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 | { |
63f4b210 | 401 | gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : 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; |
58fc1166 OU |
487 | |
488 | /* Fill the stats id string for the vcpu */ | |
489 | snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d", | |
490 | task_pid_nr(current), id); | |
fb3f0f51 | 491 | } |
fb3f0f51 | 492 | |
27592ae8 | 493 | static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
4543bdc0 SC |
494 | { |
495 | kvm_arch_vcpu_destroy(vcpu); | |
5593473a | 496 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
e529ef66 | 497 | |
9941d224 SC |
498 | /* |
499 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
500 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
501 | * are already gone. | |
502 | */ | |
503 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
504 | ||
8bd826d6 | 505 | free_page((unsigned long)vcpu->run); |
e529ef66 | 506 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 | 507 | } |
27592ae8 MZ |
508 | |
509 | void kvm_destroy_vcpus(struct kvm *kvm) | |
510 | { | |
46808a4c | 511 | unsigned long i; |
27592ae8 MZ |
512 | struct kvm_vcpu *vcpu; |
513 | ||
514 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
515 | kvm_vcpu_destroy(vcpu); | |
c5b07754 | 516 | xa_erase(&kvm->vcpu_array, i); |
27592ae8 MZ |
517 | } |
518 | ||
519 | atomic_set(&kvm->online_vcpus, 0); | |
520 | } | |
521 | EXPORT_SYMBOL_GPL(kvm_destroy_vcpus); | |
4543bdc0 | 522 | |
e930bffe AA |
523 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
524 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
525 | { | |
526 | return container_of(mn, struct kvm, mmu_notifier); | |
527 | } | |
528 | ||
e649b3f0 ET |
529 | static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn, |
530 | struct mm_struct *mm, | |
531 | unsigned long start, unsigned long end) | |
532 | { | |
533 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
534 | int idx; | |
535 | ||
536 | idx = srcu_read_lock(&kvm->srcu); | |
537 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
538 | srcu_read_unlock(&kvm->srcu, idx); | |
539 | } | |
540 | ||
3039bcc7 SC |
541 | typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range); |
542 | ||
f922bd9b SC |
543 | typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start, |
544 | unsigned long end); | |
545 | ||
683412cc MZ |
546 | typedef void (*on_unlock_fn_t)(struct kvm *kvm); |
547 | ||
3039bcc7 SC |
548 | struct kvm_hva_range { |
549 | unsigned long start; | |
550 | unsigned long end; | |
551 | pte_t pte; | |
552 | hva_handler_t handler; | |
f922bd9b | 553 | on_lock_fn_t on_lock; |
683412cc | 554 | on_unlock_fn_t on_unlock; |
3039bcc7 SC |
555 | bool flush_on_ret; |
556 | bool may_block; | |
557 | }; | |
558 | ||
f922bd9b SC |
559 | /* |
560 | * Use a dedicated stub instead of NULL to indicate that there is no callback | |
561 | * function/handler. The compiler technically can't guarantee that a real | |
562 | * function will have a non-zero address, and so it will generate code to | |
563 | * check for !NULL, whereas comparing against a stub will be elided at compile | |
564 | * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9). | |
565 | */ | |
566 | static void kvm_null_fn(void) | |
567 | { | |
568 | ||
569 | } | |
570 | #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn) | |
571 | ||
ed922739 MS |
572 | /* Iterate over each memslot intersecting [start, last] (inclusive) range */ |
573 | #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \ | |
574 | for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \ | |
575 | node; \ | |
576 | node = interval_tree_iter_next(node, start, last)) \ | |
577 | ||
3039bcc7 SC |
578 | static __always_inline int __kvm_handle_hva_range(struct kvm *kvm, |
579 | const struct kvm_hva_range *range) | |
580 | { | |
8931a454 | 581 | bool ret = false, locked = false; |
f922bd9b | 582 | struct kvm_gfn_range gfn_range; |
3039bcc7 SC |
583 | struct kvm_memory_slot *slot; |
584 | struct kvm_memslots *slots; | |
3039bcc7 SC |
585 | int i, idx; |
586 | ||
ed922739 MS |
587 | if (WARN_ON_ONCE(range->end <= range->start)) |
588 | return 0; | |
589 | ||
f922bd9b SC |
590 | /* A null handler is allowed if and only if on_lock() is provided. */ |
591 | if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) && | |
592 | IS_KVM_NULL_FN(range->handler))) | |
593 | return 0; | |
594 | ||
3039bcc7 SC |
595 | idx = srcu_read_lock(&kvm->srcu); |
596 | ||
597 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
ed922739 MS |
598 | struct interval_tree_node *node; |
599 | ||
3039bcc7 | 600 | slots = __kvm_memslots(kvm, i); |
ed922739 MS |
601 | kvm_for_each_memslot_in_hva_range(node, slots, |
602 | range->start, range->end - 1) { | |
3039bcc7 SC |
603 | unsigned long hva_start, hva_end; |
604 | ||
a54d8066 | 605 | slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]); |
3039bcc7 SC |
606 | hva_start = max(range->start, slot->userspace_addr); |
607 | hva_end = min(range->end, slot->userspace_addr + | |
608 | (slot->npages << PAGE_SHIFT)); | |
3039bcc7 SC |
609 | |
610 | /* | |
611 | * To optimize for the likely case where the address | |
612 | * range is covered by zero or one memslots, don't | |
613 | * bother making these conditional (to avoid writes on | |
614 | * the second or later invocation of the handler). | |
615 | */ | |
616 | gfn_range.pte = range->pte; | |
617 | gfn_range.may_block = range->may_block; | |
618 | ||
619 | /* | |
620 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
621 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
622 | */ | |
623 | gfn_range.start = hva_to_gfn_memslot(hva_start, slot); | |
624 | gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot); | |
625 | gfn_range.slot = slot; | |
626 | ||
8931a454 SC |
627 | if (!locked) { |
628 | locked = true; | |
629 | KVM_MMU_LOCK(kvm); | |
071064f1 PB |
630 | if (!IS_KVM_NULL_FN(range->on_lock)) |
631 | range->on_lock(kvm, range->start, range->end); | |
632 | if (IS_KVM_NULL_FN(range->handler)) | |
633 | break; | |
8931a454 | 634 | } |
3039bcc7 SC |
635 | ret |= range->handler(kvm, &gfn_range); |
636 | } | |
637 | } | |
638 | ||
6bc6db00 | 639 | if (range->flush_on_ret && ret) |
3039bcc7 SC |
640 | kvm_flush_remote_tlbs(kvm); |
641 | ||
683412cc | 642 | if (locked) { |
8931a454 | 643 | KVM_MMU_UNLOCK(kvm); |
683412cc MZ |
644 | if (!IS_KVM_NULL_FN(range->on_unlock)) |
645 | range->on_unlock(kvm); | |
646 | } | |
f922bd9b | 647 | |
3039bcc7 SC |
648 | srcu_read_unlock(&kvm->srcu, idx); |
649 | ||
650 | /* The notifiers are averse to booleans. :-( */ | |
651 | return (int)ret; | |
652 | } | |
653 | ||
654 | static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn, | |
655 | unsigned long start, | |
656 | unsigned long end, | |
657 | pte_t pte, | |
658 | hva_handler_t handler) | |
659 | { | |
660 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
661 | const struct kvm_hva_range range = { | |
662 | .start = start, | |
663 | .end = end, | |
664 | .pte = pte, | |
665 | .handler = handler, | |
f922bd9b | 666 | .on_lock = (void *)kvm_null_fn, |
683412cc | 667 | .on_unlock = (void *)kvm_null_fn, |
3039bcc7 SC |
668 | .flush_on_ret = true, |
669 | .may_block = false, | |
670 | }; | |
3039bcc7 | 671 | |
f922bd9b | 672 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 SC |
673 | } |
674 | ||
675 | static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn, | |
676 | unsigned long start, | |
677 | unsigned long end, | |
678 | hva_handler_t handler) | |
679 | { | |
680 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
681 | const struct kvm_hva_range range = { | |
682 | .start = start, | |
683 | .end = end, | |
684 | .pte = __pte(0), | |
685 | .handler = handler, | |
f922bd9b | 686 | .on_lock = (void *)kvm_null_fn, |
683412cc | 687 | .on_unlock = (void *)kvm_null_fn, |
3039bcc7 SC |
688 | .flush_on_ret = false, |
689 | .may_block = false, | |
690 | }; | |
3039bcc7 | 691 | |
f922bd9b | 692 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 | 693 | } |
3da0dd43 IE |
694 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
695 | struct mm_struct *mm, | |
696 | unsigned long address, | |
697 | pte_t pte) | |
698 | { | |
699 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
700 | ||
501b9185 SC |
701 | trace_kvm_set_spte_hva(address); |
702 | ||
c13fda23 | 703 | /* |
52ac8b35 | 704 | * .change_pte() must be surrounded by .invalidate_range_{start,end}(). |
071064f1 PB |
705 | * If mmu_notifier_count is zero, then no in-progress invalidations, |
706 | * including this one, found a relevant memslot at start(); rechecking | |
707 | * memslots here is unnecessary. Note, a false positive (count elevated | |
708 | * by a different invalidation) is sub-optimal but functionally ok. | |
c13fda23 | 709 | */ |
52ac8b35 | 710 | WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count)); |
071064f1 PB |
711 | if (!READ_ONCE(kvm->mmu_notifier_count)) |
712 | return; | |
c13fda23 | 713 | |
3039bcc7 | 714 | kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn); |
3da0dd43 IE |
715 | } |
716 | ||
edb298c6 | 717 | void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 718 | unsigned long end) |
e930bffe | 719 | { |
e930bffe AA |
720 | /* |
721 | * The count increase must become visible at unlock time as no | |
722 | * spte can be established without taking the mmu_lock and | |
723 | * count is also read inside the mmu_lock critical section. | |
724 | */ | |
725 | kvm->mmu_notifier_count++; | |
4a42d848 | 726 | if (likely(kvm->mmu_notifier_count == 1)) { |
f922bd9b SC |
727 | kvm->mmu_notifier_range_start = start; |
728 | kvm->mmu_notifier_range_end = end; | |
4a42d848 DS |
729 | } else { |
730 | /* | |
a413a625 | 731 | * Fully tracking multiple concurrent ranges has diminishing |
4a42d848 DS |
732 | * returns. Keep things simple and just find the minimal range |
733 | * which includes the current and new ranges. As there won't be | |
734 | * enough information to subtract a range after its invalidate | |
735 | * completes, any ranges invalidated concurrently will | |
736 | * accumulate and persist until all outstanding invalidates | |
737 | * complete. | |
738 | */ | |
739 | kvm->mmu_notifier_range_start = | |
f922bd9b | 740 | min(kvm->mmu_notifier_range_start, start); |
4a42d848 | 741 | kvm->mmu_notifier_range_end = |
f922bd9b | 742 | max(kvm->mmu_notifier_range_end, end); |
4a42d848 | 743 | } |
f922bd9b | 744 | } |
3039bcc7 | 745 | |
f922bd9b SC |
746 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
747 | const struct mmu_notifier_range *range) | |
748 | { | |
749 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
750 | const struct kvm_hva_range hva_range = { | |
751 | .start = range->start, | |
752 | .end = range->end, | |
753 | .pte = __pte(0), | |
754 | .handler = kvm_unmap_gfn_range, | |
755 | .on_lock = kvm_inc_notifier_count, | |
683412cc | 756 | .on_unlock = kvm_arch_guest_memory_reclaimed, |
f922bd9b SC |
757 | .flush_on_ret = true, |
758 | .may_block = mmu_notifier_range_blockable(range), | |
759 | }; | |
565f3be2 | 760 | |
f922bd9b SC |
761 | trace_kvm_unmap_hva_range(range->start, range->end); |
762 | ||
52ac8b35 PB |
763 | /* |
764 | * Prevent memslot modification between range_start() and range_end() | |
765 | * so that conditionally locking provides the same result in both | |
766 | * functions. Without that guarantee, the mmu_notifier_count | |
767 | * adjustments will be imbalanced. | |
768 | * | |
769 | * Pairs with the decrement in range_end(). | |
770 | */ | |
771 | spin_lock(&kvm->mn_invalidate_lock); | |
772 | kvm->mn_active_invalidate_count++; | |
773 | spin_unlock(&kvm->mn_invalidate_lock); | |
774 | ||
58cd407c SC |
775 | /* |
776 | * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e. | |
777 | * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring | |
778 | * each cache's lock. There are relatively few caches in existence at | |
779 | * any given time, and the caches themselves can check for hva overlap, | |
780 | * i.e. don't need to rely on memslot overlap checks for performance. | |
781 | * Because this runs without holding mmu_lock, the pfn caches must use | |
782 | * mn_active_invalidate_count (see above) instead of mmu_notifier_count. | |
783 | */ | |
982ed0de DW |
784 | gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end, |
785 | hva_range.may_block); | |
786 | ||
f922bd9b | 787 | __kvm_handle_hva_range(kvm, &hva_range); |
93065ac7 | 788 | |
e649b3f0 | 789 | return 0; |
e930bffe AA |
790 | } |
791 | ||
edb298c6 | 792 | void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 793 | unsigned long end) |
e930bffe | 794 | { |
e930bffe AA |
795 | /* |
796 | * This sequence increase will notify the kvm page fault that | |
797 | * the page that is going to be mapped in the spte could have | |
798 | * been freed. | |
799 | */ | |
800 | kvm->mmu_notifier_seq++; | |
a355aa54 | 801 | smp_wmb(); |
e930bffe AA |
802 | /* |
803 | * The above sequence increase must be visible before the | |
a355aa54 PM |
804 | * below count decrease, which is ensured by the smp_wmb above |
805 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
806 | */ |
807 | kvm->mmu_notifier_count--; | |
f922bd9b SC |
808 | } |
809 | ||
810 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
811 | const struct mmu_notifier_range *range) | |
812 | { | |
813 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
814 | const struct kvm_hva_range hva_range = { | |
815 | .start = range->start, | |
816 | .end = range->end, | |
817 | .pte = __pte(0), | |
818 | .handler = (void *)kvm_null_fn, | |
819 | .on_lock = kvm_dec_notifier_count, | |
683412cc | 820 | .on_unlock = (void *)kvm_null_fn, |
f922bd9b SC |
821 | .flush_on_ret = false, |
822 | .may_block = mmu_notifier_range_blockable(range), | |
823 | }; | |
52ac8b35 | 824 | bool wake; |
f922bd9b SC |
825 | |
826 | __kvm_handle_hva_range(kvm, &hva_range); | |
e930bffe | 827 | |
52ac8b35 PB |
828 | /* Pairs with the increment in range_start(). */ |
829 | spin_lock(&kvm->mn_invalidate_lock); | |
830 | wake = (--kvm->mn_active_invalidate_count == 0); | |
831 | spin_unlock(&kvm->mn_invalidate_lock); | |
832 | ||
833 | /* | |
834 | * There can only be one waiter, since the wait happens under | |
835 | * slots_lock. | |
836 | */ | |
837 | if (wake) | |
838 | rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait); | |
839 | ||
e930bffe AA |
840 | BUG_ON(kvm->mmu_notifier_count < 0); |
841 | } | |
842 | ||
843 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
844 | struct mm_struct *mm, | |
57128468 ALC |
845 | unsigned long start, |
846 | unsigned long end) | |
e930bffe | 847 | { |
501b9185 SC |
848 | trace_kvm_age_hva(start, end); |
849 | ||
3039bcc7 | 850 | return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn); |
e930bffe AA |
851 | } |
852 | ||
1d7715c6 VD |
853 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
854 | struct mm_struct *mm, | |
855 | unsigned long start, | |
856 | unsigned long end) | |
857 | { | |
501b9185 SC |
858 | trace_kvm_age_hva(start, end); |
859 | ||
1d7715c6 VD |
860 | /* |
861 | * Even though we do not flush TLB, this will still adversely | |
862 | * affect performance on pre-Haswell Intel EPT, where there is | |
863 | * no EPT Access Bit to clear so that we have to tear down EPT | |
864 | * tables instead. If we find this unacceptable, we can always | |
865 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
866 | * do anything on clear_young. | |
867 | * | |
868 | * Also note that currently we never issue secondary TLB flushes | |
869 | * from clear_young, leaving this job up to the regular system | |
870 | * cadence. If we find this inaccurate, we might come up with a | |
871 | * more sophisticated heuristic later. | |
872 | */ | |
3039bcc7 | 873 | return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn); |
1d7715c6 VD |
874 | } |
875 | ||
8ee53820 AA |
876 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
877 | struct mm_struct *mm, | |
878 | unsigned long address) | |
879 | { | |
501b9185 SC |
880 | trace_kvm_test_age_hva(address); |
881 | ||
3039bcc7 SC |
882 | return kvm_handle_hva_range_no_flush(mn, address, address + 1, |
883 | kvm_test_age_gfn); | |
8ee53820 AA |
884 | } |
885 | ||
85db06e5 MT |
886 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
887 | struct mm_struct *mm) | |
888 | { | |
889 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
890 | int idx; |
891 | ||
892 | idx = srcu_read_lock(&kvm->srcu); | |
683412cc | 893 | kvm_flush_shadow_all(kvm); |
eda2beda | 894 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
895 | } |
896 | ||
e930bffe | 897 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e649b3f0 | 898 | .invalidate_range = kvm_mmu_notifier_invalidate_range, |
e930bffe AA |
899 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
900 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
901 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 902 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 903 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 904 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 905 | .release = kvm_mmu_notifier_release, |
e930bffe | 906 | }; |
4c07b0a4 AK |
907 | |
908 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
909 | { | |
910 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
911 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
912 | } | |
913 | ||
914 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
915 | ||
916 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
917 | { | |
918 | return 0; | |
919 | } | |
920 | ||
e930bffe AA |
921 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
922 | ||
2fdef3a2 SS |
923 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
924 | static int kvm_pm_notifier_call(struct notifier_block *bl, | |
925 | unsigned long state, | |
926 | void *unused) | |
927 | { | |
928 | struct kvm *kvm = container_of(bl, struct kvm, pm_notifier); | |
929 | ||
930 | return kvm_arch_pm_notifier(kvm, state); | |
931 | } | |
932 | ||
933 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
934 | { | |
935 | kvm->pm_notifier.notifier_call = kvm_pm_notifier_call; | |
936 | /* Suspend KVM before we suspend ftrace, RCU, etc. */ | |
937 | kvm->pm_notifier.priority = INT_MAX; | |
938 | register_pm_notifier(&kvm->pm_notifier); | |
939 | } | |
940 | ||
941 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
942 | { | |
943 | unregister_pm_notifier(&kvm->pm_notifier); | |
944 | } | |
945 | #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
946 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
947 | { | |
948 | } | |
949 | ||
950 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
951 | { | |
952 | } | |
953 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
954 | ||
a47d2b07 PB |
955 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
956 | { | |
957 | if (!memslot->dirty_bitmap) | |
958 | return; | |
959 | ||
960 | kvfree(memslot->dirty_bitmap); | |
961 | memslot->dirty_bitmap = NULL; | |
962 | } | |
963 | ||
a54d8066 | 964 | /* This does not remove the slot from struct kvm_memslots data structures */ |
e96c81ee | 965 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 966 | { |
e96c81ee | 967 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 968 | |
e96c81ee | 969 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 970 | |
a54d8066 | 971 | kfree(slot); |
a47d2b07 PB |
972 | } |
973 | ||
974 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
975 | { | |
a54d8066 | 976 | struct hlist_node *idnode; |
a47d2b07 | 977 | struct kvm_memory_slot *memslot; |
a54d8066 | 978 | int bkt; |
a47d2b07 | 979 | |
a54d8066 MS |
980 | /* |
981 | * The same memslot objects live in both active and inactive sets, | |
982 | * arbitrarily free using index '1' so the second invocation of this | |
983 | * function isn't operating over a structure with dangling pointers | |
984 | * (even though this function isn't actually touching them). | |
985 | */ | |
986 | if (!slots->node_idx) | |
a47d2b07 PB |
987 | return; |
988 | ||
a54d8066 | 989 | hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1]) |
e96c81ee | 990 | kvm_free_memslot(kvm, memslot); |
bf3e05bc XG |
991 | } |
992 | ||
bc9e9e67 JZ |
993 | static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc) |
994 | { | |
995 | switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) { | |
996 | case KVM_STATS_TYPE_INSTANT: | |
997 | return 0444; | |
998 | case KVM_STATS_TYPE_CUMULATIVE: | |
999 | case KVM_STATS_TYPE_PEAK: | |
1000 | default: | |
1001 | return 0644; | |
1002 | } | |
1003 | } | |
1004 | ||
1005 | ||
536a6f88 JF |
1006 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
1007 | { | |
1008 | int i; | |
bc9e9e67 JZ |
1009 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
1010 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 | 1011 | |
a44a4cc1 | 1012 | if (IS_ERR(kvm->debugfs_dentry)) |
536a6f88 JF |
1013 | return; |
1014 | ||
1015 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
1016 | ||
9d5a1dce LC |
1017 | if (kvm->debugfs_stat_data) { |
1018 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
1019 | kfree(kvm->debugfs_stat_data[i]); | |
1020 | kfree(kvm->debugfs_stat_data); | |
1021 | } | |
536a6f88 JF |
1022 | } |
1023 | ||
59f82aad | 1024 | static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname) |
536a6f88 | 1025 | { |
85cd39af PB |
1026 | static DEFINE_MUTEX(kvm_debugfs_lock); |
1027 | struct dentry *dent; | |
536a6f88 JF |
1028 | char dir_name[ITOA_MAX_LEN * 2]; |
1029 | struct kvm_stat_data *stat_data; | |
bc9e9e67 | 1030 | const struct _kvm_stats_desc *pdesc; |
3165af73 | 1031 | int i, ret; |
bc9e9e67 JZ |
1032 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
1033 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
1034 | |
1035 | if (!debugfs_initialized()) | |
1036 | return 0; | |
1037 | ||
59f82aad | 1038 | snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname); |
85cd39af PB |
1039 | mutex_lock(&kvm_debugfs_lock); |
1040 | dent = debugfs_lookup(dir_name, kvm_debugfs_dir); | |
1041 | if (dent) { | |
1042 | pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name); | |
1043 | dput(dent); | |
1044 | mutex_unlock(&kvm_debugfs_lock); | |
1045 | return 0; | |
1046 | } | |
1047 | dent = debugfs_create_dir(dir_name, kvm_debugfs_dir); | |
1048 | mutex_unlock(&kvm_debugfs_lock); | |
1049 | if (IS_ERR(dent)) | |
1050 | return 0; | |
536a6f88 | 1051 | |
85cd39af | 1052 | kvm->debugfs_dentry = dent; |
536a6f88 JF |
1053 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
1054 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 1055 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
1056 | if (!kvm->debugfs_stat_data) |
1057 | return -ENOMEM; | |
1058 | ||
bc9e9e67 JZ |
1059 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
1060 | pdesc = &kvm_vm_stats_desc[i]; | |
b12ce36a | 1061 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
1062 | if (!stat_data) |
1063 | return -ENOMEM; | |
1064 | ||
1065 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1066 | stat_data->desc = pdesc; |
1067 | stat_data->kind = KVM_STAT_VM; | |
1068 | kvm->debugfs_stat_data[i] = stat_data; | |
1069 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
1070 | kvm->debugfs_dentry, stat_data, | |
1071 | &stat_fops_per_vm); | |
1072 | } | |
1073 | ||
1074 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
1075 | pdesc = &kvm_vcpu_stats_desc[i]; | |
b12ce36a | 1076 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
1077 | if (!stat_data) |
1078 | return -ENOMEM; | |
1079 | ||
1080 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1081 | stat_data->desc = pdesc; |
1082 | stat_data->kind = KVM_STAT_VCPU; | |
004d62eb | 1083 | kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data; |
bc9e9e67 | 1084 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), |
09cbcef6 MP |
1085 | kvm->debugfs_dentry, stat_data, |
1086 | &stat_fops_per_vm); | |
536a6f88 | 1087 | } |
3165af73 PX |
1088 | |
1089 | ret = kvm_arch_create_vm_debugfs(kvm); | |
1090 | if (ret) { | |
1091 | kvm_destroy_vm_debugfs(kvm); | |
1092 | return i; | |
1093 | } | |
1094 | ||
536a6f88 JF |
1095 | return 0; |
1096 | } | |
1097 | ||
1aa9b957 JS |
1098 | /* |
1099 | * Called after the VM is otherwise initialized, but just before adding it to | |
1100 | * the vm_list. | |
1101 | */ | |
1102 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
1103 | { | |
1104 | return 0; | |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * Called just after removing the VM from the vm_list, but before doing any | |
1109 | * other destruction. | |
1110 | */ | |
1111 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
1112 | { | |
1113 | } | |
1114 | ||
3165af73 PX |
1115 | /* |
1116 | * Called after per-vm debugfs created. When called kvm->debugfs_dentry should | |
1117 | * be setup already, so we can create arch-specific debugfs entries under it. | |
1118 | * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so | |
1119 | * a per-arch destroy interface is not needed. | |
1120 | */ | |
1121 | int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm) | |
1122 | { | |
1123 | return 0; | |
1124 | } | |
1125 | ||
e08b9637 | 1126 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 1127 | { |
d89f5eff | 1128 | struct kvm *kvm = kvm_arch_alloc_vm(); |
a54d8066 | 1129 | struct kvm_memslots *slots; |
9121923c | 1130 | int r = -ENOMEM; |
a54d8066 | 1131 | int i, j; |
6aa8b732 | 1132 | |
d89f5eff JK |
1133 | if (!kvm) |
1134 | return ERR_PTR(-ENOMEM); | |
1135 | ||
531810ca | 1136 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 1137 | mmgrab(current->mm); |
e9ad4ec8 PB |
1138 | kvm->mm = current->mm; |
1139 | kvm_eventfd_init(kvm); | |
1140 | mutex_init(&kvm->lock); | |
1141 | mutex_init(&kvm->irq_lock); | |
1142 | mutex_init(&kvm->slots_lock); | |
b10a038e | 1143 | mutex_init(&kvm->slots_arch_lock); |
52ac8b35 PB |
1144 | spin_lock_init(&kvm->mn_invalidate_lock); |
1145 | rcuwait_init(&kvm->mn_memslots_update_rcuwait); | |
c5b07754 | 1146 | xa_init(&kvm->vcpu_array); |
52ac8b35 | 1147 | |
982ed0de DW |
1148 | INIT_LIST_HEAD(&kvm->gpc_list); |
1149 | spin_lock_init(&kvm->gpc_lock); | |
52ac8b35 | 1150 | |
e9ad4ec8 | 1151 | INIT_LIST_HEAD(&kvm->devices); |
f502cc56 | 1152 | kvm->max_vcpus = KVM_MAX_VCPUS; |
e9ad4ec8 | 1153 | |
1e702d9a AW |
1154 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
1155 | ||
5c697c36 SC |
1156 | /* |
1157 | * Force subsequent debugfs file creations to fail if the VM directory | |
1158 | * is not created (by kvm_create_vm_debugfs()). | |
1159 | */ | |
1160 | kvm->debugfs_dentry = ERR_PTR(-ENOENT); | |
1161 | ||
f2759c08 OU |
1162 | snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d", |
1163 | task_pid_nr(current)); | |
1164 | ||
8a44119a PB |
1165 | if (init_srcu_struct(&kvm->srcu)) |
1166 | goto out_err_no_srcu; | |
1167 | if (init_srcu_struct(&kvm->irq_srcu)) | |
1168 | goto out_err_no_irq_srcu; | |
1169 | ||
e2d3fcaf | 1170 | refcount_set(&kvm->users_count, 1); |
f481b069 | 1171 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
a54d8066 MS |
1172 | for (j = 0; j < 2; j++) { |
1173 | slots = &kvm->__memslots[i][j]; | |
9121923c | 1174 | |
a54d8066 MS |
1175 | atomic_long_set(&slots->last_used_slot, (unsigned long)NULL); |
1176 | slots->hva_tree = RB_ROOT_CACHED; | |
1177 | slots->gfn_tree = RB_ROOT; | |
1178 | hash_init(slots->id_hash); | |
1179 | slots->node_idx = j; | |
1180 | ||
1181 | /* Generations must be different for each address space. */ | |
1182 | slots->generation = i; | |
1183 | } | |
1184 | ||
1185 | rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]); | |
f481b069 | 1186 | } |
00f034a1 | 1187 | |
e93f8a0f | 1188 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 1189 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 1190 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 1191 | if (!kvm->buses[i]) |
a97b0e77 | 1192 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 1193 | } |
e930bffe | 1194 | |
acd05785 DM |
1195 | kvm->max_halt_poll_ns = halt_poll_ns; |
1196 | ||
e08b9637 | 1197 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 1198 | if (r) |
a97b0e77 | 1199 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
1200 | |
1201 | r = hardware_enable_all(); | |
1202 | if (r) | |
719d93cd | 1203 | goto out_err_no_disable; |
10474ae8 | 1204 | |
c77dcacb | 1205 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 1206 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 1207 | #endif |
6aa8b732 | 1208 | |
74b5c5bf | 1209 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
1210 | if (r) |
1211 | goto out_err_no_mmu_notifier; | |
1212 | ||
1213 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
1214 | if (r) |
1215 | goto out_err; | |
1216 | ||
0d9ce162 | 1217 | mutex_lock(&kvm_lock); |
5e58cfe4 | 1218 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 1219 | mutex_unlock(&kvm_lock); |
d89f5eff | 1220 | |
2ecd9d29 | 1221 | preempt_notifier_inc(); |
2fdef3a2 | 1222 | kvm_init_pm_notifier(kvm); |
2ecd9d29 | 1223 | |
5f6de5cb DM |
1224 | /* |
1225 | * When the fd passed to this ioctl() is opened it pins the module, | |
1226 | * but try_module_get() also prevents getting a reference if the module | |
1227 | * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait"). | |
1228 | */ | |
1229 | if (!try_module_get(kvm_chardev_ops.owner)) { | |
1230 | r = -ENODEV; | |
1231 | goto out_err; | |
1232 | } | |
1233 | ||
f17abe9a | 1234 | return kvm; |
10474ae8 AG |
1235 | |
1236 | out_err: | |
1aa9b957 JS |
1237 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1238 | if (kvm->mmu_notifier.ops) | |
1239 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
1240 | #endif | |
1241 | out_err_no_mmu_notifier: | |
10474ae8 | 1242 | hardware_disable_all(); |
719d93cd | 1243 | out_err_no_disable: |
a97b0e77 | 1244 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 1245 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 1246 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 1247 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 1248 | kfree(kvm_get_bus(kvm, i)); |
8a44119a PB |
1249 | cleanup_srcu_struct(&kvm->irq_srcu); |
1250 | out_err_no_irq_srcu: | |
1251 | cleanup_srcu_struct(&kvm->srcu); | |
1252 | out_err_no_srcu: | |
d89f5eff | 1253 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 1254 | mmdrop(current->mm); |
10474ae8 | 1255 | return ERR_PTR(r); |
f17abe9a AK |
1256 | } |
1257 | ||
07f0a7bd SW |
1258 | static void kvm_destroy_devices(struct kvm *kvm) |
1259 | { | |
e6e3b5a6 | 1260 | struct kvm_device *dev, *tmp; |
07f0a7bd | 1261 | |
a28ebea2 CD |
1262 | /* |
1263 | * We do not need to take the kvm->lock here, because nobody else | |
1264 | * has a reference to the struct kvm at this point and therefore | |
1265 | * cannot access the devices list anyhow. | |
1266 | */ | |
e6e3b5a6 GT |
1267 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
1268 | list_del(&dev->vm_node); | |
07f0a7bd SW |
1269 | dev->ops->destroy(dev); |
1270 | } | |
1271 | } | |
1272 | ||
f17abe9a AK |
1273 | static void kvm_destroy_vm(struct kvm *kvm) |
1274 | { | |
e93f8a0f | 1275 | int i; |
6d4e4c4f AK |
1276 | struct mm_struct *mm = kvm->mm; |
1277 | ||
2fdef3a2 | 1278 | kvm_destroy_pm_notifier(kvm); |
286de8f6 | 1279 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 1280 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 1281 | kvm_arch_sync_events(kvm); |
0d9ce162 | 1282 | mutex_lock(&kvm_lock); |
133de902 | 1283 | list_del(&kvm->vm_list); |
0d9ce162 | 1284 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
1285 | kvm_arch_pre_destroy_vm(kvm); |
1286 | ||
399ec807 | 1287 | kvm_free_irq_routing(kvm); |
df630b8c | 1288 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 1289 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 1290 | |
4a12f951 CB |
1291 | if (bus) |
1292 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
1293 | kvm->buses[i] = NULL; |
1294 | } | |
980da6ce | 1295 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
1296 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1297 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
52ac8b35 PB |
1298 | /* |
1299 | * At this point, pending calls to invalidate_range_start() | |
1300 | * have completed but no more MMU notifiers will run, so | |
1301 | * mn_active_invalidate_count may remain unbalanced. | |
1302 | * No threads can be waiting in install_new_memslots as the | |
1303 | * last reference on KVM has been dropped, but freeing | |
1304 | * memslots would deadlock without this manual intervention. | |
1305 | */ | |
1306 | WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait)); | |
1307 | kvm->mn_active_invalidate_count = 0; | |
f00be0ca | 1308 | #else |
683412cc | 1309 | kvm_flush_shadow_all(kvm); |
5f94c174 | 1310 | #endif |
d19a9cd2 | 1311 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 1312 | kvm_destroy_devices(kvm); |
a54d8066 MS |
1313 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
1314 | kvm_free_memslots(kvm, &kvm->__memslots[i][0]); | |
1315 | kvm_free_memslots(kvm, &kvm->__memslots[i][1]); | |
1316 | } | |
820b3fcd | 1317 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
1318 | cleanup_srcu_struct(&kvm->srcu); |
1319 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 1320 | preempt_notifier_dec(); |
10474ae8 | 1321 | hardware_disable_all(); |
6d4e4c4f | 1322 | mmdrop(mm); |
5f6de5cb | 1323 | module_put(kvm_chardev_ops.owner); |
f17abe9a AK |
1324 | } |
1325 | ||
d39f13b0 IE |
1326 | void kvm_get_kvm(struct kvm *kvm) |
1327 | { | |
e3736c3e | 1328 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
1329 | } |
1330 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
1331 | ||
605c7130 PX |
1332 | /* |
1333 | * Make sure the vm is not during destruction, which is a safe version of | |
1334 | * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise. | |
1335 | */ | |
1336 | bool kvm_get_kvm_safe(struct kvm *kvm) | |
1337 | { | |
1338 | return refcount_inc_not_zero(&kvm->users_count); | |
1339 | } | |
1340 | EXPORT_SYMBOL_GPL(kvm_get_kvm_safe); | |
1341 | ||
d39f13b0 IE |
1342 | void kvm_put_kvm(struct kvm *kvm) |
1343 | { | |
e3736c3e | 1344 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
1345 | kvm_destroy_vm(kvm); |
1346 | } | |
1347 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
1348 | ||
149487bd SC |
1349 | /* |
1350 | * Used to put a reference that was taken on behalf of an object associated | |
1351 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
1352 | * of the new file descriptor fails and the reference cannot be transferred to | |
1353 | * its final owner. In such cases, the caller is still actively using @kvm and | |
1354 | * will fail miserably if the refcount unexpectedly hits zero. | |
1355 | */ | |
1356 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
1357 | { | |
1358 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
1359 | } | |
1360 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 1361 | |
f17abe9a AK |
1362 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
1363 | { | |
1364 | struct kvm *kvm = filp->private_data; | |
1365 | ||
721eecbf GH |
1366 | kvm_irqfd_release(kvm); |
1367 | ||
d39f13b0 | 1368 | kvm_put_kvm(kvm); |
6aa8b732 AK |
1369 | return 0; |
1370 | } | |
1371 | ||
515a0127 TY |
1372 | /* |
1373 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 1374 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 1375 | */ |
3c9bd400 | 1376 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 1377 | { |
37b2a651 | 1378 | unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 1379 | |
37b2a651 | 1380 | memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
1381 | if (!memslot->dirty_bitmap) |
1382 | return -ENOMEM; | |
1383 | ||
a36a57b1 TY |
1384 | return 0; |
1385 | } | |
1386 | ||
a54d8066 | 1387 | static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id) |
bf3e05bc | 1388 | { |
a54d8066 MS |
1389 | struct kvm_memslots *active = __kvm_memslots(kvm, as_id); |
1390 | int node_idx_inactive = active->node_idx ^ 1; | |
0e60b079 | 1391 | |
a54d8066 | 1392 | return &kvm->__memslots[as_id][node_idx_inactive]; |
0577d1ab SC |
1393 | } |
1394 | ||
1395 | /* | |
a54d8066 MS |
1396 | * Helper to get the address space ID when one of memslot pointers may be NULL. |
1397 | * This also serves as a sanity that at least one of the pointers is non-NULL, | |
1398 | * and that their address space IDs don't diverge. | |
0577d1ab | 1399 | */ |
a54d8066 MS |
1400 | static int kvm_memslots_get_as_id(struct kvm_memory_slot *a, |
1401 | struct kvm_memory_slot *b) | |
0577d1ab | 1402 | { |
a54d8066 MS |
1403 | if (WARN_ON_ONCE(!a && !b)) |
1404 | return 0; | |
0577d1ab | 1405 | |
a54d8066 MS |
1406 | if (!a) |
1407 | return b->as_id; | |
1408 | if (!b) | |
1409 | return a->as_id; | |
0577d1ab | 1410 | |
a54d8066 MS |
1411 | WARN_ON_ONCE(a->as_id != b->as_id); |
1412 | return a->as_id; | |
0577d1ab | 1413 | } |
efbeec70 | 1414 | |
a54d8066 MS |
1415 | static void kvm_insert_gfn_node(struct kvm_memslots *slots, |
1416 | struct kvm_memory_slot *slot) | |
0577d1ab | 1417 | { |
a54d8066 MS |
1418 | struct rb_root *gfn_tree = &slots->gfn_tree; |
1419 | struct rb_node **node, *parent; | |
1420 | int idx = slots->node_idx; | |
0577d1ab | 1421 | |
a54d8066 MS |
1422 | parent = NULL; |
1423 | for (node = &gfn_tree->rb_node; *node; ) { | |
1424 | struct kvm_memory_slot *tmp; | |
f85e2cb5 | 1425 | |
a54d8066 MS |
1426 | tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]); |
1427 | parent = *node; | |
1428 | if (slot->base_gfn < tmp->base_gfn) | |
1429 | node = &(*node)->rb_left; | |
1430 | else if (slot->base_gfn > tmp->base_gfn) | |
1431 | node = &(*node)->rb_right; | |
1432 | else | |
1433 | BUG(); | |
0577d1ab | 1434 | } |
a54d8066 MS |
1435 | |
1436 | rb_link_node(&slot->gfn_node[idx], parent, node); | |
1437 | rb_insert_color(&slot->gfn_node[idx], gfn_tree); | |
0577d1ab SC |
1438 | } |
1439 | ||
a54d8066 MS |
1440 | static void kvm_erase_gfn_node(struct kvm_memslots *slots, |
1441 | struct kvm_memory_slot *slot) | |
0577d1ab | 1442 | { |
a54d8066 MS |
1443 | rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree); |
1444 | } | |
0577d1ab | 1445 | |
a54d8066 MS |
1446 | static void kvm_replace_gfn_node(struct kvm_memslots *slots, |
1447 | struct kvm_memory_slot *old, | |
1448 | struct kvm_memory_slot *new) | |
1449 | { | |
1450 | int idx = slots->node_idx; | |
0577d1ab | 1451 | |
a54d8066 | 1452 | WARN_ON_ONCE(old->base_gfn != new->base_gfn); |
0577d1ab | 1453 | |
a54d8066 MS |
1454 | rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx], |
1455 | &slots->gfn_tree); | |
0577d1ab SC |
1456 | } |
1457 | ||
1458 | /* | |
a54d8066 | 1459 | * Replace @old with @new in the inactive memslots. |
0577d1ab | 1460 | * |
a54d8066 MS |
1461 | * With NULL @old this simply adds @new. |
1462 | * With NULL @new this simply removes @old. | |
0577d1ab | 1463 | * |
a54d8066 MS |
1464 | * If @new is non-NULL its hva_node[slots_idx] range has to be set |
1465 | * appropriately. | |
0577d1ab | 1466 | */ |
a54d8066 MS |
1467 | static void kvm_replace_memslot(struct kvm *kvm, |
1468 | struct kvm_memory_slot *old, | |
1469 | struct kvm_memory_slot *new) | |
0577d1ab | 1470 | { |
a54d8066 MS |
1471 | int as_id = kvm_memslots_get_as_id(old, new); |
1472 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); | |
1473 | int idx = slots->node_idx; | |
0577d1ab | 1474 | |
a54d8066 MS |
1475 | if (old) { |
1476 | hash_del(&old->id_node[idx]); | |
1477 | interval_tree_remove(&old->hva_node[idx], &slots->hva_tree); | |
0577d1ab | 1478 | |
a54d8066 MS |
1479 | if ((long)old == atomic_long_read(&slots->last_used_slot)) |
1480 | atomic_long_set(&slots->last_used_slot, (long)new); | |
0577d1ab | 1481 | |
a54d8066 MS |
1482 | if (!new) { |
1483 | kvm_erase_gfn_node(slots, old); | |
1e8617d3 | 1484 | return; |
a54d8066 MS |
1485 | } |
1486 | } | |
1e8617d3 | 1487 | |
a54d8066 MS |
1488 | /* |
1489 | * Initialize @new's hva range. Do this even when replacing an @old | |
1490 | * slot, kvm_copy_memslot() deliberately does not touch node data. | |
1491 | */ | |
1492 | new->hva_node[idx].start = new->userspace_addr; | |
1493 | new->hva_node[idx].last = new->userspace_addr + | |
1494 | (new->npages << PAGE_SHIFT) - 1; | |
1495 | ||
1496 | /* | |
1497 | * (Re)Add the new memslot. There is no O(1) interval_tree_replace(), | |
1498 | * hva_node needs to be swapped with remove+insert even though hva can't | |
1499 | * change when replacing an existing slot. | |
1500 | */ | |
1501 | hash_add(slots->id_hash, &new->id_node[idx], new->id); | |
1502 | interval_tree_insert(&new->hva_node[idx], &slots->hva_tree); | |
1503 | ||
1504 | /* | |
1505 | * If the memslot gfn is unchanged, rb_replace_node() can be used to | |
1506 | * switch the node in the gfn tree instead of removing the old and | |
1507 | * inserting the new as two separate operations. Replacement is a | |
1508 | * single O(1) operation versus two O(log(n)) operations for | |
1509 | * remove+insert. | |
1510 | */ | |
1511 | if (old && old->base_gfn == new->base_gfn) { | |
1512 | kvm_replace_gfn_node(slots, old, new); | |
1513 | } else { | |
1514 | if (old) | |
1515 | kvm_erase_gfn_node(slots, old); | |
1516 | kvm_insert_gfn_node(slots, new); | |
0577d1ab | 1517 | } |
bf3e05bc XG |
1518 | } |
1519 | ||
09170a49 | 1520 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 1521 | { |
4d8b81ab XG |
1522 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1523 | ||
0f8a4de3 | 1524 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
1525 | valid_flags |= KVM_MEM_READONLY; |
1526 | #endif | |
1527 | ||
1528 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1529 | return -EINVAL; |
1530 | ||
1531 | return 0; | |
1532 | } | |
1533 | ||
a54d8066 | 1534 | static void kvm_swap_active_memslots(struct kvm *kvm, int as_id) |
7ec4fb44 | 1535 | { |
a54d8066 MS |
1536 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); |
1537 | ||
1538 | /* Grab the generation from the activate memslots. */ | |
1539 | u64 gen = __kvm_memslots(kvm, as_id)->generation; | |
7ec4fb44 | 1540 | |
361209e0 SC |
1541 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1542 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1543 | |
52ac8b35 PB |
1544 | /* |
1545 | * Do not store the new memslots while there are invalidations in | |
071064f1 PB |
1546 | * progress, otherwise the locking in invalidate_range_start and |
1547 | * invalidate_range_end will be unbalanced. | |
52ac8b35 PB |
1548 | */ |
1549 | spin_lock(&kvm->mn_invalidate_lock); | |
1550 | prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
1551 | while (kvm->mn_active_invalidate_count) { | |
1552 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1553 | spin_unlock(&kvm->mn_invalidate_lock); | |
1554 | schedule(); | |
1555 | spin_lock(&kvm->mn_invalidate_lock); | |
1556 | } | |
1557 | finish_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
f481b069 | 1558 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
52ac8b35 | 1559 | spin_unlock(&kvm->mn_invalidate_lock); |
b10a038e BG |
1560 | |
1561 | /* | |
1562 | * Acquired in kvm_set_memslot. Must be released before synchronize | |
1563 | * SRCU below in order to avoid deadlock with another thread | |
1564 | * acquiring the slots_arch_lock in an srcu critical section. | |
1565 | */ | |
1566 | mutex_unlock(&kvm->slots_arch_lock); | |
1567 | ||
7ec4fb44 | 1568 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1569 | |
ee3d1570 | 1570 | /* |
361209e0 | 1571 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1572 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1573 | * the number of address spaces. This provides a unique and easily |
1574 | * identifiable generation number while the memslots are in flux. | |
1575 | */ | |
1576 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1577 | ||
1578 | /* | |
4bd518f1 PB |
1579 | * Generations must be unique even across address spaces. We do not need |
1580 | * a global counter for that, instead the generation space is evenly split | |
1581 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1582 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1583 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1584 | */ |
164bf7e5 | 1585 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 1586 | |
15248258 | 1587 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1588 | |
15248258 | 1589 | slots->generation = gen; |
7ec4fb44 GN |
1590 | } |
1591 | ||
07921665 SC |
1592 | static int kvm_prepare_memory_region(struct kvm *kvm, |
1593 | const struct kvm_memory_slot *old, | |
1594 | struct kvm_memory_slot *new, | |
1595 | enum kvm_mr_change change) | |
ddc12f2a | 1596 | { |
07921665 SC |
1597 | int r; |
1598 | ||
1599 | /* | |
1600 | * If dirty logging is disabled, nullify the bitmap; the old bitmap | |
1601 | * will be freed on "commit". If logging is enabled in both old and | |
1602 | * new, reuse the existing bitmap. If logging is enabled only in the | |
1603 | * new and KVM isn't using a ring buffer, allocate and initialize a | |
1604 | * new bitmap. | |
1605 | */ | |
244893fa SC |
1606 | if (change != KVM_MR_DELETE) { |
1607 | if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1608 | new->dirty_bitmap = NULL; | |
1609 | else if (old && old->dirty_bitmap) | |
1610 | new->dirty_bitmap = old->dirty_bitmap; | |
1611 | else if (!kvm->dirty_ring_size) { | |
1612 | r = kvm_alloc_dirty_bitmap(new); | |
1613 | if (r) | |
1614 | return r; | |
1615 | ||
1616 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1617 | bitmap_set(new->dirty_bitmap, 0, new->npages); | |
1618 | } | |
07921665 SC |
1619 | } |
1620 | ||
1621 | r = kvm_arch_prepare_memory_region(kvm, old, new, change); | |
1622 | ||
1623 | /* Free the bitmap on failure if it was allocated above. */ | |
c87661f8 | 1624 | if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap)) |
07921665 SC |
1625 | kvm_destroy_dirty_bitmap(new); |
1626 | ||
1627 | return r; | |
ddc12f2a BG |
1628 | } |
1629 | ||
07921665 SC |
1630 | static void kvm_commit_memory_region(struct kvm *kvm, |
1631 | struct kvm_memory_slot *old, | |
1632 | const struct kvm_memory_slot *new, | |
1633 | enum kvm_mr_change change) | |
ddc12f2a | 1634 | { |
07921665 SC |
1635 | /* |
1636 | * Update the total number of memslot pages before calling the arch | |
1637 | * hook so that architectures can consume the result directly. | |
1638 | */ | |
1639 | if (change == KVM_MR_DELETE) | |
1640 | kvm->nr_memslot_pages -= old->npages; | |
1641 | else if (change == KVM_MR_CREATE) | |
1642 | kvm->nr_memslot_pages += new->npages; | |
1643 | ||
1644 | kvm_arch_commit_memory_region(kvm, old, new, change); | |
1645 | ||
a54d8066 MS |
1646 | switch (change) { |
1647 | case KVM_MR_CREATE: | |
1648 | /* Nothing more to do. */ | |
1649 | break; | |
1650 | case KVM_MR_DELETE: | |
1651 | /* Free the old memslot and all its metadata. */ | |
1652 | kvm_free_memslot(kvm, old); | |
1653 | break; | |
1654 | case KVM_MR_MOVE: | |
1655 | case KVM_MR_FLAGS_ONLY: | |
1656 | /* | |
1657 | * Free the dirty bitmap as needed; the below check encompasses | |
1658 | * both the flags and whether a ring buffer is being used) | |
1659 | */ | |
1660 | if (old->dirty_bitmap && !new->dirty_bitmap) | |
1661 | kvm_destroy_dirty_bitmap(old); | |
1662 | ||
1663 | /* | |
1664 | * The final quirk. Free the detached, old slot, but only its | |
1665 | * memory, not any metadata. Metadata, including arch specific | |
1666 | * data, may be reused by @new. | |
1667 | */ | |
1668 | kfree(old); | |
1669 | break; | |
1670 | default: | |
1671 | BUG(); | |
1672 | } | |
ddc12f2a BG |
1673 | } |
1674 | ||
36947254 | 1675 | /* |
a54d8066 MS |
1676 | * Activate @new, which must be installed in the inactive slots by the caller, |
1677 | * by swapping the active slots and then propagating @new to @old once @old is | |
1678 | * unreachable and can be safely modified. | |
1679 | * | |
1680 | * With NULL @old this simply adds @new to @active (while swapping the sets). | |
1681 | * With NULL @new this simply removes @old from @active and frees it | |
1682 | * (while also swapping the sets). | |
36947254 | 1683 | */ |
a54d8066 MS |
1684 | static void kvm_activate_memslot(struct kvm *kvm, |
1685 | struct kvm_memory_slot *old, | |
1686 | struct kvm_memory_slot *new) | |
36947254 | 1687 | { |
a54d8066 | 1688 | int as_id = kvm_memslots_get_as_id(old, new); |
36947254 | 1689 | |
a54d8066 MS |
1690 | kvm_swap_active_memslots(kvm, as_id); |
1691 | ||
1692 | /* Propagate the new memslot to the now inactive memslots. */ | |
1693 | kvm_replace_memslot(kvm, old, new); | |
1694 | } | |
1695 | ||
1696 | static void kvm_copy_memslot(struct kvm_memory_slot *dest, | |
1697 | const struct kvm_memory_slot *src) | |
1698 | { | |
1699 | dest->base_gfn = src->base_gfn; | |
1700 | dest->npages = src->npages; | |
1701 | dest->dirty_bitmap = src->dirty_bitmap; | |
1702 | dest->arch = src->arch; | |
1703 | dest->userspace_addr = src->userspace_addr; | |
1704 | dest->flags = src->flags; | |
1705 | dest->id = src->id; | |
1706 | dest->as_id = src->as_id; | |
1707 | } | |
1708 | ||
1709 | static void kvm_invalidate_memslot(struct kvm *kvm, | |
1710 | struct kvm_memory_slot *old, | |
244893fa | 1711 | struct kvm_memory_slot *invalid_slot) |
a54d8066 | 1712 | { |
07921665 | 1713 | /* |
a54d8066 MS |
1714 | * Mark the current slot INVALID. As with all memslot modifications, |
1715 | * this must be done on an unreachable slot to avoid modifying the | |
1716 | * current slot in the active tree. | |
07921665 | 1717 | */ |
244893fa SC |
1718 | kvm_copy_memslot(invalid_slot, old); |
1719 | invalid_slot->flags |= KVM_MEMSLOT_INVALID; | |
1720 | kvm_replace_memslot(kvm, old, invalid_slot); | |
a54d8066 MS |
1721 | |
1722 | /* | |
1723 | * Activate the slot that is now marked INVALID, but don't propagate | |
1724 | * the slot to the now inactive slots. The slot is either going to be | |
1725 | * deleted or recreated as a new slot. | |
1726 | */ | |
1727 | kvm_swap_active_memslots(kvm, old->as_id); | |
1728 | ||
1729 | /* | |
1730 | * From this point no new shadow pages pointing to a deleted, or moved, | |
1731 | * memslot will be created. Validation of sp->gfn happens in: | |
1732 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1733 | * - kvm_is_visible_gfn (mmu_check_root) | |
1734 | */ | |
bcb63dcd | 1735 | kvm_arch_flush_shadow_memslot(kvm, old); |
683412cc | 1736 | kvm_arch_guest_memory_reclaimed(kvm); |
a54d8066 MS |
1737 | |
1738 | /* Was released by kvm_swap_active_memslots, reacquire. */ | |
1739 | mutex_lock(&kvm->slots_arch_lock); | |
1740 | ||
1741 | /* | |
1742 | * Copy the arch-specific field of the newly-installed slot back to the | |
1743 | * old slot as the arch data could have changed between releasing | |
1744 | * slots_arch_lock in install_new_memslots() and re-acquiring the lock | |
1745 | * above. Writers are required to retrieve memslots *after* acquiring | |
1746 | * slots_arch_lock, thus the active slot's data is guaranteed to be fresh. | |
1747 | */ | |
244893fa | 1748 | old->arch = invalid_slot->arch; |
a54d8066 MS |
1749 | } |
1750 | ||
1751 | static void kvm_create_memslot(struct kvm *kvm, | |
244893fa | 1752 | struct kvm_memory_slot *new) |
a54d8066 | 1753 | { |
244893fa SC |
1754 | /* Add the new memslot to the inactive set and activate. */ |
1755 | kvm_replace_memslot(kvm, NULL, new); | |
1756 | kvm_activate_memslot(kvm, NULL, new); | |
a54d8066 MS |
1757 | } |
1758 | ||
1759 | static void kvm_delete_memslot(struct kvm *kvm, | |
1760 | struct kvm_memory_slot *old, | |
1761 | struct kvm_memory_slot *invalid_slot) | |
1762 | { | |
1763 | /* | |
1764 | * Remove the old memslot (in the inactive memslots) by passing NULL as | |
244893fa | 1765 | * the "new" slot, and for the invalid version in the active slots. |
a54d8066 MS |
1766 | */ |
1767 | kvm_replace_memslot(kvm, old, NULL); | |
a54d8066 | 1768 | kvm_activate_memslot(kvm, invalid_slot, NULL); |
a54d8066 | 1769 | } |
36947254 | 1770 | |
244893fa SC |
1771 | static void kvm_move_memslot(struct kvm *kvm, |
1772 | struct kvm_memory_slot *old, | |
1773 | struct kvm_memory_slot *new, | |
1774 | struct kvm_memory_slot *invalid_slot) | |
a54d8066 | 1775 | { |
a54d8066 | 1776 | /* |
244893fa SC |
1777 | * Replace the old memslot in the inactive slots, and then swap slots |
1778 | * and replace the current INVALID with the new as well. | |
a54d8066 | 1779 | */ |
244893fa SC |
1780 | kvm_replace_memslot(kvm, old, new); |
1781 | kvm_activate_memslot(kvm, invalid_slot, new); | |
a54d8066 | 1782 | } |
36947254 | 1783 | |
a54d8066 MS |
1784 | static void kvm_update_flags_memslot(struct kvm *kvm, |
1785 | struct kvm_memory_slot *old, | |
244893fa | 1786 | struct kvm_memory_slot *new) |
a54d8066 MS |
1787 | { |
1788 | /* | |
1789 | * Similar to the MOVE case, but the slot doesn't need to be zapped as | |
1790 | * an intermediate step. Instead, the old memslot is simply replaced | |
1791 | * with a new, updated copy in both memslot sets. | |
1792 | */ | |
244893fa SC |
1793 | kvm_replace_memslot(kvm, old, new); |
1794 | kvm_activate_memslot(kvm, old, new); | |
36947254 SC |
1795 | } |
1796 | ||
cf47f50b | 1797 | static int kvm_set_memslot(struct kvm *kvm, |
a54d8066 | 1798 | struct kvm_memory_slot *old, |
ce5f0215 | 1799 | struct kvm_memory_slot *new, |
cf47f50b SC |
1800 | enum kvm_mr_change change) |
1801 | { | |
244893fa | 1802 | struct kvm_memory_slot *invalid_slot; |
cf47f50b SC |
1803 | int r; |
1804 | ||
b10a038e | 1805 | /* |
a54d8066 | 1806 | * Released in kvm_swap_active_memslots. |
b10a038e BG |
1807 | * |
1808 | * Must be held from before the current memslots are copied until | |
1809 | * after the new memslots are installed with rcu_assign_pointer, | |
a54d8066 | 1810 | * then released before the synchronize srcu in kvm_swap_active_memslots. |
b10a038e BG |
1811 | * |
1812 | * When modifying memslots outside of the slots_lock, must be held | |
1813 | * before reading the pointer to the current memslots until after all | |
1814 | * changes to those memslots are complete. | |
1815 | * | |
1816 | * These rules ensure that installing new memslots does not lose | |
1817 | * changes made to the previous memslots. | |
1818 | */ | |
1819 | mutex_lock(&kvm->slots_arch_lock); | |
1820 | ||
a54d8066 MS |
1821 | /* |
1822 | * Invalidate the old slot if it's being deleted or moved. This is | |
1823 | * done prior to actually deleting/moving the memslot to allow vCPUs to | |
1824 | * continue running by ensuring there are no mappings or shadow pages | |
1825 | * for the memslot when it is deleted/moved. Without pre-invalidation | |
1826 | * (and without a lock), a window would exist between effecting the | |
1827 | * delete/move and committing the changes in arch code where KVM or a | |
1828 | * guest could access a non-existent memslot. | |
244893fa SC |
1829 | * |
1830 | * Modifications are done on a temporary, unreachable slot. The old | |
1831 | * slot needs to be preserved in case a later step fails and the | |
1832 | * invalidation needs to be reverted. | |
a54d8066 | 1833 | */ |
cf47f50b | 1834 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
244893fa SC |
1835 | invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT); |
1836 | if (!invalid_slot) { | |
1837 | mutex_unlock(&kvm->slots_arch_lock); | |
1838 | return -ENOMEM; | |
1839 | } | |
1840 | kvm_invalidate_memslot(kvm, old, invalid_slot); | |
1841 | } | |
b10a038e | 1842 | |
a54d8066 MS |
1843 | r = kvm_prepare_memory_region(kvm, old, new, change); |
1844 | if (r) { | |
b10a038e | 1845 | /* |
a54d8066 MS |
1846 | * For DELETE/MOVE, revert the above INVALID change. No |
1847 | * modifications required since the original slot was preserved | |
1848 | * in the inactive slots. Changing the active memslots also | |
1849 | * release slots_arch_lock. | |
b10a038e | 1850 | */ |
244893fa SC |
1851 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
1852 | kvm_activate_memslot(kvm, invalid_slot, old); | |
1853 | kfree(invalid_slot); | |
1854 | } else { | |
a54d8066 | 1855 | mutex_unlock(&kvm->slots_arch_lock); |
244893fa | 1856 | } |
a54d8066 | 1857 | return r; |
cf47f50b SC |
1858 | } |
1859 | ||
bda44d84 | 1860 | /* |
a54d8066 MS |
1861 | * For DELETE and MOVE, the working slot is now active as the INVALID |
1862 | * version of the old slot. MOVE is particularly special as it reuses | |
1863 | * the old slot and returns a copy of the old slot (in working_slot). | |
1864 | * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the | |
1865 | * old slot is detached but otherwise preserved. | |
bda44d84 | 1866 | */ |
a54d8066 | 1867 | if (change == KVM_MR_CREATE) |
244893fa | 1868 | kvm_create_memslot(kvm, new); |
a54d8066 | 1869 | else if (change == KVM_MR_DELETE) |
244893fa | 1870 | kvm_delete_memslot(kvm, old, invalid_slot); |
a54d8066 | 1871 | else if (change == KVM_MR_MOVE) |
244893fa | 1872 | kvm_move_memslot(kvm, old, new, invalid_slot); |
a54d8066 | 1873 | else if (change == KVM_MR_FLAGS_ONLY) |
244893fa | 1874 | kvm_update_flags_memslot(kvm, old, new); |
a54d8066 MS |
1875 | else |
1876 | BUG(); | |
cf47f50b | 1877 | |
244893fa SC |
1878 | /* Free the temporary INVALID slot used for DELETE and MOVE. */ |
1879 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) | |
1880 | kfree(invalid_slot); | |
bda44d84 | 1881 | |
a54d8066 MS |
1882 | /* |
1883 | * No need to refresh new->arch, changes after dropping slots_arch_lock | |
a413a625 | 1884 | * will directly hit the final, active memslot. Architectures are |
a54d8066 MS |
1885 | * responsible for knowing that new->arch may be stale. |
1886 | */ | |
1887 | kvm_commit_memory_region(kvm, old, new, change); | |
cf47f50b | 1888 | |
cf47f50b | 1889 | return 0; |
cf47f50b SC |
1890 | } |
1891 | ||
44401a20 MS |
1892 | static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id, |
1893 | gfn_t start, gfn_t end) | |
5c0b4f3d | 1894 | { |
44401a20 | 1895 | struct kvm_memslot_iter iter; |
5c0b4f3d | 1896 | |
44401a20 MS |
1897 | kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) { |
1898 | if (iter.slot->id != id) | |
1899 | return true; | |
1900 | } | |
5c0b4f3d | 1901 | |
44401a20 | 1902 | return false; |
5c0b4f3d SC |
1903 | } |
1904 | ||
6aa8b732 AK |
1905 | /* |
1906 | * Allocate some memory and give it an address in the guest physical address | |
1907 | * space. | |
1908 | * | |
1909 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1910 | * |
02d5d55b | 1911 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1912 | */ |
f78e0e2e | 1913 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 1914 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 1915 | { |
244893fa | 1916 | struct kvm_memory_slot *old, *new; |
44401a20 | 1917 | struct kvm_memslots *slots; |
f64c0398 | 1918 | enum kvm_mr_change change; |
0f9bdef3 SC |
1919 | unsigned long npages; |
1920 | gfn_t base_gfn; | |
163da372 SC |
1921 | int as_id, id; |
1922 | int r; | |
6aa8b732 | 1923 | |
a50d64d6 XG |
1924 | r = check_memory_region_flags(mem); |
1925 | if (r) | |
71a4c30b | 1926 | return r; |
a50d64d6 | 1927 | |
f481b069 PB |
1928 | as_id = mem->slot >> 16; |
1929 | id = (u16)mem->slot; | |
1930 | ||
6aa8b732 | 1931 | /* General sanity checks */ |
6b285a55 SC |
1932 | if ((mem->memory_size & (PAGE_SIZE - 1)) || |
1933 | (mem->memory_size != (unsigned long)mem->memory_size)) | |
71a4c30b | 1934 | return -EINVAL; |
6aa8b732 | 1935 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 1936 | return -EINVAL; |
fa3d315a | 1937 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 1938 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 1939 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 1940 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 1941 | mem->memory_size)) |
71a4c30b | 1942 | return -EINVAL; |
f481b069 | 1943 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 1944 | return -EINVAL; |
6aa8b732 | 1945 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 1946 | return -EINVAL; |
0f9bdef3 SC |
1947 | if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES) |
1948 | return -EINVAL; | |
6aa8b732 | 1949 | |
44401a20 | 1950 | slots = __kvm_memslots(kvm, as_id); |
6aa8b732 | 1951 | |
5c0b4f3d | 1952 | /* |
7cd08553 SC |
1953 | * Note, the old memslot (and the pointer itself!) may be invalidated |
1954 | * and/or destroyed by kvm_set_memslot(). | |
5c0b4f3d | 1955 | */ |
44401a20 | 1956 | old = id_to_memslot(slots, id); |
163da372 | 1957 | |
47ea7d90 | 1958 | if (!mem->memory_size) { |
7cd08553 | 1959 | if (!old || !old->npages) |
47ea7d90 | 1960 | return -EINVAL; |
5c0b4f3d | 1961 | |
7cd08553 | 1962 | if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages)) |
47ea7d90 | 1963 | return -EIO; |
6aa8b732 | 1964 | |
244893fa | 1965 | return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE); |
47ea7d90 | 1966 | } |
5c0b4f3d | 1967 | |
0f9bdef3 SC |
1968 | base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT); |
1969 | npages = (mem->memory_size >> PAGE_SHIFT); | |
163da372 | 1970 | |
7cd08553 | 1971 | if (!old || !old->npages) { |
5c0b4f3d | 1972 | change = KVM_MR_CREATE; |
afa319a5 SC |
1973 | |
1974 | /* | |
1975 | * To simplify KVM internals, the total number of pages across | |
1976 | * all memslots must fit in an unsigned long. | |
1977 | */ | |
0f9bdef3 | 1978 | if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages) |
afa319a5 | 1979 | return -EINVAL; |
5c0b4f3d | 1980 | } else { /* Modify an existing slot. */ |
0f9bdef3 SC |
1981 | if ((mem->userspace_addr != old->userspace_addr) || |
1982 | (npages != old->npages) || | |
1983 | ((mem->flags ^ old->flags) & KVM_MEM_READONLY)) | |
71a4c30b | 1984 | return -EINVAL; |
09170a49 | 1985 | |
0f9bdef3 | 1986 | if (base_gfn != old->base_gfn) |
5c0b4f3d | 1987 | change = KVM_MR_MOVE; |
0f9bdef3 | 1988 | else if (mem->flags != old->flags) |
5c0b4f3d SC |
1989 | change = KVM_MR_FLAGS_ONLY; |
1990 | else /* Nothing to change. */ | |
1991 | return 0; | |
09170a49 | 1992 | } |
6aa8b732 | 1993 | |
44401a20 | 1994 | if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) && |
0f9bdef3 | 1995 | kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages)) |
44401a20 | 1996 | return -EEXIST; |
6aa8b732 | 1997 | |
244893fa SC |
1998 | /* Allocate a slot that will persist in the memslot. */ |
1999 | new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT); | |
2000 | if (!new) | |
2001 | return -ENOMEM; | |
3c9bd400 | 2002 | |
244893fa SC |
2003 | new->as_id = as_id; |
2004 | new->id = id; | |
2005 | new->base_gfn = base_gfn; | |
2006 | new->npages = npages; | |
2007 | new->flags = mem->flags; | |
2008 | new->userspace_addr = mem->userspace_addr; | |
6aa8b732 | 2009 | |
244893fa | 2010 | r = kvm_set_memslot(kvm, old, new, change); |
cf47f50b | 2011 | if (r) |
244893fa | 2012 | kfree(new); |
6aa8b732 | 2013 | return r; |
210c7c4d | 2014 | } |
f78e0e2e SY |
2015 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
2016 | ||
2017 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 2018 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
2019 | { |
2020 | int r; | |
2021 | ||
79fac95e | 2022 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 2023 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 2024 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
2025 | return r; |
2026 | } | |
210c7c4d IE |
2027 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
2028 | ||
7940876e SH |
2029 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
2030 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 2031 | { |
f481b069 | 2032 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 2033 | return -EINVAL; |
09170a49 | 2034 | |
47ae31e2 | 2035 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
2036 | } |
2037 | ||
0dff0846 | 2038 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
2039 | /** |
2040 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
2041 | * @kvm: pointer to kvm instance | |
2042 | * @log: slot id and address to which we copy the log | |
2043 | * @is_dirty: set to '1' if any dirty pages were found | |
2044 | * @memslot: set to the associated memslot, always valid on success | |
2045 | */ | |
2046 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
2047 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 2048 | { |
9f6b8029 | 2049 | struct kvm_memslots *slots; |
843574a3 | 2050 | int i, as_id, id; |
87bf6e7d | 2051 | unsigned long n; |
6aa8b732 AK |
2052 | unsigned long any = 0; |
2053 | ||
b2cc64c4 PX |
2054 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
2055 | if (kvm->dirty_ring_size) | |
2056 | return -ENXIO; | |
2057 | ||
2a49f61d SC |
2058 | *memslot = NULL; |
2059 | *is_dirty = 0; | |
2060 | ||
f481b069 PB |
2061 | as_id = log->slot >> 16; |
2062 | id = (u16)log->slot; | |
2063 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 2064 | return -EINVAL; |
6aa8b732 | 2065 | |
f481b069 | 2066 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 2067 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 2068 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 2069 | return -ENOENT; |
6aa8b732 | 2070 | |
2a49f61d SC |
2071 | kvm_arch_sync_dirty_log(kvm, *memslot); |
2072 | ||
2073 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 2074 | |
cd1a4a98 | 2075 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 2076 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 2077 | |
2a49f61d | 2078 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 2079 | return -EFAULT; |
6aa8b732 | 2080 | |
5bb064dc ZX |
2081 | if (any) |
2082 | *is_dirty = 1; | |
843574a3 | 2083 | return 0; |
6aa8b732 | 2084 | } |
2ba9f0d8 | 2085 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 2086 | |
0dff0846 | 2087 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 2088 | /** |
b8b00220 | 2089 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 2090 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
2091 | * @kvm: pointer to kvm instance |
2092 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
2093 | * |
2094 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
2095 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
2096 | * following order: | |
2097 | * | |
2098 | * 1. Take a snapshot of the bit and clear it if needed. | |
2099 | * 2. Write protect the corresponding page. | |
2100 | * 3. Copy the snapshot to the userspace. | |
2101 | * 4. Upon return caller flushes TLB's if needed. | |
2102 | * | |
2103 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
2104 | * entry. This is not a problem because the page is reported dirty using | |
2105 | * the snapshot taken before and step 4 ensures that writes done after | |
2106 | * exiting to userspace will be logged for the next call. | |
2107 | * | |
2108 | */ | |
0dff0846 | 2109 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 2110 | { |
9f6b8029 | 2111 | struct kvm_memslots *slots; |
ba0513b5 | 2112 | struct kvm_memory_slot *memslot; |
58d6db34 | 2113 | int i, as_id, id; |
ba0513b5 MS |
2114 | unsigned long n; |
2115 | unsigned long *dirty_bitmap; | |
2116 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2117 | bool flush; |
ba0513b5 | 2118 | |
b2cc64c4 PX |
2119 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
2120 | if (kvm->dirty_ring_size) | |
2121 | return -ENXIO; | |
2122 | ||
f481b069 PB |
2123 | as_id = log->slot >> 16; |
2124 | id = (u16)log->slot; | |
2125 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 2126 | return -EINVAL; |
ba0513b5 | 2127 | |
f481b069 PB |
2128 | slots = __kvm_memslots(kvm, as_id); |
2129 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2130 | if (!memslot || !memslot->dirty_bitmap) |
2131 | return -ENOENT; | |
ba0513b5 MS |
2132 | |
2133 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 2134 | |
0dff0846 SC |
2135 | kvm_arch_sync_dirty_log(kvm, memslot); |
2136 | ||
ba0513b5 | 2137 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 2138 | flush = false; |
2a31b9db PB |
2139 | if (kvm->manual_dirty_log_protect) { |
2140 | /* | |
2141 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
2142 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
2143 | * is some code duplication between this function and | |
2144 | * kvm_get_dirty_log, but hopefully all architecture | |
2145 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
2146 | * can be eliminated. | |
2147 | */ | |
2148 | dirty_bitmap_buffer = dirty_bitmap; | |
2149 | } else { | |
2150 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
2151 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 2152 | |
531810ca | 2153 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
2154 | for (i = 0; i < n / sizeof(long); i++) { |
2155 | unsigned long mask; | |
2156 | gfn_t offset; | |
ba0513b5 | 2157 | |
2a31b9db PB |
2158 | if (!dirty_bitmap[i]) |
2159 | continue; | |
2160 | ||
0dff0846 | 2161 | flush = true; |
2a31b9db PB |
2162 | mask = xchg(&dirty_bitmap[i], 0); |
2163 | dirty_bitmap_buffer[i] = mask; | |
2164 | ||
a67794ca LT |
2165 | offset = i * BITS_PER_LONG; |
2166 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
2167 | offset, mask); | |
2a31b9db | 2168 | } |
531810ca | 2169 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
2170 | } |
2171 | ||
0dff0846 SC |
2172 | if (flush) |
2173 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2174 | ||
2a31b9db PB |
2175 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
2176 | return -EFAULT; | |
2177 | return 0; | |
2178 | } | |
0dff0846 SC |
2179 | |
2180 | ||
2181 | /** | |
2182 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
2183 | * @kvm: kvm instance | |
2184 | * @log: slot id and address to which we copy the log | |
2185 | * | |
2186 | * Steps 1-4 below provide general overview of dirty page logging. See | |
2187 | * kvm_get_dirty_log_protect() function description for additional details. | |
2188 | * | |
2189 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
2190 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
2191 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
2192 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
2193 | * writes will be marked dirty for next log read. | |
2194 | * | |
2195 | * 1. Take a snapshot of the bit and clear it if needed. | |
2196 | * 2. Write protect the corresponding page. | |
2197 | * 3. Copy the snapshot to the userspace. | |
2198 | * 4. Flush TLB's if needed. | |
2199 | */ | |
2200 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
2201 | struct kvm_dirty_log *log) | |
2202 | { | |
2203 | int r; | |
2204 | ||
2205 | mutex_lock(&kvm->slots_lock); | |
2206 | ||
2207 | r = kvm_get_dirty_log_protect(kvm, log); | |
2208 | ||
2209 | mutex_unlock(&kvm->slots_lock); | |
2210 | return r; | |
2211 | } | |
2a31b9db PB |
2212 | |
2213 | /** | |
2214 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
2215 | * and reenable dirty page tracking for the corresponding pages. | |
2216 | * @kvm: pointer to kvm instance | |
2217 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
2218 | */ | |
0dff0846 SC |
2219 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
2220 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
2221 | { |
2222 | struct kvm_memslots *slots; | |
2223 | struct kvm_memory_slot *memslot; | |
98938aa8 | 2224 | int as_id, id; |
2a31b9db | 2225 | gfn_t offset; |
98938aa8 | 2226 | unsigned long i, n; |
2a31b9db PB |
2227 | unsigned long *dirty_bitmap; |
2228 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2229 | bool flush; |
2a31b9db | 2230 | |
b2cc64c4 PX |
2231 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
2232 | if (kvm->dirty_ring_size) | |
2233 | return -ENXIO; | |
2234 | ||
2a31b9db PB |
2235 | as_id = log->slot >> 16; |
2236 | id = (u16)log->slot; | |
2237 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
2238 | return -EINVAL; | |
2239 | ||
76d58e0f | 2240 | if (log->first_page & 63) |
2a31b9db PB |
2241 | return -EINVAL; |
2242 | ||
2243 | slots = __kvm_memslots(kvm, as_id); | |
2244 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2245 | if (!memslot || !memslot->dirty_bitmap) |
2246 | return -ENOENT; | |
2a31b9db PB |
2247 | |
2248 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 2249 | |
4ddc9204 | 2250 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
2251 | |
2252 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
2253 | log->num_pages > memslot->npages - log->first_page || |
2254 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
2255 | return -EINVAL; | |
98938aa8 | 2256 | |
0dff0846 SC |
2257 | kvm_arch_sync_dirty_log(kvm, memslot); |
2258 | ||
2259 | flush = false; | |
2a31b9db PB |
2260 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
2261 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
2262 | return -EFAULT; | |
ba0513b5 | 2263 | |
531810ca | 2264 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
2265 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
2266 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
2267 | i++, offset += BITS_PER_LONG) { |
2268 | unsigned long mask = *dirty_bitmap_buffer++; | |
2269 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
2270 | if (!mask) | |
ba0513b5 MS |
2271 | continue; |
2272 | ||
2a31b9db | 2273 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 2274 | |
2a31b9db PB |
2275 | /* |
2276 | * mask contains the bits that really have been cleared. This | |
2277 | * never includes any bits beyond the length of the memslot (if | |
2278 | * the length is not aligned to 64 pages), therefore it is not | |
2279 | * a problem if userspace sets them in log->dirty_bitmap. | |
2280 | */ | |
58d2930f | 2281 | if (mask) { |
0dff0846 | 2282 | flush = true; |
58d2930f TY |
2283 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
2284 | offset, mask); | |
2285 | } | |
ba0513b5 | 2286 | } |
531810ca | 2287 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 2288 | |
0dff0846 SC |
2289 | if (flush) |
2290 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2291 | ||
58d6db34 | 2292 | return 0; |
ba0513b5 | 2293 | } |
0dff0846 SC |
2294 | |
2295 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
2296 | struct kvm_clear_dirty_log *log) | |
2297 | { | |
2298 | int r; | |
2299 | ||
2300 | mutex_lock(&kvm->slots_lock); | |
2301 | ||
2302 | r = kvm_clear_dirty_log_protect(kvm, log); | |
2303 | ||
2304 | mutex_unlock(&kvm->slots_lock); | |
2305 | return r; | |
2306 | } | |
2307 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 2308 | |
49c7754c GN |
2309 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
2310 | { | |
2311 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
2312 | } | |
a1f4d395 | 2313 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 2314 | |
8e73485c PB |
2315 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
2316 | { | |
fe22ed82 | 2317 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
a54d8066 | 2318 | u64 gen = slots->generation; |
fe22ed82 | 2319 | struct kvm_memory_slot *slot; |
fe22ed82 | 2320 | |
a54d8066 MS |
2321 | /* |
2322 | * This also protects against using a memslot from a different address space, | |
2323 | * since different address spaces have different generation numbers. | |
2324 | */ | |
2325 | if (unlikely(gen != vcpu->last_used_slot_gen)) { | |
2326 | vcpu->last_used_slot = NULL; | |
2327 | vcpu->last_used_slot_gen = gen; | |
2328 | } | |
2329 | ||
2330 | slot = try_get_memslot(vcpu->last_used_slot, gfn); | |
fe22ed82 DM |
2331 | if (slot) |
2332 | return slot; | |
2333 | ||
2334 | /* | |
2335 | * Fall back to searching all memslots. We purposely use | |
2336 | * search_memslots() instead of __gfn_to_memslot() to avoid | |
a54d8066 | 2337 | * thrashing the VM-wide last_used_slot in kvm_memslots. |
fe22ed82 | 2338 | */ |
a54d8066 | 2339 | slot = search_memslots(slots, gfn, false); |
fe22ed82 | 2340 | if (slot) { |
a54d8066 | 2341 | vcpu->last_used_slot = slot; |
fe22ed82 DM |
2342 | return slot; |
2343 | } | |
2344 | ||
2345 | return NULL; | |
8e73485c PB |
2346 | } |
2347 | ||
33e94154 | 2348 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 2349 | { |
bf3e05bc | 2350 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 2351 | |
c36b7150 | 2352 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
2353 | } |
2354 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
2355 | ||
995decb6 VK |
2356 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
2357 | { | |
2358 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2359 | ||
2360 | return kvm_is_visible_memslot(memslot); | |
2361 | } | |
2362 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
2363 | ||
f9b84e19 | 2364 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
2365 | { |
2366 | struct vm_area_struct *vma; | |
2367 | unsigned long addr, size; | |
2368 | ||
2369 | size = PAGE_SIZE; | |
2370 | ||
42cde48b | 2371 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
2372 | if (kvm_is_error_hva(addr)) |
2373 | return PAGE_SIZE; | |
2374 | ||
d8ed45c5 | 2375 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
2376 | vma = find_vma(current->mm, addr); |
2377 | if (!vma) | |
2378 | goto out; | |
2379 | ||
2380 | size = vma_kernel_pagesize(vma); | |
2381 | ||
2382 | out: | |
d8ed45c5 | 2383 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
2384 | |
2385 | return size; | |
2386 | } | |
2387 | ||
8283e36a | 2388 | static bool memslot_is_readonly(const struct kvm_memory_slot *slot) |
4d8b81ab XG |
2389 | { |
2390 | return slot->flags & KVM_MEM_READONLY; | |
2391 | } | |
2392 | ||
8283e36a | 2393 | static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn, |
4d8b81ab | 2394 | gfn_t *nr_pages, bool write) |
539cb660 | 2395 | { |
bc6678a3 | 2396 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 2397 | return KVM_HVA_ERR_BAD; |
48987781 | 2398 | |
4d8b81ab XG |
2399 | if (memslot_is_readonly(slot) && write) |
2400 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
2401 | |
2402 | if (nr_pages) | |
2403 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
2404 | ||
4d8b81ab | 2405 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 2406 | } |
48987781 | 2407 | |
4d8b81ab XG |
2408 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2409 | gfn_t *nr_pages) | |
2410 | { | |
2411 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 2412 | } |
48987781 | 2413 | |
4d8b81ab | 2414 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 2415 | gfn_t gfn) |
4d8b81ab XG |
2416 | { |
2417 | return gfn_to_hva_many(slot, gfn, NULL); | |
2418 | } | |
2419 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
2420 | ||
48987781 XG |
2421 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
2422 | { | |
49c7754c | 2423 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 2424 | } |
0d150298 | 2425 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 2426 | |
8e73485c PB |
2427 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
2428 | { | |
2429 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
2430 | } | |
2431 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
2432 | ||
86ab8cff | 2433 | /* |
970c0d4b WY |
2434 | * Return the hva of a @gfn and the R/W attribute if possible. |
2435 | * | |
2436 | * @slot: the kvm_memory_slot which contains @gfn | |
2437 | * @gfn: the gfn to be translated | |
2438 | * @writable: used to return the read/write attribute of the @slot if the hva | |
2439 | * is valid and @writable is not NULL | |
86ab8cff | 2440 | */ |
64d83126 CD |
2441 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
2442 | gfn_t gfn, bool *writable) | |
86ab8cff | 2443 | { |
a2ac07fe GN |
2444 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
2445 | ||
2446 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
2447 | *writable = !memslot_is_readonly(slot); |
2448 | ||
a2ac07fe | 2449 | return hva; |
86ab8cff XG |
2450 | } |
2451 | ||
64d83126 CD |
2452 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
2453 | { | |
2454 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2455 | ||
2456 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2457 | } | |
2458 | ||
8e73485c PB |
2459 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
2460 | { | |
2461 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2462 | ||
2463 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2464 | } | |
2465 | ||
fafc3dba HY |
2466 | static inline int check_user_page_hwpoison(unsigned long addr) |
2467 | { | |
0d731759 | 2468 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 2469 | |
0d731759 | 2470 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
2471 | return rc == -EHWPOISON; |
2472 | } | |
2473 | ||
2fc84311 | 2474 | /* |
b9b33da2 PB |
2475 | * The fast path to get the writable pfn which will be stored in @pfn, |
2476 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 2477 | * only part that runs if we can in atomic context. |
2fc84311 | 2478 | */ |
b9b33da2 PB |
2479 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
2480 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 2481 | { |
8d4e1288 | 2482 | struct page *page[1]; |
954bbbc2 | 2483 | |
12ce13fe XG |
2484 | /* |
2485 | * Fast pin a writable pfn only if it is a write fault request | |
2486 | * or the caller allows to map a writable pfn for a read fault | |
2487 | * request. | |
2488 | */ | |
2489 | if (!(write_fault || writable)) | |
2490 | return false; | |
612819c3 | 2491 | |
dadbb612 | 2492 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 2493 | *pfn = page_to_pfn(page[0]); |
612819c3 | 2494 | |
2fc84311 XG |
2495 | if (writable) |
2496 | *writable = true; | |
2497 | return true; | |
2498 | } | |
af585b92 | 2499 | |
2fc84311 XG |
2500 | return false; |
2501 | } | |
612819c3 | 2502 | |
2fc84311 XG |
2503 | /* |
2504 | * The slow path to get the pfn of the specified host virtual address, | |
2505 | * 1 indicates success, -errno is returned if error is detected. | |
2506 | */ | |
2507 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 2508 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 2509 | { |
ce53053c AV |
2510 | unsigned int flags = FOLL_HWPOISON; |
2511 | struct page *page; | |
2fc84311 | 2512 | int npages = 0; |
612819c3 | 2513 | |
2fc84311 XG |
2514 | might_sleep(); |
2515 | ||
2516 | if (writable) | |
2517 | *writable = write_fault; | |
2518 | ||
ce53053c AV |
2519 | if (write_fault) |
2520 | flags |= FOLL_WRITE; | |
2521 | if (async) | |
2522 | flags |= FOLL_NOWAIT; | |
d4944b0e | 2523 | |
ce53053c | 2524 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
2525 | if (npages != 1) |
2526 | return npages; | |
2527 | ||
2528 | /* map read fault as writable if possible */ | |
12ce13fe | 2529 | if (unlikely(!write_fault) && writable) { |
ce53053c | 2530 | struct page *wpage; |
2fc84311 | 2531 | |
dadbb612 | 2532 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 2533 | *writable = true; |
ce53053c AV |
2534 | put_page(page); |
2535 | page = wpage; | |
612819c3 | 2536 | } |
887c08ac | 2537 | } |
ce53053c | 2538 | *pfn = page_to_pfn(page); |
2fc84311 XG |
2539 | return npages; |
2540 | } | |
539cb660 | 2541 | |
4d8b81ab XG |
2542 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
2543 | { | |
2544 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
2545 | return false; | |
2e2e3738 | 2546 | |
4d8b81ab XG |
2547 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
2548 | return false; | |
887c08ac | 2549 | |
4d8b81ab XG |
2550 | return true; |
2551 | } | |
bf998156 | 2552 | |
f8be156b NP |
2553 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
2554 | { | |
b14b2690 SC |
2555 | struct page *page = kvm_pfn_to_refcounted_page(pfn); |
2556 | ||
2557 | if (!page) | |
f8be156b | 2558 | return 1; |
b14b2690 SC |
2559 | |
2560 | return get_page_unless_zero(page); | |
f8be156b NP |
2561 | } |
2562 | ||
92176a8e | 2563 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1625566e XT |
2564 | unsigned long addr, bool write_fault, |
2565 | bool *writable, kvm_pfn_t *p_pfn) | |
92176a8e | 2566 | { |
a9545779 | 2567 | kvm_pfn_t pfn; |
bd2fae8d PB |
2568 | pte_t *ptep; |
2569 | spinlock_t *ptl; | |
add6a0cd PB |
2570 | int r; |
2571 | ||
9fd6dad1 | 2572 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2573 | if (r) { |
2574 | /* | |
2575 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
2576 | * not call the fault handler, so do it here. | |
2577 | */ | |
2578 | bool unlocked = false; | |
64019a2e | 2579 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
2580 | (write_fault ? FAULT_FLAG_WRITE : 0), |
2581 | &unlocked); | |
a8387d0b PB |
2582 | if (unlocked) |
2583 | return -EAGAIN; | |
add6a0cd PB |
2584 | if (r) |
2585 | return r; | |
2586 | ||
9fd6dad1 | 2587 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2588 | if (r) |
2589 | return r; | |
bd2fae8d | 2590 | } |
add6a0cd | 2591 | |
bd2fae8d PB |
2592 | if (write_fault && !pte_write(*ptep)) { |
2593 | pfn = KVM_PFN_ERR_RO_FAULT; | |
2594 | goto out; | |
add6a0cd PB |
2595 | } |
2596 | ||
a340b3e2 | 2597 | if (writable) |
bd2fae8d PB |
2598 | *writable = pte_write(*ptep); |
2599 | pfn = pte_pfn(*ptep); | |
add6a0cd PB |
2600 | |
2601 | /* | |
2602 | * Get a reference here because callers of *hva_to_pfn* and | |
2603 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
2604 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
36c3ce6c | 2605 | * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will |
add6a0cd PB |
2606 | * simply do nothing for reserved pfns. |
2607 | * | |
2608 | * Whoever called remap_pfn_range is also going to call e.g. | |
2609 | * unmap_mapping_range before the underlying pages are freed, | |
2610 | * causing a call to our MMU notifier. | |
f8be156b NP |
2611 | * |
2612 | * Certain IO or PFNMAP mappings can be backed with valid | |
2613 | * struct pages, but be allocated without refcounting e.g., | |
2614 | * tail pages of non-compound higher order allocations, which | |
2615 | * would then underflow the refcount when the caller does the | |
2616 | * required put_page. Don't allow those pages here. | |
add6a0cd | 2617 | */ |
f8be156b NP |
2618 | if (!kvm_try_get_pfn(pfn)) |
2619 | r = -EFAULT; | |
add6a0cd | 2620 | |
bd2fae8d PB |
2621 | out: |
2622 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 2623 | *p_pfn = pfn; |
f8be156b NP |
2624 | |
2625 | return r; | |
92176a8e PB |
2626 | } |
2627 | ||
12ce13fe XG |
2628 | /* |
2629 | * Pin guest page in memory and return its pfn. | |
2630 | * @addr: host virtual address which maps memory to the guest | |
2631 | * @atomic: whether this function can sleep | |
2632 | * @async: whether this function need to wait IO complete if the | |
2633 | * host page is not in the memory | |
2634 | * @write_fault: whether we should get a writable host page | |
2635 | * @writable: whether it allows to map a writable host page for !@write_fault | |
2636 | * | |
2637 | * The function will map a writable host page for these two cases: | |
2638 | * 1): @write_fault = true | |
2639 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
2640 | * whether the mapping is writable. | |
2641 | */ | |
982ed0de DW |
2642 | kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2643 | bool write_fault, bool *writable) | |
2fc84311 XG |
2644 | { |
2645 | struct vm_area_struct *vma; | |
943dfea8 | 2646 | kvm_pfn_t pfn; |
92176a8e | 2647 | int npages, r; |
2e2e3738 | 2648 | |
2fc84311 XG |
2649 | /* we can do it either atomically or asynchronously, not both */ |
2650 | BUG_ON(atomic && async); | |
8d4e1288 | 2651 | |
b9b33da2 | 2652 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
2653 | return pfn; |
2654 | ||
2655 | if (atomic) | |
2656 | return KVM_PFN_ERR_FAULT; | |
2657 | ||
2658 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
2659 | if (npages == 1) | |
2660 | return pfn; | |
8d4e1288 | 2661 | |
d8ed45c5 | 2662 | mmap_read_lock(current->mm); |
2fc84311 XG |
2663 | if (npages == -EHWPOISON || |
2664 | (!async && check_user_page_hwpoison(addr))) { | |
2665 | pfn = KVM_PFN_ERR_HWPOISON; | |
2666 | goto exit; | |
2667 | } | |
2668 | ||
a8387d0b | 2669 | retry: |
fc98c03b | 2670 | vma = vma_lookup(current->mm, addr); |
2fc84311 XG |
2671 | |
2672 | if (vma == NULL) | |
2673 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 2674 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
1625566e | 2675 | r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn); |
a8387d0b PB |
2676 | if (r == -EAGAIN) |
2677 | goto retry; | |
92176a8e PB |
2678 | if (r < 0) |
2679 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 2680 | } else { |
4d8b81ab | 2681 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
2682 | *async = true; |
2683 | pfn = KVM_PFN_ERR_FAULT; | |
2684 | } | |
2685 | exit: | |
d8ed45c5 | 2686 | mmap_read_unlock(current->mm); |
2e2e3738 | 2687 | return pfn; |
35149e21 AL |
2688 | } |
2689 | ||
8283e36a | 2690 | kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn, |
ba049e93 | 2691 | bool atomic, bool *async, bool write_fault, |
4a42d848 | 2692 | bool *writable, hva_t *hva) |
887c08ac | 2693 | { |
4d8b81ab XG |
2694 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
2695 | ||
4a42d848 DS |
2696 | if (hva) |
2697 | *hva = addr; | |
2698 | ||
b2740d35 PB |
2699 | if (addr == KVM_HVA_ERR_RO_BAD) { |
2700 | if (writable) | |
2701 | *writable = false; | |
4d8b81ab | 2702 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 2703 | } |
4d8b81ab | 2704 | |
b2740d35 PB |
2705 | if (kvm_is_error_hva(addr)) { |
2706 | if (writable) | |
2707 | *writable = false; | |
81c52c56 | 2708 | return KVM_PFN_NOSLOT; |
b2740d35 | 2709 | } |
4d8b81ab XG |
2710 | |
2711 | /* Do not map writable pfn in the readonly memslot. */ | |
2712 | if (writable && memslot_is_readonly(slot)) { | |
2713 | *writable = false; | |
2714 | writable = NULL; | |
2715 | } | |
2716 | ||
2717 | return hva_to_pfn(addr, atomic, async, write_fault, | |
2718 | writable); | |
887c08ac | 2719 | } |
3520469d | 2720 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 2721 | |
ba049e93 | 2722 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
2723 | bool *writable) |
2724 | { | |
e37afc6e | 2725 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
4a42d848 | 2726 | write_fault, writable, NULL); |
612819c3 MT |
2727 | } |
2728 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
2729 | ||
8283e36a | 2730 | kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2731 | { |
4a42d848 | 2732 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL); |
506f0d6f | 2733 | } |
e37afc6e | 2734 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 2735 | |
8283e36a | 2736 | kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2737 | { |
4a42d848 | 2738 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL); |
506f0d6f | 2739 | } |
037d92dc | 2740 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 2741 | |
ba049e93 | 2742 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2743 | { |
2744 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2745 | } | |
2746 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
2747 | ||
ba049e93 | 2748 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
2749 | { |
2750 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
2751 | } | |
2752 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
2753 | ||
ba049e93 | 2754 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2755 | { |
2756 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2757 | } | |
2758 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
2759 | ||
d9ef13c2 PB |
2760 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2761 | struct page **pages, int nr_pages) | |
48987781 XG |
2762 | { |
2763 | unsigned long addr; | |
076b925d | 2764 | gfn_t entry = 0; |
48987781 | 2765 | |
d9ef13c2 | 2766 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
2767 | if (kvm_is_error_hva(addr)) |
2768 | return -1; | |
2769 | ||
2770 | if (entry < nr_pages) | |
2771 | return 0; | |
2772 | ||
dadbb612 | 2773 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
2774 | } |
2775 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
2776 | ||
b1624f99 SC |
2777 | /* |
2778 | * Do not use this helper unless you are absolutely certain the gfn _must_ be | |
2779 | * backed by 'struct page'. A valid example is if the backing memslot is | |
2780 | * controlled by KVM. Note, if the returned page is valid, it's refcount has | |
2781 | * been elevated by gfn_to_pfn(). | |
2782 | */ | |
35149e21 AL |
2783 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
2784 | { | |
b14b2690 | 2785 | struct page *page; |
ba049e93 | 2786 | kvm_pfn_t pfn; |
2e2e3738 AL |
2787 | |
2788 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 2789 | |
81c52c56 | 2790 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 2791 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2792 | |
b14b2690 SC |
2793 | page = kvm_pfn_to_refcounted_page(pfn); |
2794 | if (!page) | |
6cede2e6 | 2795 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2796 | |
b14b2690 | 2797 | return page; |
954bbbc2 AK |
2798 | } |
2799 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
2800 | ||
357a18ad | 2801 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty) |
91724814 | 2802 | { |
91724814 BO |
2803 | if (dirty) |
2804 | kvm_release_pfn_dirty(pfn); | |
2805 | else | |
2806 | kvm_release_pfn_clean(pfn); | |
2807 | } | |
2808 | ||
357a18ad | 2809 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
e45adf66 KA |
2810 | { |
2811 | kvm_pfn_t pfn; | |
2812 | void *hva = NULL; | |
2813 | struct page *page = KVM_UNMAPPED_PAGE; | |
2814 | ||
2815 | if (!map) | |
2816 | return -EINVAL; | |
2817 | ||
357a18ad | 2818 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
e45adf66 KA |
2819 | if (is_error_noslot_pfn(pfn)) |
2820 | return -EINVAL; | |
2821 | ||
2822 | if (pfn_valid(pfn)) { | |
2823 | page = pfn_to_page(pfn); | |
357a18ad | 2824 | hva = kmap(page); |
d30b214d | 2825 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2826 | } else { |
357a18ad | 2827 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
d30b214d | 2828 | #endif |
e45adf66 KA |
2829 | } |
2830 | ||
2831 | if (!hva) | |
2832 | return -EFAULT; | |
2833 | ||
2834 | map->page = page; | |
2835 | map->hva = hva; | |
2836 | map->pfn = pfn; | |
2837 | map->gfn = gfn; | |
2838 | ||
2839 | return 0; | |
2840 | } | |
e45adf66 KA |
2841 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); |
2842 | ||
357a18ad | 2843 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) |
e45adf66 KA |
2844 | { |
2845 | if (!map) | |
2846 | return; | |
2847 | ||
2848 | if (!map->hva) | |
2849 | return; | |
2850 | ||
357a18ad DW |
2851 | if (map->page != KVM_UNMAPPED_PAGE) |
2852 | kunmap(map->page); | |
eb1f2f38 | 2853 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2854 | else |
357a18ad | 2855 | memunmap(map->hva); |
eb1f2f38 | 2856 | #endif |
e45adf66 | 2857 | |
91724814 | 2858 | if (dirty) |
357a18ad | 2859 | kvm_vcpu_mark_page_dirty(vcpu, map->gfn); |
91724814 | 2860 | |
357a18ad | 2861 | kvm_release_pfn(map->pfn, dirty); |
e45adf66 KA |
2862 | |
2863 | map->hva = NULL; | |
2864 | map->page = NULL; | |
2865 | } | |
2866 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); | |
2867 | ||
8e1c6914 | 2868 | static bool kvm_is_ad_tracked_page(struct page *page) |
8e73485c | 2869 | { |
8e1c6914 SC |
2870 | /* |
2871 | * Per page-flags.h, pages tagged PG_reserved "should in general not be | |
2872 | * touched (e.g. set dirty) except by its owner". | |
2873 | */ | |
2874 | return !PageReserved(page); | |
2875 | } | |
8e73485c | 2876 | |
8e1c6914 SC |
2877 | static void kvm_set_page_dirty(struct page *page) |
2878 | { | |
2879 | if (kvm_is_ad_tracked_page(page)) | |
2880 | SetPageDirty(page); | |
2881 | } | |
8e73485c | 2882 | |
8e1c6914 SC |
2883 | static void kvm_set_page_accessed(struct page *page) |
2884 | { | |
2885 | if (kvm_is_ad_tracked_page(page)) | |
2886 | mark_page_accessed(page); | |
8e73485c | 2887 | } |
8e73485c | 2888 | |
b4231d61 IE |
2889 | void kvm_release_page_clean(struct page *page) |
2890 | { | |
32cad84f XG |
2891 | WARN_ON(is_error_page(page)); |
2892 | ||
8e1c6914 SC |
2893 | kvm_set_page_accessed(page); |
2894 | put_page(page); | |
b4231d61 IE |
2895 | } |
2896 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
2897 | ||
ba049e93 | 2898 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 2899 | { |
b14b2690 SC |
2900 | struct page *page; |
2901 | ||
2902 | if (is_error_noslot_pfn(pfn)) | |
2903 | return; | |
2904 | ||
2905 | page = kvm_pfn_to_refcounted_page(pfn); | |
2906 | if (!page) | |
2907 | return; | |
2908 | ||
2909 | kvm_release_page_clean(page); | |
35149e21 AL |
2910 | } |
2911 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
2912 | ||
b4231d61 | 2913 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 2914 | { |
a2766325 XG |
2915 | WARN_ON(is_error_page(page)); |
2916 | ||
8e1c6914 SC |
2917 | kvm_set_page_dirty(page); |
2918 | kvm_release_page_clean(page); | |
35149e21 AL |
2919 | } |
2920 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
2921 | ||
f7a6509f | 2922 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2923 | { |
b14b2690 SC |
2924 | struct page *page; |
2925 | ||
2926 | if (is_error_noslot_pfn(pfn)) | |
2927 | return; | |
2928 | ||
2929 | page = kvm_pfn_to_refcounted_page(pfn); | |
2930 | if (!page) | |
2931 | return; | |
2932 | ||
2933 | kvm_release_page_dirty(page); | |
35149e21 | 2934 | } |
f7a6509f | 2935 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 2936 | |
8e1c6914 SC |
2937 | /* |
2938 | * Note, checking for an error/noslot pfn is the caller's responsibility when | |
2939 | * directly marking a page dirty/accessed. Unlike the "release" helpers, the | |
2940 | * "set" helpers are not to be used when the pfn might point at garbage. | |
2941 | */ | |
ba049e93 | 2942 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2943 | { |
8e1c6914 SC |
2944 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
2945 | return; | |
2946 | ||
2947 | if (pfn_valid(pfn)) | |
2948 | kvm_set_page_dirty(pfn_to_page(pfn)); | |
8a7ae055 | 2949 | } |
35149e21 AL |
2950 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
2951 | ||
ba049e93 | 2952 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 2953 | { |
8e1c6914 SC |
2954 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
2955 | return; | |
2956 | ||
2957 | if (pfn_valid(pfn)) | |
2958 | kvm_set_page_accessed(pfn_to_page(pfn)); | |
35149e21 AL |
2959 | } |
2960 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
2961 | ||
195aefde IE |
2962 | static int next_segment(unsigned long len, int offset) |
2963 | { | |
2964 | if (len > PAGE_SIZE - offset) | |
2965 | return PAGE_SIZE - offset; | |
2966 | else | |
2967 | return len; | |
2968 | } | |
2969 | ||
8e73485c PB |
2970 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
2971 | void *data, int offset, int len) | |
195aefde | 2972 | { |
e0506bcb IE |
2973 | int r; |
2974 | unsigned long addr; | |
195aefde | 2975 | |
8e73485c | 2976 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
2977 | if (kvm_is_error_hva(addr)) |
2978 | return -EFAULT; | |
3180a7fc | 2979 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 2980 | if (r) |
195aefde | 2981 | return -EFAULT; |
195aefde IE |
2982 | return 0; |
2983 | } | |
8e73485c PB |
2984 | |
2985 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
2986 | int len) | |
2987 | { | |
2988 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2989 | ||
2990 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2991 | } | |
195aefde IE |
2992 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
2993 | ||
8e73485c PB |
2994 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
2995 | int offset, int len) | |
2996 | { | |
2997 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2998 | ||
2999 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
3000 | } | |
3001 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
3002 | ||
195aefde IE |
3003 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
3004 | { | |
3005 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3006 | int seg; | |
3007 | int offset = offset_in_page(gpa); | |
3008 | int ret; | |
3009 | ||
3010 | while ((seg = next_segment(len, offset)) != 0) { | |
3011 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
3012 | if (ret < 0) | |
3013 | return ret; | |
3014 | offset = 0; | |
3015 | len -= seg; | |
3016 | data += seg; | |
3017 | ++gfn; | |
3018 | } | |
3019 | return 0; | |
3020 | } | |
3021 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
3022 | ||
8e73485c | 3023 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 3024 | { |
7ec54588 | 3025 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 3026 | int seg; |
7ec54588 | 3027 | int offset = offset_in_page(gpa); |
8e73485c PB |
3028 | int ret; |
3029 | ||
3030 | while ((seg = next_segment(len, offset)) != 0) { | |
3031 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
3032 | if (ret < 0) | |
3033 | return ret; | |
3034 | offset = 0; | |
3035 | len -= seg; | |
3036 | data += seg; | |
3037 | ++gfn; | |
3038 | } | |
3039 | return 0; | |
3040 | } | |
3041 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 3042 | |
8e73485c PB |
3043 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
3044 | void *data, int offset, unsigned long len) | |
3045 | { | |
3046 | int r; | |
3047 | unsigned long addr; | |
3048 | ||
3049 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
3050 | if (kvm_is_error_hva(addr)) |
3051 | return -EFAULT; | |
0aac03f0 | 3052 | pagefault_disable(); |
3180a7fc | 3053 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 3054 | pagefault_enable(); |
7ec54588 MT |
3055 | if (r) |
3056 | return -EFAULT; | |
3057 | return 0; | |
3058 | } | |
7ec54588 | 3059 | |
8e73485c PB |
3060 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
3061 | void *data, unsigned long len) | |
3062 | { | |
3063 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3064 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3065 | int offset = offset_in_page(gpa); | |
3066 | ||
3067 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
3068 | } | |
3069 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
3070 | ||
28bd726a PX |
3071 | static int __kvm_write_guest_page(struct kvm *kvm, |
3072 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 3073 | const void *data, int offset, int len) |
195aefde | 3074 | { |
e0506bcb IE |
3075 | int r; |
3076 | unsigned long addr; | |
195aefde | 3077 | |
251eb841 | 3078 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
3079 | if (kvm_is_error_hva(addr)) |
3080 | return -EFAULT; | |
8b0cedff | 3081 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 3082 | if (r) |
195aefde | 3083 | return -EFAULT; |
28bd726a | 3084 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
3085 | return 0; |
3086 | } | |
8e73485c PB |
3087 | |
3088 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
3089 | const void *data, int offset, int len) | |
3090 | { | |
3091 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
3092 | ||
28bd726a | 3093 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 3094 | } |
195aefde IE |
3095 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
3096 | ||
8e73485c PB |
3097 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3098 | const void *data, int offset, int len) | |
3099 | { | |
3100 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3101 | ||
28bd726a | 3102 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
3103 | } |
3104 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
3105 | ||
195aefde IE |
3106 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
3107 | unsigned long len) | |
3108 | { | |
3109 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3110 | int seg; | |
3111 | int offset = offset_in_page(gpa); | |
3112 | int ret; | |
3113 | ||
3114 | while ((seg = next_segment(len, offset)) != 0) { | |
3115 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
3116 | if (ret < 0) | |
3117 | return ret; | |
3118 | offset = 0; | |
3119 | len -= seg; | |
3120 | data += seg; | |
3121 | ++gfn; | |
3122 | } | |
3123 | return 0; | |
3124 | } | |
ff651cb6 | 3125 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 3126 | |
8e73485c PB |
3127 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
3128 | unsigned long len) | |
3129 | { | |
3130 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3131 | int seg; | |
3132 | int offset = offset_in_page(gpa); | |
3133 | int ret; | |
3134 | ||
3135 | while ((seg = next_segment(len, offset)) != 0) { | |
3136 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
3137 | if (ret < 0) | |
3138 | return ret; | |
3139 | offset = 0; | |
3140 | len -= seg; | |
3141 | data += seg; | |
3142 | ++gfn; | |
3143 | } | |
3144 | return 0; | |
3145 | } | |
3146 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
3147 | ||
5a2d4365 PB |
3148 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
3149 | struct gfn_to_hva_cache *ghc, | |
3150 | gpa_t gpa, unsigned long len) | |
49c7754c | 3151 | { |
49c7754c | 3152 | int offset = offset_in_page(gpa); |
8f964525 AH |
3153 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
3154 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
3155 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
3156 | gfn_t nr_pages_avail; | |
49c7754c | 3157 | |
6ad1e29f | 3158 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 3159 | ghc->generation = slots->generation; |
6ad1e29f SC |
3160 | |
3161 | if (start_gfn > end_gfn) { | |
3162 | ghc->hva = KVM_HVA_ERR_BAD; | |
3163 | return -EINVAL; | |
3164 | } | |
f1b9dd5e JM |
3165 | |
3166 | /* | |
3167 | * If the requested region crosses two memslots, we still | |
3168 | * verify that the entire region is valid here. | |
3169 | */ | |
6ad1e29f | 3170 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
3171 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
3172 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
3173 | &nr_pages_avail); | |
3174 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 3175 | return -EFAULT; |
f1b9dd5e JM |
3176 | } |
3177 | ||
3178 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 3179 | if (nr_pages_needed == 1) |
49c7754c | 3180 | ghc->hva += offset; |
f1b9dd5e | 3181 | else |
8f964525 | 3182 | ghc->memslot = NULL; |
f1b9dd5e | 3183 | |
6ad1e29f SC |
3184 | ghc->gpa = gpa; |
3185 | ghc->len = len; | |
3186 | return 0; | |
49c7754c | 3187 | } |
5a2d4365 | 3188 | |
4e335d9e | 3189 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
3190 | gpa_t gpa, unsigned long len) |
3191 | { | |
4e335d9e | 3192 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
3193 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
3194 | } | |
4e335d9e | 3195 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 3196 | |
4e335d9e | 3197 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
3198 | void *data, unsigned int offset, |
3199 | unsigned long len) | |
49c7754c | 3200 | { |
4e335d9e | 3201 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 3202 | int r; |
4ec6e863 | 3203 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 3204 | |
5f25e71e PB |
3205 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3206 | return -EINVAL; | |
8f964525 | 3207 | |
dc9ce71e SC |
3208 | if (slots->generation != ghc->generation) { |
3209 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3210 | return -EFAULT; | |
3211 | } | |
8f964525 | 3212 | |
49c7754c GN |
3213 | if (kvm_is_error_hva(ghc->hva)) |
3214 | return -EFAULT; | |
3215 | ||
fcfbc617 SC |
3216 | if (unlikely(!ghc->memslot)) |
3217 | return kvm_write_guest(kvm, gpa, data, len); | |
3218 | ||
4ec6e863 | 3219 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
3220 | if (r) |
3221 | return -EFAULT; | |
28bd726a | 3222 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
3223 | |
3224 | return 0; | |
3225 | } | |
4e335d9e | 3226 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 3227 | |
4e335d9e PB |
3228 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3229 | void *data, unsigned long len) | |
4ec6e863 | 3230 | { |
4e335d9e | 3231 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 3232 | } |
4e335d9e | 3233 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 3234 | |
0958f0ce VK |
3235 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3236 | void *data, unsigned int offset, | |
3237 | unsigned long len) | |
e03b644f | 3238 | { |
4e335d9e | 3239 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 3240 | int r; |
0958f0ce | 3241 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 3242 | |
5f25e71e PB |
3243 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3244 | return -EINVAL; | |
8f964525 | 3245 | |
dc9ce71e SC |
3246 | if (slots->generation != ghc->generation) { |
3247 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3248 | return -EFAULT; | |
3249 | } | |
8f964525 | 3250 | |
e03b644f GN |
3251 | if (kvm_is_error_hva(ghc->hva)) |
3252 | return -EFAULT; | |
3253 | ||
fcfbc617 | 3254 | if (unlikely(!ghc->memslot)) |
0958f0ce | 3255 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 3256 | |
0958f0ce | 3257 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
3258 | if (r) |
3259 | return -EFAULT; | |
3260 | ||
3261 | return 0; | |
3262 | } | |
0958f0ce VK |
3263 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
3264 | ||
3265 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
3266 | void *data, unsigned long len) | |
3267 | { | |
3268 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
3269 | } | |
4e335d9e | 3270 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 3271 | |
195aefde IE |
3272 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
3273 | { | |
2f541442 | 3274 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
3275 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3276 | int seg; | |
3277 | int offset = offset_in_page(gpa); | |
3278 | int ret; | |
3279 | ||
bfda0e84 | 3280 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 3281 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
3282 | if (ret < 0) |
3283 | return ret; | |
3284 | offset = 0; | |
3285 | len -= seg; | |
3286 | ++gfn; | |
3287 | } | |
3288 | return 0; | |
3289 | } | |
3290 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
3291 | ||
28bd726a | 3292 | void mark_page_dirty_in_slot(struct kvm *kvm, |
8283e36a | 3293 | const struct kvm_memory_slot *memslot, |
28bd726a | 3294 | gfn_t gfn) |
6aa8b732 | 3295 | { |
2efd61a6 DW |
3296 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); |
3297 | ||
e09fccb5 | 3298 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
2efd61a6 DW |
3299 | if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm)) |
3300 | return; | |
e09fccb5 | 3301 | #endif |
2efd61a6 | 3302 | |
044c59c4 | 3303 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 3304 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 3305 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 3306 | |
fb04a1ed | 3307 | if (kvm->dirty_ring_size) |
2efd61a6 | 3308 | kvm_dirty_ring_push(&vcpu->dirty_ring, |
fb04a1ed PX |
3309 | slot, rel_gfn); |
3310 | else | |
3311 | set_bit_le(rel_gfn, memslot->dirty_bitmap); | |
6aa8b732 AK |
3312 | } |
3313 | } | |
a6a0b05d | 3314 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 3315 | |
49c7754c GN |
3316 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
3317 | { | |
3318 | struct kvm_memory_slot *memslot; | |
3319 | ||
3320 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 3321 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 3322 | } |
2ba9f0d8 | 3323 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 3324 | |
8e73485c PB |
3325 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
3326 | { | |
3327 | struct kvm_memory_slot *memslot; | |
3328 | ||
3329 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 3330 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
3331 | } |
3332 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
3333 | ||
20b7035c JS |
3334 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
3335 | { | |
3336 | if (!vcpu->sigset_active) | |
3337 | return; | |
3338 | ||
3339 | /* | |
3340 | * This does a lockless modification of ->real_blocked, which is fine | |
3341 | * because, only current can change ->real_blocked and all readers of | |
3342 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
3343 | * of ->blocked. | |
3344 | */ | |
3345 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
3346 | } | |
3347 | ||
3348 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
3349 | { | |
3350 | if (!vcpu->sigset_active) | |
3351 | return; | |
3352 | ||
3353 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
3354 | sigemptyset(¤t->real_blocked); | |
3355 | } | |
3356 | ||
aca6ff29 WL |
3357 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
3358 | { | |
dee339b5 | 3359 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 3360 | |
2cbd7824 | 3361 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 3362 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3363 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
3364 | if (!grow) |
3365 | goto out; | |
3366 | ||
dee339b5 NW |
3367 | val *= grow; |
3368 | if (val < grow_start) | |
3369 | val = grow_start; | |
aca6ff29 | 3370 | |
258785ef DM |
3371 | if (val > vcpu->kvm->max_halt_poll_ns) |
3372 | val = vcpu->kvm->max_halt_poll_ns; | |
313f636d | 3373 | |
aca6ff29 | 3374 | vcpu->halt_poll_ns = val; |
7fa08e71 | 3375 | out: |
2cbd7824 | 3376 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3377 | } |
3378 | ||
3379 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
3380 | { | |
ae232ea4 | 3381 | unsigned int old, val, shrink, grow_start; |
aca6ff29 | 3382 | |
2cbd7824 | 3383 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 3384 | shrink = READ_ONCE(halt_poll_ns_shrink); |
ae232ea4 | 3385 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3386 | if (shrink == 0) |
aca6ff29 WL |
3387 | val = 0; |
3388 | else | |
6b6de68c | 3389 | val /= shrink; |
aca6ff29 | 3390 | |
ae232ea4 SS |
3391 | if (val < grow_start) |
3392 | val = 0; | |
3393 | ||
aca6ff29 | 3394 | vcpu->halt_poll_ns = val; |
2cbd7824 | 3395 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3396 | } |
3397 | ||
f7819512 PB |
3398 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
3399 | { | |
50c28f21 JS |
3400 | int ret = -EINTR; |
3401 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
3402 | ||
f7819512 PB |
3403 | if (kvm_arch_vcpu_runnable(vcpu)) { |
3404 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 3405 | goto out; |
f7819512 PB |
3406 | } |
3407 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 3408 | goto out; |
f7819512 | 3409 | if (signal_pending(current)) |
50c28f21 | 3410 | goto out; |
084071d5 MT |
3411 | if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu)) |
3412 | goto out; | |
f7819512 | 3413 | |
50c28f21 JS |
3414 | ret = 0; |
3415 | out: | |
3416 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
3417 | return ret; | |
f7819512 PB |
3418 | } |
3419 | ||
fac42688 SC |
3420 | /* |
3421 | * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is | |
3422 | * pending. This is mostly used when halting a vCPU, but may also be used | |
3423 | * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI. | |
3424 | */ | |
3425 | bool kvm_vcpu_block(struct kvm_vcpu *vcpu) | |
cb953129 | 3426 | { |
fac42688 SC |
3427 | struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); |
3428 | bool waited = false; | |
3429 | ||
c3858335 JZ |
3430 | vcpu->stat.generic.blocking = 1; |
3431 | ||
18869f26 | 3432 | preempt_disable(); |
fac42688 | 3433 | kvm_arch_vcpu_blocking(vcpu); |
fac42688 | 3434 | prepare_to_rcuwait(wait); |
18869f26 ML |
3435 | preempt_enable(); |
3436 | ||
fac42688 SC |
3437 | for (;;) { |
3438 | set_current_state(TASK_INTERRUPTIBLE); | |
3439 | ||
3440 | if (kvm_vcpu_check_block(vcpu) < 0) | |
3441 | break; | |
3442 | ||
3443 | waited = true; | |
3444 | schedule(); | |
3445 | } | |
fac42688 | 3446 | |
18869f26 ML |
3447 | preempt_disable(); |
3448 | finish_rcuwait(wait); | |
fac42688 | 3449 | kvm_arch_vcpu_unblocking(vcpu); |
18869f26 | 3450 | preempt_enable(); |
fac42688 | 3451 | |
c3858335 JZ |
3452 | vcpu->stat.generic.blocking = 0; |
3453 | ||
fac42688 SC |
3454 | return waited; |
3455 | } | |
3456 | ||
29e72893 SC |
3457 | static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start, |
3458 | ktime_t end, bool success) | |
cb953129 | 3459 | { |
30c94347 | 3460 | struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic; |
29e72893 SC |
3461 | u64 poll_ns = ktime_to_ns(ktime_sub(end, start)); |
3462 | ||
30c94347 SC |
3463 | ++vcpu->stat.generic.halt_attempted_poll; |
3464 | ||
3465 | if (success) { | |
3466 | ++vcpu->stat.generic.halt_successful_poll; | |
3467 | ||
3468 | if (!vcpu_valid_wakeup(vcpu)) | |
3469 | ++vcpu->stat.generic.halt_poll_invalid; | |
3470 | ||
3471 | stats->halt_poll_success_ns += poll_ns; | |
3472 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns); | |
3473 | } else { | |
3474 | stats->halt_poll_fail_ns += poll_ns; | |
3475 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns); | |
3476 | } | |
cb953129 DM |
3477 | } |
3478 | ||
b6958ce4 | 3479 | /* |
fac42688 SC |
3480 | * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt |
3481 | * polling is enabled, busy wait for a short time before blocking to avoid the | |
3482 | * expensive block+unblock sequence if a wake event arrives soon after the vCPU | |
3483 | * is halted. | |
b6958ce4 | 3484 | */ |
91b99ea7 | 3485 | void kvm_vcpu_halt(struct kvm_vcpu *vcpu) |
d3bef15f | 3486 | { |
6f390916 | 3487 | bool halt_poll_allowed = !kvm_arch_no_poll(vcpu); |
8df6a61c | 3488 | bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns; |
cb953129 | 3489 | ktime_t start, cur, poll_end; |
f7819512 | 3490 | bool waited = false; |
91b99ea7 | 3491 | u64 halt_ns; |
07ab0f8d | 3492 | |
cb953129 | 3493 | start = cur = poll_end = ktime_get(); |
8df6a61c | 3494 | if (do_halt_poll) { |
109a9826 | 3495 | ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns); |
f95ef0cd | 3496 | |
f7819512 PB |
3497 | do { |
3498 | /* | |
3499 | * This sets KVM_REQ_UNHALT if an interrupt | |
3500 | * arrives. | |
3501 | */ | |
30c94347 | 3502 | if (kvm_vcpu_check_block(vcpu) < 0) |
f7819512 | 3503 | goto out; |
74775654 | 3504 | cpu_relax(); |
cb953129 | 3505 | poll_end = cur = ktime_get(); |
6bd5b743 | 3506 | } while (kvm_vcpu_can_poll(cur, stop)); |
f7819512 | 3507 | } |
e5c239cf | 3508 | |
fac42688 | 3509 | waited = kvm_vcpu_block(vcpu); |
8ccba534 | 3510 | |
f7819512 | 3511 | cur = ktime_get(); |
87bcc5fa JZ |
3512 | if (waited) { |
3513 | vcpu->stat.generic.halt_wait_ns += | |
3514 | ktime_to_ns(cur) - ktime_to_ns(poll_end); | |
8ccba534 JZ |
3515 | KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist, |
3516 | ktime_to_ns(cur) - ktime_to_ns(poll_end)); | |
87bcc5fa | 3517 | } |
f7819512 | 3518 | out: |
91b99ea7 SC |
3519 | /* The total time the vCPU was "halted", including polling time. */ |
3520 | halt_ns = ktime_to_ns(cur) - ktime_to_ns(start); | |
aca6ff29 | 3521 | |
29e72893 SC |
3522 | /* |
3523 | * Note, halt-polling is considered successful so long as the vCPU was | |
3524 | * never actually scheduled out, i.e. even if the wake event arrived | |
3525 | * after of the halt-polling loop itself, but before the full wait. | |
3526 | */ | |
8df6a61c | 3527 | if (do_halt_poll) |
29e72893 | 3528 | update_halt_poll_stats(vcpu, start, poll_end, !waited); |
cb953129 | 3529 | |
6f390916 | 3530 | if (halt_poll_allowed) { |
44551b2f | 3531 | if (!vcpu_valid_wakeup(vcpu)) { |
aca6ff29 | 3532 | shrink_halt_poll_ns(vcpu); |
acd05785 | 3533 | } else if (vcpu->kvm->max_halt_poll_ns) { |
91b99ea7 | 3534 | if (halt_ns <= vcpu->halt_poll_ns) |
44551b2f WL |
3535 | ; |
3536 | /* we had a long block, shrink polling */ | |
acd05785 | 3537 | else if (vcpu->halt_poll_ns && |
91b99ea7 | 3538 | halt_ns > vcpu->kvm->max_halt_poll_ns) |
44551b2f WL |
3539 | shrink_halt_poll_ns(vcpu); |
3540 | /* we had a short halt and our poll time is too small */ | |
acd05785 | 3541 | else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && |
91b99ea7 | 3542 | halt_ns < vcpu->kvm->max_halt_poll_ns) |
44551b2f WL |
3543 | grow_halt_poll_ns(vcpu); |
3544 | } else { | |
3545 | vcpu->halt_poll_ns = 0; | |
3546 | } | |
3547 | } | |
aca6ff29 | 3548 | |
91b99ea7 | 3549 | trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu)); |
b6958ce4 | 3550 | } |
91b99ea7 | 3551 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
b6958ce4 | 3552 | |
178f02ff | 3553 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 3554 | { |
d92a5d1c | 3555 | if (__kvm_vcpu_wake_up(vcpu)) { |
d73eb57b | 3556 | WRITE_ONCE(vcpu->ready, true); |
0193cc90 | 3557 | ++vcpu->stat.generic.halt_wakeup; |
178f02ff | 3558 | return true; |
b6d33834 CD |
3559 | } |
3560 | ||
178f02ff | 3561 | return false; |
dd1a4cc1 RK |
3562 | } |
3563 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
3564 | ||
0266c894 | 3565 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
3566 | /* |
3567 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
3568 | */ | |
3569 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
3570 | { | |
85b64045 | 3571 | int me, cpu; |
dd1a4cc1 | 3572 | |
178f02ff RK |
3573 | if (kvm_vcpu_wake_up(vcpu)) |
3574 | return; | |
3575 | ||
aefdc2ed PB |
3576 | me = get_cpu(); |
3577 | /* | |
3578 | * The only state change done outside the vcpu mutex is IN_GUEST_MODE | |
3579 | * to EXITING_GUEST_MODE. Therefore the moderately expensive "should | |
3580 | * kick" check does not need atomic operations if kvm_vcpu_kick is used | |
3581 | * within the vCPU thread itself. | |
3582 | */ | |
3583 | if (vcpu == __this_cpu_read(kvm_running_vcpu)) { | |
3584 | if (vcpu->mode == IN_GUEST_MODE) | |
3585 | WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE); | |
3586 | goto out; | |
3587 | } | |
3588 | ||
85b64045 SC |
3589 | /* |
3590 | * Note, the vCPU could get migrated to a different pCPU at any point | |
3591 | * after kvm_arch_vcpu_should_kick(), which could result in sending an | |
3592 | * IPI to the previous pCPU. But, that's ok because the purpose of the | |
3593 | * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the | |
3594 | * vCPU also requires it to leave IN_GUEST_MODE. | |
3595 | */ | |
85b64045 SC |
3596 | if (kvm_arch_vcpu_should_kick(vcpu)) { |
3597 | cpu = READ_ONCE(vcpu->cpu); | |
3598 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
b6d33834 | 3599 | smp_send_reschedule(cpu); |
85b64045 | 3600 | } |
aefdc2ed | 3601 | out: |
b6d33834 CD |
3602 | put_cpu(); |
3603 | } | |
a20ed54d | 3604 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 3605 | #endif /* !CONFIG_S390 */ |
b6d33834 | 3606 | |
fa93384f | 3607 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
3608 | { |
3609 | struct pid *pid; | |
3610 | struct task_struct *task = NULL; | |
fa93384f | 3611 | int ret = 0; |
41628d33 KW |
3612 | |
3613 | rcu_read_lock(); | |
3614 | pid = rcu_dereference(target->pid); | |
3615 | if (pid) | |
27fbe64b | 3616 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
3617 | rcu_read_unlock(); |
3618 | if (!task) | |
c45c528e | 3619 | return ret; |
c45c528e | 3620 | ret = yield_to(task, 1); |
41628d33 | 3621 | put_task_struct(task); |
c45c528e R |
3622 | |
3623 | return ret; | |
41628d33 KW |
3624 | } |
3625 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
3626 | ||
06e48c51 R |
3627 | /* |
3628 | * Helper that checks whether a VCPU is eligible for directed yield. | |
3629 | * Most eligible candidate to yield is decided by following heuristics: | |
3630 | * | |
3631 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
3632 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 3633 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
3634 | * |
3635 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
3636 | * chance last time (mostly it has become eligible now since we have probably | |
3637 | * yielded to lockholder in last iteration. This is done by toggling | |
3638 | * @dy_eligible each time a VCPU checked for eligibility.) | |
3639 | * | |
3640 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
3641 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
3642 | * burning. Giving priority for a potential lock-holder increases lock | |
3643 | * progress. | |
3644 | * | |
3645 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
3646 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
3647 | * and continue with next VCPU and so on. | |
3648 | */ | |
7940876e | 3649 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 3650 | { |
4a55dd72 | 3651 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
3652 | bool eligible; |
3653 | ||
3654 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 3655 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
3656 | |
3657 | if (vcpu->spin_loop.in_spin_loop) | |
3658 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
3659 | ||
3660 | return eligible; | |
4a55dd72 SW |
3661 | #else |
3662 | return true; | |
06e48c51 | 3663 | #endif |
4a55dd72 | 3664 | } |
c45c528e | 3665 | |
17e433b5 WL |
3666 | /* |
3667 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
3668 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
3669 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
3670 | */ | |
3671 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
3672 | { | |
3673 | return kvm_arch_vcpu_runnable(vcpu); | |
3674 | } | |
3675 | ||
3676 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
3677 | { | |
3678 | if (kvm_arch_dy_runnable(vcpu)) | |
3679 | return true; | |
3680 | ||
3681 | #ifdef CONFIG_KVM_ASYNC_PF | |
3682 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
3683 | return true; | |
3684 | #endif | |
3685 | ||
3686 | return false; | |
3687 | } | |
3688 | ||
52acd22f WL |
3689 | bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
3690 | { | |
3691 | return false; | |
3692 | } | |
3693 | ||
199b5763 | 3694 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 3695 | { |
217ece61 RR |
3696 | struct kvm *kvm = me->kvm; |
3697 | struct kvm_vcpu *vcpu; | |
3698 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
46808a4c | 3699 | unsigned long i; |
217ece61 | 3700 | int yielded = 0; |
c45c528e | 3701 | int try = 3; |
217ece61 | 3702 | int pass; |
d255f4f2 | 3703 | |
4c088493 | 3704 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
3705 | /* |
3706 | * We boost the priority of a VCPU that is runnable but not | |
3707 | * currently running, because it got preempted by something | |
3708 | * else and called schedule in __vcpu_run. Hopefully that | |
3709 | * VCPU is holding the lock that we need and will release it. | |
3710 | * We approximate round-robin by starting at the last boosted VCPU. | |
3711 | */ | |
c45c528e | 3712 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 3713 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 3714 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
3715 | i = last_boosted_vcpu; |
3716 | continue; | |
3717 | } else if (pass && i > last_boosted_vcpu) | |
3718 | break; | |
d73eb57b | 3719 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 3720 | continue; |
217ece61 RR |
3721 | if (vcpu == me) |
3722 | continue; | |
d92a5d1c | 3723 | if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu)) |
217ece61 | 3724 | continue; |
046ddeed | 3725 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
52acd22f WL |
3726 | !kvm_arch_dy_has_pending_interrupt(vcpu) && |
3727 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 3728 | continue; |
06e48c51 R |
3729 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
3730 | continue; | |
c45c528e R |
3731 | |
3732 | yielded = kvm_vcpu_yield_to(vcpu); | |
3733 | if (yielded > 0) { | |
217ece61 | 3734 | kvm->last_boosted_vcpu = i; |
217ece61 | 3735 | break; |
c45c528e R |
3736 | } else if (yielded < 0) { |
3737 | try--; | |
3738 | if (!try) | |
3739 | break; | |
217ece61 | 3740 | } |
217ece61 RR |
3741 | } |
3742 | } | |
4c088493 | 3743 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
3744 | |
3745 | /* Ensure vcpu is not eligible during next spinloop */ | |
3746 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
3747 | } |
3748 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
3749 | ||
fb04a1ed PX |
3750 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
3751 | { | |
dc70ec21 | 3752 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
fb04a1ed PX |
3753 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && |
3754 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
3755 | kvm->dirty_ring_size / PAGE_SIZE); | |
3756 | #else | |
3757 | return false; | |
3758 | #endif | |
3759 | } | |
3760 | ||
1499fa80 | 3761 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 3762 | { |
11bac800 | 3763 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
3764 | struct page *page; |
3765 | ||
e4a533a4 | 3766 | if (vmf->pgoff == 0) |
039576c0 | 3767 | page = virt_to_page(vcpu->run); |
09566765 | 3768 | #ifdef CONFIG_X86 |
e4a533a4 | 3769 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 3770 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 3771 | #endif |
4b4357e0 | 3772 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3773 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
3774 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 3775 | #endif |
fb04a1ed PX |
3776 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
3777 | page = kvm_dirty_ring_get_page( | |
3778 | &vcpu->dirty_ring, | |
3779 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 3780 | else |
5b1c1493 | 3781 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 3782 | get_page(page); |
e4a533a4 | 3783 | vmf->page = page; |
3784 | return 0; | |
9a2bb7f4 AK |
3785 | } |
3786 | ||
f0f37e2f | 3787 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 3788 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
3789 | }; |
3790 | ||
3791 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
3792 | { | |
fb04a1ed | 3793 | struct kvm_vcpu *vcpu = file->private_data; |
11476d27 | 3794 | unsigned long pages = vma_pages(vma); |
fb04a1ed PX |
3795 | |
3796 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
3797 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
3798 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
3799 | return -EINVAL; | |
3800 | ||
9a2bb7f4 AK |
3801 | vma->vm_ops = &kvm_vcpu_vm_ops; |
3802 | return 0; | |
3803 | } | |
3804 | ||
bccf2150 AK |
3805 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
3806 | { | |
3807 | struct kvm_vcpu *vcpu = filp->private_data; | |
3808 | ||
66c0b394 | 3809 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
3810 | return 0; |
3811 | } | |
3812 | ||
70375c2d | 3813 | static const struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
3814 | .release = kvm_vcpu_release, |
3815 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 3816 | .mmap = kvm_vcpu_mmap, |
6038f373 | 3817 | .llseek = noop_llseek, |
7ddfd3e0 | 3818 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
3819 | }; |
3820 | ||
3821 | /* | |
3822 | * Allocates an inode for the vcpu. | |
3823 | */ | |
3824 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
3825 | { | |
e46b4692 MY |
3826 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
3827 | ||
3828 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
3829 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
3830 | } |
3831 | ||
e36de87d VP |
3832 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
3833 | static int vcpu_get_pid(void *data, u64 *val) | |
3834 | { | |
3835 | struct kvm_vcpu *vcpu = (struct kvm_vcpu *) data; | |
3836 | *val = pid_nr(rcu_access_pointer(vcpu->pid)); | |
3837 | return 0; | |
3838 | } | |
3839 | ||
3840 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n"); | |
3841 | ||
3e7093d0 | 3842 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 3843 | { |
d56f5136 | 3844 | struct dentry *debugfs_dentry; |
45b5939e | 3845 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 3846 | |
45b5939e | 3847 | if (!debugfs_initialized()) |
3e7093d0 | 3848 | return; |
45b5939e LC |
3849 | |
3850 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
3851 | debugfs_dentry = debugfs_create_dir(dir_name, |
3852 | vcpu->kvm->debugfs_dentry); | |
e36de87d VP |
3853 | debugfs_create_file("pid", 0444, debugfs_dentry, vcpu, |
3854 | &vcpu_get_pid_fops); | |
45b5939e | 3855 | |
d56f5136 | 3856 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
45b5939e | 3857 | } |
e36de87d | 3858 | #endif |
45b5939e | 3859 | |
c5ea7660 AK |
3860 | /* |
3861 | * Creates some virtual cpus. Good luck creating more than one. | |
3862 | */ | |
73880c80 | 3863 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
3864 | { |
3865 | int r; | |
e09fefde | 3866 | struct kvm_vcpu *vcpu; |
8bd826d6 | 3867 | struct page *page; |
c5ea7660 | 3868 | |
a1c42dde | 3869 | if (id >= KVM_MAX_VCPU_IDS) |
338c7dba AH |
3870 | return -EINVAL; |
3871 | ||
6c7caebc | 3872 | mutex_lock(&kvm->lock); |
f502cc56 | 3873 | if (kvm->created_vcpus >= kvm->max_vcpus) { |
6c7caebc PB |
3874 | mutex_unlock(&kvm->lock); |
3875 | return -EINVAL; | |
3876 | } | |
3877 | ||
1d5e740d ZG |
3878 | r = kvm_arch_vcpu_precreate(kvm, id); |
3879 | if (r) { | |
3880 | mutex_unlock(&kvm->lock); | |
3881 | return r; | |
3882 | } | |
3883 | ||
6c7caebc PB |
3884 | kvm->created_vcpus++; |
3885 | mutex_unlock(&kvm->lock); | |
3886 | ||
85f47930 | 3887 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
e529ef66 SC |
3888 | if (!vcpu) { |
3889 | r = -ENOMEM; | |
6c7caebc PB |
3890 | goto vcpu_decrement; |
3891 | } | |
c5ea7660 | 3892 | |
fcd97ad5 | 3893 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 3894 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
3895 | if (!page) { |
3896 | r = -ENOMEM; | |
e529ef66 | 3897 | goto vcpu_free; |
8bd826d6 SC |
3898 | } |
3899 | vcpu->run = page_address(page); | |
3900 | ||
3901 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
3902 | |
3903 | r = kvm_arch_vcpu_create(vcpu); | |
3904 | if (r) | |
8bd826d6 | 3905 | goto vcpu_free_run_page; |
e529ef66 | 3906 | |
fb04a1ed PX |
3907 | if (kvm->dirty_ring_size) { |
3908 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
3909 | id, kvm->dirty_ring_size); | |
3910 | if (r) | |
3911 | goto arch_vcpu_destroy; | |
3912 | } | |
3913 | ||
11ec2804 | 3914 | mutex_lock(&kvm->lock); |
e09fefde DH |
3915 | if (kvm_get_vcpu_by_id(kvm, id)) { |
3916 | r = -EEXIST; | |
3917 | goto unlock_vcpu_destroy; | |
3918 | } | |
73880c80 | 3919 | |
8750e72a | 3920 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
c5b07754 MZ |
3921 | r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT); |
3922 | BUG_ON(r == -EBUSY); | |
3923 | if (r) | |
3924 | goto unlock_vcpu_destroy; | |
c5ea7660 | 3925 | |
fb3f0f51 | 3926 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 3927 | kvm_get_kvm(kvm); |
bccf2150 | 3928 | r = create_vcpu_fd(vcpu); |
73880c80 | 3929 | if (r < 0) { |
c5b07754 | 3930 | xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx); |
149487bd | 3931 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 3932 | goto unlock_vcpu_destroy; |
73880c80 GN |
3933 | } |
3934 | ||
dd489240 | 3935 | /* |
c5b07754 MZ |
3936 | * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu |
3937 | * pointer before kvm->online_vcpu's incremented value. | |
dd489240 | 3938 | */ |
73880c80 GN |
3939 | smp_wmb(); |
3940 | atomic_inc(&kvm->online_vcpus); | |
3941 | ||
73880c80 | 3942 | mutex_unlock(&kvm->lock); |
42897d86 | 3943 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 3944 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 3945 | return r; |
39c3b86e | 3946 | |
d780592b | 3947 | unlock_vcpu_destroy: |
7d8fece6 | 3948 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
3949 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
3950 | arch_vcpu_destroy: | |
d40ccc62 | 3951 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
3952 | vcpu_free_run_page: |
3953 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
3954 | vcpu_free: |
3955 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
3956 | vcpu_decrement: |
3957 | mutex_lock(&kvm->lock); | |
3958 | kvm->created_vcpus--; | |
3959 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
3960 | return r; |
3961 | } | |
3962 | ||
1961d276 AK |
3963 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
3964 | { | |
3965 | if (sigset) { | |
3966 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
3967 | vcpu->sigset_active = 1; | |
3968 | vcpu->sigset = *sigset; | |
3969 | } else | |
3970 | vcpu->sigset_active = 0; | |
3971 | return 0; | |
3972 | } | |
3973 | ||
ce55c049 JZ |
3974 | static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer, |
3975 | size_t size, loff_t *offset) | |
3976 | { | |
3977 | struct kvm_vcpu *vcpu = file->private_data; | |
3978 | ||
3979 | return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header, | |
3980 | &kvm_vcpu_stats_desc[0], &vcpu->stat, | |
3981 | sizeof(vcpu->stat), user_buffer, size, offset); | |
3982 | } | |
3983 | ||
3984 | static const struct file_operations kvm_vcpu_stats_fops = { | |
3985 | .read = kvm_vcpu_stats_read, | |
3986 | .llseek = noop_llseek, | |
3987 | }; | |
3988 | ||
3989 | static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu) | |
3990 | { | |
3991 | int fd; | |
3992 | struct file *file; | |
3993 | char name[15 + ITOA_MAX_LEN + 1]; | |
3994 | ||
3995 | snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id); | |
3996 | ||
3997 | fd = get_unused_fd_flags(O_CLOEXEC); | |
3998 | if (fd < 0) | |
3999 | return fd; | |
4000 | ||
4001 | file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY); | |
4002 | if (IS_ERR(file)) { | |
4003 | put_unused_fd(fd); | |
4004 | return PTR_ERR(file); | |
4005 | } | |
4006 | file->f_mode |= FMODE_PREAD; | |
4007 | fd_install(fd, file); | |
4008 | ||
4009 | return fd; | |
4010 | } | |
4011 | ||
bccf2150 AK |
4012 | static long kvm_vcpu_ioctl(struct file *filp, |
4013 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 4014 | { |
bccf2150 | 4015 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 4016 | void __user *argp = (void __user *)arg; |
313a3dc7 | 4017 | int r; |
fa3795a7 DH |
4018 | struct kvm_fpu *fpu = NULL; |
4019 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 4020 | |
f4d31653 | 4021 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
6d4e4c4f | 4022 | return -EIO; |
2122ff5e | 4023 | |
2ea75be3 DM |
4024 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
4025 | return -EINVAL; | |
4026 | ||
2122ff5e | 4027 | /* |
5cb0944c PB |
4028 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
4029 | * execution; mutex_lock() would break them. | |
2122ff5e | 4030 | */ |
5cb0944c PB |
4031 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
4032 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 4033 | return r; |
2122ff5e | 4034 | |
ec7660cc CD |
4035 | if (mutex_lock_killable(&vcpu->mutex)) |
4036 | return -EINTR; | |
6aa8b732 | 4037 | switch (ioctl) { |
0e4524a5 CB |
4038 | case KVM_RUN: { |
4039 | struct pid *oldpid; | |
f0fe5108 AK |
4040 | r = -EINVAL; |
4041 | if (arg) | |
4042 | goto out; | |
0e4524a5 | 4043 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 4044 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 4045 | /* The thread running this VCPU changed. */ |
bd2a6394 | 4046 | struct pid *newpid; |
f95ef0cd | 4047 | |
bd2a6394 CD |
4048 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
4049 | if (r) | |
4050 | break; | |
4051 | ||
4052 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
4053 | rcu_assign_pointer(vcpu->pid, newpid); |
4054 | if (oldpid) | |
4055 | synchronize_rcu(); | |
4056 | put_pid(oldpid); | |
4057 | } | |
1b94f6f8 | 4058 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 4059 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 4060 | break; |
0e4524a5 | 4061 | } |
6aa8b732 | 4062 | case KVM_GET_REGS: { |
3e4bb3ac | 4063 | struct kvm_regs *kvm_regs; |
6aa8b732 | 4064 | |
3e4bb3ac | 4065 | r = -ENOMEM; |
b12ce36a | 4066 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 4067 | if (!kvm_regs) |
6aa8b732 | 4068 | goto out; |
3e4bb3ac XZ |
4069 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
4070 | if (r) | |
4071 | goto out_free1; | |
6aa8b732 | 4072 | r = -EFAULT; |
3e4bb3ac XZ |
4073 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
4074 | goto out_free1; | |
6aa8b732 | 4075 | r = 0; |
3e4bb3ac XZ |
4076 | out_free1: |
4077 | kfree(kvm_regs); | |
6aa8b732 AK |
4078 | break; |
4079 | } | |
4080 | case KVM_SET_REGS: { | |
3e4bb3ac | 4081 | struct kvm_regs *kvm_regs; |
6aa8b732 | 4082 | |
ff5c2c03 SL |
4083 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
4084 | if (IS_ERR(kvm_regs)) { | |
4085 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 4086 | goto out; |
ff5c2c03 | 4087 | } |
3e4bb3ac | 4088 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 4089 | kfree(kvm_regs); |
6aa8b732 AK |
4090 | break; |
4091 | } | |
4092 | case KVM_GET_SREGS: { | |
b12ce36a BG |
4093 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
4094 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
4095 | r = -ENOMEM; |
4096 | if (!kvm_sregs) | |
4097 | goto out; | |
4098 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
4099 | if (r) |
4100 | goto out; | |
4101 | r = -EFAULT; | |
fa3795a7 | 4102 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
4103 | goto out; |
4104 | r = 0; | |
4105 | break; | |
4106 | } | |
4107 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
4108 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
4109 | if (IS_ERR(kvm_sregs)) { | |
4110 | r = PTR_ERR(kvm_sregs); | |
18595411 | 4111 | kvm_sregs = NULL; |
6aa8b732 | 4112 | goto out; |
ff5c2c03 | 4113 | } |
fa3795a7 | 4114 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
4115 | break; |
4116 | } | |
62d9f0db MT |
4117 | case KVM_GET_MP_STATE: { |
4118 | struct kvm_mp_state mp_state; | |
4119 | ||
4120 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
4121 | if (r) | |
4122 | goto out; | |
4123 | r = -EFAULT; | |
893bdbf1 | 4124 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
4125 | goto out; |
4126 | r = 0; | |
4127 | break; | |
4128 | } | |
4129 | case KVM_SET_MP_STATE: { | |
4130 | struct kvm_mp_state mp_state; | |
4131 | ||
4132 | r = -EFAULT; | |
893bdbf1 | 4133 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
4134 | goto out; |
4135 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
4136 | break; |
4137 | } | |
6aa8b732 AK |
4138 | case KVM_TRANSLATE: { |
4139 | struct kvm_translation tr; | |
4140 | ||
4141 | r = -EFAULT; | |
893bdbf1 | 4142 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 4143 | goto out; |
8b006791 | 4144 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
4145 | if (r) |
4146 | goto out; | |
4147 | r = -EFAULT; | |
893bdbf1 | 4148 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
4149 | goto out; |
4150 | r = 0; | |
4151 | break; | |
4152 | } | |
d0bfb940 JK |
4153 | case KVM_SET_GUEST_DEBUG: { |
4154 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
4155 | |
4156 | r = -EFAULT; | |
893bdbf1 | 4157 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 4158 | goto out; |
d0bfb940 | 4159 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
4160 | break; |
4161 | } | |
1961d276 AK |
4162 | case KVM_SET_SIGNAL_MASK: { |
4163 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4164 | struct kvm_signal_mask kvm_sigmask; | |
4165 | sigset_t sigset, *p; | |
4166 | ||
4167 | p = NULL; | |
4168 | if (argp) { | |
4169 | r = -EFAULT; | |
4170 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4171 | sizeof(kvm_sigmask))) |
1961d276 AK |
4172 | goto out; |
4173 | r = -EINVAL; | |
893bdbf1 | 4174 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
4175 | goto out; |
4176 | r = -EFAULT; | |
4177 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 4178 | sizeof(sigset))) |
1961d276 AK |
4179 | goto out; |
4180 | p = &sigset; | |
4181 | } | |
376d41ff | 4182 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
4183 | break; |
4184 | } | |
b8836737 | 4185 | case KVM_GET_FPU: { |
b12ce36a | 4186 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
4187 | r = -ENOMEM; |
4188 | if (!fpu) | |
4189 | goto out; | |
4190 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
4191 | if (r) |
4192 | goto out; | |
4193 | r = -EFAULT; | |
fa3795a7 | 4194 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
4195 | goto out; |
4196 | r = 0; | |
4197 | break; | |
4198 | } | |
4199 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
4200 | fpu = memdup_user(argp, sizeof(*fpu)); |
4201 | if (IS_ERR(fpu)) { | |
4202 | r = PTR_ERR(fpu); | |
18595411 | 4203 | fpu = NULL; |
b8836737 | 4204 | goto out; |
ff5c2c03 | 4205 | } |
fa3795a7 | 4206 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
4207 | break; |
4208 | } | |
ce55c049 JZ |
4209 | case KVM_GET_STATS_FD: { |
4210 | r = kvm_vcpu_ioctl_get_stats_fd(vcpu); | |
4211 | break; | |
4212 | } | |
bccf2150 | 4213 | default: |
313a3dc7 | 4214 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
4215 | } |
4216 | out: | |
ec7660cc | 4217 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
4218 | kfree(fpu); |
4219 | kfree(kvm_sregs); | |
bccf2150 AK |
4220 | return r; |
4221 | } | |
4222 | ||
de8e5d74 | 4223 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
4224 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
4225 | unsigned int ioctl, unsigned long arg) | |
4226 | { | |
4227 | struct kvm_vcpu *vcpu = filp->private_data; | |
4228 | void __user *argp = compat_ptr(arg); | |
4229 | int r; | |
4230 | ||
f4d31653 | 4231 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
1dda606c AG |
4232 | return -EIO; |
4233 | ||
4234 | switch (ioctl) { | |
4235 | case KVM_SET_SIGNAL_MASK: { | |
4236 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4237 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
4238 | sigset_t sigset; |
4239 | ||
4240 | if (argp) { | |
4241 | r = -EFAULT; | |
4242 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4243 | sizeof(kvm_sigmask))) |
1dda606c AG |
4244 | goto out; |
4245 | r = -EINVAL; | |
3968cf62 | 4246 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
4247 | goto out; |
4248 | r = -EFAULT; | |
1393b4aa PB |
4249 | if (get_compat_sigset(&sigset, |
4250 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 4251 | goto out; |
760a9a30 AC |
4252 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
4253 | } else | |
4254 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
4255 | break; |
4256 | } | |
4257 | default: | |
4258 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
4259 | } | |
4260 | ||
4261 | out: | |
4262 | return r; | |
4263 | } | |
4264 | #endif | |
4265 | ||
a1cd3f08 CLG |
4266 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
4267 | { | |
4268 | struct kvm_device *dev = filp->private_data; | |
4269 | ||
4270 | if (dev->ops->mmap) | |
4271 | return dev->ops->mmap(dev, vma); | |
4272 | ||
4273 | return -ENODEV; | |
4274 | } | |
4275 | ||
852b6d57 SW |
4276 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
4277 | int (*accessor)(struct kvm_device *dev, | |
4278 | struct kvm_device_attr *attr), | |
4279 | unsigned long arg) | |
4280 | { | |
4281 | struct kvm_device_attr attr; | |
4282 | ||
4283 | if (!accessor) | |
4284 | return -EPERM; | |
4285 | ||
4286 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
4287 | return -EFAULT; | |
4288 | ||
4289 | return accessor(dev, &attr); | |
4290 | } | |
4291 | ||
4292 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
4293 | unsigned long arg) | |
4294 | { | |
4295 | struct kvm_device *dev = filp->private_data; | |
4296 | ||
f4d31653 | 4297 | if (dev->kvm->mm != current->mm || dev->kvm->vm_dead) |
ddba9180 SC |
4298 | return -EIO; |
4299 | ||
852b6d57 SW |
4300 | switch (ioctl) { |
4301 | case KVM_SET_DEVICE_ATTR: | |
4302 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
4303 | case KVM_GET_DEVICE_ATTR: | |
4304 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
4305 | case KVM_HAS_DEVICE_ATTR: | |
4306 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
4307 | default: | |
4308 | if (dev->ops->ioctl) | |
4309 | return dev->ops->ioctl(dev, ioctl, arg); | |
4310 | ||
4311 | return -ENOTTY; | |
4312 | } | |
4313 | } | |
4314 | ||
852b6d57 SW |
4315 | static int kvm_device_release(struct inode *inode, struct file *filp) |
4316 | { | |
4317 | struct kvm_device *dev = filp->private_data; | |
4318 | struct kvm *kvm = dev->kvm; | |
4319 | ||
2bde9b3e CLG |
4320 | if (dev->ops->release) { |
4321 | mutex_lock(&kvm->lock); | |
4322 | list_del(&dev->vm_node); | |
4323 | dev->ops->release(dev); | |
4324 | mutex_unlock(&kvm->lock); | |
4325 | } | |
4326 | ||
852b6d57 SW |
4327 | kvm_put_kvm(kvm); |
4328 | return 0; | |
4329 | } | |
4330 | ||
4331 | static const struct file_operations kvm_device_fops = { | |
4332 | .unlocked_ioctl = kvm_device_ioctl, | |
4333 | .release = kvm_device_release, | |
7ddfd3e0 | 4334 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 4335 | .mmap = kvm_device_mmap, |
852b6d57 SW |
4336 | }; |
4337 | ||
4338 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
4339 | { | |
4340 | if (filp->f_op != &kvm_device_fops) | |
4341 | return NULL; | |
4342 | ||
4343 | return filp->private_data; | |
4344 | } | |
4345 | ||
8538cb22 | 4346 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 4347 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
4348 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
4349 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 4350 | #endif |
d60eacb0 WD |
4351 | }; |
4352 | ||
8538cb22 | 4353 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
4354 | { |
4355 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
4356 | return -ENOSPC; | |
4357 | ||
4358 | if (kvm_device_ops_table[type] != NULL) | |
4359 | return -EEXIST; | |
4360 | ||
4361 | kvm_device_ops_table[type] = ops; | |
4362 | return 0; | |
4363 | } | |
4364 | ||
571ee1b6 WL |
4365 | void kvm_unregister_device_ops(u32 type) |
4366 | { | |
4367 | if (kvm_device_ops_table[type] != NULL) | |
4368 | kvm_device_ops_table[type] = NULL; | |
4369 | } | |
4370 | ||
852b6d57 SW |
4371 | static int kvm_ioctl_create_device(struct kvm *kvm, |
4372 | struct kvm_create_device *cd) | |
4373 | { | |
8538cb22 | 4374 | const struct kvm_device_ops *ops = NULL; |
852b6d57 SW |
4375 | struct kvm_device *dev; |
4376 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 4377 | int type; |
852b6d57 SW |
4378 | int ret; |
4379 | ||
d60eacb0 WD |
4380 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
4381 | return -ENODEV; | |
4382 | ||
1d487e9b PB |
4383 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
4384 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 4385 | if (ops == NULL) |
852b6d57 | 4386 | return -ENODEV; |
852b6d57 SW |
4387 | |
4388 | if (test) | |
4389 | return 0; | |
4390 | ||
b12ce36a | 4391 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
4392 | if (!dev) |
4393 | return -ENOMEM; | |
4394 | ||
4395 | dev->ops = ops; | |
4396 | dev->kvm = kvm; | |
852b6d57 | 4397 | |
a28ebea2 | 4398 | mutex_lock(&kvm->lock); |
1d487e9b | 4399 | ret = ops->create(dev, type); |
852b6d57 | 4400 | if (ret < 0) { |
a28ebea2 | 4401 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
4402 | kfree(dev); |
4403 | return ret; | |
4404 | } | |
a28ebea2 CD |
4405 | list_add(&dev->vm_node, &kvm->devices); |
4406 | mutex_unlock(&kvm->lock); | |
852b6d57 | 4407 | |
023e9fdd CD |
4408 | if (ops->init) |
4409 | ops->init(dev); | |
4410 | ||
cfa39381 | 4411 | kvm_get_kvm(kvm); |
24009b05 | 4412 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 4413 | if (ret < 0) { |
149487bd | 4414 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
4415 | mutex_lock(&kvm->lock); |
4416 | list_del(&dev->vm_node); | |
e8bc2427 AK |
4417 | if (ops->release) |
4418 | ops->release(dev); | |
a28ebea2 | 4419 | mutex_unlock(&kvm->lock); |
e8bc2427 AK |
4420 | if (ops->destroy) |
4421 | ops->destroy(dev); | |
852b6d57 SW |
4422 | return ret; |
4423 | } | |
4424 | ||
852b6d57 SW |
4425 | cd->fd = ret; |
4426 | return 0; | |
4427 | } | |
4428 | ||
92b591a4 AG |
4429 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
4430 | { | |
4431 | switch (arg) { | |
4432 | case KVM_CAP_USER_MEMORY: | |
4433 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
4434 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
4435 | case KVM_CAP_INTERNAL_ERROR_DATA: |
4436 | #ifdef CONFIG_HAVE_KVM_MSI | |
4437 | case KVM_CAP_SIGNAL_MSI: | |
4438 | #endif | |
297e2105 | 4439 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 4440 | case KVM_CAP_IRQFD: |
92b591a4 AG |
4441 | case KVM_CAP_IRQFD_RESAMPLE: |
4442 | #endif | |
e9ea5069 | 4443 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 4444 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 4445 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 4446 | case KVM_CAP_HALT_POLL: |
92b591a4 | 4447 | return 1; |
4b4357e0 | 4448 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
4449 | case KVM_CAP_COALESCED_MMIO: |
4450 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
4451 | case KVM_CAP_COALESCED_PIO: |
4452 | return 1; | |
30422558 | 4453 | #endif |
3c9bd400 JZ |
4454 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4455 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
4456 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
4457 | #endif | |
92b591a4 AG |
4458 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4459 | case KVM_CAP_IRQ_ROUTING: | |
4460 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
4461 | #endif |
4462 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
4463 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
4464 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 4465 | #endif |
c110ae57 PB |
4466 | case KVM_CAP_NR_MEMSLOTS: |
4467 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed | 4468 | case KVM_CAP_DIRTY_LOG_RING: |
dc70ec21 | 4469 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
fb04a1ed PX |
4470 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); |
4471 | #else | |
4472 | return 0; | |
4473 | #endif | |
ce55c049 | 4474 | case KVM_CAP_BINARY_STATS_FD: |
d495f942 | 4475 | case KVM_CAP_SYSTEM_EVENT_DATA: |
ce55c049 | 4476 | return 1; |
92b591a4 AG |
4477 | default: |
4478 | break; | |
4479 | } | |
4480 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
4481 | } | |
4482 | ||
fb04a1ed PX |
4483 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
4484 | { | |
4485 | int r; | |
4486 | ||
4487 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
4488 | return -EINVAL; | |
4489 | ||
4490 | /* the size should be power of 2 */ | |
4491 | if (!size || (size & (size - 1))) | |
4492 | return -EINVAL; | |
4493 | ||
4494 | /* Should be bigger to keep the reserved entries, or a page */ | |
4495 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
4496 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
4497 | return -EINVAL; | |
4498 | ||
4499 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
4500 | sizeof(struct kvm_dirty_gfn)) | |
4501 | return -E2BIG; | |
4502 | ||
4503 | /* We only allow it to set once */ | |
4504 | if (kvm->dirty_ring_size) | |
4505 | return -EINVAL; | |
4506 | ||
4507 | mutex_lock(&kvm->lock); | |
4508 | ||
4509 | if (kvm->created_vcpus) { | |
4510 | /* We don't allow to change this value after vcpu created */ | |
4511 | r = -EINVAL; | |
4512 | } else { | |
4513 | kvm->dirty_ring_size = size; | |
4514 | r = 0; | |
4515 | } | |
4516 | ||
4517 | mutex_unlock(&kvm->lock); | |
4518 | return r; | |
4519 | } | |
4520 | ||
4521 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
4522 | { | |
46808a4c | 4523 | unsigned long i; |
fb04a1ed PX |
4524 | struct kvm_vcpu *vcpu; |
4525 | int cleared = 0; | |
4526 | ||
4527 | if (!kvm->dirty_ring_size) | |
4528 | return -EINVAL; | |
4529 | ||
4530 | mutex_lock(&kvm->slots_lock); | |
4531 | ||
4532 | kvm_for_each_vcpu(i, vcpu, kvm) | |
4533 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
4534 | ||
4535 | mutex_unlock(&kvm->slots_lock); | |
4536 | ||
4537 | if (cleared) | |
4538 | kvm_flush_remote_tlbs(kvm); | |
4539 | ||
4540 | return cleared; | |
4541 | } | |
4542 | ||
e5d83c74 PB |
4543 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4544 | struct kvm_enable_cap *cap) | |
4545 | { | |
4546 | return -EINVAL; | |
4547 | } | |
4548 | ||
4549 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
4550 | struct kvm_enable_cap *cap) | |
4551 | { | |
4552 | switch (cap->cap) { | |
2a31b9db | 4553 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
4554 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
4555 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
4556 | ||
4557 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
4558 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
4559 | ||
4560 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
4561 | return -EINVAL; |
4562 | kvm->manual_dirty_log_protect = cap->args[0]; | |
4563 | return 0; | |
3c9bd400 | 4564 | } |
2a31b9db | 4565 | #endif |
acd05785 DM |
4566 | case KVM_CAP_HALT_POLL: { |
4567 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
4568 | return -EINVAL; | |
4569 | ||
4570 | kvm->max_halt_poll_ns = cap->args[0]; | |
4571 | return 0; | |
4572 | } | |
fb04a1ed PX |
4573 | case KVM_CAP_DIRTY_LOG_RING: |
4574 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); | |
e5d83c74 PB |
4575 | default: |
4576 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
4577 | } | |
4578 | } | |
4579 | ||
fcfe1bae JZ |
4580 | static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer, |
4581 | size_t size, loff_t *offset) | |
4582 | { | |
4583 | struct kvm *kvm = file->private_data; | |
4584 | ||
4585 | return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header, | |
4586 | &kvm_vm_stats_desc[0], &kvm->stat, | |
4587 | sizeof(kvm->stat), user_buffer, size, offset); | |
4588 | } | |
4589 | ||
4590 | static const struct file_operations kvm_vm_stats_fops = { | |
4591 | .read = kvm_vm_stats_read, | |
4592 | .llseek = noop_llseek, | |
4593 | }; | |
4594 | ||
4595 | static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm) | |
4596 | { | |
4597 | int fd; | |
4598 | struct file *file; | |
4599 | ||
4600 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4601 | if (fd < 0) | |
4602 | return fd; | |
4603 | ||
4604 | file = anon_inode_getfile("kvm-vm-stats", | |
4605 | &kvm_vm_stats_fops, kvm, O_RDONLY); | |
4606 | if (IS_ERR(file)) { | |
4607 | put_unused_fd(fd); | |
4608 | return PTR_ERR(file); | |
4609 | } | |
4610 | file->f_mode |= FMODE_PREAD; | |
4611 | fd_install(fd, file); | |
4612 | ||
4613 | return fd; | |
4614 | } | |
4615 | ||
bccf2150 AK |
4616 | static long kvm_vm_ioctl(struct file *filp, |
4617 | unsigned int ioctl, unsigned long arg) | |
4618 | { | |
4619 | struct kvm *kvm = filp->private_data; | |
4620 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 4621 | int r; |
bccf2150 | 4622 | |
f4d31653 | 4623 | if (kvm->mm != current->mm || kvm->vm_dead) |
6d4e4c4f | 4624 | return -EIO; |
bccf2150 AK |
4625 | switch (ioctl) { |
4626 | case KVM_CREATE_VCPU: | |
4627 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 4628 | break; |
e5d83c74 PB |
4629 | case KVM_ENABLE_CAP: { |
4630 | struct kvm_enable_cap cap; | |
4631 | ||
4632 | r = -EFAULT; | |
4633 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4634 | goto out; | |
4635 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
4636 | break; | |
4637 | } | |
6fc138d2 IE |
4638 | case KVM_SET_USER_MEMORY_REGION: { |
4639 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
4640 | ||
4641 | r = -EFAULT; | |
4642 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 4643 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
4644 | goto out; |
4645 | ||
47ae31e2 | 4646 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
4647 | break; |
4648 | } | |
4649 | case KVM_GET_DIRTY_LOG: { | |
4650 | struct kvm_dirty_log log; | |
4651 | ||
4652 | r = -EFAULT; | |
893bdbf1 | 4653 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 4654 | goto out; |
2c6f5df9 | 4655 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
4656 | break; |
4657 | } | |
2a31b9db PB |
4658 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4659 | case KVM_CLEAR_DIRTY_LOG: { | |
4660 | struct kvm_clear_dirty_log log; | |
4661 | ||
4662 | r = -EFAULT; | |
4663 | if (copy_from_user(&log, argp, sizeof(log))) | |
4664 | goto out; | |
4665 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4666 | break; | |
4667 | } | |
4668 | #endif | |
4b4357e0 | 4669 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
4670 | case KVM_REGISTER_COALESCED_MMIO: { |
4671 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4672 | |
5f94c174 | 4673 | r = -EFAULT; |
893bdbf1 | 4674 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4675 | goto out; |
5f94c174 | 4676 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4677 | break; |
4678 | } | |
4679 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
4680 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4681 | |
5f94c174 | 4682 | r = -EFAULT; |
893bdbf1 | 4683 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4684 | goto out; |
5f94c174 | 4685 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4686 | break; |
4687 | } | |
4688 | #endif | |
721eecbf GH |
4689 | case KVM_IRQFD: { |
4690 | struct kvm_irqfd data; | |
4691 | ||
4692 | r = -EFAULT; | |
893bdbf1 | 4693 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 4694 | goto out; |
d4db2935 | 4695 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
4696 | break; |
4697 | } | |
d34e6b17 GH |
4698 | case KVM_IOEVENTFD: { |
4699 | struct kvm_ioeventfd data; | |
4700 | ||
4701 | r = -EFAULT; | |
893bdbf1 | 4702 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
4703 | goto out; |
4704 | r = kvm_ioeventfd(kvm, &data); | |
4705 | break; | |
4706 | } | |
07975ad3 JK |
4707 | #ifdef CONFIG_HAVE_KVM_MSI |
4708 | case KVM_SIGNAL_MSI: { | |
4709 | struct kvm_msi msi; | |
4710 | ||
4711 | r = -EFAULT; | |
893bdbf1 | 4712 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
4713 | goto out; |
4714 | r = kvm_send_userspace_msi(kvm, &msi); | |
4715 | break; | |
4716 | } | |
23d43cf9 CD |
4717 | #endif |
4718 | #ifdef __KVM_HAVE_IRQ_LINE | |
4719 | case KVM_IRQ_LINE_STATUS: | |
4720 | case KVM_IRQ_LINE: { | |
4721 | struct kvm_irq_level irq_event; | |
4722 | ||
4723 | r = -EFAULT; | |
893bdbf1 | 4724 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
4725 | goto out; |
4726 | ||
aa2fbe6d YZ |
4727 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
4728 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
4729 | if (r) |
4730 | goto out; | |
4731 | ||
4732 | r = -EFAULT; | |
4733 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 4734 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
4735 | goto out; |
4736 | } | |
4737 | ||
4738 | r = 0; | |
4739 | break; | |
4740 | } | |
73880c80 | 4741 | #endif |
aa8d5944 AG |
4742 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4743 | case KVM_SET_GSI_ROUTING: { | |
4744 | struct kvm_irq_routing routing; | |
4745 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 4746 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
4747 | |
4748 | r = -EFAULT; | |
4749 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
4750 | goto out; | |
4751 | r = -EINVAL; | |
5c0aea0e DH |
4752 | if (!kvm_arch_can_set_irq_routing(kvm)) |
4753 | goto out; | |
caf1ff26 | 4754 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
4755 | goto out; |
4756 | if (routing.flags) | |
4757 | goto out; | |
f8c1b85b | 4758 | if (routing.nr) { |
f8c1b85b | 4759 | urouting = argp; |
7ec28e26 DE |
4760 | entries = vmemdup_user(urouting->entries, |
4761 | array_size(sizeof(*entries), | |
4762 | routing.nr)); | |
4763 | if (IS_ERR(entries)) { | |
4764 | r = PTR_ERR(entries); | |
4765 | goto out; | |
4766 | } | |
f8c1b85b | 4767 | } |
aa8d5944 AG |
4768 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
4769 | routing.flags); | |
7ec28e26 | 4770 | kvfree(entries); |
aa8d5944 AG |
4771 | break; |
4772 | } | |
4773 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
4774 | case KVM_CREATE_DEVICE: { |
4775 | struct kvm_create_device cd; | |
4776 | ||
4777 | r = -EFAULT; | |
4778 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
4779 | goto out; | |
4780 | ||
4781 | r = kvm_ioctl_create_device(kvm, &cd); | |
4782 | if (r) | |
4783 | goto out; | |
4784 | ||
4785 | r = -EFAULT; | |
4786 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
4787 | goto out; | |
4788 | ||
4789 | r = 0; | |
4790 | break; | |
4791 | } | |
92b591a4 AG |
4792 | case KVM_CHECK_EXTENSION: |
4793 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
4794 | break; | |
fb04a1ed PX |
4795 | case KVM_RESET_DIRTY_RINGS: |
4796 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
4797 | break; | |
fcfe1bae JZ |
4798 | case KVM_GET_STATS_FD: |
4799 | r = kvm_vm_ioctl_get_stats_fd(kvm); | |
4800 | break; | |
f17abe9a | 4801 | default: |
1fe779f8 | 4802 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
4803 | } |
4804 | out: | |
4805 | return r; | |
4806 | } | |
4807 | ||
de8e5d74 | 4808 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
4809 | struct compat_kvm_dirty_log { |
4810 | __u32 slot; | |
4811 | __u32 padding1; | |
4812 | union { | |
4813 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4814 | __u64 padding2; | |
4815 | }; | |
4816 | }; | |
4817 | ||
8750f9bb PB |
4818 | struct compat_kvm_clear_dirty_log { |
4819 | __u32 slot; | |
4820 | __u32 num_pages; | |
4821 | __u64 first_page; | |
4822 | union { | |
4823 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4824 | __u64 padding2; | |
4825 | }; | |
4826 | }; | |
4827 | ||
6ff5894c AB |
4828 | static long kvm_vm_compat_ioctl(struct file *filp, |
4829 | unsigned int ioctl, unsigned long arg) | |
4830 | { | |
4831 | struct kvm *kvm = filp->private_data; | |
4832 | int r; | |
4833 | ||
f4d31653 | 4834 | if (kvm->mm != current->mm || kvm->vm_dead) |
6ff5894c AB |
4835 | return -EIO; |
4836 | switch (ioctl) { | |
8750f9bb PB |
4837 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4838 | case KVM_CLEAR_DIRTY_LOG: { | |
4839 | struct compat_kvm_clear_dirty_log compat_log; | |
4840 | struct kvm_clear_dirty_log log; | |
4841 | ||
4842 | if (copy_from_user(&compat_log, (void __user *)arg, | |
4843 | sizeof(compat_log))) | |
4844 | return -EFAULT; | |
4845 | log.slot = compat_log.slot; | |
4846 | log.num_pages = compat_log.num_pages; | |
4847 | log.first_page = compat_log.first_page; | |
4848 | log.padding2 = compat_log.padding2; | |
4849 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4850 | ||
4851 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4852 | break; | |
4853 | } | |
4854 | #endif | |
6ff5894c AB |
4855 | case KVM_GET_DIRTY_LOG: { |
4856 | struct compat_kvm_dirty_log compat_log; | |
4857 | struct kvm_dirty_log log; | |
4858 | ||
6ff5894c AB |
4859 | if (copy_from_user(&compat_log, (void __user *)arg, |
4860 | sizeof(compat_log))) | |
f6a3b168 | 4861 | return -EFAULT; |
6ff5894c AB |
4862 | log.slot = compat_log.slot; |
4863 | log.padding1 = compat_log.padding1; | |
4864 | log.padding2 = compat_log.padding2; | |
4865 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4866 | ||
4867 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
4868 | break; |
4869 | } | |
4870 | default: | |
4871 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
4872 | } | |
6ff5894c AB |
4873 | return r; |
4874 | } | |
4875 | #endif | |
4876 | ||
70375c2d | 4877 | static const struct file_operations kvm_vm_fops = { |
f17abe9a AK |
4878 | .release = kvm_vm_release, |
4879 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 4880 | .llseek = noop_llseek, |
7ddfd3e0 | 4881 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
4882 | }; |
4883 | ||
54526d1f NT |
4884 | bool file_is_kvm(struct file *file) |
4885 | { | |
4886 | return file && file->f_op == &kvm_vm_fops; | |
4887 | } | |
4888 | EXPORT_SYMBOL_GPL(file_is_kvm); | |
4889 | ||
e08b9637 | 4890 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 4891 | { |
59f82aad | 4892 | char fdname[ITOA_MAX_LEN + 1]; |
20020f4c | 4893 | int r, fd; |
f17abe9a | 4894 | struct kvm *kvm; |
506cfba9 | 4895 | struct file *file; |
f17abe9a | 4896 | |
20020f4c OU |
4897 | fd = get_unused_fd_flags(O_CLOEXEC); |
4898 | if (fd < 0) | |
4899 | return fd; | |
4900 | ||
59f82aad OU |
4901 | snprintf(fdname, sizeof(fdname), "%d", fd); |
4902 | ||
e08b9637 | 4903 | kvm = kvm_create_vm(type); |
20020f4c OU |
4904 | if (IS_ERR(kvm)) { |
4905 | r = PTR_ERR(kvm); | |
4906 | goto put_fd; | |
4907 | } | |
4908 | ||
4b4357e0 | 4909 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 4910 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
4911 | if (r < 0) |
4912 | goto put_kvm; | |
6ce5a090 | 4913 | #endif |
506cfba9 AV |
4914 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
4915 | if (IS_ERR(file)) { | |
78588335 ME |
4916 | r = PTR_ERR(file); |
4917 | goto put_kvm; | |
506cfba9 | 4918 | } |
536a6f88 | 4919 | |
525df861 PB |
4920 | /* |
4921 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
4922 | * already set, with ->release() being kvm_vm_release(). In error | |
4923 | * cases it will be called by the final fput(file) and will take | |
4924 | * care of doing kvm_put_kvm(kvm). | |
4925 | */ | |
59f82aad | 4926 | if (kvm_create_vm_debugfs(kvm, fdname) < 0) { |
506cfba9 | 4927 | fput(file); |
20020f4c OU |
4928 | r = -ENOMEM; |
4929 | goto put_fd; | |
536a6f88 | 4930 | } |
286de8f6 | 4931 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 4932 | |
20020f4c OU |
4933 | fd_install(fd, file); |
4934 | return fd; | |
78588335 ME |
4935 | |
4936 | put_kvm: | |
4937 | kvm_put_kvm(kvm); | |
20020f4c OU |
4938 | put_fd: |
4939 | put_unused_fd(fd); | |
78588335 | 4940 | return r; |
f17abe9a AK |
4941 | } |
4942 | ||
4943 | static long kvm_dev_ioctl(struct file *filp, | |
4944 | unsigned int ioctl, unsigned long arg) | |
4945 | { | |
07c45a36 | 4946 | long r = -EINVAL; |
f17abe9a AK |
4947 | |
4948 | switch (ioctl) { | |
4949 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
4950 | if (arg) |
4951 | goto out; | |
f17abe9a AK |
4952 | r = KVM_API_VERSION; |
4953 | break; | |
4954 | case KVM_CREATE_VM: | |
e08b9637 | 4955 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 4956 | break; |
018d00d2 | 4957 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 4958 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 4959 | break; |
07c45a36 | 4960 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
4961 | if (arg) |
4962 | goto out; | |
adb1ff46 AK |
4963 | r = PAGE_SIZE; /* struct kvm_run */ |
4964 | #ifdef CONFIG_X86 | |
4965 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 4966 | #endif |
4b4357e0 | 4967 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 4968 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 4969 | #endif |
07c45a36 | 4970 | break; |
d4c9ff2d FEL |
4971 | case KVM_TRACE_ENABLE: |
4972 | case KVM_TRACE_PAUSE: | |
4973 | case KVM_TRACE_DISABLE: | |
2023a29c | 4974 | r = -EOPNOTSUPP; |
d4c9ff2d | 4975 | break; |
6aa8b732 | 4976 | default: |
043405e1 | 4977 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
4978 | } |
4979 | out: | |
4980 | return r; | |
4981 | } | |
4982 | ||
6aa8b732 | 4983 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 4984 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 4985 | .llseek = noop_llseek, |
7ddfd3e0 | 4986 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
4987 | }; |
4988 | ||
4989 | static struct miscdevice kvm_dev = { | |
bbe4432e | 4990 | KVM_MINOR, |
6aa8b732 AK |
4991 | "kvm", |
4992 | &kvm_chardev_ops, | |
4993 | }; | |
4994 | ||
75b7127c | 4995 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
4996 | { |
4997 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 4998 | int r; |
1b6c0168 | 4999 | |
7f59f492 | 5000 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 5001 | return; |
10474ae8 | 5002 | |
7f59f492 | 5003 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 5004 | |
13a34e06 | 5005 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
5006 | |
5007 | if (r) { | |
5008 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
5009 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 5010 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 5011 | } |
1b6c0168 AK |
5012 | } |
5013 | ||
8c18b2d2 | 5014 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 5015 | { |
4a937f96 | 5016 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
5017 | if (kvm_usage_count) |
5018 | hardware_enable_nolock(NULL); | |
4a937f96 | 5019 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 5020 | return 0; |
75b7127c TY |
5021 | } |
5022 | ||
5023 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
5024 | { |
5025 | int cpu = raw_smp_processor_id(); | |
5026 | ||
7f59f492 | 5027 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 5028 | return; |
7f59f492 | 5029 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 5030 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
5031 | } |
5032 | ||
8c18b2d2 | 5033 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 5034 | { |
4a937f96 | 5035 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
5036 | if (kvm_usage_count) |
5037 | hardware_disable_nolock(NULL); | |
4a937f96 | 5038 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 5039 | return 0; |
75b7127c TY |
5040 | } |
5041 | ||
10474ae8 AG |
5042 | static void hardware_disable_all_nolock(void) |
5043 | { | |
5044 | BUG_ON(!kvm_usage_count); | |
5045 | ||
5046 | kvm_usage_count--; | |
5047 | if (!kvm_usage_count) | |
75b7127c | 5048 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
5049 | } |
5050 | ||
5051 | static void hardware_disable_all(void) | |
5052 | { | |
4a937f96 | 5053 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 5054 | hardware_disable_all_nolock(); |
4a937f96 | 5055 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
5056 | } |
5057 | ||
5058 | static int hardware_enable_all(void) | |
5059 | { | |
5060 | int r = 0; | |
5061 | ||
4a937f96 | 5062 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
5063 | |
5064 | kvm_usage_count++; | |
5065 | if (kvm_usage_count == 1) { | |
5066 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 5067 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
5068 | |
5069 | if (atomic_read(&hardware_enable_failed)) { | |
5070 | hardware_disable_all_nolock(); | |
5071 | r = -EBUSY; | |
5072 | } | |
5073 | } | |
5074 | ||
4a937f96 | 5075 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
5076 | |
5077 | return r; | |
5078 | } | |
5079 | ||
9a2b85c6 | 5080 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 5081 | void *v) |
9a2b85c6 | 5082 | { |
8e1c1815 SY |
5083 | /* |
5084 | * Some (well, at least mine) BIOSes hang on reboot if | |
5085 | * in vmx root mode. | |
5086 | * | |
5087 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
5088 | */ | |
1170adc6 | 5089 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 5090 | kvm_rebooting = true; |
75b7127c | 5091 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
5092 | return NOTIFY_OK; |
5093 | } | |
5094 | ||
5095 | static struct notifier_block kvm_reboot_notifier = { | |
5096 | .notifier_call = kvm_reboot, | |
5097 | .priority = 0, | |
5098 | }; | |
5099 | ||
e93f8a0f | 5100 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
5101 | { |
5102 | int i; | |
5103 | ||
5104 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 5105 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
5106 | |
5107 | kvm_iodevice_destructor(pos); | |
5108 | } | |
e93f8a0f | 5109 | kfree(bus); |
2eeb2e94 GH |
5110 | } |
5111 | ||
c21fbff1 | 5112 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 5113 | const struct kvm_io_range *r2) |
743eeb0b | 5114 | { |
8f4216c7 JW |
5115 | gpa_t addr1 = r1->addr; |
5116 | gpa_t addr2 = r2->addr; | |
5117 | ||
5118 | if (addr1 < addr2) | |
743eeb0b | 5119 | return -1; |
8f4216c7 JW |
5120 | |
5121 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
5122 | * accept any overlapping write. Any order is acceptable for | |
5123 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
5124 | * we process all of them. | |
5125 | */ | |
5126 | if (r2->len) { | |
5127 | addr1 += r1->len; | |
5128 | addr2 += r2->len; | |
5129 | } | |
5130 | ||
5131 | if (addr1 > addr2) | |
743eeb0b | 5132 | return 1; |
8f4216c7 | 5133 | |
743eeb0b SL |
5134 | return 0; |
5135 | } | |
5136 | ||
a343c9b7 PB |
5137 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
5138 | { | |
c21fbff1 | 5139 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
5140 | } |
5141 | ||
39369f7a | 5142 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
5143 | gpa_t addr, int len) |
5144 | { | |
5145 | struct kvm_io_range *range, key; | |
5146 | int off; | |
5147 | ||
5148 | key = (struct kvm_io_range) { | |
5149 | .addr = addr, | |
5150 | .len = len, | |
5151 | }; | |
5152 | ||
5153 | range = bsearch(&key, bus->range, bus->dev_count, | |
5154 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
5155 | if (range == NULL) | |
5156 | return -ENOENT; | |
5157 | ||
5158 | off = range - bus->range; | |
5159 | ||
c21fbff1 | 5160 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
5161 | off--; |
5162 | ||
5163 | return off; | |
5164 | } | |
5165 | ||
e32edf4f | 5166 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
5167 | struct kvm_io_range *range, const void *val) |
5168 | { | |
5169 | int idx; | |
5170 | ||
5171 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
5172 | if (idx < 0) | |
5173 | return -EOPNOTSUPP; | |
5174 | ||
5175 | while (idx < bus->dev_count && | |
c21fbff1 | 5176 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5177 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5178 | range->len, val)) |
5179 | return idx; | |
5180 | idx++; | |
5181 | } | |
5182 | ||
5183 | return -EOPNOTSUPP; | |
5184 | } | |
5185 | ||
bda9020e | 5186 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 5187 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 5188 | int len, const void *val) |
2eeb2e94 | 5189 | { |
90d83dc3 | 5190 | struct kvm_io_bus *bus; |
743eeb0b | 5191 | struct kvm_io_range range; |
126a5af5 | 5192 | int r; |
743eeb0b SL |
5193 | |
5194 | range = (struct kvm_io_range) { | |
5195 | .addr = addr, | |
5196 | .len = len, | |
5197 | }; | |
90d83dc3 | 5198 | |
e32edf4f | 5199 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5200 | if (!bus) |
5201 | return -ENOMEM; | |
e32edf4f | 5202 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5203 | return r < 0 ? r : 0; |
5204 | } | |
a2420107 | 5205 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
5206 | |
5207 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
5208 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
5209 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
5210 | { |
5211 | struct kvm_io_bus *bus; | |
5212 | struct kvm_io_range range; | |
5213 | ||
5214 | range = (struct kvm_io_range) { | |
5215 | .addr = addr, | |
5216 | .len = len, | |
5217 | }; | |
5218 | ||
e32edf4f | 5219 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5220 | if (!bus) |
5221 | return -ENOMEM; | |
126a5af5 CH |
5222 | |
5223 | /* First try the device referenced by cookie. */ | |
5224 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 5225 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 5226 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
5227 | val)) |
5228 | return cookie; | |
5229 | ||
5230 | /* | |
5231 | * cookie contained garbage; fall back to search and return the | |
5232 | * correct cookie value. | |
5233 | */ | |
e32edf4f | 5234 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5235 | } |
5236 | ||
e32edf4f NN |
5237 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
5238 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
5239 | { |
5240 | int idx; | |
5241 | ||
5242 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
5243 | if (idx < 0) |
5244 | return -EOPNOTSUPP; | |
5245 | ||
5246 | while (idx < bus->dev_count && | |
c21fbff1 | 5247 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5248 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5249 | range->len, val)) |
5250 | return idx; | |
743eeb0b SL |
5251 | idx++; |
5252 | } | |
5253 | ||
bda9020e MT |
5254 | return -EOPNOTSUPP; |
5255 | } | |
2eeb2e94 | 5256 | |
bda9020e | 5257 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 5258 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 5259 | int len, void *val) |
bda9020e | 5260 | { |
90d83dc3 | 5261 | struct kvm_io_bus *bus; |
743eeb0b | 5262 | struct kvm_io_range range; |
126a5af5 | 5263 | int r; |
743eeb0b SL |
5264 | |
5265 | range = (struct kvm_io_range) { | |
5266 | .addr = addr, | |
5267 | .len = len, | |
5268 | }; | |
e93f8a0f | 5269 | |
e32edf4f | 5270 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5271 | if (!bus) |
5272 | return -ENOMEM; | |
e32edf4f | 5273 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
5274 | return r < 0 ? r : 0; |
5275 | } | |
743eeb0b | 5276 | |
79fac95e | 5277 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
5278 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
5279 | int len, struct kvm_io_device *dev) | |
6c474694 | 5280 | { |
d4c67a7a | 5281 | int i; |
e93f8a0f | 5282 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 5283 | struct kvm_io_range range; |
090b7aff | 5284 | |
4a12f951 | 5285 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
5286 | if (!bus) |
5287 | return -ENOMEM; | |
5288 | ||
6ea34c9b AK |
5289 | /* exclude ioeventfd which is limited by maximum fd */ |
5290 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 5291 | return -ENOSPC; |
2eeb2e94 | 5292 | |
90952cd3 | 5293 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 5294 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
5295 | if (!new_bus) |
5296 | return -ENOMEM; | |
d4c67a7a GH |
5297 | |
5298 | range = (struct kvm_io_range) { | |
5299 | .addr = addr, | |
5300 | .len = len, | |
5301 | .dev = dev, | |
5302 | }; | |
5303 | ||
5304 | for (i = 0; i < bus->dev_count; i++) | |
5305 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
5306 | break; | |
5307 | ||
5308 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
5309 | new_bus->dev_count++; | |
5310 | new_bus->range[i] = range; | |
5311 | memcpy(new_bus->range + i + 1, bus->range + i, | |
5312 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
5313 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
5314 | synchronize_srcu_expedited(&kvm->srcu); | |
5315 | kfree(bus); | |
090b7aff GH |
5316 | |
5317 | return 0; | |
5318 | } | |
5319 | ||
5d3c4c79 SC |
5320 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5321 | struct kvm_io_device *dev) | |
090b7aff | 5322 | { |
f6588660 | 5323 | int i, j; |
e93f8a0f | 5324 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 5325 | |
7c896d37 SC |
5326 | lockdep_assert_held(&kvm->slots_lock); |
5327 | ||
4a12f951 | 5328 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 5329 | if (!bus) |
5d3c4c79 | 5330 | return 0; |
df630b8c | 5331 | |
7c896d37 | 5332 | for (i = 0; i < bus->dev_count; i++) { |
a1300716 | 5333 | if (bus->range[i].dev == dev) { |
090b7aff GH |
5334 | break; |
5335 | } | |
7c896d37 | 5336 | } |
e93f8a0f | 5337 | |
90db1043 | 5338 | if (i == bus->dev_count) |
5d3c4c79 | 5339 | return 0; |
a1300716 | 5340 | |
90952cd3 | 5341 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 5342 | GFP_KERNEL_ACCOUNT); |
f6588660 | 5343 | if (new_bus) { |
871c433b | 5344 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
5345 | new_bus->dev_count--; |
5346 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 5347 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
2ee37574 SC |
5348 | } |
5349 | ||
5350 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
5351 | synchronize_srcu_expedited(&kvm->srcu); | |
5352 | ||
5353 | /* Destroy the old bus _after_ installing the (null) bus. */ | |
5354 | if (!new_bus) { | |
90db1043 | 5355 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
f6588660 RK |
5356 | for (j = 0; j < bus->dev_count; j++) { |
5357 | if (j == i) | |
5358 | continue; | |
5359 | kvm_iodevice_destructor(bus->range[j].dev); | |
5360 | } | |
90db1043 | 5361 | } |
a1300716 | 5362 | |
e93f8a0f | 5363 | kfree(bus); |
5d3c4c79 | 5364 | return new_bus ? 0 : -ENOMEM; |
2eeb2e94 GH |
5365 | } |
5366 | ||
8a39d006 AP |
5367 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5368 | gpa_t addr) | |
5369 | { | |
5370 | struct kvm_io_bus *bus; | |
5371 | int dev_idx, srcu_idx; | |
5372 | struct kvm_io_device *iodev = NULL; | |
5373 | ||
5374 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
5375 | ||
5376 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
5377 | if (!bus) |
5378 | goto out_unlock; | |
8a39d006 AP |
5379 | |
5380 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
5381 | if (dev_idx < 0) | |
5382 | goto out_unlock; | |
5383 | ||
5384 | iodev = bus->range[dev_idx].dev; | |
5385 | ||
5386 | out_unlock: | |
5387 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
5388 | ||
5389 | return iodev; | |
5390 | } | |
5391 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
5392 | ||
536a6f88 JF |
5393 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
5394 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
5395 | const char *fmt) | |
5396 | { | |
5397 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5398 | inode->i_private; | |
5399 | ||
605c7130 PX |
5400 | /* |
5401 | * The debugfs files are a reference to the kvm struct which | |
5402 | * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe | |
5403 | * avoids the race between open and the removal of the debugfs directory. | |
536a6f88 | 5404 | */ |
605c7130 | 5405 | if (!kvm_get_kvm_safe(stat_data->kvm)) |
536a6f88 JF |
5406 | return -ENOENT; |
5407 | ||
833b45de | 5408 | if (simple_attr_open(inode, file, get, |
bc9e9e67 | 5409 | kvm_stats_debugfs_mode(stat_data->desc) & 0222 |
09cbcef6 MP |
5410 | ? set : NULL, |
5411 | fmt)) { | |
536a6f88 JF |
5412 | kvm_put_kvm(stat_data->kvm); |
5413 | return -ENOMEM; | |
5414 | } | |
5415 | ||
5416 | return 0; | |
5417 | } | |
5418 | ||
5419 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
5420 | { | |
5421 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5422 | inode->i_private; | |
5423 | ||
5424 | simple_attr_release(inode, file); | |
5425 | kvm_put_kvm(stat_data->kvm); | |
5426 | ||
5427 | return 0; | |
5428 | } | |
5429 | ||
09cbcef6 | 5430 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 5431 | { |
bc9e9e67 | 5432 | *val = *(u64 *)((void *)(&kvm->stat) + offset); |
536a6f88 | 5433 | |
09cbcef6 MP |
5434 | return 0; |
5435 | } | |
5436 | ||
5437 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
5438 | { | |
bc9e9e67 | 5439 | *(u64 *)((void *)(&kvm->stat) + offset) = 0; |
536a6f88 JF |
5440 | |
5441 | return 0; | |
5442 | } | |
5443 | ||
09cbcef6 | 5444 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 5445 | { |
46808a4c | 5446 | unsigned long i; |
09cbcef6 | 5447 | struct kvm_vcpu *vcpu; |
ce35ef27 | 5448 | |
09cbcef6 | 5449 | *val = 0; |
ce35ef27 | 5450 | |
09cbcef6 | 5451 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5452 | *val += *(u64 *)((void *)(&vcpu->stat) + offset); |
ce35ef27 SJS |
5453 | |
5454 | return 0; | |
5455 | } | |
5456 | ||
09cbcef6 | 5457 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 5458 | { |
46808a4c | 5459 | unsigned long i; |
09cbcef6 | 5460 | struct kvm_vcpu *vcpu; |
536a6f88 | 5461 | |
09cbcef6 | 5462 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5463 | *(u64 *)((void *)(&vcpu->stat) + offset) = 0; |
09cbcef6 MP |
5464 | |
5465 | return 0; | |
5466 | } | |
536a6f88 | 5467 | |
09cbcef6 | 5468 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 5469 | { |
09cbcef6 | 5470 | int r = -EFAULT; |
536a6f88 | 5471 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
536a6f88 | 5472 | |
bc9e9e67 | 5473 | switch (stat_data->kind) { |
09cbcef6 MP |
5474 | case KVM_STAT_VM: |
5475 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5476 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5477 | break; |
5478 | case KVM_STAT_VCPU: | |
5479 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5480 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5481 | break; |
5482 | } | |
536a6f88 | 5483 | |
09cbcef6 | 5484 | return r; |
536a6f88 JF |
5485 | } |
5486 | ||
09cbcef6 | 5487 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 5488 | { |
09cbcef6 | 5489 | int r = -EFAULT; |
ce35ef27 | 5490 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
ce35ef27 SJS |
5491 | |
5492 | if (val) | |
5493 | return -EINVAL; | |
5494 | ||
bc9e9e67 | 5495 | switch (stat_data->kind) { |
09cbcef6 MP |
5496 | case KVM_STAT_VM: |
5497 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5498 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5499 | break; |
5500 | case KVM_STAT_VCPU: | |
5501 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5502 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5503 | break; |
5504 | } | |
ce35ef27 | 5505 | |
09cbcef6 | 5506 | return r; |
ce35ef27 SJS |
5507 | } |
5508 | ||
09cbcef6 | 5509 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
5510 | { |
5511 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
5512 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
5513 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
5514 | } |
5515 | ||
09cbcef6 MP |
5516 | static const struct file_operations stat_fops_per_vm = { |
5517 | .owner = THIS_MODULE, | |
5518 | .open = kvm_stat_data_open, | |
536a6f88 | 5519 | .release = kvm_debugfs_release, |
09cbcef6 MP |
5520 | .read = simple_attr_read, |
5521 | .write = simple_attr_write, | |
5522 | .llseek = no_llseek, | |
536a6f88 JF |
5523 | }; |
5524 | ||
8b88b099 | 5525 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
5526 | { |
5527 | unsigned offset = (long)_offset; | |
ba1389b7 | 5528 | struct kvm *kvm; |
536a6f88 | 5529 | u64 tmp_val; |
ba1389b7 | 5530 | |
8b88b099 | 5531 | *val = 0; |
0d9ce162 | 5532 | mutex_lock(&kvm_lock); |
536a6f88 | 5533 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5534 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
5535 | *val += tmp_val; |
5536 | } | |
0d9ce162 | 5537 | mutex_unlock(&kvm_lock); |
8b88b099 | 5538 | return 0; |
ba1389b7 AK |
5539 | } |
5540 | ||
ce35ef27 SJS |
5541 | static int vm_stat_clear(void *_offset, u64 val) |
5542 | { | |
5543 | unsigned offset = (long)_offset; | |
5544 | struct kvm *kvm; | |
ce35ef27 SJS |
5545 | |
5546 | if (val) | |
5547 | return -EINVAL; | |
5548 | ||
0d9ce162 | 5549 | mutex_lock(&kvm_lock); |
ce35ef27 | 5550 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5551 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 5552 | } |
0d9ce162 | 5553 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5554 | |
5555 | return 0; | |
5556 | } | |
5557 | ||
5558 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
bc9e9e67 | 5559 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n"); |
ba1389b7 | 5560 | |
8b88b099 | 5561 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
5562 | { |
5563 | unsigned offset = (long)_offset; | |
1165f5fe | 5564 | struct kvm *kvm; |
536a6f88 | 5565 | u64 tmp_val; |
1165f5fe | 5566 | |
8b88b099 | 5567 | *val = 0; |
0d9ce162 | 5568 | mutex_lock(&kvm_lock); |
536a6f88 | 5569 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5570 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
5571 | *val += tmp_val; |
5572 | } | |
0d9ce162 | 5573 | mutex_unlock(&kvm_lock); |
8b88b099 | 5574 | return 0; |
1165f5fe AK |
5575 | } |
5576 | ||
ce35ef27 SJS |
5577 | static int vcpu_stat_clear(void *_offset, u64 val) |
5578 | { | |
5579 | unsigned offset = (long)_offset; | |
5580 | struct kvm *kvm; | |
ce35ef27 SJS |
5581 | |
5582 | if (val) | |
5583 | return -EINVAL; | |
5584 | ||
0d9ce162 | 5585 | mutex_lock(&kvm_lock); |
ce35ef27 | 5586 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5587 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 5588 | } |
0d9ce162 | 5589 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5590 | |
5591 | return 0; | |
5592 | } | |
5593 | ||
5594 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
5595 | "%llu\n"); | |
bc9e9e67 | 5596 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n"); |
1165f5fe | 5597 | |
286de8f6 CI |
5598 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
5599 | { | |
5600 | struct kobj_uevent_env *env; | |
286de8f6 CI |
5601 | unsigned long long created, active; |
5602 | ||
5603 | if (!kvm_dev.this_device || !kvm) | |
5604 | return; | |
5605 | ||
0d9ce162 | 5606 | mutex_lock(&kvm_lock); |
286de8f6 CI |
5607 | if (type == KVM_EVENT_CREATE_VM) { |
5608 | kvm_createvm_count++; | |
5609 | kvm_active_vms++; | |
5610 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
5611 | kvm_active_vms--; | |
5612 | } | |
5613 | created = kvm_createvm_count; | |
5614 | active = kvm_active_vms; | |
0d9ce162 | 5615 | mutex_unlock(&kvm_lock); |
286de8f6 | 5616 | |
b12ce36a | 5617 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
5618 | if (!env) |
5619 | return; | |
5620 | ||
5621 | add_uevent_var(env, "CREATED=%llu", created); | |
5622 | add_uevent_var(env, "COUNT=%llu", active); | |
5623 | ||
fdeaf7e3 | 5624 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 5625 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
5626 | kvm->userspace_pid = task_pid_nr(current); |
5627 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 5628 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
5629 | } |
5630 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 5631 | |
a44a4cc1 | 5632 | if (!IS_ERR(kvm->debugfs_dentry)) { |
b12ce36a | 5633 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
5634 | |
5635 | if (p) { | |
5636 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
5637 | if (!IS_ERR(tmp)) | |
5638 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
5639 | kfree(p); | |
286de8f6 CI |
5640 | } |
5641 | } | |
5642 | /* no need for checks, since we are adding at most only 5 keys */ | |
5643 | env->envp[env->envp_idx++] = NULL; | |
5644 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
5645 | kfree(env); | |
286de8f6 CI |
5646 | } |
5647 | ||
929f45e3 | 5648 | static void kvm_init_debug(void) |
6aa8b732 | 5649 | { |
bc9e9e67 JZ |
5650 | const struct file_operations *fops; |
5651 | const struct _kvm_stats_desc *pdesc; | |
5652 | int i; | |
6aa8b732 | 5653 | |
76f7c879 | 5654 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 5655 | |
bc9e9e67 JZ |
5656 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
5657 | pdesc = &kvm_vm_stats_desc[i]; | |
5658 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5659 | fops = &vm_stat_fops; | |
5660 | else | |
5661 | fops = &vm_stat_readonly_fops; | |
5662 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5663 | kvm_debugfs_dir, | |
5664 | (void *)(long)pdesc->desc.offset, fops); | |
5665 | } | |
5666 | ||
5667 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
5668 | pdesc = &kvm_vcpu_stats_desc[i]; | |
5669 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5670 | fops = &vcpu_stat_fops; | |
5671 | else | |
5672 | fops = &vcpu_stat_readonly_fops; | |
5673 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5674 | kvm_debugfs_dir, | |
5675 | (void *)(long)pdesc->desc.offset, fops); | |
4f69b680 | 5676 | } |
6aa8b732 AK |
5677 | } |
5678 | ||
fb3600cc | 5679 | static int kvm_suspend(void) |
59ae6c6b | 5680 | { |
10474ae8 | 5681 | if (kvm_usage_count) |
75b7127c | 5682 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
5683 | return 0; |
5684 | } | |
5685 | ||
fb3600cc | 5686 | static void kvm_resume(void) |
59ae6c6b | 5687 | { |
ca84d1a2 | 5688 | if (kvm_usage_count) { |
4cb9a998 | 5689 | lockdep_assert_not_held(&kvm_count_lock); |
75b7127c | 5690 | hardware_enable_nolock(NULL); |
ca84d1a2 | 5691 | } |
59ae6c6b AK |
5692 | } |
5693 | ||
fb3600cc | 5694 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
5695 | .suspend = kvm_suspend, |
5696 | .resume = kvm_resume, | |
5697 | }; | |
5698 | ||
15ad7146 AK |
5699 | static inline |
5700 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
5701 | { | |
5702 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
5703 | } | |
5704 | ||
5705 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
5706 | { | |
5707 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 5708 | |
046ddeed | 5709 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 5710 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 5711 | |
7495e22b | 5712 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 5713 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 5714 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
5715 | } |
5716 | ||
5717 | static void kvm_sched_out(struct preempt_notifier *pn, | |
5718 | struct task_struct *next) | |
5719 | { | |
5720 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
5721 | ||
3ba9f93b | 5722 | if (current->on_rq) { |
046ddeed | 5723 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
5724 | WRITE_ONCE(vcpu->ready, true); |
5725 | } | |
e9b11c17 | 5726 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
5727 | __this_cpu_write(kvm_running_vcpu, NULL); |
5728 | } | |
5729 | ||
5730 | /** | |
5731 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
5732 | * |
5733 | * We can disable preemption locally around accessing the per-CPU variable, | |
5734 | * and use the resolved vcpu pointer after enabling preemption again, | |
5735 | * because even if the current thread is migrated to another CPU, reading | |
5736 | * the per-CPU value later will give us the same value as we update the | |
5737 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
5738 | */ |
5739 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
5740 | { | |
1f03b2bc MZ |
5741 | struct kvm_vcpu *vcpu; |
5742 | ||
5743 | preempt_disable(); | |
5744 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
5745 | preempt_enable(); | |
5746 | ||
5747 | return vcpu; | |
7495e22b | 5748 | } |
379a3c8e | 5749 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
5750 | |
5751 | /** | |
5752 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
5753 | */ | |
5754 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
5755 | { | |
5756 | return &kvm_running_vcpu; | |
15ad7146 AK |
5757 | } |
5758 | ||
e1bfc245 SC |
5759 | #ifdef CONFIG_GUEST_PERF_EVENTS |
5760 | static unsigned int kvm_guest_state(void) | |
5761 | { | |
5762 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
5763 | unsigned int state; | |
5764 | ||
5765 | if (!kvm_arch_pmi_in_guest(vcpu)) | |
5766 | return 0; | |
5767 | ||
5768 | state = PERF_GUEST_ACTIVE; | |
5769 | if (!kvm_arch_vcpu_in_kernel(vcpu)) | |
5770 | state |= PERF_GUEST_USER; | |
5771 | ||
5772 | return state; | |
5773 | } | |
5774 | ||
5775 | static unsigned long kvm_guest_get_ip(void) | |
5776 | { | |
5777 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
5778 | ||
5779 | /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */ | |
5780 | if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu))) | |
5781 | return 0; | |
5782 | ||
5783 | return kvm_arch_vcpu_get_ip(vcpu); | |
5784 | } | |
5785 | ||
5786 | static struct perf_guest_info_callbacks kvm_guest_cbs = { | |
5787 | .state = kvm_guest_state, | |
5788 | .get_ip = kvm_guest_get_ip, | |
5789 | .handle_intel_pt_intr = NULL, | |
5790 | }; | |
5791 | ||
5792 | void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void)) | |
5793 | { | |
5794 | kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler; | |
5795 | perf_register_guest_info_callbacks(&kvm_guest_cbs); | |
5796 | } | |
5797 | void kvm_unregister_perf_callbacks(void) | |
5798 | { | |
5799 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); | |
5800 | } | |
5801 | #endif | |
5802 | ||
b9904085 SC |
5803 | struct kvm_cpu_compat_check { |
5804 | void *opaque; | |
5805 | int *ret; | |
5806 | }; | |
5807 | ||
5808 | static void check_processor_compat(void *data) | |
f257d6dc | 5809 | { |
b9904085 SC |
5810 | struct kvm_cpu_compat_check *c = data; |
5811 | ||
5812 | *c->ret = kvm_arch_check_processor_compat(c->opaque); | |
f257d6dc SC |
5813 | } |
5814 | ||
0ee75bea | 5815 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 5816 | struct module *module) |
6aa8b732 | 5817 | { |
b9904085 | 5818 | struct kvm_cpu_compat_check c; |
6aa8b732 | 5819 | int r; |
002c7f7c | 5820 | int cpu; |
6aa8b732 | 5821 | |
f8c16bba ZX |
5822 | r = kvm_arch_init(opaque); |
5823 | if (r) | |
d2308784 | 5824 | goto out_fail; |
cb498ea2 | 5825 | |
7dac16c3 AH |
5826 | /* |
5827 | * kvm_arch_init makes sure there's at most one caller | |
5828 | * for architectures that support multiple implementations, | |
5829 | * like intel and amd on x86. | |
36343f6e PB |
5830 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
5831 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 5832 | */ |
36343f6e PB |
5833 | r = kvm_irqfd_init(); |
5834 | if (r) | |
5835 | goto out_irqfd; | |
7dac16c3 | 5836 | |
8437a617 | 5837 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
5838 | r = -ENOMEM; |
5839 | goto out_free_0; | |
5840 | } | |
5841 | ||
b9904085 | 5842 | r = kvm_arch_hardware_setup(opaque); |
6aa8b732 | 5843 | if (r < 0) |
faf0be22 | 5844 | goto out_free_1; |
6aa8b732 | 5845 | |
b9904085 SC |
5846 | c.ret = &r; |
5847 | c.opaque = opaque; | |
002c7f7c | 5848 | for_each_online_cpu(cpu) { |
b9904085 | 5849 | smp_call_function_single(cpu, check_processor_compat, &c, 1); |
002c7f7c | 5850 | if (r < 0) |
faf0be22 | 5851 | goto out_free_2; |
002c7f7c YS |
5852 | } |
5853 | ||
73c1b41e | 5854 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 5855 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 5856 | if (r) |
d2308784 | 5857 | goto out_free_2; |
6aa8b732 AK |
5858 | register_reboot_notifier(&kvm_reboot_notifier); |
5859 | ||
c16f862d | 5860 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
5861 | if (!vcpu_align) |
5862 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
5863 | kvm_vcpu_cache = |
5864 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
5865 | SLAB_ACCOUNT, | |
5866 | offsetof(struct kvm_vcpu, arch), | |
ce55c049 JZ |
5867 | offsetofend(struct kvm_vcpu, stats_id) |
5868 | - offsetof(struct kvm_vcpu, arch), | |
46515736 | 5869 | NULL); |
c16f862d RR |
5870 | if (!kvm_vcpu_cache) { |
5871 | r = -ENOMEM; | |
fb3600cc | 5872 | goto out_free_3; |
c16f862d RR |
5873 | } |
5874 | ||
baff59cc VK |
5875 | for_each_possible_cpu(cpu) { |
5876 | if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu), | |
5877 | GFP_KERNEL, cpu_to_node(cpu))) { | |
5878 | r = -ENOMEM; | |
5879 | goto out_free_4; | |
5880 | } | |
5881 | } | |
5882 | ||
af585b92 GN |
5883 | r = kvm_async_pf_init(); |
5884 | if (r) | |
baff59cc | 5885 | goto out_free_5; |
af585b92 | 5886 | |
6aa8b732 AK |
5887 | kvm_chardev_ops.owner = module; |
5888 | ||
5889 | r = misc_register(&kvm_dev); | |
5890 | if (r) { | |
1170adc6 | 5891 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 5892 | goto out_unreg; |
6aa8b732 AK |
5893 | } |
5894 | ||
fb3600cc RW |
5895 | register_syscore_ops(&kvm_syscore_ops); |
5896 | ||
15ad7146 AK |
5897 | kvm_preempt_ops.sched_in = kvm_sched_in; |
5898 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
5899 | ||
929f45e3 | 5900 | kvm_init_debug(); |
0ea4ed8e | 5901 | |
3c3c29fd PB |
5902 | r = kvm_vfio_ops_init(); |
5903 | WARN_ON(r); | |
5904 | ||
c7addb90 | 5905 | return 0; |
6aa8b732 | 5906 | |
af585b92 GN |
5907 | out_unreg: |
5908 | kvm_async_pf_deinit(); | |
baff59cc VK |
5909 | out_free_5: |
5910 | for_each_possible_cpu(cpu) | |
5911 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
5912 | out_free_4: | |
c16f862d | 5913 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 5914 | out_free_3: |
6aa8b732 | 5915 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5916 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 5917 | out_free_2: |
e9b11c17 | 5918 | kvm_arch_hardware_unsetup(); |
faf0be22 | 5919 | out_free_1: |
7f59f492 | 5920 | free_cpumask_var(cpus_hardware_enabled); |
d2308784 | 5921 | out_free_0: |
a0f155e9 | 5922 | kvm_irqfd_exit(); |
36343f6e | 5923 | out_irqfd: |
7dac16c3 AH |
5924 | kvm_arch_exit(); |
5925 | out_fail: | |
6aa8b732 AK |
5926 | return r; |
5927 | } | |
cb498ea2 | 5928 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 5929 | |
cb498ea2 | 5930 | void kvm_exit(void) |
6aa8b732 | 5931 | { |
baff59cc VK |
5932 | int cpu; |
5933 | ||
4bd33b56 | 5934 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 5935 | misc_deregister(&kvm_dev); |
baff59cc VK |
5936 | for_each_possible_cpu(cpu) |
5937 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
c16f862d | 5938 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 5939 | kvm_async_pf_deinit(); |
fb3600cc | 5940 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 5941 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5942 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 5943 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 5944 | kvm_arch_hardware_unsetup(); |
f8c16bba | 5945 | kvm_arch_exit(); |
a0f155e9 | 5946 | kvm_irqfd_exit(); |
7f59f492 | 5947 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 5948 | kvm_vfio_ops_exit(); |
6aa8b732 | 5949 | } |
cb498ea2 | 5950 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
5951 | |
5952 | struct kvm_vm_worker_thread_context { | |
5953 | struct kvm *kvm; | |
5954 | struct task_struct *parent; | |
5955 | struct completion init_done; | |
5956 | kvm_vm_thread_fn_t thread_fn; | |
5957 | uintptr_t data; | |
5958 | int err; | |
5959 | }; | |
5960 | ||
5961 | static int kvm_vm_worker_thread(void *context) | |
5962 | { | |
5963 | /* | |
5964 | * The init_context is allocated on the stack of the parent thread, so | |
5965 | * we have to locally copy anything that is needed beyond initialization | |
5966 | */ | |
5967 | struct kvm_vm_worker_thread_context *init_context = context; | |
e45cce30 | 5968 | struct task_struct *parent; |
c57c8046 JS |
5969 | struct kvm *kvm = init_context->kvm; |
5970 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
5971 | uintptr_t data = init_context->data; | |
5972 | int err; | |
5973 | ||
5974 | err = kthread_park(current); | |
5975 | /* kthread_park(current) is never supposed to return an error */ | |
5976 | WARN_ON(err != 0); | |
5977 | if (err) | |
5978 | goto init_complete; | |
5979 | ||
5980 | err = cgroup_attach_task_all(init_context->parent, current); | |
5981 | if (err) { | |
5982 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
5983 | __func__, err); | |
5984 | goto init_complete; | |
5985 | } | |
5986 | ||
5987 | set_user_nice(current, task_nice(init_context->parent)); | |
5988 | ||
5989 | init_complete: | |
5990 | init_context->err = err; | |
5991 | complete(&init_context->init_done); | |
5992 | init_context = NULL; | |
5993 | ||
5994 | if (err) | |
e45cce30 | 5995 | goto out; |
c57c8046 JS |
5996 | |
5997 | /* Wait to be woken up by the spawner before proceeding. */ | |
5998 | kthread_parkme(); | |
5999 | ||
6000 | if (!kthread_should_stop()) | |
6001 | err = thread_fn(kvm, data); | |
6002 | ||
e45cce30 VS |
6003 | out: |
6004 | /* | |
6005 | * Move kthread back to its original cgroup to prevent it lingering in | |
6006 | * the cgroup of the VM process, after the latter finishes its | |
6007 | * execution. | |
6008 | * | |
6009 | * kthread_stop() waits on the 'exited' completion condition which is | |
6010 | * set in exit_mm(), via mm_release(), in do_exit(). However, the | |
6011 | * kthread is removed from the cgroup in the cgroup_exit() which is | |
6012 | * called after the exit_mm(). This causes the kthread_stop() to return | |
6013 | * before the kthread actually quits the cgroup. | |
6014 | */ | |
6015 | rcu_read_lock(); | |
6016 | parent = rcu_dereference(current->real_parent); | |
6017 | get_task_struct(parent); | |
6018 | rcu_read_unlock(); | |
6019 | cgroup_attach_task_all(parent, current); | |
6020 | put_task_struct(parent); | |
6021 | ||
c57c8046 JS |
6022 | return err; |
6023 | } | |
6024 | ||
6025 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
6026 | uintptr_t data, const char *name, | |
6027 | struct task_struct **thread_ptr) | |
6028 | { | |
6029 | struct kvm_vm_worker_thread_context init_context = {}; | |
6030 | struct task_struct *thread; | |
6031 | ||
6032 | *thread_ptr = NULL; | |
6033 | init_context.kvm = kvm; | |
6034 | init_context.parent = current; | |
6035 | init_context.thread_fn = thread_fn; | |
6036 | init_context.data = data; | |
6037 | init_completion(&init_context.init_done); | |
6038 | ||
6039 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
6040 | "%s-%d", name, task_pid_nr(current)); | |
6041 | if (IS_ERR(thread)) | |
6042 | return PTR_ERR(thread); | |
6043 | ||
6044 | /* kthread_run is never supposed to return NULL */ | |
6045 | WARN_ON(thread == NULL); | |
6046 | ||
6047 | wait_for_completion(&init_context.init_done); | |
6048 | ||
6049 | if (!init_context.err) | |
6050 | *thread_ptr = thread; | |
6051 | ||
6052 | return init_context.err; | |
6053 | } |