm_arch_commit_memory_region(kvm, old, new, change); | |
1638 | ||
a54d8066 MS |
1639 | switch (change) { |
1640 | case KVM_MR_CREATE: | |
1641 | /* Nothing more to do. */ | |
1642 | break; | |
1643 | case KVM_MR_DELETE: | |
1644 | /* Free the old memslot and all its metadata. */ | |
1645 | kvm_free_memslot(kvm, old); | |
1646 | break; | |
1647 | case KVM_MR_MOVE: | |
1648 | case KVM_MR_FLAGS_ONLY: | |
1649 | /* | |
1650 | * Free the dirty bitmap as needed; the below check encompasses | |
1651 | * both the flags and whether a ring buffer is being used) | |
1652 | */ | |
1653 | if (old->dirty_bitmap && !new->dirty_bitmap) | |
1654 | kvm_destroy_dirty_bitmap(old); | |
1655 | ||
1656 | /* | |
1657 | * The final quirk. Free the detached, old slot, but only its | |
1658 | * memory, not any metadata. Metadata, including arch specific | |
1659 | * data, may be reused by @new. | |
1660 | */ | |
1661 | kfree(old); | |
1662 | break; | |
1663 | default: | |
1664 | BUG(); | |
1665 | } | |
ddc12f2a BG |
1666 | } |
1667 | ||
36947254 | 1668 | /* |
a54d8066 MS |
1669 | * Activate @new, which must be installed in the inactive slots by the caller, |
1670 | * by swapping the active slots and then propagating @new to @old once @old is | |
1671 | * unreachable and can be safely modified. | |
1672 | * | |
1673 | * With NULL @old this simply adds @new to @active (while swapping the sets). | |
1674 | * With NULL @new this simply removes @old from @active and frees it | |
1675 | * (while also swapping the sets). | |
36947254 | 1676 | */ |
a54d8066 MS |
1677 | static void kvm_activate_memslot(struct kvm *kvm, |
1678 | struct kvm_memory_slot *old, | |
1679 | struct kvm_memory_slot *new) | |
36947254 | 1680 | { |
a54d8066 | 1681 | int as_id = kvm_memslots_get_as_id(old, new); |
36947254 | 1682 | |
a54d8066 MS |
1683 | kvm_swap_active_memslots(kvm, as_id); |
1684 | ||
1685 | /* Propagate the new memslot to the now inactive memslots. */ | |
1686 | kvm_replace_memslot(kvm, old, new); | |
1687 | } | |
1688 | ||
1689 | static void kvm_copy_memslot(struct kvm_memory_slot *dest, | |
1690 | const struct kvm_memory_slot *src) | |
1691 | { | |
1692 | dest->base_gfn = src->base_gfn; | |
1693 | dest->npages = src->npages; | |
1694 | dest->dirty_bitmap = src->dirty_bitmap; | |
1695 | dest->arch = src->arch; | |
1696 | dest->userspace_addr = src->userspace_addr; | |
1697 | dest->flags = src->flags; | |
1698 | dest->id = src->id; | |
1699 | dest->as_id = src->as_id; | |
1700 | } | |
1701 | ||
1702 | static void kvm_invalidate_memslot(struct kvm *kvm, | |
1703 | struct kvm_memory_slot *old, | |
244893fa | 1704 | struct kvm_memory_slot *invalid_slot) |
a54d8066 | 1705 | { |
07921665 | 1706 | /* |
a54d8066 MS |
1707 | * Mark the current slot INVALID. As with all memslot modifications, |
1708 | * this must be done on an unreachable slot to avoid modifying the | |
1709 | * current slot in the active tree. | |
07921665 | 1710 | */ |
244893fa SC |
1711 | kvm_copy_memslot(invalid_slot, old); |
1712 | invalid_slot->flags |= KVM_MEMSLOT_INVALID; | |
1713 | kvm_replace_memslot(kvm, old, invalid_slot); | |
a54d8066 MS |
1714 | |
1715 | /* | |
1716 | * Activate the slot that is now marked INVALID, but don't propagate | |
1717 | * the slot to the now inactive slots. The slot is either going to be | |
1718 | * deleted or recreated as a new slot. | |
1719 | */ | |
1720 | kvm_swap_active_memslots(kvm, old->as_id); | |
1721 | ||
1722 | /* | |
1723 | * From this point no new shadow pages pointing to a deleted, or moved, | |
1724 | * memslot will be created. Validation of sp->gfn happens in: | |
1725 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1726 | * - kvm_is_visible_gfn (mmu_check_root) | |
1727 | */ | |
bcb63dcd | 1728 | kvm_arch_flush_shadow_memslot(kvm, old); |
683412cc | 1729 | kvm_arch_guest_memory_reclaimed(kvm); |
a54d8066 MS |
1730 | |
1731 | /* Was released by kvm_swap_active_memslots, reacquire. */ | |
1732 | mutex_lock(&kvm->slots_arch_lock); | |
1733 | ||
1734 | /* | |
1735 | * Copy the arch-specific field of the newly-installed slot back to the | |
1736 | * old slot as the arch data could have changed between releasing | |
1737 | * slots_arch_lock in install_new_memslots() and re-acquiring the lock | |
1738 | * above. Writers are required to retrieve memslots *after* acquiring | |
1739 | * slots_arch_lock, thus the active slot's data is guaranteed to be fresh. | |
1740 | */ | |
244893fa | 1741 | old->arch = invalid_slot->arch; |
a54d8066 MS |
1742 | } |
1743 | ||
1744 | static void kvm_create_memslot(struct kvm *kvm, | |
244893fa | 1745 | struct kvm_memory_slot *new) |
a54d8066 | 1746 | { |
244893fa SC |
1747 | /* Add the new memslot to the inactive set and activate. */ |
1748 | kvm_replace_memslot(kvm, NULL, new); | |
1749 | kvm_activate_memslot(kvm, NULL, new); | |
a54d8066 MS |
1750 | } |
1751 | ||
1752 | static void kvm_delete_memslot(struct kvm *kvm, | |
1753 | struct kvm_memory_slot *old, | |
1754 | struct kvm_memory_slot *invalid_slot) | |
1755 | { | |
1756 | /* | |
1757 | * Remove the old memslot (in the inactive memslots) by passing NULL as | |
244893fa | 1758 | * the "new" slot, and for the invalid version in the active slots. |
a54d8066 MS |
1759 | */ |
1760 | kvm_replace_memslot(kvm, old, NULL); | |
a54d8066 | 1761 | kvm_activate_memslot(kvm, invalid_slot, NULL); |
a54d8066 | 1762 | } |
36947254 | 1763 | |
244893fa SC |
1764 | static void kvm_move_memslot(struct kvm *kvm, |
1765 | struct kvm_memory_slot *old, | |
1766 | struct kvm_memory_slot *new, | |
1767 | struct kvm_memory_slot *invalid_slot) | |
a54d8066 | 1768 | { |
a54d8066 | 1769 | /* |
244893fa SC |
1770 | * Replace the old memslot in the inactive slots, and then swap slots |
1771 | * and replace the current INVALID with the new as well. | |
a54d8066 | 1772 | */ |
244893fa SC |
1773 | kvm_replace_memslot(kvm, old, new); |
1774 | kvm_activate_memslot(kvm, invalid_slot, new); | |
a54d8066 | 1775 | } |
36947254 | 1776 | |
a54d8066 MS |
1777 | static void kvm_update_flags_memslot(struct kvm *kvm, |
1778 | struct kvm_memory_slot *old, | |
244893fa | 1779 | struct kvm_memory_slot *new) |
a54d8066 MS |
1780 | { |
1781 | /* | |
1782 | * Similar to the MOVE case, but the slot doesn't need to be zapped as | |
1783 | * an intermediate step. Instead, the old memslot is simply replaced | |
1784 | * with a new, updated copy in both memslot sets. | |
1785 | */ | |
244893fa SC |
1786 | kvm_replace_memslot(kvm, old, new); |
1787 | kvm_activate_memslot(kvm, old, new); | |
36947254 SC |
1788 | } |
1789 | ||
cf47f50b | 1790 | static int kvm_set_memslot(struct kvm *kvm, |
a54d8066 | 1791 | struct kvm_memory_slot *old, |
ce5f0215 | 1792 | struct kvm_memory_slot *new, |
cf47f50b SC |
1793 | enum kvm_mr_change change) |
1794 | { | |
244893fa | 1795 | struct kvm_memory_slot *invalid_slot; |
cf47f50b SC |
1796 | int r; |
1797 | ||
b10a038e | 1798 | /* |
a54d8066 | 1799 | * Released in kvm_swap_active_memslots. |
b10a038e BG |
1800 | * |
1801 | * Must be held from before the current memslots are copied until | |
1802 | * after the new memslots are installed with rcu_assign_pointer, | |
a54d8066 | 1803 | * then released before the synchronize srcu in kvm_swap_active_memslots. |
b10a038e BG |
1804 | * |
1805 | * When modifying memslots outside of the slots_lock, must be held | |
1806 | * before reading the pointer to the current memslots until after all | |
1807 | * changes to those memslots are complete. | |
1808 | * | |
1809 | * These rules ensure that installing new memslots does not lose | |
1810 | * changes made to the previous memslots. | |
1811 | */ | |
1812 | mutex_lock(&kvm->slots_arch_lock); | |
1813 | ||
a54d8066 MS |
1814 | /* |
1815 | * Invalidate the old slot if it's being deleted or moved. This is | |
1816 | * done prior to actually deleting/moving the memslot to allow vCPUs to | |
1817 | * continue running by ensuring there are no mappings or shadow pages | |
1818 | * for the memslot when it is deleted/moved. Without pre-invalidation | |
1819 | * (and without a lock), a window would exist between effecting the | |
1820 | * delete/move and committing the changes in arch code where KVM or a | |
1821 | * guest could access a non-existent memslot. | |
244893fa SC |
1822 | * |
1823 | * Modifications are done on a temporary, unreachable slot. The old | |
1824 | * slot needs to be preserved in case a later step fails and the | |
1825 | * invalidation needs to be reverted. | |
a54d8066 | 1826 | */ |
cf47f50b | 1827 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
244893fa SC |
1828 | invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT); |
1829 | if (!invalid_slot) { | |
1830 | mutex_unlock(&kvm->slots_arch_lock); | |
1831 | return -ENOMEM; | |
1832 | } | |
1833 | kvm_invalidate_memslot(kvm, old, invalid_slot); | |
1834 | } | |
b10a038e | 1835 | |
a54d8066 MS |
1836 | r = kvm_prepare_memory_region(kvm, old, new, change); |
1837 | if (r) { | |
b10a038e | 1838 | /* |
a54d8066 MS |
1839 | * For DELETE/MOVE, revert the above INVALID change. No |
1840 | * modifications required since the original slot was preserved | |
1841 | * in the inactive slots. Changing the active memslots also | |
1842 | * release slots_arch_lock. | |
b10a038e | 1843 | */ |
244893fa SC |
1844 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
1845 | kvm_activate_memslot(kvm, invalid_slot, old); | |
1846 | kfree(invalid_slot); | |
1847 | } else { | |
a54d8066 | 1848 | mutex_unlock(&kvm->slots_arch_lock); |
244893fa | 1849 | } |
a54d8066 | 1850 | return r; |
cf47f50b SC |
1851 | } |
1852 | ||
bda44d84 | 1853 | /* |
a54d8066 MS |
1854 | * For DELETE and MOVE, the working slot is now active as the INVALID |
1855 | * version of the old slot. MOVE is particularly special as it reuses | |
1856 | * the old slot and returns a copy of the old slot (in working_slot). | |
1857 | * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the | |
1858 | * old slot is detached but otherwise preserved. | |
bda44d84 | 1859 | */ |
a54d8066 | 1860 | if (change == KVM_MR_CREATE) |
244893fa | 1861 | kvm_create_memslot(kvm, new); |
a54d8066 | 1862 | else if (change == KVM_MR_DELETE) |
244893fa | 1863 | kvm_delete_memslot(kvm, old, invalid_slot); |
a54d8066 | 1864 | else if (change == KVM_MR_MOVE) |
244893fa | 1865 | kvm_move_memslot(kvm, old, new, invalid_slot); |
a54d8066 | 1866 | else if (change == KVM_MR_FLAGS_ONLY) |
244893fa | 1867 | kvm_update_flags_memslot(kvm, old, new); |
a54d8066 MS |
1868 | else |
1869 | BUG(); | |
cf47f50b | 1870 | |
244893fa SC |
1871 | /* Free the temporary INVALID slot used for DELETE and MOVE. */ |
1872 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) | |
1873 | kfree(invalid_slot); | |
bda44d84 | 1874 | |
a54d8066 MS |
1875 | /* |
1876 | * No need to refresh new->arch, changes after dropping slots_arch_lock | |
a413a625 | 1877 | * will directly hit the final, active memslot. Architectures are |
a54d8066 MS |
1878 | * responsible for knowing that new->arch may be stale. |
1879 | */ | |
1880 | kvm_commit_memory_region(kvm, old, new, change); | |
cf47f50b | 1881 | |
cf47f50b | 1882 | return 0; |
cf47f50b SC |
1883 | } |
1884 | ||
44401a20 MS |
1885 | static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id, |
1886 | gfn_t start, gfn_t end) | |
5c0b4f3d | 1887 | { |
44401a20 | 1888 | struct kvm_memslot_iter iter; |
5c0b4f3d | 1889 | |
44401a20 MS |
1890 | kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) { |
1891 | if (iter.slot->id != id) | |
1892 | return true; | |
1893 | } | |
5c0b4f3d | 1894 | |
44401a20 | 1895 | return false; |
5c0b4f3d SC |
1896 | } |
1897 | ||
6aa8b732 AK |
1898 | /* |
1899 | * Allocate some memory and give it an address in the guest physical address | |
1900 | * space. | |
1901 | * | |
1902 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1903 | * |
02d5d55b | 1904 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1905 | */ |
f78e0e2e | 1906 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 1907 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 1908 | { |
244893fa | 1909 | struct kvm_memory_slot *old, *new; |
44401a20 | 1910 | struct kvm_memslots *slots; |
f64c0398 | 1911 | enum kvm_mr_change change; |
0f9bdef3 SC |
1912 | unsigned long npages; |
1913 | gfn_t base_gfn; | |
163da372 SC |
1914 | int as_id, id; |
1915 | int r; | |
6aa8b732 | 1916 | |
a50d64d6 XG |
1917 | r = check_memory_region_flags(mem); |
1918 | if (r) | |
71a4c30b | 1919 | return r; |
a50d64d6 | 1920 | |
f481b069 PB |
1921 | as_id = mem->slot >> 16; |
1922 | id = (u16)mem->slot; | |
1923 | ||
6aa8b732 | 1924 | /* General sanity checks */ |
6b285a55 SC |
1925 | if ((mem->memory_size & (PAGE_SIZE - 1)) || |
1926 | (mem->memory_size != (unsigned long)mem->memory_size)) | |
71a4c30b | 1927 | return -EINVAL; |
6aa8b732 | 1928 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 1929 | return -EINVAL; |
fa3d315a | 1930 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 1931 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 1932 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 1933 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 1934 | mem->memory_size)) |
71a4c30b | 1935 | return -EINVAL; |
f481b069 | 1936 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 1937 | return -EINVAL; |
6aa8b732 | 1938 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 1939 | return -EINVAL; |
0f9bdef3 SC |
1940 | if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES) |
1941 | return -EINVAL; | |
6aa8b732 | 1942 | |
44401a20 | 1943 | slots = __kvm_memslots(kvm, as_id); |
6aa8b732 | 1944 | |
5c0b4f3d | 1945 | /* |
7cd08553 SC |
1946 | * Note, the old memslot (and the pointer itself!) may be invalidated |
1947 | * and/or destroyed by kvm_set_memslot(). | |
5c0b4f3d | 1948 | */ |
44401a20 | 1949 | old = id_to_memslot(slots, id); |
163da372 | 1950 | |
47ea7d90 | 1951 | if (!mem->memory_size) { |
7cd08553 | 1952 | if (!old || !old->npages) |
47ea7d90 | 1953 | return -EINVAL; |
5c0b4f3d | 1954 | |
7cd08553 | 1955 | if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages)) |
47ea7d90 | 1956 | return -EIO; |
6aa8b732 | 1957 | |
244893fa | 1958 | return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE); |
47ea7d90 | 1959 | } |
5c0b4f3d | 1960 | |
0f9bdef3 SC |
1961 | base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT); |
1962 | npages = (mem->memory_size >> PAGE_SHIFT); | |
163da372 | 1963 | |
7cd08553 | 1964 | if (!old || !old->npages) { |
5c0b4f3d | 1965 | change = KVM_MR_CREATE; |
afa319a5 SC |
1966 | |
1967 | /* | |
1968 | * To simplify KVM internals, the total number of pages across | |
1969 | * all memslots must fit in an unsigned long. | |
1970 | */ | |
0f9bdef3 | 1971 | if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages) |
afa319a5 | 1972 | return -EINVAL; |
5c0b4f3d | 1973 | } else { /* Modify an existing slot. */ |
0f9bdef3 SC |
1974 | if ((mem->userspace_addr != old->userspace_addr) || |
1975 | (npages != old->npages) || | |
1976 | ((mem->flags ^ old->flags) & KVM_MEM_READONLY)) | |
71a4c30b | 1977 | return -EINVAL; |
09170a49 | 1978 | |
0f9bdef3 | 1979 | if (base_gfn != old->base_gfn) |
5c0b4f3d | 1980 | change = KVM_MR_MOVE; |
0f9bdef3 | 1981 | else if (mem->flags != old->flags) |
5c0b4f3d SC |
1982 | change = KVM_MR_FLAGS_ONLY; |
1983 | else /* Nothing to change. */ | |
1984 | return 0; | |
09170a49 | 1985 | } |
6aa8b732 | 1986 | |
44401a20 | 1987 | if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) && |
0f9bdef3 | 1988 | kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages)) |
44401a20 | 1989 | return -EEXIST; |
6aa8b732 | 1990 | |
244893fa SC |
1991 | /* Allocate a slot that will persist in the memslot. */ |
1992 | new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT); | |
1993 | if (!new) | |
1994 | return -ENOMEM; | |
3c9bd400 | 1995 | |
244893fa SC |
1996 | new->as_id = as_id; |
1997 | new->id = id; | |
1998 | new->base_gfn = base_gfn; | |
1999 | new->npages = npages; | |
2000 | new->flags = mem->flags; | |
2001 | new->userspace_addr = mem->userspace_addr; | |
6aa8b732 | 2002 | |
244893fa | 2003 | r = kvm_set_memslot(kvm, old, new, change); |
cf47f50b | 2004 | if (r) |
244893fa | 2005 | kfree(new); |
6aa8b732 | 2006 | return r; |
210c7c4d | 2007 | } |
f78e0e2e SY |
2008 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
2009 | ||
2010 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 2011 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
2012 | { |
2013 | int r; | |
2014 | ||
79fac95e | 2015 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 2016 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 2017 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
2018 | return r; |
2019 | } | |
210c7c4d IE |
2020 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
2021 | ||
7940876e SH |
2022 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
2023 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 2024 | { |
f481b069 | 2025 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 2026 | return -EINVAL; |
09170a49 | 2027 | |
47ae31e2 | 2028 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
2029 | } |
2030 | ||
0dff0846 | 2031 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
2032 | /** |
2033 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
2034 | * @kvm: pointer to kvm instance | |
2035 | * @log: slot id and address to which we copy the log | |
2036 | * @is_dirty: set to '1' if any dirty pages were found | |
2037 | * @memslot: set to the associated memslot, always valid on success | |
2038 | */ | |
2039 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
2040 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 2041 | { |
9f6b8029 | 2042 | struct kvm_memslots *slots; |
843574a3 | 2043 | int i, as_id, id; |
87bf6e7d | 2044 | unsigned long n; |
6aa8b732 AK |
2045 | unsigned long any = 0; |
2046 | ||
b2cc64c4 PX |
2047 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
2048 | if (kvm->dirty_ring_size) | |
2049 | return -ENXIO; | |
2050 | ||
2a49f61d SC |
2051 | *memslot = NULL; |
2052 | *is_dirty = 0; | |
2053 | ||
f481b069 PB |
2054 | as_id = log->slot >> 16; |
2055 | id = (u16)log->slot; | |
2056 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 2057 | return -EINVAL; |
6aa8b732 | 2058 | |
f481b069 | 2059 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 2060 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 2061 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 2062 | return -ENOENT; |
6aa8b732 | 2063 | |
2a49f61d SC |
2064 | kvm_arch_sync_dirty_log(kvm, *memslot); |
2065 | ||
2066 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 2067 | |
cd1a4a98 | 2068 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 2069 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 2070 | |
2a49f61d | 2071 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 2072 | return -EFAULT; |
6aa8b732 | 2073 | |
5bb064dc ZX |
2074 | if (any) |
2075 | *is_dirty = 1; | |
843574a3 | 2076 | return 0; |
6aa8b732 | 2077 | } |
2ba9f0d8 | 2078 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 2079 | |
0dff0846 | 2080 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 2081 | /** |
b8b00220 | 2082 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 2083 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
2084 | * @kvm: pointer to kvm instance |
2085 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
2086 | * |
2087 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
2088 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
2089 | * following order: | |
2090 | * | |
2091 | * 1. Take a snapshot of the bit and clear it if needed. | |
2092 | * 2. Write protect the corresponding page. | |
2093 | * 3. Copy the snapshot to the userspace. | |
2094 | * 4. Upon return caller flushes TLB's if needed. | |
2095 | * | |
2096 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
2097 | * entry. This is not a problem because the page is reported dirty using | |
2098 | * the snapshot taken before and step 4 ensures that writes done after | |
2099 | * exiting to userspace will be logged for the next call. | |
2100 | * | |
2101 | */ | |
0dff0846 | 2102 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 2103 | { |
9f6b8029 | 2104 | struct kvm_memslots *slots; |
ba0513b5 | 2105 | struct kvm_memory_slot *memslot; |
58d6db34 | 2106 | int i, as_id, id; |
ba0513b5 MS |
2107 | unsigned long n; |
2108 | unsigned long *dirty_bitmap; | |
2109 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2110 | bool flush; |
ba0513b5 | 2111 | |
b2cc64c4 PX |
2112 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
2113 | if (kvm->dirty_ring_size) | |
2114 | return -ENXIO; | |
2115 | ||
f481b069 PB |
2116 | as_id = log->slot >> 16; |
2117 | id = (u16)log->slot; | |
2118 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 2119 | return -EINVAL; |
ba0513b5 | 2120 | |
f481b069 PB |
2121 | slots = __kvm_memslots(kvm, as_id); |
2122 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2123 | if (!memslot || !memslot->dirty_bitmap) |
2124 | return -ENOENT; | |
ba0513b5 MS |
2125 | |
2126 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 2127 | |
0dff0846 SC |
2128 | kvm_arch_sync_dirty_log(kvm, memslot); |
2129 | ||
ba0513b5 | 2130 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 2131 | flush = false; |
2a31b9db PB |
2132 | if (kvm->manual_dirty_log_protect) { |
2133 | /* | |
2134 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
2135 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
2136 | * is some code duplication between this function and | |
2137 | * kvm_get_dirty_log, but hopefully all architecture | |
2138 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
2139 | * can be eliminated. | |
2140 | */ | |
2141 | dirty_bitmap_buffer = dirty_bitmap; | |
2142 | } else { | |
2143 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
2144 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 2145 | |
531810ca | 2146 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
2147 | for (i = 0; i < n / sizeof(long); i++) { |
2148 | unsigned long mask; | |
2149 | gfn_t offset; | |
ba0513b5 | 2150 | |
2a31b9db PB |
2151 | if (!dirty_bitmap[i]) |
2152 | continue; | |
2153 | ||
0dff0846 | 2154 | flush = true; |
2a31b9db PB |
2155 | mask = xchg(&dirty_bitmap[i], 0); |
2156 | dirty_bitmap_buffer[i] = mask; | |
2157 | ||
a67794ca LT |
2158 | offset = i * BITS_PER_LONG; |
2159 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
2160 | offset, mask); | |
2a31b9db | 2161 | } |
531810ca | 2162 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
2163 | } |
2164 | ||
0dff0846 SC |
2165 | if (flush) |
2166 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2167 | ||
2a31b9db PB |
2168 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
2169 | return -EFAULT; | |
2170 | return 0; | |
2171 | } | |
0dff0846 SC |
2172 | |
2173 | ||
2174 | /** | |
2175 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
2176 | * @kvm: kvm instance | |
2177 | * @log: slot id and address to which we copy the log | |
2178 | * | |
2179 | * Steps 1-4 below provide general overview of dirty page logging. See | |
2180 | * kvm_get_dirty_log_protect() function description for additional details. | |
2181 | * | |
2182 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
2183 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
2184 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
2185 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
2186 | * writes will be marked dirty for next log read. | |
2187 | * | |
2188 | * 1. Take a snapshot of the bit and clear it if needed. | |
2189 | * 2. Write protect the corresponding page. | |
2190 | * 3. Copy the snapshot to the userspace. | |
2191 | * 4. Flush TLB's if needed. | |
2192 | */ | |
2193 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
2194 | struct kvm_dirty_log *log) | |
2195 | { | |
2196 | int r; | |
2197 | ||
2198 | mutex_lock(&kvm->slots_lock); | |
2199 | ||
2200 | r = kvm_get_dirty_log_protect(kvm, log); | |
2201 | ||
2202 | mutex_unlock(&kvm->slots_lock); | |
2203 | return r; | |
2204 | } | |
2a31b9db PB |
2205 | |
2206 | /** | |
2207 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
2208 | * and reenable dirty page tracking for the corresponding pages. | |
2209 | * @kvm: pointer to kvm instance | |
2210 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
2211 | */ | |
0dff0846 SC |
2212 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
2213 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
2214 | { |
2215 | struct kvm_memslots *slots; | |
2216 | struct kvm_memory_slot *memslot; | |
98938aa8 | 2217 | int as_id, id; |
2a31b9db | 2218 | gfn_t offset; |
98938aa8 | 2219 | unsigned long i, n; |
2a31b9db PB |
2220 | unsigned long *dirty_bitmap; |
2221 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2222 | bool flush; |
2a31b9db | 2223 | |
b2cc64c4 PX |
2224 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
2225 | if (kvm->dirty_ring_size) | |
2226 | return -ENXIO; | |
2227 | ||
2a31b9db PB |
2228 | as_id = log->slot >> 16; |
2229 | id = (u16)log->slot; | |
2230 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
2231 | return -EINVAL; | |
2232 | ||
76d58e0f | 2233 | if (log->first_page & 63) |
2a31b9db PB |
2234 | return -EINVAL; |
2235 | ||
2236 | slots = __kvm_memslots(kvm, as_id); | |
2237 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2238 | if (!memslot || !memslot->dirty_bitmap) |
2239 | return -ENOENT; | |
2a31b9db PB |
2240 | |
2241 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 2242 | |
4ddc9204 | 2243 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
2244 | |
2245 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
2246 | log->num_pages > memslot->npages - log->first_page || |
2247 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
2248 | return -EINVAL; | |
98938aa8 | 2249 | |
0dff0846 SC |
2250 | kvm_arch_sync_dirty_log(kvm, memslot); |
2251 | ||
2252 | flush = false; | |
2a31b9db PB |
2253 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
2254 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
2255 | return -EFAULT; | |
ba0513b5 | 2256 | |
531810ca | 2257 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
2258 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
2259 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
2260 | i++, offset += BITS_PER_LONG) { |
2261 | unsigned long mask = *dirty_bitmap_buffer++; | |
2262 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
2263 | if (!mask) | |
ba0513b5 MS |
2264 | continue; |
2265 | ||
2a31b9db | 2266 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 2267 | |
2a31b9db PB |
2268 | /* |
2269 | * mask contains the bits that really have been cleared. This | |
2270 | * never includes any bits beyond the length of the memslot (if | |
2271 | * the length is not aligned to 64 pages), therefore it is not | |
2272 | * a problem if userspace sets them in log->dirty_bitmap. | |
2273 | */ | |
58d2930f | 2274 | if (mask) { |
0dff0846 | 2275 | flush = true; |
58d2930f TY |
2276 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
2277 | offset, mask); | |
2278 | } | |
ba0513b5 | 2279 | } |
531810ca | 2280 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 2281 | |
0dff0846 SC |
2282 | if (flush) |
2283 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2284 | ||
58d6db34 | 2285 | return 0; |
ba0513b5 | 2286 | } |
0dff0846 SC |
2287 | |
2288 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
2289 | struct kvm_clear_dirty_log *log) | |
2290 | { | |
2291 | int r; | |
2292 | ||
2293 | mutex_lock(&kvm->slots_lock); | |
2294 | ||
2295 | r = kvm_clear_dirty_log_protect(kvm, log); | |
2296 | ||
2297 | mutex_unlock(&kvm->slots_lock); | |
2298 | return r; | |
2299 | } | |
2300 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 2301 | |
49c7754c GN |
2302 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
2303 | { | |
2304 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
2305 | } | |
a1f4d395 | 2306 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 2307 | |
8e73485c PB |
2308 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
2309 | { | |
fe22ed82 | 2310 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
a54d8066 | 2311 | u64 gen = slots->generation; |
fe22ed82 | 2312 | struct kvm_memory_slot *slot; |
fe22ed82 | 2313 | |
a54d8066 MS |
2314 | /* |
2315 | * This also protects against using a memslot from a different address space, | |
2316 | * since different address spaces have different generation numbers. | |
2317 | */ | |
2318 | if (unlikely(gen != vcpu->last_used_slot_gen)) { | |
2319 | vcpu->last_used_slot = NULL; | |
2320 | vcpu->last_used_slot_gen = gen; | |
2321 | } | |
2322 | ||
2323 | slot = try_get_memslot(vcpu->last_used_slot, gfn); | |
fe22ed82 DM |
2324 | if (slot) |
2325 | return slot; | |
2326 | ||
2327 | /* | |
2328 | * Fall back to searching all memslots. We purposely use | |
2329 | * search_memslots() instead of __gfn_to_memslot() to avoid | |
a54d8066 | 2330 | * thrashing the VM-wide last_used_slot in kvm_memslots. |
fe22ed82 | 2331 | */ |
a54d8066 | 2332 | slot = search_memslots(slots, gfn, false); |
fe22ed82 | 2333 | if (slot) { |
a54d8066 | 2334 | vcpu->last_used_slot = slot; |
fe22ed82 DM |
2335 | return slot; |
2336 | } | |
2337 | ||
2338 | return NULL; | |
8e73485c PB |
2339 | } |
2340 | ||
33e94154 | 2341 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 2342 | { |
bf3e05bc | 2343 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 2344 | |
c36b7150 | 2345 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
2346 | } |
2347 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
2348 | ||
995decb6 VK |
2349 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
2350 | { | |
2351 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2352 | ||
2353 | return kvm_is_visible_memslot(memslot); | |
2354 | } | |
2355 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
2356 | ||
f9b84e19 | 2357 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
2358 | { |
2359 | struct vm_area_struct *vma; | |
2360 | unsigned long addr, size; | |
2361 | ||
2362 | size = PAGE_SIZE; | |
2363 | ||
42cde48b | 2364 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
2365 | if (kvm_is_error_hva(addr)) |
2366 | return PAGE_SIZE; | |
2367 | ||
d8ed45c5 | 2368 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
2369 | vma = find_vma(current->mm, addr); |
2370 | if (!vma) | |
2371 | goto out; | |
2372 | ||
2373 | size = vma_kernel_pagesize(vma); | |
2374 | ||
2375 | out: | |
d8ed45c5 | 2376 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
2377 | |
2378 | return size; | |
2379 | } | |
2380 | ||
8283e36a | 2381 | static bool memslot_is_readonly(const struct kvm_memory_slot *slot) |
4d8b81ab XG |
2382 | { |
2383 | return slot->flags & KVM_MEM_READONLY; | |
2384 | } | |
2385 | ||
8283e36a | 2386 | static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn, |
4d8b81ab | 2387 | gfn_t *nr_pages, bool write) |
539cb660 | 2388 | { |
bc6678a3 | 2389 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 2390 | return KVM_HVA_ERR_BAD; |
48987781 | 2391 | |
4d8b81ab XG |
2392 | if (memslot_is_readonly(slot) && write) |
2393 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
2394 | |
2395 | if (nr_pages) | |
2396 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
2397 | ||
4d8b81ab | 2398 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 2399 | } |
48987781 | 2400 | |
4d8b81ab XG |
2401 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2402 | gfn_t *nr_pages) | |
2403 | { | |
2404 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 2405 | } |
48987781 | 2406 | |
4d8b81ab | 2407 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 2408 | gfn_t gfn) |
4d8b81ab XG |
2409 | { |
2410 | return gfn_to_hva_many(slot, gfn, NULL); | |
2411 | } | |
2412 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
2413 | ||
48987781 XG |
2414 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
2415 | { | |
49c7754c | 2416 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 2417 | } |
0d150298 | 2418 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 2419 | |
8e73485c PB |
2420 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
2421 | { | |
2422 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
2423 | } | |
2424 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
2425 | ||
86ab8cff | 2426 | /* |
970c0d4b WY |
2427 | * Return the hva of a @gfn and the R/W attribute if possible. |
2428 | * | |
2429 | * @slot: the kvm_memory_slot which contains @gfn | |
2430 | * @gfn: the gfn to be translated | |
2431 | * @writable: used to return the read/write attribute of the @slot if the hva | |
2432 | * is valid and @writable is not NULL | |
86ab8cff | 2433 | */ |
64d83126 CD |
2434 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
2435 | gfn_t gfn, bool *writable) | |
86ab8cff | 2436 | { |
a2ac07fe GN |
2437 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
2438 | ||
2439 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
2440 | *writable = !memslot_is_readonly(slot); |
2441 | ||
a2ac07fe | 2442 | return hva; |
86ab8cff XG |
2443 | } |
2444 | ||
64d83126 CD |
2445 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
2446 | { | |
2447 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2448 | ||
2449 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2450 | } | |
2451 | ||
8e73485c PB |
2452 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
2453 | { | |
2454 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2455 | ||
2456 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2457 | } | |
2458 | ||
fafc3dba HY |
2459 | static inline int check_user_page_hwpoison(unsigned long addr) |
2460 | { | |
0d731759 | 2461 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 2462 | |
0d731759 | 2463 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
2464 | return rc == -EHWPOISON; |
2465 | } | |
2466 | ||
2fc84311 | 2467 | /* |
b9b33da2 PB |
2468 | * The fast path to get the writable pfn which will be stored in @pfn, |
2469 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 2470 | * only part that runs if we can in atomic context. |
2fc84311 | 2471 | */ |
b9b33da2 PB |
2472 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
2473 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 2474 | { |
8d4e1288 | 2475 | struct page *page[1]; |
954bbbc2 | 2476 | |
12ce13fe XG |
2477 | /* |
2478 | * Fast pin a writable pfn only if it is a write fault request | |
2479 | * or the caller allows to map a writable pfn for a read fault | |
2480 | * request. | |
2481 | */ | |
2482 | if (!(write_fault || writable)) | |
2483 | return false; | |
612819c3 | 2484 | |
dadbb612 | 2485 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 2486 | *pfn = page_to_pfn(page[0]); |
612819c3 | 2487 | |
2fc84311 XG |
2488 | if (writable) |
2489 | *writable = true; | |
2490 | return true; | |
2491 | } | |
af585b92 | 2492 | |
2fc84311 XG |
2493 | return false; |
2494 | } | |
612819c3 | 2495 | |
2fc84311 XG |
2496 | /* |
2497 | * The slow path to get the pfn of the specified host virtual address, | |
2498 | * 1 indicates success, -errno is returned if error is detected. | |
2499 | */ | |
2500 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 2501 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 2502 | { |
ce53053c AV |
2503 | unsigned int flags = FOLL_HWPOISON; |
2504 | struct page *page; | |
2fc84311 | 2505 | int npages = 0; |
612819c3 | 2506 | |
2fc84311 XG |
2507 | might_sleep(); |
2508 | ||
2509 | if (writable) | |
2510 | *writable = write_fault; | |
2511 | ||
ce53053c AV |
2512 | if (write_fault) |
2513 | flags |= FOLL_WRITE; | |
2514 | if (async) | |
2515 | flags |= FOLL_NOWAIT; | |
d4944b0e | 2516 | |
ce53053c | 2517 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
2518 | if (npages != 1) |
2519 | return npages; | |
2520 | ||
2521 | /* map read fault as writable if possible */ | |
12ce13fe | 2522 | if (unlikely(!write_fault) && writable) { |
ce53053c | 2523 | struct page *wpage; |
2fc84311 | 2524 | |
dadbb612 | 2525 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 2526 | *writable = true; |
ce53053c AV |
2527 | put_page(page); |
2528 | page = wpage; | |
612819c3 | 2529 | } |
887c08ac | 2530 | } |
ce53053c | 2531 | *pfn = page_to_pfn(page); |
2fc84311 XG |
2532 | return npages; |
2533 | } | |
539cb660 | 2534 | |
4d8b81ab XG |
2535 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
2536 | { | |
2537 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
2538 | return false; | |
2e2e3738 | 2539 | |
4d8b81ab XG |
2540 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
2541 | return false; | |
887c08ac | 2542 | |
4d8b81ab XG |
2543 | return true; |
2544 | } | |
bf998156 | 2545 | |
f8be156b NP |
2546 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
2547 | { | |
b14b2690 SC |
2548 | struct page *page = kvm_pfn_to_refcounted_page(pfn); |
2549 | ||
2550 | if (!page) | |
f8be156b | 2551 | return 1; |
b14b2690 SC |
2552 | |
2553 | return get_page_unless_zero(page); | |
f8be156b NP |
2554 | } |
2555 | ||
92176a8e | 2556 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1625566e XT |
2557 | unsigned long addr, bool write_fault, |
2558 | bool *writable, kvm_pfn_t *p_pfn) | |
92176a8e | 2559 | { |
a9545779 | 2560 | kvm_pfn_t pfn; |
bd2fae8d PB |
2561 | pte_t *ptep; |
2562 | spinlock_t *ptl; | |
add6a0cd PB |
2563 | int r; |
2564 | ||
9fd6dad1 | 2565 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2566 | if (r) { |
2567 | /* | |
2568 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
2569 | * not call the fault handler, so do it here. | |
2570 | */ | |
2571 | bool unlocked = false; | |
64019a2e | 2572 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
2573 | (write_fault ? FAULT_FLAG_WRITE : 0), |
2574 | &unlocked); | |
a8387d0b PB |
2575 | if (unlocked) |
2576 | return -EAGAIN; | |
add6a0cd PB |
2577 | if (r) |
2578 | return r; | |
2579 | ||
9fd6dad1 | 2580 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2581 | if (r) |
2582 | return r; | |
bd2fae8d | 2583 | } |
add6a0cd | 2584 | |
bd2fae8d PB |
2585 | if (write_fault && !pte_write(*ptep)) { |
2586 | pfn = KVM_PFN_ERR_RO_FAULT; | |
2587 | goto out; | |
add6a0cd PB |
2588 | } |
2589 | ||
a340b3e2 | 2590 | if (writable) |
bd2fae8d PB |
2591 | *writable = pte_write(*ptep); |
2592 | pfn = pte_pfn(*ptep); | |
add6a0cd PB |
2593 | |
2594 | /* | |
2595 | * Get a reference here because callers of *hva_to_pfn* and | |
2596 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
2597 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
36c3ce6c | 2598 | * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will |
add6a0cd PB |
2599 | * simply do nothing for reserved pfns. |
2600 | * | |
2601 | * Whoever called remap_pfn_range is also going to call e.g. | |
2602 | * unmap_mapping_range before the underlying pages are freed, | |
2603 | * causing a call to our MMU notifier. | |
f8be156b NP |
2604 | * |
2605 | * Certain IO or PFNMAP mappings can be backed with valid | |
2606 | * struct pages, but be allocated without refcounting e.g., | |
2607 | * tail pages of non-compound higher order allocations, which | |
2608 | * would then underflow the refcount when the caller does the | |
2609 | * required put_page. Don't allow those pages here. | |
add6a0cd | 2610 | */ |
f8be156b NP |
2611 | if (!kvm_try_get_pfn(pfn)) |
2612 | r = -EFAULT; | |
add6a0cd | 2613 | |
bd2fae8d PB |
2614 | out: |
2615 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 2616 | *p_pfn = pfn; |
f8be156b NP |
2617 | |
2618 | return r; | |
92176a8e PB |
2619 | } |
2620 | ||
12ce13fe XG |
2621 | /* |
2622 | * Pin guest page in memory and return its pfn. | |
2623 | * @addr: host virtual address which maps memory to the guest | |
2624 | * @atomic: whether this function can sleep | |
2625 | * @async: whether this function need to wait IO complete if the | |
2626 | * host page is not in the memory | |
2627 | * @write_fault: whether we should get a writable host page | |
2628 | * @writable: whether it allows to map a writable host page for !@write_fault | |
2629 | * | |
2630 | * The function will map a writable host page for these two cases: | |
2631 | * 1): @write_fault = true | |
2632 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
2633 | * whether the mapping is writable. | |
2634 | */ | |
982ed0de DW |
2635 | kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2636 | bool write_fault, bool *writable) | |
2fc84311 XG |
2637 | { |
2638 | struct vm_area_struct *vma; | |
943dfea8 | 2639 | kvm_pfn_t pfn; |
92176a8e | 2640 | int npages, r; |
2e2e3738 | 2641 | |
2fc84311 XG |
2642 | /* we can do it either atomically or asynchronously, not both */ |
2643 | BUG_ON(atomic && async); | |
8d4e1288 | 2644 | |
b9b33da2 | 2645 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
2646 | return pfn; |
2647 | ||
2648 | if (atomic) | |
2649 | return KVM_PFN_ERR_FAULT; | |
2650 | ||
2651 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
2652 | if (npages == 1) | |
2653 | return pfn; | |
8d4e1288 | 2654 | |
d8ed45c5 | 2655 | mmap_read_lock(current->mm); |
2fc84311 XG |
2656 | if (npages == -EHWPOISON || |
2657 | (!async && check_user_page_hwpoison(addr))) { | |
2658 | pfn = KVM_PFN_ERR_HWPOISON; | |
2659 | goto exit; | |
2660 | } | |
2661 | ||
a8387d0b | 2662 | retry: |
fc98c03b | 2663 | vma = vma_lookup(current->mm, addr); |
2fc84311 XG |
2664 | |
2665 | if (vma == NULL) | |
2666 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 2667 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
1625566e | 2668 | r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn); |
a8387d0b PB |
2669 | if (r == -EAGAIN) |
2670 | goto retry; | |
92176a8e PB |
2671 | if (r < 0) |
2672 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 2673 | } else { |
4d8b81ab | 2674 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
2675 | *async = true; |
2676 | pfn = KVM_PFN_ERR_FAULT; | |
2677 | } | |
2678 | exit: | |
d8ed45c5 | 2679 | mmap_read_unlock(current->mm); |
2e2e3738 | 2680 | return pfn; |
35149e21 AL |
2681 | } |
2682 | ||
8283e36a | 2683 | kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn, |
ba049e93 | 2684 | bool atomic, bool *async, bool write_fault, |
4a42d848 | 2685 | bool *writable, hva_t *hva) |
887c08ac | 2686 | { |
4d8b81ab XG |
2687 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
2688 | ||
4a42d848 DS |
2689 | if (hva) |
2690 | *hva = addr; | |
2691 | ||
b2740d35 PB |
2692 | if (addr == KVM_HVA_ERR_RO_BAD) { |
2693 | if (writable) | |
2694 | *writable = false; | |
4d8b81ab | 2695 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 2696 | } |
4d8b81ab | 2697 | |
b2740d35 PB |
2698 | if (kvm_is_error_hva(addr)) { |
2699 | if (writable) | |
2700 | *writable = false; | |
81c52c56 | 2701 | return KVM_PFN_NOSLOT; |
b2740d35 | 2702 | } |
4d8b81ab XG |
2703 | |
2704 | /* Do not map writable pfn in the readonly memslot. */ | |
2705 | if (writable && memslot_is_readonly(slot)) { | |
2706 | *writable = false; | |
2707 | writable = NULL; | |
2708 | } | |
2709 | ||
2710 | return hva_to_pfn(addr, atomic, async, write_fault, | |
2711 | writable); | |
887c08ac | 2712 | } |
3520469d | 2713 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 2714 | |
ba049e93 | 2715 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
2716 | bool *writable) |
2717 | { | |
e37afc6e | 2718 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
4a42d848 | 2719 | write_fault, writable, NULL); |
612819c3 MT |
2720 | } |
2721 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
2722 | ||
8283e36a | 2723 | kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2724 | { |
4a42d848 | 2725 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL); |
506f0d6f | 2726 | } |
e37afc6e | 2727 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 2728 | |
8283e36a | 2729 | kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2730 | { |
4a42d848 | 2731 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL); |
506f0d6f | 2732 | } |
037d92dc | 2733 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 2734 | |
ba049e93 | 2735 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2736 | { |
2737 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2738 | } | |
2739 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
2740 | ||
ba049e93 | 2741 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
2742 | { |
2743 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
2744 | } | |
2745 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
2746 | ||
ba049e93 | 2747 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2748 | { |
2749 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2750 | } | |
2751 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
2752 | ||
d9ef13c2 PB |
2753 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2754 | struct page **pages, int nr_pages) | |
48987781 XG |
2755 | { |
2756 | unsigned long addr; | |
076b925d | 2757 | gfn_t entry = 0; |
48987781 | 2758 | |
d9ef13c2 | 2759 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
2760 | if (kvm_is_error_hva(addr)) |
2761 | return -1; | |
2762 | ||
2763 | if (entry < nr_pages) | |
2764 | return 0; | |
2765 | ||
dadbb612 | 2766 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
2767 | } |
2768 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
2769 | ||
b1624f99 SC |
2770 | /* |
2771 | * Do not use this helper unless you are absolutely certain the gfn _must_ be | |
2772 | * backed by 'struct page'. A valid example is if the backing memslot is | |
2773 | * controlled by KVM. Note, if the returned page is valid, it's refcount has | |
2774 | * been elevated by gfn_to_pfn(). | |
2775 | */ | |
2776 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) | |
a2766325 | 2777 | { |
b14b2690 | 2778 | struct page *page; |
b1624f99 SC |
2779 | kvm_pfn_t pfn; |
2780 | ||
2781 | pfn = gfn_to_pfn(kvm, gfn); | |
2782 | ||
81c52c56 | 2783 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 2784 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2785 | |
b14b2690 SC |
2786 | page = kvm_pfn_to_refcounted_page(pfn); |
2787 | if (!page) | |
6cede2e6 | 2788 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2789 | |
b14b2690 | 2790 | return page; |
a2766325 | 2791 | } |
954bbbc2 AK |
2792 | EXPORT_SYMBOL_GPL(gfn_to_page); |
2793 | ||
357a18ad | 2794 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty) |
91724814 | 2795 | { |
91724814 BO |
2796 | if (dirty) |
2797 | kvm_release_pfn_dirty(pfn); | |
2798 | else | |
2799 | kvm_release_pfn_clean(pfn); | |
2800 | } | |
2801 | ||
357a18ad | 2802 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
e45adf66 KA |
2803 | { |
2804 | kvm_pfn_t pfn; | |
2805 | void *hva = NULL; | |
2806 | struct page *page = KVM_UNMAPPED_PAGE; | |
2807 | ||
2808 | if (!map) | |
2809 | return -EINVAL; | |
2810 | ||
357a18ad | 2811 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
e45adf66 KA |
2812 | if (is_error_noslot_pfn(pfn)) |
2813 | return -EINVAL; | |
2814 | ||
2815 | if (pfn_valid(pfn)) { | |
2816 | page = pfn_to_page(pfn); | |
357a18ad | 2817 | hva = kmap(page); |
d30b214d | 2818 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2819 | } else { |
357a18ad | 2820 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
d30b214d | 2821 | #endif |
e45adf66 KA |
2822 | } |
2823 | ||
2824 | if (!hva) | |
2825 | return -EFAULT; | |
2826 | ||
2827 | map->page = page; | |
2828 | map->hva = hva; | |
2829 | map->pfn = pfn; | |
2830 | map->gfn = gfn; | |
2831 | ||
2832 | return 0; | |
2833 | } | |
e45adf66 KA |
2834 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); |
2835 | ||
357a18ad | 2836 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) |
e45adf66 KA |
2837 | { |
2838 | if (!map) | |
2839 | return; | |
2840 | ||
2841 | if (!map->hva) | |
2842 | return; | |
2843 | ||
357a18ad DW |
2844 | if (map->page != KVM_UNMAPPED_PAGE) |
2845 | kunmap(map->page); | |
eb1f2f38 | 2846 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2847 | else |
357a18ad | 2848 | memunmap(map->hva); |
eb1f2f38 | 2849 | #endif |
e45adf66 | 2850 | |
91724814 | 2851 | if (dirty) |
357a18ad | 2852 | kvm_vcpu_mark_page_dirty(vcpu, map->gfn); |
91724814 | 2853 | |
357a18ad | 2854 | kvm_release_pfn(map->pfn, dirty); |
e45adf66 KA |
2855 | |
2856 | map->hva = NULL; | |
2857 | map->page = NULL; | |
2858 | } | |
2859 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); | |
2860 | ||
8e1c6914 SC |
2861 | static bool kvm_is_ad_tracked_page(struct page *page) |
2862 | { | |
2863 | /* | |
2864 | * Per page-flags.h, pages tagged PG_reserved "should in general not be | |
2865 | * touched (e.g. set dirty) except by its owner". | |
2866 | */ | |
2867 | return !PageReserved(page); | |
2868 | } | |
2869 | ||
2870 | static void kvm_set_page_dirty(struct page *page) | |
2871 | { | |
2872 | if (kvm_is_ad_tracked_page(page)) | |
2873 | SetPageDirty(page); | |
2874 | } | |
2875 | ||
2876 | static void kvm_set_page_accessed(struct page *page) | |
2877 | { | |
2878 | if (kvm_is_ad_tracked_page(page)) | |
2879 | mark_page_accessed(page); | |
2880 | } | |
2881 | ||
b4231d61 IE |
2882 | void kvm_release_page_clean(struct page *page) |
2883 | { | |
32cad84f XG |
2884 | WARN_ON(is_error_page(page)); |
2885 | ||
8e1c6914 SC |
2886 | kvm_set_page_accessed(page); |
2887 | put_page(page); | |
b4231d61 IE |
2888 | } |
2889 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
2890 | ||
ba049e93 | 2891 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 2892 | { |
b14b2690 SC |
2893 | struct page *page; |
2894 | ||
2895 | if (is_error_noslot_pfn(pfn)) | |
2896 | return; | |
2897 | ||
2898 | page = kvm_pfn_to_refcounted_page(pfn); | |
2899 | if (!page) | |
2900 | return; | |
2901 | ||
2902 | kvm_release_page_clean(page); | |
35149e21 AL |
2903 | } |
2904 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
2905 | ||
b4231d61 | 2906 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 2907 | { |
a2766325 XG |
2908 | WARN_ON(is_error_page(page)); |
2909 | ||
8e1c6914 SC |
2910 | kvm_set_page_dirty(page); |
2911 | kvm_release_page_clean(page); | |
35149e21 AL |
2912 | } |
2913 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
2914 | ||
f7a6509f | 2915 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2916 | { |
b14b2690 SC |
2917 | struct page *page; |
2918 | ||
2919 | if (is_error_noslot_pfn(pfn)) | |
2920 | return; | |
2921 | ||
2922 | page = kvm_pfn_to_refcounted_page(pfn); | |
2923 | if (!page) | |
2924 | return; | |
2925 | ||
2926 | kvm_release_page_dirty(page); | |
35149e21 | 2927 | } |
f7a6509f | 2928 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 2929 | |
8e1c6914 SC |
2930 | /* |
2931 | * Note, checking for an error/noslot pfn is the caller's responsibility when | |
2932 | * directly marking a page dirty/accessed. Unlike the "release" helpers, the | |
2933 | * "set" helpers are not to be used when the pfn might point at garbage. | |
2934 | */ | |
ba049e93 | 2935 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2936 | { |
8e1c6914 SC |
2937 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
2938 | return; | |
2939 | ||
2940 | if (pfn_valid(pfn)) | |
2941 | kvm_set_page_dirty(pfn_to_page(pfn)); | |
8a7ae055 | 2942 | } |
35149e21 AL |
2943 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
2944 | ||
ba049e93 | 2945 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 2946 | { |
8e1c6914 SC |
2947 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
2948 | return; | |
2949 | ||
2950 | if (pfn_valid(pfn)) | |
2951 | kvm_set_page_accessed(pfn_to_page(pfn)); | |
35149e21 AL |
2952 | } |
2953 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
2954 | ||
195aefde IE |
2955 | static int next_segment(unsigned long len, int offset) |
2956 | { | |
2957 | if (len > PAGE_SIZE - offset) | |
2958 | return PAGE_SIZE - offset; | |
2959 | else | |
2960 | return len; | |
2961 | } | |
2962 | ||
8e73485c PB |
2963 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
2964 | void *data, int offset, int len) | |
195aefde | 2965 | { |
e0506bcb IE |
2966 | int r; |
2967 | unsigned long addr; | |
195aefde | 2968 | |
8e73485c | 2969 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
2970 | if (kvm_is_error_hva(addr)) |
2971 | return -EFAULT; | |
3180a7fc | 2972 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 2973 | if (r) |
195aefde | 2974 | return -EFAULT; |
195aefde IE |
2975 | return 0; |
2976 | } | |
8e73485c PB |
2977 | |
2978 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
2979 | int len) | |
2980 | { | |
2981 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2982 | ||
2983 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2984 | } | |
195aefde IE |
2985 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
2986 | ||
8e73485c PB |
2987 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
2988 | int offset, int len) | |
2989 | { | |
2990 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2991 | ||
2992 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2993 | } | |
2994 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
2995 | ||
195aefde IE |
2996 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
2997 | { | |
2998 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2999 | int seg; | |
3000 | int offset = offset_in_page(gpa); | |
3001 | int ret; | |
3002 | ||
3003 | while ((seg = next_segment(len, offset)) != 0) { | |
3004 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
3005 | if (ret < 0) | |
3006 | return ret; | |
3007 | offset = 0; | |
3008 | len -= seg; | |
3009 | data += seg; | |
3010 | ++gfn; | |
3011 | } | |
3012 | return 0; | |
3013 | } | |
3014 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
3015 | ||
8e73485c | 3016 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 3017 | { |
7ec54588 | 3018 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 3019 | int seg; |
7ec54588 | 3020 | int offset = offset_in_page(gpa); |
8e73485c PB |
3021 | int ret; |
3022 | ||
3023 | while ((seg = next_segment(len, offset)) != 0) { | |
3024 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
3025 | if (ret < 0) | |
3026 | return ret; | |
3027 | offset = 0; | |
3028 | len -= seg; | |
3029 | data += seg; | |
3030 | ++gfn; | |
3031 | } | |
3032 | return 0; | |
3033 | } | |
3034 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 3035 | |
8e73485c PB |
3036 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
3037 | void *data, int offset, unsigned long len) | |
3038 | { | |
3039 | int r; | |
3040 | unsigned long addr; | |
3041 | ||
3042 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
3043 | if (kvm_is_error_hva(addr)) |
3044 | return -EFAULT; | |
0aac03f0 | 3045 | pagefault_disable(); |
3180a7fc | 3046 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 3047 | pagefault_enable(); |
7ec54588 MT |
3048 | if (r) |
3049 | return -EFAULT; | |
3050 | return 0; | |
3051 | } | |
7ec54588 | 3052 | |
8e73485c PB |
3053 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
3054 | void *data, unsigned long len) | |
3055 | { | |
3056 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3057 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3058 | int offset = offset_in_page(gpa); | |
3059 | ||
3060 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
3061 | } | |
3062 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
3063 | ||
28bd726a PX |
3064 | static int __kvm_write_guest_page(struct kvm *kvm, |
3065 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 3066 | const void *data, int offset, int len) |
195aefde | 3067 | { |
e0506bcb IE |
3068 | int r; |
3069 | unsigned long addr; | |
195aefde | 3070 | |
251eb841 | 3071 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
3072 | if (kvm_is_error_hva(addr)) |
3073 | return -EFAULT; | |
8b0cedff | 3074 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 3075 | if (r) |
195aefde | 3076 | return -EFAULT; |
28bd726a | 3077 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
3078 | return 0; |
3079 | } | |
8e73485c PB |
3080 | |
3081 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
3082 | const void *data, int offset, int len) | |
3083 | { | |
3084 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
3085 | ||
28bd726a | 3086 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 3087 | } |
195aefde IE |
3088 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
3089 | ||
8e73485c PB |
3090 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3091 | const void *data, int offset, int len) | |
3092 | { | |
3093 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3094 | ||
28bd726a | 3095 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
3096 | } |
3097 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
3098 | ||
195aefde IE |
3099 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
3100 | unsigned long len) | |
3101 | { | |
3102 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3103 | int seg; | |
3104 | int offset = offset_in_page(gpa); | |
3105 | int ret; | |
3106 | ||
3107 | while ((seg = next_segment(len, offset)) != 0) { | |
3108 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
3109 | if (ret < 0) | |
3110 | return ret; | |
3111 | offset = 0; | |
3112 | len -= seg; | |
3113 | data += seg; | |
3114 | ++gfn; | |
3115 | } | |
3116 | return 0; | |
3117 | } | |
ff651cb6 | 3118 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 3119 | |
8e73485c PB |
3120 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
3121 | unsigned long len) | |
3122 | { | |
3123 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3124 | int seg; | |
3125 | int offset = offset_in_page(gpa); | |
3126 | int ret; | |
3127 | ||
3128 | while ((seg = next_segment(len, offset)) != 0) { | |
3129 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
3130 | if (ret < 0) | |
3131 | return ret; | |
3132 | offset = 0; | |
3133 | len -= seg; | |
3134 | data += seg; | |
3135 | ++gfn; | |
3136 | } | |
3137 | return 0; | |
3138 | } | |
3139 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
3140 | ||
5a2d4365 PB |
3141 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
3142 | struct gfn_to_hva_cache *ghc, | |
3143 | gpa_t gpa, unsigned long len) | |
49c7754c | 3144 | { |
49c7754c | 3145 | int offset = offset_in_page(gpa); |
8f964525 AH |
3146 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
3147 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
3148 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
3149 | gfn_t nr_pages_avail; | |
49c7754c | 3150 | |
6ad1e29f | 3151 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 3152 | ghc->generation = slots->generation; |
6ad1e29f SC |
3153 | |
3154 | if (start_gfn > end_gfn) { | |
3155 | ghc->hva = KVM_HVA_ERR_BAD; | |
3156 | return -EINVAL; | |
3157 | } | |
f1b9dd5e JM |
3158 | |
3159 | /* | |
3160 | * If the requested region crosses two memslots, we still | |
3161 | * verify that the entire region is valid here. | |
3162 | */ | |
6ad1e29f | 3163 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
3164 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
3165 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
3166 | &nr_pages_avail); | |
3167 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 3168 | return -EFAULT; |
f1b9dd5e JM |
3169 | } |
3170 | ||
3171 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 3172 | if (nr_pages_needed == 1) |
49c7754c | 3173 | ghc->hva += offset; |
f1b9dd5e | 3174 | else |
8f964525 | 3175 | ghc->memslot = NULL; |
f1b9dd5e | 3176 | |
6ad1e29f SC |
3177 | ghc->gpa = gpa; |
3178 | ghc->len = len; | |
3179 | return 0; | |
49c7754c | 3180 | } |
5a2d4365 | 3181 | |
4e335d9e | 3182 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
3183 | gpa_t gpa, unsigned long len) |
3184 | { | |
4e335d9e | 3185 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
3186 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
3187 | } | |
4e335d9e | 3188 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 3189 | |
4e335d9e | 3190 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
3191 | void *data, unsigned int offset, |
3192 | unsigned long len) | |
49c7754c | 3193 | { |
4e335d9e | 3194 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 3195 | int r; |
4ec6e863 | 3196 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 3197 | |
5f25e71e PB |
3198 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3199 | return -EINVAL; | |
8f964525 | 3200 | |
dc9ce71e SC |
3201 | if (slots->generation != ghc->generation) { |
3202 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3203 | return -EFAULT; | |
3204 | } | |
8f964525 | 3205 | |
49c7754c GN |
3206 | if (kvm_is_error_hva(ghc->hva)) |
3207 | return -EFAULT; | |
3208 | ||
fcfbc617 SC |
3209 | if (unlikely(!ghc->memslot)) |
3210 | return kvm_write_guest(kvm, gpa, data, len); | |
3211 | ||
4ec6e863 | 3212 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
3213 | if (r) |
3214 | return -EFAULT; | |
28bd726a | 3215 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
3216 | |
3217 | return 0; | |
3218 | } | |
4e335d9e | 3219 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 3220 | |
4e335d9e PB |
3221 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3222 | void *data, unsigned long len) | |
4ec6e863 | 3223 | { |
4e335d9e | 3224 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 3225 | } |
4e335d9e | 3226 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 3227 | |
0958f0ce VK |
3228 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3229 | void *data, unsigned int offset, | |
3230 | unsigned long len) | |
e03b644f | 3231 | { |
4e335d9e | 3232 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 3233 | int r; |
0958f0ce | 3234 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 3235 | |
5f25e71e PB |
3236 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3237 | return -EINVAL; | |
8f964525 | 3238 | |
dc9ce71e SC |
3239 | if (slots->generation != ghc->generation) { |
3240 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3241 | return -EFAULT; | |
3242 | } | |
8f964525 | 3243 | |
e03b644f GN |
3244 | if (kvm_is_error_hva(ghc->hva)) |
3245 | return -EFAULT; | |
3246 | ||
fcfbc617 | 3247 | if (unlikely(!ghc->memslot)) |
0958f0ce | 3248 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 3249 | |
0958f0ce | 3250 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
3251 | if (r) |
3252 | return -EFAULT; | |
3253 | ||
3254 | return 0; | |
3255 | } | |
0958f0ce VK |
3256 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
3257 | ||
3258 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
3259 | void *data, unsigned long len) | |
3260 | { | |
3261 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
3262 | } | |
4e335d9e | 3263 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 3264 | |
195aefde IE |
3265 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
3266 | { | |
2f541442 | 3267 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
3268 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3269 | int seg; | |
3270 | int offset = offset_in_page(gpa); | |
3271 | int ret; | |
3272 | ||
bfda0e84 | 3273 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 3274 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
3275 | if (ret < 0) |
3276 | return ret; | |
3277 | offset = 0; | |
3278 | len -= seg; | |
3279 | ++gfn; | |
3280 | } | |
3281 | return 0; | |
3282 | } | |
3283 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
3284 | ||
28bd726a | 3285 | void mark_page_dirty_in_slot(struct kvm *kvm, |
8283e36a | 3286 | const struct kvm_memory_slot *memslot, |
28bd726a | 3287 | gfn_t gfn) |
6aa8b732 | 3288 | { |
2efd61a6 DW |
3289 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); |
3290 | ||
e09fccb5 | 3291 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
2efd61a6 DW |
3292 | if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm)) |
3293 | return; | |
e09fccb5 | 3294 | #endif |
2efd61a6 | 3295 | |
044c59c4 | 3296 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 3297 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 3298 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 3299 | |
fb04a1ed | 3300 | if (kvm->dirty_ring_size) |
2efd61a6 | 3301 | kvm_dirty_ring_push(&vcpu->dirty_ring, |
fb04a1ed PX |
3302 | slot, rel_gfn); |
3303 | else | |
3304 | set_bit_le(rel_gfn, memslot->dirty_bitmap); | |
6aa8b732 AK |
3305 | } |
3306 | } | |
a6a0b05d | 3307 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 3308 | |
49c7754c GN |
3309 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
3310 | { | |
3311 | struct kvm_memory_slot *memslot; | |
3312 | ||
3313 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 3314 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 3315 | } |
2ba9f0d8 | 3316 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 3317 | |
8e73485c PB |
3318 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
3319 | { | |
3320 | struct kvm_memory_slot *memslot; | |
3321 | ||
3322 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 3323 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
3324 | } |
3325 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
3326 | ||
20b7035c JS |
3327 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
3328 | { | |
3329 | if (!vcpu->sigset_active) | |
3330 | return; | |
3331 | ||
3332 | /* | |
3333 | * This does a lockless modification of ->real_blocked, which is fine | |
3334 | * because, only current can change ->real_blocked and all readers of | |
3335 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
3336 | * of ->blocked. | |
3337 | */ | |
3338 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
3339 | } | |
3340 | ||
3341 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
3342 | { | |
3343 | if (!vcpu->sigset_active) | |
3344 | return; | |
3345 | ||
3346 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
3347 | sigemptyset(¤t->real_blocked); | |
3348 | } | |
3349 | ||
aca6ff29 WL |
3350 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
3351 | { | |
dee339b5 | 3352 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 3353 | |
2cbd7824 | 3354 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 3355 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3356 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
3357 | if (!grow) |
3358 | goto out; | |
3359 | ||
dee339b5 NW |
3360 | val *= grow; |
3361 | if (val < grow_start) | |
3362 | val = grow_start; | |
aca6ff29 | 3363 | |
258785ef DM |
3364 | if (val > vcpu->kvm->max_halt_poll_ns) |
3365 | val = vcpu->kvm->max_halt_poll_ns; | |
313f636d | 3366 | |
aca6ff29 | 3367 | vcpu->halt_poll_ns = val; |
7fa08e71 | 3368 | out: |
2cbd7824 | 3369 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3370 | } |
3371 | ||
3372 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
3373 | { | |
ae232ea4 | 3374 | unsigned int old, val, shrink, grow_start; |
aca6ff29 | 3375 | |
2cbd7824 | 3376 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 3377 | shrink = READ_ONCE(halt_poll_ns_shrink); |
ae232ea4 | 3378 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3379 | if (shrink == 0) |
aca6ff29 WL |
3380 | val = 0; |
3381 | else | |
6b6de68c | 3382 | val /= shrink; |
aca6ff29 | 3383 | |
ae232ea4 SS |
3384 | if (val < grow_start) |
3385 | val = 0; | |
3386 | ||
aca6ff29 | 3387 | vcpu->halt_poll_ns = val; |
2cbd7824 | 3388 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3389 | } |
3390 | ||
f7819512 PB |
3391 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
3392 | { | |
50c28f21 JS |
3393 | int ret = -EINTR; |
3394 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
3395 | ||
f7819512 PB |
3396 | if (kvm_arch_vcpu_runnable(vcpu)) { |
3397 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 3398 | goto out; |
f7819512 PB |
3399 | } |
3400 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 3401 | goto out; |
f7819512 | 3402 | if (signal_pending(current)) |
50c28f21 | 3403 | goto out; |
084071d5 MT |
3404 | if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu)) |
3405 | goto out; | |
f7819512 | 3406 | |
50c28f21 JS |
3407 | ret = 0; |
3408 | out: | |
3409 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
3410 | return ret; | |
f7819512 PB |
3411 | } |
3412 | ||
fac42688 SC |
3413 | /* |
3414 | * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is | |
3415 | * pending. This is mostly used when halting a vCPU, but may also be used | |
3416 | * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI. | |
3417 | */ | |
3418 | bool kvm_vcpu_block(struct kvm_vcpu *vcpu) | |
cb953129 | 3419 | { |
fac42688 SC |
3420 | struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); |
3421 | bool waited = false; | |
3422 | ||
c3858335 JZ |
3423 | vcpu->stat.generic.blocking = 1; |
3424 | ||
18869f26 | 3425 | preempt_disable(); |
fac42688 | 3426 | kvm_arch_vcpu_blocking(vcpu); |
fac42688 | 3427 | prepare_to_rcuwait(wait); |
18869f26 ML |
3428 | preempt_enable(); |
3429 | ||
fac42688 SC |
3430 | for (;;) { |
3431 | set_current_state(TASK_INTERRUPTIBLE); | |
3432 | ||
3433 | if (kvm_vcpu_check_block(vcpu) < 0) | |
3434 | break; | |
3435 | ||
3436 | waited = true; | |
3437 | schedule(); | |
3438 | } | |
fac42688 | 3439 | |
18869f26 ML |
3440 | preempt_disable(); |
3441 | finish_rcuwait(wait); | |
fac42688 | 3442 | kvm_arch_vcpu_unblocking(vcpu); |
18869f26 | 3443 | preempt_enable(); |
fac42688 | 3444 | |
c3858335 JZ |
3445 | vcpu->stat.generic.blocking = 0; |
3446 | ||
fac42688 SC |
3447 | return waited; |
3448 | } | |
3449 | ||
29e72893 SC |
3450 | static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start, |
3451 | ktime_t end, bool success) | |
cb953129 | 3452 | { |
30c94347 | 3453 | struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic; |
29e72893 SC |
3454 | u64 poll_ns = ktime_to_ns(ktime_sub(end, start)); |
3455 | ||
30c94347 SC |
3456 | ++vcpu->stat.generic.halt_attempted_poll; |
3457 | ||
3458 | if (success) { | |
3459 | ++vcpu->stat.generic.halt_successful_poll; | |
3460 | ||
3461 | if (!vcpu_valid_wakeup(vcpu)) | |
3462 | ++vcpu->stat.generic.halt_poll_invalid; | |
3463 | ||
3464 | stats->halt_poll_success_ns += poll_ns; | |
3465 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns); | |
3466 | } else { | |
3467 | stats->halt_poll_fail_ns += poll_ns; | |
3468 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns); | |
3469 | } | |
cb953129 DM |
3470 | } |
3471 | ||
b6958ce4 | 3472 | /* |
fac42688 SC |
3473 | * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt |
3474 | * polling is enabled, busy wait for a short time before blocking to avoid the | |
3475 | * expensive block+unblock sequence if a wake event arrives soon after the vCPU | |
3476 | * is halted. | |
b6958ce4 | 3477 | */ |
91b99ea7 | 3478 | void kvm_vcpu_halt(struct kvm_vcpu *vcpu) |
d3bef15f | 3479 | { |
6f390916 | 3480 | bool halt_poll_allowed = !kvm_arch_no_poll(vcpu); |
8df6a61c | 3481 | bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns; |
cb953129 | 3482 | ktime_t start, cur, poll_end; |
f7819512 | 3483 | bool waited = false; |
91b99ea7 | 3484 | u64 halt_ns; |
07ab0f8d | 3485 | |
cb953129 | 3486 | start = cur = poll_end = ktime_get(); |
8df6a61c | 3487 | if (do_halt_poll) { |
109a9826 | 3488 | ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns); |
f95ef0cd | 3489 | |
f7819512 PB |
3490 | do { |
3491 | /* | |
3492 | * This sets KVM_REQ_UNHALT if an interrupt | |
3493 | * arrives. | |
3494 | */ | |
30c94347 | 3495 | if (kvm_vcpu_check_block(vcpu) < 0) |
f7819512 | 3496 | goto out; |
74775654 | 3497 | cpu_relax(); |
cb953129 | 3498 | poll_end = cur = ktime_get(); |
6bd5b743 | 3499 | } while (kvm_vcpu_can_poll(cur, stop)); |
f7819512 | 3500 | } |
e5c239cf | 3501 | |
fac42688 | 3502 | waited = kvm_vcpu_block(vcpu); |
8ccba534 | 3503 | |
f7819512 | 3504 | cur = ktime_get(); |
87bcc5fa JZ |
3505 | if (waited) { |
3506 | vcpu->stat.generic.halt_wait_ns += | |
3507 | ktime_to_ns(cur) - ktime_to_ns(poll_end); | |
8ccba534 JZ |
3508 | KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist, |
3509 | ktime_to_ns(cur) - ktime_to_ns(poll_end)); | |
87bcc5fa | 3510 | } |
f7819512 | 3511 | out: |
91b99ea7 SC |
3512 | /* The total time the vCPU was "halted", including polling time. */ |
3513 | halt_ns = ktime_to_ns(cur) - ktime_to_ns(start); | |
aca6ff29 | 3514 | |
29e72893 SC |
3515 | /* |
3516 | * Note, halt-polling is considered successful so long as the vCPU was | |
3517 | * never actually scheduled out, i.e. even if the wake event arrived | |
3518 | * after of the halt-polling loop itself, but before the full wait. | |
3519 | */ | |
8df6a61c | 3520 | if (do_halt_poll) |
29e72893 | 3521 | update_halt_poll_stats(vcpu, start, poll_end, !waited); |
cb953129 | 3522 | |
6f390916 | 3523 | if (halt_poll_allowed) { |
44551b2f | 3524 | if (!vcpu_valid_wakeup(vcpu)) { |
aca6ff29 | 3525 | shrink_halt_poll_ns(vcpu); |
acd05785 | 3526 | } else if (vcpu->kvm->max_halt_poll_ns) { |
91b99ea7 | 3527 | if (halt_ns <= vcpu->halt_poll_ns) |
44551b2f WL |
3528 | ; |
3529 | /* we had a long block, shrink polling */ | |
acd05785 | 3530 | else if (vcpu->halt_poll_ns && |
91b99ea7 | 3531 | halt_ns > vcpu->kvm->max_halt_poll_ns) |
44551b2f WL |
3532 | shrink_halt_poll_ns(vcpu); |
3533 | /* we had a short halt and our poll time is too small */ | |
acd05785 | 3534 | else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && |
91b99ea7 | 3535 | halt_ns < vcpu->kvm->max_halt_poll_ns) |
44551b2f WL |
3536 | grow_halt_poll_ns(vcpu); |
3537 | } else { | |
3538 | vcpu->halt_poll_ns = 0; | |
3539 | } | |
3540 | } | |
aca6ff29 | 3541 | |
91b99ea7 | 3542 | trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu)); |
b6958ce4 | 3543 | } |
91b99ea7 | 3544 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
b6958ce4 | 3545 | |
178f02ff | 3546 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 3547 | { |
d92a5d1c | 3548 | if (__kvm_vcpu_wake_up(vcpu)) { |
d73eb57b | 3549 | WRITE_ONCE(vcpu->ready, true); |
0193cc90 | 3550 | ++vcpu->stat.generic.halt_wakeup; |
178f02ff | 3551 | return true; |
b6d33834 CD |
3552 | } |
3553 | ||
178f02ff | 3554 | return false; |
dd1a4cc1 RK |
3555 | } |
3556 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
3557 | ||
0266c894 | 3558 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
3559 | /* |
3560 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
3561 | */ | |
3562 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
3563 | { | |
85b64045 | 3564 | int me, cpu; |
dd1a4cc1 | 3565 | |
178f02ff RK |
3566 | if (kvm_vcpu_wake_up(vcpu)) |
3567 | return; | |
3568 | ||
aefdc2ed PB |
3569 | me = get_cpu(); |
3570 | /* | |
3571 | * The only state change done outside the vcpu mutex is IN_GUEST_MODE | |
3572 | * to EXITING_GUEST_MODE. Therefore the moderately expensive "should | |
3573 | * kick" check does not need atomic operations if kvm_vcpu_kick is used | |
3574 | * within the vCPU thread itself. | |
3575 | */ | |
3576 | if (vcpu == __this_cpu_read(kvm_running_vcpu)) { | |
3577 | if (vcpu->mode == IN_GUEST_MODE) | |
3578 | WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE); | |
3579 | goto out; | |
3580 | } | |
3581 | ||
85b64045 SC |
3582 | /* |
3583 | * Note, the vCPU could get migrated to a different pCPU at any point | |
3584 | * after kvm_arch_vcpu_should_kick(), which could result in sending an | |
3585 | * IPI to the previous pCPU. But, that's ok because the purpose of the | |
3586 | * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the | |
3587 | * vCPU also requires it to leave IN_GUEST_MODE. | |
3588 | */ | |
85b64045 SC |
3589 | if (kvm_arch_vcpu_should_kick(vcpu)) { |
3590 | cpu = READ_ONCE(vcpu->cpu); | |
3591 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
b6d33834 | 3592 | smp_send_reschedule(cpu); |
85b64045 | 3593 | } |
aefdc2ed | 3594 | out: |
b6d33834 CD |
3595 | put_cpu(); |
3596 | } | |
a20ed54d | 3597 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 3598 | #endif /* !CONFIG_S390 */ |
b6d33834 | 3599 | |
fa93384f | 3600 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
3601 | { |
3602 | struct pid *pid; | |
3603 | struct task_struct *task = NULL; | |
fa93384f | 3604 | int ret = 0; |
41628d33 KW |
3605 | |
3606 | rcu_read_lock(); | |
3607 | pid = rcu_dereference(target->pid); | |
3608 | if (pid) | |
27fbe64b | 3609 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
3610 | rcu_read_unlock(); |
3611 | if (!task) | |
c45c528e | 3612 | return ret; |
c45c528e | 3613 | ret = yield_to(task, 1); |
41628d33 | 3614 | put_task_struct(task); |
c45c528e R |
3615 | |
3616 | return ret; | |
41628d33 KW |
3617 | } |
3618 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
3619 | ||
06e48c51 R |
3620 | /* |
3621 | * Helper that checks whether a VCPU is eligible for directed yield. | |
3622 | * Most eligible candidate to yield is decided by following heuristics: | |
3623 | * | |
3624 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
3625 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 3626 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
3627 | * |
3628 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
3629 | * chance last time (mostly it has become eligible now since we have probably | |
3630 | * yielded to lockholder in last iteration. This is done by toggling | |
3631 | * @dy_eligible each time a VCPU checked for eligibility.) | |
3632 | * | |
3633 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
3634 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
3635 | * burning. Giving priority for a potential lock-holder increases lock | |
3636 | * progress. | |
3637 | * | |
3638 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
3639 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
3640 | * and continue with next VCPU and so on. | |
3641 | */ | |
7940876e | 3642 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 3643 | { |
4a55dd72 | 3644 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
3645 | bool eligible; |
3646 | ||
3647 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 3648 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
3649 | |
3650 | if (vcpu->spin_loop.in_spin_loop) | |
3651 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
3652 | ||
3653 | return eligible; | |
4a55dd72 SW |
3654 | #else |
3655 | return true; | |
06e48c51 | 3656 | #endif |
4a55dd72 | 3657 | } |
c45c528e | 3658 | |
17e433b5 WL |
3659 | /* |
3660 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
3661 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
3662 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
3663 | */ | |
3664 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
3665 | { | |
3666 | return kvm_arch_vcpu_runnable(vcpu); | |
3667 | } | |
3668 | ||
3669 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
3670 | { | |
3671 | if (kvm_arch_dy_runnable(vcpu)) | |
3672 | return true; | |
3673 | ||
3674 | #ifdef CONFIG_KVM_ASYNC_PF | |
3675 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
3676 | return true; | |
3677 | #endif | |
3678 | ||
3679 | return false; | |
3680 | } | |
3681 | ||
52acd22f WL |
3682 | bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
3683 | { | |
3684 | return false; | |
3685 | } | |
3686 | ||
199b5763 | 3687 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 3688 | { |
217ece61 RR |
3689 | struct kvm *kvm = me->kvm; |
3690 | struct kvm_vcpu *vcpu; | |
3691 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
46808a4c | 3692 | unsigned long i; |
217ece61 | 3693 | int yielded = 0; |
c45c528e | 3694 | int try = 3; |
217ece61 | 3695 | int pass; |
d255f4f2 | 3696 | |
4c088493 | 3697 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
3698 | /* |
3699 | * We boost the priority of a VCPU that is runnable but not | |
3700 | * currently running, because it got preempted by something | |
3701 | * else and called schedule in __vcpu_run. Hopefully that | |
3702 | * VCPU is holding the lock that we need and will release it. | |
3703 | * We approximate round-robin by starting at the last boosted VCPU. | |
3704 | */ | |
c45c528e | 3705 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 3706 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 3707 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
3708 | i = last_boosted_vcpu; |
3709 | continue; | |
3710 | } else if (pass && i > last_boosted_vcpu) | |
3711 | break; | |
d73eb57b | 3712 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 3713 | continue; |
217ece61 RR |
3714 | if (vcpu == me) |
3715 | continue; | |
d92a5d1c | 3716 | if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu)) |
217ece61 | 3717 | continue; |
046ddeed | 3718 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
52acd22f WL |
3719 | !kvm_arch_dy_has_pending_interrupt(vcpu) && |
3720 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 3721 | continue; |
06e48c51 R |
3722 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
3723 | continue; | |
c45c528e R |
3724 | |
3725 | yielded = kvm_vcpu_yield_to(vcpu); | |
3726 | if (yielded > 0) { | |
217ece61 | 3727 | kvm->last_boosted_vcpu = i; |
217ece61 | 3728 | break; |
c45c528e R |
3729 | } else if (yielded < 0) { |
3730 | try--; | |
3731 | if (!try) | |
3732 | break; | |
217ece61 | 3733 | } |
217ece61 RR |
3734 | } |
3735 | } | |
4c088493 | 3736 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
3737 | |
3738 | /* Ensure vcpu is not eligible during next spinloop */ | |
3739 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
3740 | } |
3741 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
3742 | ||
fb04a1ed PX |
3743 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
3744 | { | |
dc70ec21 | 3745 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
fb04a1ed PX |
3746 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && |
3747 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
3748 | kvm->dirty_ring_size / PAGE_SIZE); | |
3749 | #else | |
3750 | return false; | |
3751 | #endif | |
3752 | } | |
3753 | ||
1499fa80 | 3754 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 3755 | { |
11bac800 | 3756 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
3757 | struct page *page; |
3758 | ||
e4a533a4 | 3759 | if (vmf->pgoff == 0) |
039576c0 | 3760 | page = virt_to_page(vcpu->run); |
09566765 | 3761 | #ifdef CONFIG_X86 |
e4a533a4 | 3762 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 3763 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 3764 | #endif |
4b4357e0 | 3765 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3766 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
3767 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 3768 | #endif |
fb04a1ed PX |
3769 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
3770 | page = kvm_dirty_ring_get_page( | |
3771 | &vcpu->dirty_ring, | |
3772 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 3773 | else |
5b1c1493 | 3774 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 3775 | get_page(page); |
e4a533a4 NP |
3776 | vmf->page = page; |
3777 | return 0; | |
9a2bb7f4 AK |
3778 | } |
3779 | ||
f0f37e2f | 3780 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 3781 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
3782 | }; |
3783 | ||
3784 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
3785 | { | |
fb04a1ed | 3786 | struct kvm_vcpu *vcpu = file->private_data; |
11476d27 | 3787 | unsigned long pages = vma_pages(vma); |
fb04a1ed PX |
3788 | |
3789 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
3790 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
3791 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
3792 | return -EINVAL; | |
3793 | ||
9a2bb7f4 AK |
3794 | vma->vm_ops = &kvm_vcpu_vm_ops; |
3795 | return 0; | |
3796 | } | |
3797 | ||
bccf2150 AK |
3798 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
3799 | { | |
3800 | struct kvm_vcpu *vcpu = filp->private_data; | |
3801 | ||
66c0b394 | 3802 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
3803 | return 0; |
3804 | } | |
3805 | ||
70375c2d | 3806 | static const struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
3807 | .release = kvm_vcpu_release, |
3808 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 3809 | .mmap = kvm_vcpu_mmap, |
6038f373 | 3810 | .llseek = noop_llseek, |
7ddfd3e0 | 3811 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
3812 | }; |
3813 | ||
3814 | /* | |
3815 | * Allocates an inode for the vcpu. | |
3816 | */ | |
3817 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
3818 | { | |
e46b4692 MY |
3819 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
3820 | ||
3821 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
3822 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
3823 | } |
3824 | ||
3e7093d0 | 3825 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 3826 | { |
741cbbae | 3827 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
d56f5136 | 3828 | struct dentry *debugfs_dentry; |
45b5939e | 3829 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 3830 | |
45b5939e | 3831 | if (!debugfs_initialized()) |
3e7093d0 | 3832 | return; |
45b5939e LC |
3833 | |
3834 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
3835 | debugfs_dentry = debugfs_create_dir(dir_name, |
3836 | vcpu->kvm->debugfs_dentry); | |
45b5939e | 3837 | |
d56f5136 | 3838 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
741cbbae | 3839 | #endif |
45b5939e LC |
3840 | } |
3841 | ||
c5ea7660 AK |
3842 | /* |
3843 | * Creates some virtual cpus. Good luck creating more than one. | |
3844 | */ | |
73880c80 | 3845 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
3846 | { |
3847 | int r; | |
e09fefde | 3848 | struct kvm_vcpu *vcpu; |
8bd826d6 | 3849 | struct page *page; |
c5ea7660 | 3850 | |
a1c42dde | 3851 | if (id >= KVM_MAX_VCPU_IDS) |
338c7dba AH |
3852 | return -EINVAL; |
3853 | ||
6c7caebc | 3854 | mutex_lock(&kvm->lock); |
f502cc56 | 3855 | if (kvm->created_vcpus >= kvm->max_vcpus) { |
6c7caebc PB |
3856 | mutex_unlock(&kvm->lock); |
3857 | return -EINVAL; | |
3858 | } | |
3859 | ||
1d5e740d ZG |
3860 | r = kvm_arch_vcpu_precreate(kvm, id); |
3861 | if (r) { | |
3862 | mutex_unlock(&kvm->lock); | |
3863 | return r; | |
3864 | } | |
3865 | ||
6c7caebc PB |
3866 | kvm->created_vcpus++; |
3867 | mutex_unlock(&kvm->lock); | |
3868 | ||
85f47930 | 3869 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
e529ef66 SC |
3870 | if (!vcpu) { |
3871 | r = -ENOMEM; | |
6c7caebc PB |
3872 | goto vcpu_decrement; |
3873 | } | |
c5ea7660 | 3874 | |
fcd97ad5 | 3875 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 3876 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
3877 | if (!page) { |
3878 | r = -ENOMEM; | |
e529ef66 | 3879 | goto vcpu_free; |
8bd826d6 SC |
3880 | } |
3881 | vcpu->run = page_address(page); | |
3882 | ||
3883 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
3884 | |
3885 | r = kvm_arch_vcpu_create(vcpu); | |
3886 | if (r) | |
8bd826d6 | 3887 | goto vcpu_free_run_page; |
e529ef66 | 3888 | |
fb04a1ed PX |
3889 | if (kvm->dirty_ring_size) { |
3890 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
3891 | id, kvm->dirty_ring_size); | |
3892 | if (r) | |
3893 | goto arch_vcpu_destroy; | |
3894 | } | |
3895 | ||
11ec2804 | 3896 | mutex_lock(&kvm->lock); |
e09fefde DH |
3897 | if (kvm_get_vcpu_by_id(kvm, id)) { |
3898 | r = -EEXIST; | |
3899 | goto unlock_vcpu_destroy; | |
3900 | } | |
73880c80 | 3901 | |
8750e72a | 3902 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
c5b07754 MZ |
3903 | r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT); |
3904 | BUG_ON(r == -EBUSY); | |
3905 | if (r) | |
3906 | goto unlock_vcpu_destroy; | |
c5ea7660 | 3907 | |
ce55c049 JZ |
3908 | /* Fill the stats id string for the vcpu */ |
3909 | snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d", | |
3910 | task_pid_nr(current), id); | |
3911 | ||
fb3f0f51 | 3912 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 3913 | kvm_get_kvm(kvm); |
bccf2150 | 3914 | r = create_vcpu_fd(vcpu); |
73880c80 | 3915 | if (r < 0) { |
c5b07754 | 3916 | xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx); |
149487bd | 3917 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 3918 | goto unlock_vcpu_destroy; |
73880c80 GN |
3919 | } |
3920 | ||
dd489240 | 3921 | /* |
c5b07754 MZ |
3922 | * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu |
3923 | * pointer before kvm->online_vcpu's incremented value. | |
dd489240 | 3924 | */ |
73880c80 GN |
3925 | smp_wmb(); |
3926 | atomic_inc(&kvm->online_vcpus); | |
3927 | ||
73880c80 | 3928 | mutex_unlock(&kvm->lock); |
42897d86 | 3929 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 3930 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 3931 | return r; |
39c3b86e | 3932 | |
d780592b | 3933 | unlock_vcpu_destroy: |
7d8fece6 | 3934 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
3935 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
3936 | arch_vcpu_destroy: | |
d40ccc62 | 3937 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
3938 | vcpu_free_run_page: |
3939 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
3940 | vcpu_free: |
3941 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
3942 | vcpu_decrement: |
3943 | mutex_lock(&kvm->lock); | |
3944 | kvm->created_vcpus--; | |
3945 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
3946 | return r; |
3947 | } | |
3948 | ||
1961d276 AK |
3949 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
3950 | { | |
3951 | if (sigset) { | |
3952 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
3953 | vcpu->sigset_active = 1; | |
3954 | vcpu->sigset = *sigset; | |
3955 | } else | |
3956 | vcpu->sigset_active = 0; | |
3957 | return 0; | |
3958 | } | |
3959 | ||
ce55c049 JZ |
3960 | static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer, |
3961 | size_t size, loff_t *offset) | |
3962 | { | |
3963 | struct kvm_vcpu *vcpu = file->private_data; | |
3964 | ||
3965 | return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header, | |
3966 | &kvm_vcpu_stats_desc[0], &vcpu->stat, | |
3967 | sizeof(vcpu->stat), user_buffer, size, offset); | |
3968 | } | |
3969 | ||
3970 | static const struct file_operations kvm_vcpu_stats_fops = { | |
3971 | .read = kvm_vcpu_stats_read, | |
3972 | .llseek = noop_llseek, | |
3973 | }; | |
3974 | ||
3975 | static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu) | |
3976 | { | |
3977 | int fd; | |
3978 | struct file *file; | |
3979 | char name[15 + ITOA_MAX_LEN + 1]; | |
3980 | ||
3981 | snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id); | |
3982 | ||
3983 | fd = get_unused_fd_flags(O_CLOEXEC); | |
3984 | if (fd < 0) | |
3985 | return fd; | |
3986 | ||
3987 | file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY); | |
3988 | if (IS_ERR(file)) { | |
3989 | put_unused_fd(fd); | |
3990 | return PTR_ERR(file); | |
3991 | } | |
3992 | file->f_mode |= FMODE_PREAD; | |
3993 | fd_install(fd, file); | |
3994 | ||
3995 | return fd; | |
3996 | } | |
3997 | ||
bccf2150 AK |
3998 | static long kvm_vcpu_ioctl(struct file *filp, |
3999 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 4000 | { |
bccf2150 | 4001 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 4002 | void __user *argp = (void __user *)arg; |
313a3dc7 | 4003 | int r; |
fa3795a7 DH |
4004 | struct kvm_fpu *fpu = NULL; |
4005 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 4006 | |
f4d31653 | 4007 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
6d4e4c4f | 4008 | return -EIO; |
2122ff5e | 4009 | |
2ea75be3 DM |
4010 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
4011 | return -EINVAL; | |
4012 | ||
2122ff5e | 4013 | /* |
5cb0944c PB |
4014 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
4015 | * execution; mutex_lock() would break them. | |
2122ff5e | 4016 | */ |
5cb0944c PB |
4017 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
4018 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 4019 | return r; |
2122ff5e | 4020 | |
ec7660cc CD |
4021 | if (mutex_lock_killable(&vcpu->mutex)) |
4022 | return -EINTR; | |
6aa8b732 | 4023 | switch (ioctl) { |
0e4524a5 CB |
4024 | case KVM_RUN: { |
4025 | struct pid *oldpid; | |
f0fe5108 AK |
4026 | r = -EINVAL; |
4027 | if (arg) | |
4028 | goto out; | |
0e4524a5 | 4029 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 4030 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 4031 | /* The thread running this VCPU changed. */ |
bd2a6394 | 4032 | struct pid *newpid; |
f95ef0cd | 4033 | |
bd2a6394 CD |
4034 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
4035 | if (r) | |
4036 | break; | |
4037 | ||
4038 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
4039 | rcu_assign_pointer(vcpu->pid, newpid); |
4040 | if (oldpid) | |
4041 | synchronize_rcu(); | |
4042 | put_pid(oldpid); | |
4043 | } | |
1b94f6f8 | 4044 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 4045 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 4046 | break; |
0e4524a5 | 4047 | } |
6aa8b732 | 4048 | case KVM_GET_REGS: { |
3e4bb3ac | 4049 | struct kvm_regs *kvm_regs; |
6aa8b732 | 4050 | |
3e4bb3ac | 4051 | r = -ENOMEM; |
b12ce36a | 4052 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 4053 | if (!kvm_regs) |
6aa8b732 | 4054 | goto out; |
3e4bb3ac XZ |
4055 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
4056 | if (r) | |
4057 | goto out_free1; | |
6aa8b732 | 4058 | r = -EFAULT; |
3e4bb3ac XZ |
4059 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
4060 | goto out_free1; | |
6aa8b732 | 4061 | r = 0; |
3e4bb3ac XZ |
4062 | out_free1: |
4063 | kfree(kvm_regs); | |
6aa8b732 AK |
4064 | break; |
4065 | } | |
4066 | case KVM_SET_REGS: { | |
3e4bb3ac | 4067 | struct kvm_regs *kvm_regs; |
6aa8b732 | 4068 | |
ff5c2c03 SL |
4069 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
4070 | if (IS_ERR(kvm_regs)) { | |
4071 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 4072 | goto out; |
ff5c2c03 | 4073 | } |
3e4bb3ac | 4074 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 4075 | kfree(kvm_regs); |
6aa8b732 AK |
4076 | break; |
4077 | } | |
4078 | case KVM_GET_SREGS: { | |
b12ce36a BG |
4079 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
4080 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
4081 | r = -ENOMEM; |
4082 | if (!kvm_sregs) | |
4083 | goto out; | |
4084 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
4085 | if (r) |
4086 | goto out; | |
4087 | r = -EFAULT; | |
fa3795a7 | 4088 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
4089 | goto out; |
4090 | r = 0; | |
4091 | break; | |
4092 | } | |
4093 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
4094 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
4095 | if (IS_ERR(kvm_sregs)) { | |
4096 | r = PTR_ERR(kvm_sregs); | |
18595411 | 4097 | kvm_sregs = NULL; |
6aa8b732 | 4098 | goto out; |
ff5c2c03 | 4099 | } |
fa3795a7 | 4100 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
4101 | break; |
4102 | } | |
62d9f0db MT |
4103 | case KVM_GET_MP_STATE: { |
4104 | struct kvm_mp_state mp_state; | |
4105 | ||
4106 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
4107 | if (r) | |
4108 | goto out; | |
4109 | r = -EFAULT; | |
893bdbf1 | 4110 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
4111 | goto out; |
4112 | r = 0; | |
4113 | break; | |
4114 | } | |
4115 | case KVM_SET_MP_STATE: { | |
4116 | struct kvm_mp_state mp_state; | |
4117 | ||
4118 | r = -EFAULT; | |
893bdbf1 | 4119 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
4120 | goto out; |
4121 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
4122 | break; |
4123 | } | |
6aa8b732 AK |
4124 | case KVM_TRANSLATE: { |
4125 | struct kvm_translation tr; | |
4126 | ||
4127 | r = -EFAULT; | |
893bdbf1 | 4128 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 4129 | goto out; |
8b006791 | 4130 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
4131 | if (r) |
4132 | goto out; | |
4133 | r = -EFAULT; | |
893bdbf1 | 4134 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
4135 | goto out; |
4136 | r = 0; | |
4137 | break; | |
4138 | } | |
d0bfb940 JK |
4139 | case KVM_SET_GUEST_DEBUG: { |
4140 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
4141 | |
4142 | r = -EFAULT; | |
893bdbf1 | 4143 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 4144 | goto out; |
d0bfb940 | 4145 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
4146 | break; |
4147 | } | |
1961d276 AK |
4148 | case KVM_SET_SIGNAL_MASK: { |
4149 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4150 | struct kvm_signal_mask kvm_sigmask; | |
4151 | sigset_t sigset, *p; | |
4152 | ||
4153 | p = NULL; | |
4154 | if (argp) { | |
4155 | r = -EFAULT; | |
4156 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4157 | sizeof(kvm_sigmask))) |
1961d276 AK |
4158 | goto out; |
4159 | r = -EINVAL; | |
893bdbf1 | 4160 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
4161 | goto out; |
4162 | r = -EFAULT; | |
4163 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 4164 | sizeof(sigset))) |
1961d276 AK |
4165 | goto out; |
4166 | p = &sigset; | |
4167 | } | |
376d41ff | 4168 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
4169 | break; |
4170 | } | |
b8836737 | 4171 | case KVM_GET_FPU: { |
b12ce36a | 4172 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
4173 | r = -ENOMEM; |
4174 | if (!fpu) | |
4175 | goto out; | |
4176 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
4177 | if (r) |
4178 | goto out; | |
4179 | r = -EFAULT; | |
fa3795a7 | 4180 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
4181 | goto out; |
4182 | r = 0; | |
4183 | break; | |
4184 | } | |
4185 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
4186 | fpu = memdup_user(argp, sizeof(*fpu)); |
4187 | if (IS_ERR(fpu)) { | |
4188 | r = PTR_ERR(fpu); | |
18595411 | 4189 | fpu = NULL; |
b8836737 | 4190 | goto out; |
ff5c2c03 | 4191 | } |
fa3795a7 | 4192 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
4193 | break; |
4194 | } | |
ce55c049 JZ |
4195 | case KVM_GET_STATS_FD: { |
4196 | r = kvm_vcpu_ioctl_get_stats_fd(vcpu); | |
4197 | break; | |
4198 | } | |
bccf2150 | 4199 | default: |
313a3dc7 | 4200 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
4201 | } |
4202 | out: | |
ec7660cc | 4203 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
4204 | kfree(fpu); |
4205 | kfree(kvm_sregs); | |
bccf2150 AK |
4206 | return r; |
4207 | } | |
4208 | ||
de8e5d74 | 4209 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
4210 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
4211 | unsigned int ioctl, unsigned long arg) | |
4212 | { | |
4213 | struct kvm_vcpu *vcpu = filp->private_data; | |
4214 | void __user *argp = compat_ptr(arg); | |
4215 | int r; | |
4216 | ||
f4d31653 | 4217 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
1dda606c AG |
4218 | return -EIO; |
4219 | ||
4220 | switch (ioctl) { | |
4221 | case KVM_SET_SIGNAL_MASK: { | |
4222 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4223 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
4224 | sigset_t sigset; |
4225 | ||
4226 | if (argp) { | |
4227 | r = -EFAULT; | |
4228 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4229 | sizeof(kvm_sigmask))) |
1dda606c AG |
4230 | goto out; |
4231 | r = -EINVAL; | |
3968cf62 | 4232 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
4233 | goto out; |
4234 | r = -EFAULT; | |
1393b4aa PB |
4235 | if (get_compat_sigset(&sigset, |
4236 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 4237 | goto out; |
760a9a30 AC |
4238 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
4239 | } else | |
4240 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
4241 | break; |
4242 | } | |
4243 | default: | |
4244 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
4245 | } | |
4246 | ||
4247 | out: | |
4248 | return r; | |
4249 | } | |
4250 | #endif | |
4251 | ||
a1cd3f08 CLG |
4252 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
4253 | { | |
4254 | struct kvm_device *dev = filp->private_data; | |
4255 | ||
4256 | if (dev->ops->mmap) | |
4257 | return dev->ops->mmap(dev, vma); | |
4258 | ||
4259 | return -ENODEV; | |
4260 | } | |
4261 | ||
852b6d57 SW |
4262 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
4263 | int (*accessor)(struct kvm_device *dev, | |
4264 | struct kvm_device_attr *attr), | |
4265 | unsigned long arg) | |
4266 | { | |
4267 | struct kvm_device_attr attr; | |
4268 | ||
4269 | if (!accessor) | |
4270 | return -EPERM; | |
4271 | ||
4272 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
4273 | return -EFAULT; | |
4274 | ||
4275 | return accessor(dev, &attr); | |
4276 | } | |
4277 | ||
4278 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
4279 | unsigned long arg) | |
4280 | { | |
4281 | struct kvm_device *dev = filp->private_data; | |
4282 | ||
f4d31653 | 4283 | if (dev->kvm->mm != current->mm || dev->kvm->vm_dead) |
ddba9180 SC |
4284 | return -EIO; |
4285 | ||
852b6d57 SW |
4286 | switch (ioctl) { |
4287 | case KVM_SET_DEVICE_ATTR: | |
4288 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
4289 | case KVM_GET_DEVICE_ATTR: | |
4290 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
4291 | case KVM_HAS_DEVICE_ATTR: | |
4292 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
4293 | default: | |
4294 | if (dev->ops->ioctl) | |
4295 | return dev->ops->ioctl(dev, ioctl, arg); | |
4296 | ||
4297 | return -ENOTTY; | |
4298 | } | |
4299 | } | |
4300 | ||
852b6d57 SW |
4301 | static int kvm_device_release(struct inode *inode, struct file *filp) |
4302 | { | |
4303 | struct kvm_device *dev = filp->private_data; | |
4304 | struct kvm *kvm = dev->kvm; | |
4305 | ||
2bde9b3e CLG |
4306 | if (dev->ops->release) { |
4307 | mutex_lock(&kvm->lock); | |
4308 | list_del(&dev->vm_node); | |
4309 | dev->ops->release(dev); | |
4310 | mutex_unlock(&kvm->lock); | |
4311 | } | |
4312 | ||
852b6d57 SW |
4313 | kvm_put_kvm(kvm); |
4314 | return 0; | |
4315 | } | |
4316 | ||
4317 | static const struct file_operations kvm_device_fops = { | |
4318 | .unlocked_ioctl = kvm_device_ioctl, | |
4319 | .release = kvm_device_release, | |
7ddfd3e0 | 4320 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 4321 | .mmap = kvm_device_mmap, |
852b6d57 SW |
4322 | }; |
4323 | ||
4324 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
4325 | { | |
4326 | if (filp->f_op != &kvm_device_fops) | |
4327 | return NULL; | |
4328 | ||
4329 | return filp->private_data; | |
4330 | } | |
4331 | ||
8538cb22 | 4332 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 4333 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
4334 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
4335 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 4336 | #endif |
d60eacb0 WD |
4337 | }; |
4338 | ||
8538cb22 | 4339 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
4340 | { |
4341 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
4342 | return -ENOSPC; | |
4343 | ||
4344 | if (kvm_device_ops_table[type] != NULL) | |
4345 | return -EEXIST; | |
4346 | ||
4347 | kvm_device_ops_table[type] = ops; | |
4348 | return 0; | |
4349 | } | |
4350 | ||
571ee1b6 WL |
4351 | void kvm_unregister_device_ops(u32 type) |
4352 | { | |
4353 | if (kvm_device_ops_table[type] != NULL) | |
4354 | kvm_device_ops_table[type] = NULL; | |
4355 | } | |
4356 | ||
852b6d57 SW |
4357 | static int kvm_ioctl_create_device(struct kvm *kvm, |
4358 | struct kvm_create_device *cd) | |
4359 | { | |
8538cb22 | 4360 | const struct kvm_device_ops *ops = NULL; |
852b6d57 SW |
4361 | struct kvm_device *dev; |
4362 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 4363 | int type; |
852b6d57 SW |
4364 | int ret; |
4365 | ||
d60eacb0 WD |
4366 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
4367 | return -ENODEV; | |
4368 | ||
1d487e9b PB |
4369 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
4370 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 4371 | if (ops == NULL) |
852b6d57 | 4372 | return -ENODEV; |
852b6d57 SW |
4373 | |
4374 | if (test) | |
4375 | return 0; | |
4376 | ||
b12ce36a | 4377 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
4378 | if (!dev) |
4379 | return -ENOMEM; | |
4380 | ||
4381 | dev->ops = ops; | |
4382 | dev->kvm = kvm; | |
852b6d57 | 4383 | |
a28ebea2 | 4384 | mutex_lock(&kvm->lock); |
1d487e9b | 4385 | ret = ops->create(dev, type); |
852b6d57 | 4386 | if (ret < 0) { |
a28ebea2 | 4387 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
4388 | kfree(dev); |
4389 | return ret; | |
4390 | } | |
a28ebea2 CD |
4391 | list_add(&dev->vm_node, &kvm->devices); |
4392 | mutex_unlock(&kvm->lock); | |
852b6d57 | 4393 | |
023e9fdd CD |
4394 | if (ops->init) |
4395 | ops->init(dev); | |
4396 | ||
cfa39381 | 4397 | kvm_get_kvm(kvm); |
24009b05 | 4398 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 4399 | if (ret < 0) { |
149487bd | 4400 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
4401 | mutex_lock(&kvm->lock); |
4402 | list_del(&dev->vm_node); | |
e8bc2427 AK |
4403 | if (ops->release) |
4404 | ops->release(dev); | |
a28ebea2 | 4405 | mutex_unlock(&kvm->lock); |
e8bc2427 AK |
4406 | if (ops->destroy) |
4407 | ops->destroy(dev); | |
852b6d57 SW |
4408 | return ret; |
4409 | } | |
4410 | ||
852b6d57 SW |
4411 | cd->fd = ret; |
4412 | return 0; | |
4413 | } | |
4414 | ||
92b591a4 AG |
4415 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
4416 | { | |
4417 | switch (arg) { | |
4418 | case KVM_CAP_USER_MEMORY: | |
4419 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
4420 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
4421 | case KVM_CAP_INTERNAL_ERROR_DATA: |
4422 | #ifdef CONFIG_HAVE_KVM_MSI | |
4423 | case KVM_CAP_SIGNAL_MSI: | |
4424 | #endif | |
297e2105 | 4425 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 4426 | case KVM_CAP_IRQFD: |
92b591a4 AG |
4427 | case KVM_CAP_IRQFD_RESAMPLE: |
4428 | #endif | |
e9ea5069 | 4429 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 4430 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 4431 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 4432 | case KVM_CAP_HALT_POLL: |
92b591a4 | 4433 | return 1; |
4b4357e0 | 4434 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
4435 | case KVM_CAP_COALESCED_MMIO: |
4436 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
4437 | case KVM_CAP_COALESCED_PIO: |
4438 | return 1; | |
30422558 | 4439 | #endif |
3c9bd400 JZ |
4440 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4441 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
4442 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
4443 | #endif | |
92b591a4 AG |
4444 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4445 | case KVM_CAP_IRQ_ROUTING: | |
4446 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
4447 | #endif |
4448 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
4449 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
4450 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 4451 | #endif |
c110ae57 PB |
4452 | case KVM_CAP_NR_MEMSLOTS: |
4453 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed | 4454 | case KVM_CAP_DIRTY_LOG_RING: |
dc70ec21 | 4455 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
fb04a1ed PX |
4456 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); |
4457 | #else | |
4458 | return 0; | |
4459 | #endif | |
ce55c049 | 4460 | case KVM_CAP_BINARY_STATS_FD: |
d495f942 | 4461 | case KVM_CAP_SYSTEM_EVENT_DATA: |
ce55c049 | 4462 | return 1; |
92b591a4 AG |
4463 | default: |
4464 | break; | |
4465 | } | |
4466 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
4467 | } | |
4468 | ||
fb04a1ed PX |
4469 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
4470 | { | |
4471 | int r; | |
4472 | ||
4473 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
4474 | return -EINVAL; | |
4475 | ||
4476 | /* the size should be power of 2 */ | |
4477 | if (!size || (size & (size - 1))) | |
4478 | return -EINVAL; | |
4479 | ||
4480 | /* Should be bigger to keep the reserved entries, or a page */ | |
4481 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
4482 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
4483 | return -EINVAL; | |
4484 | ||
4485 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
4486 | sizeof(struct kvm_dirty_gfn)) | |
4487 | return -E2BIG; | |
4488 | ||
4489 | /* We only allow it to set once */ | |
4490 | if (kvm->dirty_ring_size) | |
4491 | return -EINVAL; | |
4492 | ||
4493 | mutex_lock(&kvm->lock); | |
4494 | ||
4495 | if (kvm->created_vcpus) { | |
4496 | /* We don't allow to change this value after vcpu created */ | |
4497 | r = -EINVAL; | |
4498 | } else { | |
4499 | kvm->dirty_ring_size = size; | |
4500 | r = 0; | |
4501 | } | |
4502 | ||
4503 | mutex_unlock(&kvm->lock); | |
4504 | return r; | |
4505 | } | |
4506 | ||
4507 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
4508 | { | |
46808a4c | 4509 | unsigned long i; |
fb04a1ed PX |
4510 | struct kvm_vcpu *vcpu; |
4511 | int cleared = 0; | |
4512 | ||
4513 | if (!kvm->dirty_ring_size) | |
4514 | return -EINVAL; | |
4515 | ||
4516 | mutex_lock(&kvm->slots_lock); | |
4517 | ||
4518 | kvm_for_each_vcpu(i, vcpu, kvm) | |
4519 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
4520 | ||
4521 | mutex_unlock(&kvm->slots_lock); | |
4522 | ||
4523 | if (cleared) | |
4524 | kvm_flush_remote_tlbs(kvm); | |
4525 | ||
4526 | return cleared; | |
4527 | } | |
4528 | ||
e5d83c74 PB |
4529 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4530 | struct kvm_enable_cap *cap) | |
4531 | { | |
4532 | return -EINVAL; | |
4533 | } | |
4534 | ||
4535 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
4536 | struct kvm_enable_cap *cap) | |
4537 | { | |
4538 | switch (cap->cap) { | |
2a31b9db | 4539 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
4540 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
4541 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
4542 | ||
4543 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
4544 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
4545 | ||
4546 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
4547 | return -EINVAL; |
4548 | kvm->manual_dirty_log_protect = cap->args[0]; | |
4549 | return 0; | |
3c9bd400 | 4550 | } |
2a31b9db | 4551 | #endif |
acd05785 DM |
4552 | case KVM_CAP_HALT_POLL: { |
4553 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
4554 | return -EINVAL; | |
4555 | ||
4556 | kvm->max_halt_poll_ns = cap->args[0]; | |
4557 | return 0; | |
4558 | } | |
fb04a1ed PX |
4559 | case KVM_CAP_DIRTY_LOG_RING: |
4560 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); | |
e5d83c74 PB |
4561 | default: |
4562 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
4563 | } | |
4564 | } | |
4565 | ||
fcfe1bae JZ |
4566 | static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer, |
4567 | size_t size, loff_t *offset) | |
4568 | { | |
4569 | struct kvm *kvm = file->private_data; | |
4570 | ||
4571 | return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header, | |
4572 | &kvm_vm_stats_desc[0], &kvm->stat, | |
4573 | sizeof(kvm->stat), user_buffer, size, offset); | |
4574 | } | |
4575 | ||
4576 | static const struct file_operations kvm_vm_stats_fops = { | |
4577 | .read = kvm_vm_stats_read, | |
4578 | .llseek = noop_llseek, | |
4579 | }; | |
4580 | ||
4581 | static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm) | |
4582 | { | |
4583 | int fd; | |
4584 | struct file *file; | |
4585 | ||
4586 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4587 | if (fd < 0) | |
4588 | return fd; | |
4589 | ||
4590 | file = anon_inode_getfile("kvm-vm-stats", | |
4591 | &kvm_vm_stats_fops, kvm, O_RDONLY); | |
4592 | if (IS_ERR(file)) { | |
4593 | put_unused_fd(fd); | |
4594 | return PTR_ERR(file); | |
4595 | } | |
4596 | file->f_mode |= FMODE_PREAD; | |
4597 | fd_install(fd, file); | |
4598 | ||
4599 | return fd; | |
4600 | } | |
4601 | ||
bccf2150 AK |
4602 | static long kvm_vm_ioctl(struct file *filp, |
4603 | unsigned int ioctl, unsigned long arg) | |
4604 | { | |
4605 | struct kvm *kvm = filp->private_data; | |
4606 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 4607 | int r; |
bccf2150 | 4608 | |
f4d31653 | 4609 | if (kvm->mm != current->mm || kvm->vm_dead) |
6d4e4c4f | 4610 | return -EIO; |
bccf2150 AK |
4611 | switch (ioctl) { |
4612 | case KVM_CREATE_VCPU: | |
4613 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 4614 | break; |
e5d83c74 PB |
4615 | case KVM_ENABLE_CAP: { |
4616 | struct kvm_enable_cap cap; | |
4617 | ||
4618 | r = -EFAULT; | |
4619 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4620 | goto out; | |
4621 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
4622 | break; | |
4623 | } | |
6fc138d2 IE |
4624 | case KVM_SET_USER_MEMORY_REGION: { |
4625 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
4626 | ||
4627 | r = -EFAULT; | |
4628 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 4629 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
4630 | goto out; |
4631 | ||
47ae31e2 | 4632 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
4633 | break; |
4634 | } | |
4635 | case KVM_GET_DIRTY_LOG: { | |
4636 | struct kvm_dirty_log log; | |
4637 | ||
4638 | r = -EFAULT; | |
893bdbf1 | 4639 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 4640 | goto out; |
2c6f5df9 | 4641 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
4642 | break; |
4643 | } | |
2a31b9db PB |
4644 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4645 | case KVM_CLEAR_DIRTY_LOG: { | |
4646 | struct kvm_clear_dirty_log log; | |
4647 | ||
4648 | r = -EFAULT; | |
4649 | if (copy_from_user(&log, argp, sizeof(log))) | |
4650 | goto out; | |
4651 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4652 | break; | |
4653 | } | |
4654 | #endif | |
4b4357e0 | 4655 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
4656 | case KVM_REGISTER_COALESCED_MMIO: { |
4657 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4658 | |
5f94c174 | 4659 | r = -EFAULT; |
893bdbf1 | 4660 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4661 | goto out; |
5f94c174 | 4662 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4663 | break; |
4664 | } | |
4665 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
4666 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4667 | |
5f94c174 | 4668 | r = -EFAULT; |
893bdbf1 | 4669 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4670 | goto out; |
5f94c174 | 4671 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4672 | break; |
4673 | } | |
4674 | #endif | |
721eecbf GH |
4675 | case KVM_IRQFD: { |
4676 | struct kvm_irqfd data; | |
4677 | ||
4678 | r = -EFAULT; | |
893bdbf1 | 4679 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 4680 | goto out; |
d4db2935 | 4681 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
4682 | break; |
4683 | } | |
d34e6b17 GH |
4684 | case KVM_IOEVENTFD: { |
4685 | struct kvm_ioeventfd data; | |
4686 | ||
4687 | r = -EFAULT; | |
893bdbf1 | 4688 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
4689 | goto out; |
4690 | r = kvm_ioeventfd(kvm, &data); | |
4691 | break; | |
4692 | } | |
07975ad3 JK |
4693 | #ifdef CONFIG_HAVE_KVM_MSI |
4694 | case KVM_SIGNAL_MSI: { | |
4695 | struct kvm_msi msi; | |
4696 | ||
4697 | r = -EFAULT; | |
893bdbf1 | 4698 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
4699 | goto out; |
4700 | r = kvm_send_userspace_msi(kvm, &msi); | |
4701 | break; | |
4702 | } | |
23d43cf9 CD |
4703 | #endif |
4704 | #ifdef __KVM_HAVE_IRQ_LINE | |
4705 | case KVM_IRQ_LINE_STATUS: | |
4706 | case KVM_IRQ_LINE: { | |
4707 | struct kvm_irq_level irq_event; | |
4708 | ||
4709 | r = -EFAULT; | |
893bdbf1 | 4710 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
4711 | goto out; |
4712 | ||
aa2fbe6d YZ |
4713 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
4714 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
4715 | if (r) |
4716 | goto out; | |
4717 | ||
4718 | r = -EFAULT; | |
4719 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 4720 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
4721 | goto out; |
4722 | } | |
4723 | ||
4724 | r = 0; | |
4725 | break; | |
4726 | } | |
73880c80 | 4727 | #endif |
aa8d5944 AG |
4728 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4729 | case KVM_SET_GSI_ROUTING: { | |
4730 | struct kvm_irq_routing routing; | |
4731 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 4732 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
4733 | |
4734 | r = -EFAULT; | |
4735 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
4736 | goto out; | |
4737 | r = -EINVAL; | |
5c0aea0e DH |
4738 | if (!kvm_arch_can_set_irq_routing(kvm)) |
4739 | goto out; | |
caf1ff26 | 4740 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
4741 | goto out; |
4742 | if (routing.flags) | |
4743 | goto out; | |
f8c1b85b | 4744 | if (routing.nr) { |
f8c1b85b | 4745 | urouting = argp; |
7ec28e26 DE |
4746 | entries = vmemdup_user(urouting->entries, |
4747 | array_size(sizeof(*entries), | |
4748 | routing.nr)); | |
4749 | if (IS_ERR(entries)) { | |
4750 | r = PTR_ERR(entries); | |
4751 | goto out; | |
4752 | } | |
f8c1b85b | 4753 | } |
aa8d5944 AG |
4754 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
4755 | routing.flags); | |
7ec28e26 | 4756 | kvfree(entries); |
aa8d5944 AG |
4757 | break; |
4758 | } | |
4759 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
4760 | case KVM_CREATE_DEVICE: { |
4761 | struct kvm_create_device cd; | |
4762 | ||
4763 | r = -EFAULT; | |
4764 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
4765 | goto out; | |
4766 | ||
4767 | r = kvm_ioctl_create_device(kvm, &cd); | |
4768 | if (r) | |
4769 | goto out; | |
4770 | ||
4771 | r = -EFAULT; | |
4772 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
4773 | goto out; | |
4774 | ||
4775 | r = 0; | |
4776 | break; | |
4777 | } | |
92b591a4 AG |
4778 | case KVM_CHECK_EXTENSION: |
4779 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
4780 | break; | |
fb04a1ed PX |
4781 | case KVM_RESET_DIRTY_RINGS: |
4782 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
4783 | break; | |
fcfe1bae JZ |
4784 | case KVM_GET_STATS_FD: |
4785 | r = kvm_vm_ioctl_get_stats_fd(kvm); | |
4786 | break; | |
f17abe9a | 4787 | default: |
1fe779f8 | 4788 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
4789 | } |
4790 | out: | |
4791 | return r; | |
4792 | } | |
4793 | ||
de8e5d74 | 4794 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
4795 | struct compat_kvm_dirty_log { |
4796 | __u32 slot; | |
4797 | __u32 padding1; | |
4798 | union { | |
4799 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4800 | __u64 padding2; | |
4801 | }; | |
4802 | }; | |
4803 | ||
8750f9bb PB |
4804 | struct compat_kvm_clear_dirty_log { |
4805 | __u32 slot; | |
4806 | __u32 num_pages; | |
4807 | __u64 first_page; | |
4808 | union { | |
4809 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4810 | __u64 padding2; | |
4811 | }; | |
4812 | }; | |
4813 | ||
6ff5894c AB |
4814 | static long kvm_vm_compat_ioctl(struct file *filp, |
4815 | unsigned int ioctl, unsigned long arg) | |
4816 | { | |
4817 | struct kvm *kvm = filp->private_data; | |
4818 | int r; | |
4819 | ||
f4d31653 | 4820 | if (kvm->mm != current->mm || kvm->vm_dead) |
6ff5894c AB |
4821 | return -EIO; |
4822 | switch (ioctl) { | |
8750f9bb PB |
4823 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4824 | case KVM_CLEAR_DIRTY_LOG: { | |
4825 | struct compat_kvm_clear_dirty_log compat_log; | |
4826 | struct kvm_clear_dirty_log log; | |
4827 | ||
4828 | if (copy_from_user(&compat_log, (void __user *)arg, | |
4829 | sizeof(compat_log))) | |
4830 | return -EFAULT; | |
4831 | log.slot = compat_log.slot; | |
4832 | log.num_pages = compat_log.num_pages; | |
4833 | log.first_page = compat_log.first_page; | |
4834 | log.padding2 = compat_log.padding2; | |
4835 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4836 | ||
4837 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4838 | break; | |
4839 | } | |
4840 | #endif | |
6ff5894c AB |
4841 | case KVM_GET_DIRTY_LOG: { |
4842 | struct compat_kvm_dirty_log compat_log; | |
4843 | struct kvm_dirty_log log; | |
4844 | ||
6ff5894c AB |
4845 | if (copy_from_user(&compat_log, (void __user *)arg, |
4846 | sizeof(compat_log))) | |
f6a3b168 | 4847 | return -EFAULT; |
6ff5894c AB |
4848 | log.slot = compat_log.slot; |
4849 | log.padding1 = compat_log.padding1; | |
4850 | log.padding2 = compat_log.padding2; | |
4851 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4852 | ||
4853 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
4854 | break; |
4855 | } | |
4856 | default: | |
4857 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
4858 | } | |
6ff5894c AB |
4859 | return r; |
4860 | } | |
4861 | #endif | |
4862 | ||
70375c2d | 4863 | static const struct file_operations kvm_vm_fops = { |
f17abe9a AK |
4864 | .release = kvm_vm_release, |
4865 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 4866 | .llseek = noop_llseek, |
7ddfd3e0 | 4867 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
4868 | }; |
4869 | ||
54526d1f NT |
4870 | bool file_is_kvm(struct file *file) |
4871 | { | |
4872 | return file && file->f_op == &kvm_vm_fops; | |
4873 | } | |
4874 | EXPORT_SYMBOL_GPL(file_is_kvm); | |
4875 | ||
e08b9637 | 4876 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 4877 | { |
aac87636 | 4878 | int r; |
f17abe9a | 4879 | struct kvm *kvm; |
506cfba9 | 4880 | struct file *file; |
f17abe9a | 4881 | |
e08b9637 | 4882 | kvm = kvm_create_vm(type); |
d6d28168 AK |
4883 | if (IS_ERR(kvm)) |
4884 | return PTR_ERR(kvm); | |
4b4357e0 | 4885 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 4886 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
4887 | if (r < 0) |
4888 | goto put_kvm; | |
6ce5a090 | 4889 | #endif |
506cfba9 | 4890 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
4891 | if (r < 0) |
4892 | goto put_kvm; | |
4893 | ||
fcfe1bae JZ |
4894 | snprintf(kvm->stats_id, sizeof(kvm->stats_id), |
4895 | "kvm-%d", task_pid_nr(current)); | |
4896 | ||
506cfba9 AV |
4897 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
4898 | if (IS_ERR(file)) { | |
4899 | put_unused_fd(r); | |
78588335 ME |
4900 | r = PTR_ERR(file); |
4901 | goto put_kvm; | |
506cfba9 | 4902 | } |
536a6f88 | 4903 | |
525df861 PB |
4904 | /* |
4905 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
4906 | * already set, with ->release() being kvm_vm_release(). In error | |
4907 | * cases it will be called by the final fput(file) and will take | |
4908 | * care of doing kvm_put_kvm(kvm). | |
4909 | */ | |
536a6f88 | 4910 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
4911 | put_unused_fd(r); |
4912 | fput(file); | |
536a6f88 JF |
4913 | return -ENOMEM; |
4914 | } | |
286de8f6 | 4915 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 4916 | |
506cfba9 | 4917 | fd_install(r, file); |
aac87636 | 4918 | return r; |
78588335 ME |
4919 | |
4920 | put_kvm: | |
4921 | kvm_put_kvm(kvm); | |
4922 | return r; | |
f17abe9a AK |
4923 | } |
4924 | ||
4925 | static long kvm_dev_ioctl(struct file *filp, | |
4926 | unsigned int ioctl, unsigned long arg) | |
4927 | { | |
07c45a36 | 4928 | long r = -EINVAL; |
f17abe9a AK |
4929 | |
4930 | switch (ioctl) { | |
4931 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
4932 | if (arg) |
4933 | goto out; | |
f17abe9a AK |
4934 | r = KVM_API_VERSION; |
4935 | break; | |
4936 | case KVM_CREATE_VM: | |
e08b9637 | 4937 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 4938 | break; |
018d00d2 | 4939 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 4940 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 4941 | break; |
07c45a36 | 4942 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
4943 | if (arg) |
4944 | goto out; | |
adb1ff46 AK |
4945 | r = PAGE_SIZE; /* struct kvm_run */ |
4946 | #ifdef CONFIG_X86 | |
4947 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 4948 | #endif |
4b4357e0 | 4949 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 4950 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 4951 | #endif |
07c45a36 | 4952 | break; |
d4c9ff2d FEL |
4953 | case KVM_TRACE_ENABLE: |
4954 | case KVM_TRACE_PAUSE: | |
4955 | case KVM_TRACE_DISABLE: | |
2023a29c | 4956 | r = -EOPNOTSUPP; |
d4c9ff2d | 4957 | break; |
6aa8b732 | 4958 | default: |
043405e1 | 4959 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
4960 | } |
4961 | out: | |
4962 | return r; | |
4963 | } | |
4964 | ||
6aa8b732 | 4965 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 4966 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 4967 | .llseek = noop_llseek, |
7ddfd3e0 | 4968 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
4969 | }; |
4970 | ||
4971 | static struct miscdevice kvm_dev = { | |
bbe4432e | 4972 | KVM_MINOR, |
6aa8b732 AK |
4973 | "kvm", |
4974 | &kvm_chardev_ops, | |
4975 | }; | |
4976 | ||
75b7127c | 4977 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
4978 | { |
4979 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 4980 | int r; |
1b6c0168 | 4981 | |
7f59f492 | 4982 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4983 | return; |
10474ae8 | 4984 | |
7f59f492 | 4985 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 4986 | |
13a34e06 | 4987 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
4988 | |
4989 | if (r) { | |
4990 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
4991 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 4992 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 4993 | } |
1b6c0168 AK |
4994 | } |
4995 | ||
8c18b2d2 | 4996 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 4997 | { |
4a937f96 | 4998 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4999 | if (kvm_usage_count) |
5000 | hardware_enable_nolock(NULL); | |
4a937f96 | 5001 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 5002 | return 0; |
75b7127c TY |
5003 | } |
5004 | ||
5005 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
5006 | { |
5007 | int cpu = raw_smp_processor_id(); | |
5008 | ||
7f59f492 | 5009 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 5010 | return; |
7f59f492 | 5011 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 5012 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
5013 | } |
5014 | ||
8c18b2d2 | 5015 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 5016 | { |
4a937f96 | 5017 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
5018 | if (kvm_usage_count) |
5019 | hardware_disable_nolock(NULL); | |
4a937f96 | 5020 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 5021 | return 0; |
75b7127c TY |
5022 | } |
5023 | ||
10474ae8 AG |
5024 | static void hardware_disable_all_nolock(void) |
5025 | { | |
5026 | BUG_ON(!kvm_usage_count); | |
5027 | ||
5028 | kvm_usage_count--; | |
5029 | if (!kvm_usage_count) | |
75b7127c | 5030 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
5031 | } |
5032 | ||
5033 | static void hardware_disable_all(void) | |
5034 | { | |
4a937f96 | 5035 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 5036 | hardware_disable_all_nolock(); |
4a937f96 | 5037 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
5038 | } |
5039 | ||
5040 | static int hardware_enable_all(void) | |
5041 | { | |
5042 | int r = 0; | |
5043 | ||
4a937f96 | 5044 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
5045 | |
5046 | kvm_usage_count++; | |
5047 | if (kvm_usage_count == 1) { | |
5048 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 5049 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
5050 | |
5051 | if (atomic_read(&hardware_enable_failed)) { | |
5052 | hardware_disable_all_nolock(); | |
5053 | r = -EBUSY; | |
5054 | } | |
5055 | } | |
5056 | ||
4a937f96 | 5057 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
5058 | |
5059 | return r; | |
5060 | } | |
5061 | ||
9a2b85c6 | 5062 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 5063 | void *v) |
9a2b85c6 | 5064 | { |
8e1c1815 SY |
5065 | /* |
5066 | * Some (well, at least mine) BIOSes hang on reboot if | |
5067 | * in vmx root mode. | |
5068 | * | |
5069 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
5070 | */ | |
1170adc6 | 5071 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 5072 | kvm_rebooting = true; |
75b7127c | 5073 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
5074 | return NOTIFY_OK; |
5075 | } | |
5076 | ||
5077 | static struct notifier_block kvm_reboot_notifier = { | |
5078 | .notifier_call = kvm_reboot, | |
5079 | .priority = 0, | |
5080 | }; | |
5081 | ||
e93f8a0f | 5082 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
5083 | { |
5084 | int i; | |
5085 | ||
5086 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 5087 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
5088 | |
5089 | kvm_iodevice_destructor(pos); | |
5090 | } | |
e93f8a0f | 5091 | kfree(bus); |
2eeb2e94 GH |
5092 | } |
5093 | ||
c21fbff1 | 5094 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 5095 | const struct kvm_io_range *r2) |
743eeb0b | 5096 | { |
8f4216c7 JW |
5097 | gpa_t addr1 = r1->addr; |
5098 | gpa_t addr2 = r2->addr; | |
5099 | ||
5100 | if (addr1 < addr2) | |
743eeb0b | 5101 | return -1; |
8f4216c7 JW |
5102 | |
5103 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
5104 | * accept any overlapping write. Any order is acceptable for | |
5105 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
5106 | * we process all of them. | |
5107 | */ | |
5108 | if (r2->len) { | |
5109 | addr1 += r1->len; | |
5110 | addr2 += r2->len; | |
5111 | } | |
5112 | ||
5113 | if (addr1 > addr2) | |
743eeb0b | 5114 | return 1; |
8f4216c7 | 5115 | |
743eeb0b SL |
5116 | return 0; |
5117 | } | |
5118 | ||
a343c9b7 PB |
5119 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
5120 | { | |
c21fbff1 | 5121 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
5122 | } |
5123 | ||
39369f7a | 5124 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
5125 | gpa_t addr, int len) |
5126 | { | |
5127 | struct kvm_io_range *range, key; | |
5128 | int off; | |
5129 | ||
5130 | key = (struct kvm_io_range) { | |
5131 | .addr = addr, | |
5132 | .len = len, | |
5133 | }; | |
5134 | ||
5135 | range = bsearch(&key, bus->range, bus->dev_count, | |
5136 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
5137 | if (range == NULL) | |
5138 | return -ENOENT; | |
5139 | ||
5140 | off = range - bus->range; | |
5141 | ||
c21fbff1 | 5142 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
5143 | off--; |
5144 | ||
5145 | return off; | |
5146 | } | |
5147 | ||
e32edf4f | 5148 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
5149 | struct kvm_io_range *range, const void *val) |
5150 | { | |
5151 | int idx; | |
5152 | ||
5153 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
5154 | if (idx < 0) | |
5155 | return -EOPNOTSUPP; | |
5156 | ||
5157 | while (idx < bus->dev_count && | |
c21fbff1 | 5158 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5159 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5160 | range->len, val)) |
5161 | return idx; | |
5162 | idx++; | |
5163 | } | |
5164 | ||
5165 | return -EOPNOTSUPP; | |
5166 | } | |
5167 | ||
bda9020e | 5168 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 5169 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 5170 | int len, const void *val) |
2eeb2e94 | 5171 | { |
90d83dc3 | 5172 | struct kvm_io_bus *bus; |
743eeb0b | 5173 | struct kvm_io_range range; |
126a5af5 | 5174 | int r; |
743eeb0b SL |
5175 | |
5176 | range = (struct kvm_io_range) { | |
5177 | .addr = addr, | |
5178 | .len = len, | |
5179 | }; | |
90d83dc3 | 5180 | |
e32edf4f | 5181 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5182 | if (!bus) |
5183 | return -ENOMEM; | |
e32edf4f | 5184 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5185 | return r < 0 ? r : 0; |
5186 | } | |
a2420107 | 5187 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
5188 | |
5189 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
5190 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
5191 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
5192 | { |
5193 | struct kvm_io_bus *bus; | |
5194 | struct kvm_io_range range; | |
5195 | ||
5196 | range = (struct kvm_io_range) { | |
5197 | .addr = addr, | |
5198 | .len = len, | |
5199 | }; | |
5200 | ||
e32edf4f | 5201 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5202 | if (!bus) |
5203 | return -ENOMEM; | |
126a5af5 CH |
5204 | |
5205 | /* First try the device referenced by cookie. */ | |
5206 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 5207 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 5208 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
5209 | val)) |
5210 | return cookie; | |
5211 | ||
5212 | /* | |
5213 | * cookie contained garbage; fall back to search and return the | |
5214 | * correct cookie value. | |
5215 | */ | |
e32edf4f | 5216 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5217 | } |
5218 | ||
e32edf4f NN |
5219 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
5220 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
5221 | { |
5222 | int idx; | |
5223 | ||
5224 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
5225 | if (idx < 0) |
5226 | return -EOPNOTSUPP; | |
5227 | ||
5228 | while (idx < bus->dev_count && | |
c21fbff1 | 5229 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5230 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5231 | range->len, val)) |
5232 | return idx; | |
743eeb0b SL |
5233 | idx++; |
5234 | } | |
5235 | ||
bda9020e MT |
5236 | return -EOPNOTSUPP; |
5237 | } | |
2eeb2e94 | 5238 | |
bda9020e | 5239 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 5240 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 5241 | int len, void *val) |
bda9020e | 5242 | { |
90d83dc3 | 5243 | struct kvm_io_bus *bus; |
743eeb0b | 5244 | struct kvm_io_range range; |
126a5af5 | 5245 | int r; |
743eeb0b SL |
5246 | |
5247 | range = (struct kvm_io_range) { | |
5248 | .addr = addr, | |
5249 | .len = len, | |
5250 | }; | |
e93f8a0f | 5251 | |
e32edf4f | 5252 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5253 | if (!bus) |
5254 | return -ENOMEM; | |
e32edf4f | 5255 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
5256 | return r < 0 ? r : 0; |
5257 | } | |
743eeb0b | 5258 | |
79fac95e | 5259 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
5260 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
5261 | int len, struct kvm_io_device *dev) | |
6c474694 | 5262 | { |
d4c67a7a | 5263 | int i; |
e93f8a0f | 5264 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 5265 | struct kvm_io_range range; |
090b7aff | 5266 | |
4a12f951 | 5267 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
5268 | if (!bus) |
5269 | return -ENOMEM; | |
5270 | ||
6ea34c9b AK |
5271 | /* exclude ioeventfd which is limited by maximum fd */ |
5272 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 5273 | return -ENOSPC; |
2eeb2e94 | 5274 | |
90952cd3 | 5275 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 5276 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
5277 | if (!new_bus) |
5278 | return -ENOMEM; | |
d4c67a7a GH |
5279 | |
5280 | range = (struct kvm_io_range) { | |
5281 | .addr = addr, | |
5282 | .len = len, | |
5283 | .dev = dev, | |
5284 | }; | |
5285 | ||
5286 | for (i = 0; i < bus->dev_count; i++) | |
5287 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
5288 | break; | |
5289 | ||
5290 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
5291 | new_bus->dev_count++; | |
5292 | new_bus->range[i] = range; | |
5293 | memcpy(new_bus->range + i + 1, bus->range + i, | |
5294 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
5295 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
5296 | synchronize_srcu_expedited(&kvm->srcu); | |
5297 | kfree(bus); | |
090b7aff GH |
5298 | |
5299 | return 0; | |
5300 | } | |
5301 | ||
5d3c4c79 SC |
5302 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5303 | struct kvm_io_device *dev) | |
090b7aff | 5304 | { |
f6588660 | 5305 | int i, j; |
e93f8a0f | 5306 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 5307 | |
7c896d37 SC |
5308 | lockdep_assert_held(&kvm->slots_lock); |
5309 | ||
4a12f951 | 5310 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 5311 | if (!bus) |
5d3c4c79 | 5312 | return 0; |
df630b8c | 5313 | |
7c896d37 | 5314 | for (i = 0; i < bus->dev_count; i++) { |
a1300716 | 5315 | if (bus->range[i].dev == dev) { |
090b7aff GH |
5316 | break; |
5317 | } | |
7c896d37 | 5318 | } |
e93f8a0f | 5319 | |
90db1043 | 5320 | if (i == bus->dev_count) |
5d3c4c79 | 5321 | return 0; |
a1300716 | 5322 | |
90952cd3 | 5323 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 5324 | GFP_KERNEL_ACCOUNT); |
f6588660 | 5325 | if (new_bus) { |
871c433b | 5326 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
5327 | new_bus->dev_count--; |
5328 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 5329 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
2ee37574 SC |
5330 | } |
5331 | ||
5332 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
5333 | synchronize_srcu_expedited(&kvm->srcu); | |
5334 | ||
5335 | /* Destroy the old bus _after_ installing the (null) bus. */ | |
5336 | if (!new_bus) { | |
90db1043 | 5337 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
f6588660 RK |
5338 | for (j = 0; j < bus->dev_count; j++) { |
5339 | if (j == i) | |
5340 | continue; | |
5341 | kvm_iodevice_destructor(bus->range[j].dev); | |
5342 | } | |
90db1043 | 5343 | } |
a1300716 | 5344 | |
e93f8a0f | 5345 | kfree(bus); |
5d3c4c79 | 5346 | return new_bus ? 0 : -ENOMEM; |
2eeb2e94 GH |
5347 | } |
5348 | ||
8a39d006 AP |
5349 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5350 | gpa_t addr) | |
5351 | { | |
5352 | struct kvm_io_bus *bus; | |
5353 | int dev_idx, srcu_idx; | |
5354 | struct kvm_io_device *iodev = NULL; | |
5355 | ||
5356 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
5357 | ||
5358 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
5359 | if (!bus) |
5360 | goto out_unlock; | |
8a39d006 AP |
5361 | |
5362 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
5363 | if (dev_idx < 0) | |
5364 | goto out_unlock; | |
5365 | ||
5366 | iodev = bus->range[dev_idx].dev; | |
5367 | ||
5368 | out_unlock: | |
5369 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
5370 | ||
5371 | return iodev; | |
5372 | } | |
5373 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
5374 | ||
536a6f88 JF |
5375 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
5376 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
5377 | const char *fmt) | |
5378 | { | |
5379 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5380 | inode->i_private; | |
5381 | ||
605c7130 PX |
5382 | /* |
5383 | * The debugfs files are a reference to the kvm struct which | |
5384 | * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe | |
5385 | * avoids the race between open and the removal of the debugfs directory. | |
536a6f88 | 5386 | */ |
605c7130 | 5387 | if (!kvm_get_kvm_safe(stat_data->kvm)) |
536a6f88 JF |
5388 | return -ENOENT; |
5389 | ||
833b45de | 5390 | if (simple_attr_open(inode, file, get, |
bc9e9e67 | 5391 | kvm_stats_debugfs_mode(stat_data->desc) & 0222 |
09cbcef6 MP |
5392 | ? set : NULL, |
5393 | fmt)) { | |
536a6f88 JF |
5394 | kvm_put_kvm(stat_data->kvm); |
5395 | return -ENOMEM; | |
5396 | } | |
5397 | ||
5398 | return 0; | |
5399 | } | |
5400 | ||
5401 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
5402 | { | |
5403 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5404 | inode->i_private; | |
5405 | ||
5406 | simple_attr_release(inode, file); | |
5407 | kvm_put_kvm(stat_data->kvm); | |
5408 | ||
5409 | return 0; | |
5410 | } | |
5411 | ||
09cbcef6 | 5412 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 5413 | { |
bc9e9e67 | 5414 | *val = *(u64 *)((void *)(&kvm->stat) + offset); |
536a6f88 | 5415 | |
09cbcef6 MP |
5416 | return 0; |
5417 | } | |
5418 | ||
5419 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
5420 | { | |
bc9e9e67 | 5421 | *(u64 *)((void *)(&kvm->stat) + offset) = 0; |
536a6f88 JF |
5422 | |
5423 | return 0; | |
5424 | } | |
5425 | ||
09cbcef6 | 5426 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 5427 | { |
46808a4c | 5428 | unsigned long i; |
09cbcef6 | 5429 | struct kvm_vcpu *vcpu; |
ce35ef27 | 5430 | |
09cbcef6 | 5431 | *val = 0; |
ce35ef27 | 5432 | |
09cbcef6 | 5433 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5434 | *val += *(u64 *)((void *)(&vcpu->stat) + offset); |
ce35ef27 SJS |
5435 | |
5436 | return 0; | |
5437 | } | |
5438 | ||
09cbcef6 | 5439 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 5440 | { |
46808a4c | 5441 | unsigned long i; |
09cbcef6 | 5442 | struct kvm_vcpu *vcpu; |
536a6f88 | 5443 | |
09cbcef6 | 5444 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5445 | *(u64 *)((void *)(&vcpu->stat) + offset) = 0; |
09cbcef6 MP |
5446 | |
5447 | return 0; | |
5448 | } | |
536a6f88 | 5449 | |
09cbcef6 | 5450 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 5451 | { |
09cbcef6 | 5452 | int r = -EFAULT; |
536a6f88 | 5453 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
536a6f88 | 5454 | |
bc9e9e67 | 5455 | switch (stat_data->kind) { |
09cbcef6 MP |
5456 | case KVM_STAT_VM: |
5457 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5458 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5459 | break; |
5460 | case KVM_STAT_VCPU: | |
5461 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5462 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5463 | break; |
5464 | } | |
536a6f88 | 5465 | |
09cbcef6 | 5466 | return r; |
536a6f88 JF |
5467 | } |
5468 | ||
09cbcef6 | 5469 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 5470 | { |
09cbcef6 | 5471 | int r = -EFAULT; |
ce35ef27 | 5472 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
ce35ef27 SJS |
5473 | |
5474 | if (val) | |
5475 | return -EINVAL; | |
5476 | ||
bc9e9e67 | 5477 | switch (stat_data->kind) { |
09cbcef6 MP |
5478 | case KVM_STAT_VM: |
5479 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5480 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5481 | break; |
5482 | case KVM_STAT_VCPU: | |
5483 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5484 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5485 | break; |
5486 | } | |
ce35ef27 | 5487 | |
09cbcef6 | 5488 | return r; |
ce35ef27 SJS |
5489 | } |
5490 | ||
09cbcef6 | 5491 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
5492 | { |
5493 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
5494 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
5495 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
5496 | } |
5497 | ||
09cbcef6 MP |
5498 | static const struct file_operations stat_fops_per_vm = { |
5499 | .owner = THIS_MODULE, | |
5500 | .open = kvm_stat_data_open, | |
536a6f88 | 5501 | .release = kvm_debugfs_release, |
09cbcef6 MP |
5502 | .read = simple_attr_read, |
5503 | .write = simple_attr_write, | |
5504 | .llseek = no_llseek, | |
536a6f88 JF |
5505 | }; |
5506 | ||
8b88b099 | 5507 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
5508 | { |
5509 | unsigned offset = (long)_offset; | |
ba1389b7 | 5510 | struct kvm *kvm; |
536a6f88 | 5511 | u64 tmp_val; |
ba1389b7 | 5512 | |
8b88b099 | 5513 | *val = 0; |
0d9ce162 | 5514 | mutex_lock(&kvm_lock); |
536a6f88 | 5515 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5516 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
5517 | *val += tmp_val; |
5518 | } | |
0d9ce162 | 5519 | mutex_unlock(&kvm_lock); |
8b88b099 | 5520 | return 0; |
ba1389b7 AK |
5521 | } |
5522 | ||
ce35ef27 SJS |
5523 | static int vm_stat_clear(void *_offset, u64 val) |
5524 | { | |
5525 | unsigned offset = (long)_offset; | |
5526 | struct kvm *kvm; | |
ce35ef27 SJS |
5527 | |
5528 | if (val) | |
5529 | return -EINVAL; | |
5530 | ||
0d9ce162 | 5531 | mutex_lock(&kvm_lock); |
ce35ef27 | 5532 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5533 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 5534 | } |
0d9ce162 | 5535 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5536 | |
5537 | return 0; | |
5538 | } | |
5539 | ||
5540 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
bc9e9e67 | 5541 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n"); |
ba1389b7 | 5542 | |
8b88b099 | 5543 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
5544 | { |
5545 | unsigned offset = (long)_offset; | |
1165f5fe | 5546 | struct kvm *kvm; |
536a6f88 | 5547 | u64 tmp_val; |
1165f5fe | 5548 | |
8b88b099 | 5549 | *val = 0; |
0d9ce162 | 5550 | mutex_lock(&kvm_lock); |
536a6f88 | 5551 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5552 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
5553 | *val += tmp_val; |
5554 | } | |
0d9ce162 | 5555 | mutex_unlock(&kvm_lock); |
8b88b099 | 5556 | return 0; |
1165f5fe AK |
5557 | } |
5558 | ||
ce35ef27 SJS |
5559 | static int vcpu_stat_clear(void *_offset, u64 val) |
5560 | { | |
5561 | unsigned offset = (long)_offset; | |
5562 | struct kvm *kvm; | |
ce35ef27 SJS |
5563 | |
5564 | if (val) | |
5565 | return -EINVAL; | |
5566 | ||
0d9ce162 | 5567 | mutex_lock(&kvm_lock); |
ce35ef27 | 5568 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5569 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 5570 | } |
0d9ce162 | 5571 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5572 | |
5573 | return 0; | |
5574 | } | |
5575 | ||
5576 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
5577 | "%llu\n"); | |
bc9e9e67 | 5578 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n"); |
1165f5fe | 5579 | |
286de8f6 CI |
5580 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
5581 | { | |
5582 | struct kobj_uevent_env *env; | |
286de8f6 CI |
5583 | unsigned long long created, active; |
5584 | ||
5585 | if (!kvm_dev.this_device || !kvm) | |
5586 | return; | |
5587 | ||
0d9ce162 | 5588 | mutex_lock(&kvm_lock); |
286de8f6 CI |
5589 | if (type == KVM_EVENT_CREATE_VM) { |
5590 | kvm_createvm_count++; | |
5591 | kvm_active_vms++; | |
5592 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
5593 | kvm_active_vms--; | |
5594 | } | |
5595 | created = kvm_createvm_count; | |
5596 | active = kvm_active_vms; | |
0d9ce162 | 5597 | mutex_unlock(&kvm_lock); |
286de8f6 | 5598 | |
b12ce36a | 5599 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
5600 | if (!env) |
5601 | return; | |
5602 | ||
5603 | add_uevent_var(env, "CREATED=%llu", created); | |
5604 | add_uevent_var(env, "COUNT=%llu", active); | |
5605 | ||
fdeaf7e3 | 5606 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 5607 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
5608 | kvm->userspace_pid = task_pid_nr(current); |
5609 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 5610 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
5611 | } |
5612 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 5613 | |
a44a4cc1 | 5614 | if (!IS_ERR(kvm->debugfs_dentry)) { |
b12ce36a | 5615 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
5616 | |
5617 | if (p) { | |
5618 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
5619 | if (!IS_ERR(tmp)) | |
5620 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
5621 | kfree(p); | |
286de8f6 CI |
5622 | } |
5623 | } | |
5624 | /* no need for checks, since we are adding at most only 5 keys */ | |
5625 | env->envp[env->envp_idx++] = NULL; | |
5626 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
5627 | kfree(env); | |
286de8f6 CI |
5628 | } |
5629 | ||
929f45e3 | 5630 | static void kvm_init_debug(void) |
6aa8b732 | 5631 | { |
bc9e9e67 JZ |
5632 | const struct file_operations *fops; |
5633 | const struct _kvm_stats_desc *pdesc; | |
5634 | int i; | |
6aa8b732 | 5635 | |
76f7c879 | 5636 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 5637 | |
bc9e9e67 JZ |
5638 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
5639 | pdesc = &kvm_vm_stats_desc[i]; | |
5640 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5641 | fops = &vm_stat_fops; | |
5642 | else | |
5643 | fops = &vm_stat_readonly_fops; | |
5644 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5645 | kvm_debugfs_dir, | |
5646 | (void *)(long)pdesc->desc.offset, fops); | |
5647 | } | |
5648 | ||
5649 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
5650 | pdesc = &kvm_vcpu_stats_desc[i]; | |
5651 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5652 | fops = &vcpu_stat_fops; | |
5653 | else | |
5654 | fops = &vcpu_stat_readonly_fops; | |
5655 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5656 | kvm_debugfs_dir, | |
5657 | (void *)(long)pdesc->desc.offset, fops); | |
4f69b680 | 5658 | } |
6aa8b732 AK |
5659 | } |
5660 | ||
fb3600cc | 5661 | static int kvm_suspend(void) |
59ae6c6b | 5662 | { |
10474ae8 | 5663 | if (kvm_usage_count) |
75b7127c | 5664 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
5665 | return 0; |
5666 | } | |
5667 | ||
fb3600cc | 5668 | static void kvm_resume(void) |
59ae6c6b | 5669 | { |
ca84d1a2 | 5670 | if (kvm_usage_count) { |
4cb9a998 | 5671 | lockdep_assert_not_held(&kvm_count_lock); |
75b7127c | 5672 | hardware_enable_nolock(NULL); |
ca84d1a2 | 5673 | } |
59ae6c6b AK |
5674 | } |
5675 | ||
fb3600cc | 5676 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
5677 | .suspend = kvm_suspend, |
5678 | .resume = kvm_resume, | |
5679 | }; | |
5680 | ||
15ad7146 AK |
5681 | static inline |
5682 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
5683 | { | |
5684 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
5685 | } | |
5686 | ||
5687 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
5688 | { | |
5689 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 5690 | |
046ddeed | 5691 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 5692 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 5693 | |
7495e22b | 5694 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 5695 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 5696 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
5697 | } |
5698 | ||
5699 | static void kvm_sched_out(struct preempt_notifier *pn, | |
5700 | struct task_struct *next) | |
5701 | { | |
5702 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
5703 | ||
3ba9f93b | 5704 | if (current->on_rq) { |
046ddeed | 5705 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
5706 | WRITE_ONCE(vcpu->ready, true); |
5707 | } | |
e9b11c17 | 5708 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
5709 | __this_cpu_write(kvm_running_vcpu, NULL); |
5710 | } | |
5711 | ||
5712 | /** | |
5713 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
5714 | * |
5715 | * We can disable preemption locally around accessing the per-CPU variable, | |
5716 | * and use the resolved vcpu pointer after enabling preemption again, | |
5717 | * because even if the current thread is migrated to another CPU, reading | |
5718 | * the per-CPU value later will give us the same value as we update the | |
5719 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
5720 | */ |
5721 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
5722 | { | |
1f03b2bc MZ |
5723 | struct kvm_vcpu *vcpu; |
5724 | ||
5725 | preempt_disable(); | |
5726 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
5727 | preempt_enable(); | |
5728 | ||
5729 | return vcpu; | |
7495e22b | 5730 | } |
379a3c8e | 5731 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
5732 | |
5733 | /** | |
5734 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
5735 | */ | |
5736 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
5737 | { | |
5738 | return &kvm_running_vcpu; | |
15ad7146 AK |
5739 | } |
5740 | ||
e1bfc245 SC |
5741 | #ifdef CONFIG_GUEST_PERF_EVENTS |
5742 | static unsigned int kvm_guest_state(void) | |
5743 | { | |
5744 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
5745 | unsigned int state; | |
5746 | ||
5747 | if (!kvm_arch_pmi_in_guest(vcpu)) | |
5748 | return 0; | |
5749 | ||
5750 | state = PERF_GUEST_ACTIVE; | |
5751 | if (!kvm_arch_vcpu_in_kernel(vcpu)) | |
5752 | state |= PERF_GUEST_USER; | |
5753 | ||
5754 | return state; | |
5755 | } | |
5756 | ||
5757 | static unsigned long kvm_guest_get_ip(void) | |
5758 | { | |
5759 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
5760 | ||
5761 | /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */ | |
5762 | if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu))) | |
5763 | return 0; | |
5764 | ||
5765 | return kvm_arch_vcpu_get_ip(vcpu); | |
5766 | } | |
5767 | ||
5768 | static struct perf_guest_info_callbacks kvm_guest_cbs = { | |
5769 | .state = kvm_guest_state, | |
5770 | .get_ip = kvm_guest_get_ip, | |
5771 | .handle_intel_pt_intr = NULL, | |
5772 | }; | |
5773 | ||
5774 | void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void)) | |
5775 | { | |
5776 | kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler; | |
5777 | perf_register_guest_info_callbacks(&kvm_guest_cbs); | |
5778 | } | |
5779 | void kvm_unregister_perf_callbacks(void) | |
5780 | { | |
5781 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); | |
5782 | } | |
5783 | #endif | |
5784 | ||
b9904085 SC |
5785 | struct kvm_cpu_compat_check { |
5786 | void *opaque; | |
5787 | int *ret; | |
5788 | }; | |
5789 | ||
5790 | static void check_processor_compat(void *data) | |
f257d6dc | 5791 | { |
b9904085 SC |
5792 | struct kvm_cpu_compat_check *c = data; |
5793 | ||
5794 | *c->ret = kvm_arch_check_processor_compat(c->opaque); | |
f257d6dc SC |
5795 | } |
5796 | ||
0ee75bea | 5797 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 5798 | struct module *module) |
6aa8b732 | 5799 | { |
b9904085 | 5800 | struct kvm_cpu_compat_check c; |
6aa8b732 | 5801 | int r; |
002c7f7c | 5802 | int cpu; |
6aa8b732 | 5803 | |
f8c16bba ZX |
5804 | r = kvm_arch_init(opaque); |
5805 | if (r) | |
d2308784 | 5806 | goto out_fail; |
cb498ea2 | 5807 | |
7dac16c3 AH |
5808 | /* |
5809 | * kvm_arch_init makes sure there's at most one caller | |
5810 | * for architectures that support multiple implementations, | |
5811 | * like intel and amd on x86. | |
36343f6e PB |
5812 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
5813 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 5814 | */ |
36343f6e PB |
5815 | r = kvm_irqfd_init(); |
5816 | if (r) | |
5817 | goto out_irqfd; | |
7dac16c3 | 5818 | |
8437a617 | 5819 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
5820 | r = -ENOMEM; |
5821 | goto out_free_0; | |
5822 | } | |
5823 | ||
b9904085 | 5824 | r = kvm_arch_hardware_setup(opaque); |
6aa8b732 | 5825 | if (r < 0) |
faf0be22 | 5826 | goto out_free_1; |
6aa8b732 | 5827 | |
b9904085 SC |
5828 | c.ret = &r; |
5829 | c.opaque = opaque; | |
002c7f7c | 5830 | for_each_online_cpu(cpu) { |
b9904085 | 5831 | smp_call_function_single(cpu, check_processor_compat, &c, 1); |
002c7f7c | 5832 | if (r < 0) |
faf0be22 | 5833 | goto out_free_2; |
002c7f7c YS |
5834 | } |
5835 | ||
73c1b41e | 5836 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 5837 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 5838 | if (r) |
d2308784 | 5839 | goto out_free_2; |
6aa8b732 AK |
5840 | register_reboot_notifier(&kvm_reboot_notifier); |
5841 | ||
c16f862d | 5842 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
5843 | if (!vcpu_align) |
5844 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
5845 | kvm_vcpu_cache = |
5846 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
5847 | SLAB_ACCOUNT, | |
5848 | offsetof(struct kvm_vcpu, arch), | |
ce55c049 JZ |
5849 | offsetofend(struct kvm_vcpu, stats_id) |
5850 | - offsetof(struct kvm_vcpu, arch), | |
46515736 | 5851 | NULL); |
c16f862d RR |
5852 | if (!kvm_vcpu_cache) { |
5853 | r = -ENOMEM; | |
fb3600cc | 5854 | goto out_free_3; |
c16f862d RR |
5855 | } |
5856 | ||
baff59cc VK |
5857 | for_each_possible_cpu(cpu) { |
5858 | if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu), | |
5859 | GFP_KERNEL, cpu_to_node(cpu))) { | |
5860 | r = -ENOMEM; | |
5861 | goto out_free_4; | |
5862 | } | |
5863 | } | |
5864 | ||
af585b92 GN |
5865 | r = kvm_async_pf_init(); |
5866 | if (r) | |
baff59cc | 5867 | goto out_free_5; |
af585b92 | 5868 | |
6aa8b732 AK |
5869 | kvm_chardev_ops.owner = module; |
5870 | ||
5871 | r = misc_register(&kvm_dev); | |
5872 | if (r) { | |
1170adc6 | 5873 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 5874 | goto out_unreg; |
6aa8b732 AK |
5875 | } |
5876 | ||
fb3600cc RW |
5877 | register_syscore_ops(&kvm_syscore_ops); |
5878 | ||
15ad7146 AK |
5879 | kvm_preempt_ops.sched_in = kvm_sched_in; |
5880 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
5881 | ||
929f45e3 | 5882 | kvm_init_debug(); |
0ea4ed8e | 5883 | |
3c3c29fd PB |
5884 | r = kvm_vfio_ops_init(); |
5885 | WARN_ON(r); | |
5886 | ||
c7addb90 | 5887 | return 0; |
6aa8b732 | 5888 | |
af585b92 GN |
5889 | out_unreg: |
5890 | kvm_async_pf_deinit(); | |
baff59cc VK |
5891 | out_free_5: |
5892 | for_each_possible_cpu(cpu) | |
5893 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
5894 | out_free_4: | |
c16f862d | 5895 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 5896 | out_free_3: |
6aa8b732 | 5897 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5898 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 5899 | out_free_2: |
e9b11c17 | 5900 | kvm_arch_hardware_unsetup(); |
faf0be22 | 5901 | out_free_1: |
7f59f492 | 5902 | free_cpumask_var(cpus_hardware_enabled); |
d2308784 | 5903 | out_free_0: |
a0f155e9 | 5904 | kvm_irqfd_exit(); |
36343f6e | 5905 | out_irqfd: |
7dac16c3 AH |
5906 | kvm_arch_exit(); |
5907 | out_fail: | |
6aa8b732 AK |
5908 | return r; |
5909 | } | |
cb498ea2 | 5910 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 5911 | |
cb498ea2 | 5912 | void kvm_exit(void) |
6aa8b732 | 5913 | { |
baff59cc VK |
5914 | int cpu; |
5915 | ||
4bd33b56 | 5916 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 5917 | misc_deregister(&kvm_dev); |
baff59cc VK |
5918 | for_each_possible_cpu(cpu) |
5919 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
c16f862d | 5920 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 5921 | kvm_async_pf_deinit(); |
fb3600cc | 5922 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 5923 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5924 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 5925 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 5926 | kvm_arch_hardware_unsetup(); |
f8c16bba | 5927 | kvm_arch_exit(); |
a0f155e9 | 5928 | kvm_irqfd_exit(); |
7f59f492 | 5929 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 5930 | kvm_vfio_ops_exit(); |
6aa8b732 | 5931 | } |
cb498ea2 | 5932 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
5933 | |
5934 | struct kvm_vm_worker_thread_context { | |
5935 | struct kvm *kvm; | |
5936 | struct task_struct *parent; | |
5937 | struct completion init_done; | |
5938 | kvm_vm_thread_fn_t thread_fn; | |
5939 | uintptr_t data; | |
5940 | int err; | |
5941 | }; | |
5942 | ||
5943 | static int kvm_vm_worker_thread(void *context) | |
5944 | { | |
5945 | /* | |
5946 | * The init_context is allocated on the stack of the parent thread, so | |
5947 | * we have to locally copy anything that is needed beyond initialization | |
5948 | */ | |
5949 | struct kvm_vm_worker_thread_context *init_context = context; | |
e45cce30 | 5950 | struct task_struct *parent; |
c57c8046 JS |
5951 | struct kvm *kvm = init_context->kvm; |
5952 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
5953 | uintptr_t data = init_context->data; | |
5954 | int err; | |
5955 | ||
5956 | err = kthread_park(current); | |
5957 | /* kthread_park(current) is never supposed to return an error */ | |
5958 | WARN_ON(err != 0); | |
5959 | if (err) | |
5960 | goto init_complete; | |
5961 | ||
5962 | err = cgroup_attach_task_all(init_context->parent, current); | |
5963 | if (err) { | |
5964 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
5965 | __func__, err); | |
5966 | goto init_complete; | |
5967 | } | |
5968 | ||
5969 | set_user_nice(current, task_nice(init_context->parent)); | |
5970 | ||
5971 | init_complete: | |
5972 | init_context->err = err; | |
5973 | complete(&init_context->init_done); | |
5974 | init_context = NULL; | |
5975 | ||
5976 | if (err) | |
e45cce30 | 5977 | goto out; |
c57c8046 JS |
5978 | |
5979 | /* Wait to be woken up by the spawner before proceeding. */ | |
5980 | kthread_parkme(); | |
5981 | ||
5982 | if (!kthread_should_stop()) | |
5983 | err = thread_fn(kvm, data); | |
5984 | ||
e45cce30 VS |
5985 | out: |
5986 | /* | |
5987 | * Move kthread back to its original cgroup to prevent it lingering in | |
5988 | * the cgroup of the VM process, after the latter finishes its | |
5989 | * execution. | |
5990 | * | |
5991 | * kthread_stop() waits on the 'exited' completion condition which is | |
5992 | * set in exit_mm(), via mm_release(), in do_exit(). However, the | |
5993 | * kthread is removed from the cgroup in the cgroup_exit() which is | |
5994 | * called after the exit_mm(). This causes the kthread_stop() to return | |
5995 | * before the kthread actually quits the cgroup. | |
5996 | */ | |
5997 | rcu_read_lock(); | |
5998 | parent = rcu_dereference(current->real_parent); | |
5999 | get_task_struct(parent); | |
6000 | rcu_read_unlock(); | |
6001 | cgroup_attach_task_all(parent, current); | |
6002 | put_task_struct(parent); | |
6003 | ||
c57c8046 JS |
6004 | return err; |
6005 | } | |
6006 | ||
6007 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
6008 | uintptr_t data, const char *name, | |
6009 | struct task_struct **thread_ptr) | |
6010 | { | |
6011 | struct kvm_vm_worker_thread_context init_context = {}; | |
6012 | struct task_struct *thread; | |
6013 | ||
6014 | *thread_ptr = NULL; | |
6015 | init_context.kvm = kvm; | |
6016 | init_context.parent = current; | |
6017 | init_context.thread_fn = thread_fn; | |
6018 | init_context.data = data; | |
6019 | init_completion(&init_context.init_done); | |
6020 | ||
6021 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
6022 | "%s-%d", name, task_pid_nr(current)); | |
6023 | if (IS_ERR(thread)) | |
6024 | return PTR_ERR(thread); | |
6025 | ||
6026 | /* kthread_run is never supposed to return NULL */ | |
6027 | WARN_ON(thread == NULL); | |
6028 | ||
6029 | wait_for_completion(&init_context.init_done); | |
6030 | ||
6031 | if (!init_context.err) | |
6032 | *thread_ptr = thread; | |
6033 | ||
6034 | return init_context.err; | |
6035 | } |