Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
6aa8b732 AK |
2 | /* |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * This module enables machines with Intel VT-x extensions to run virtual | |
6 | * machines without emulation or binary translation. | |
7 | * | |
8 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 9 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
10 | * |
11 | * Authors: | |
12 | * Avi Kivity <avi@qumranet.com> | |
13 | * Yaniv Kamay <yaniv@qumranet.com> | |
6aa8b732 AK |
14 | */ |
15 | ||
af669ac6 | 16 | #include <kvm/iodev.h> |
6aa8b732 | 17 | |
edf88417 | 18 | #include <linux/kvm_host.h> |
6aa8b732 AK |
19 | #include <linux/kvm.h> |
20 | #include <linux/module.h> | |
21 | #include <linux/errno.h> | |
6aa8b732 | 22 | #include <linux/percpu.h> |
6aa8b732 AK |
23 | #include <linux/mm.h> |
24 | #include <linux/miscdevice.h> | |
25 | #include <linux/vmalloc.h> | |
6aa8b732 | 26 | #include <linux/reboot.h> |
6aa8b732 AK |
27 | #include <linux/debugfs.h> |
28 | #include <linux/highmem.h> | |
29 | #include <linux/file.h> | |
fb3600cc | 30 | #include <linux/syscore_ops.h> |
774c47f1 | 31 | #include <linux/cpu.h> |
174cd4b1 | 32 | #include <linux/sched/signal.h> |
6e84f315 | 33 | #include <linux/sched/mm.h> |
03441a34 | 34 | #include <linux/sched/stat.h> |
d9e368d6 AK |
35 | #include <linux/cpumask.h> |
36 | #include <linux/smp.h> | |
d6d28168 | 37 | #include <linux/anon_inodes.h> |
04d2cc77 | 38 | #include <linux/profile.h> |
7aa81cc0 | 39 | #include <linux/kvm_para.h> |
6fc138d2 | 40 | #include <linux/pagemap.h> |
8d4e1288 | 41 | #include <linux/mman.h> |
35149e21 | 42 | #include <linux/swap.h> |
e56d532f | 43 | #include <linux/bitops.h> |
547de29e | 44 | #include <linux/spinlock.h> |
6ff5894c | 45 | #include <linux/compat.h> |
bc6678a3 | 46 | #include <linux/srcu.h> |
8f0b1ab6 | 47 | #include <linux/hugetlb.h> |
5a0e3ad6 | 48 | #include <linux/slab.h> |
743eeb0b SL |
49 | #include <linux/sort.h> |
50 | #include <linux/bsearch.h> | |
c011d23b | 51 | #include <linux/io.h> |
2eb06c30 | 52 | #include <linux/lockdep.h> |
c57c8046 | 53 | #include <linux/kthread.h> |
2fdef3a2 | 54 | #include <linux/suspend.h> |
6aa8b732 | 55 | |
e495606d | 56 | #include <asm/processor.h> |
2ea75be3 | 57 | #include <asm/ioctl.h> |
7c0f6ba6 | 58 | #include <linux/uaccess.h> |
6aa8b732 | 59 | |
5f94c174 | 60 | #include "coalesced_mmio.h" |
af585b92 | 61 | #include "async_pf.h" |
982ed0de | 62 | #include "kvm_mm.h" |
3c3c29fd | 63 | #include "vfio.h" |
5f94c174 | 64 | |
4c8c3c7f VS |
65 | #include <trace/events/ipi.h> |
66 | ||
229456fc MT |
67 | #define CREATE_TRACE_POINTS |
68 | #include <trace/events/kvm.h> | |
69 | ||
fb04a1ed PX |
70 | #include <linux/kvm_dirty_ring.h> |
71 | ||
4c8c3c7f | 72 | |
536a6f88 JF |
73 | /* Worst case buffer size needed for holding an integer. */ |
74 | #define ITOA_MAX_LEN 12 | |
75 | ||
6aa8b732 AK |
76 | MODULE_AUTHOR("Qumranet"); |
77 | MODULE_LICENSE("GPL"); | |
78 | ||
920552b2 | 79 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 80 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
039c5d1b | 81 | module_param(halt_poll_ns, uint, 0644); |
ec76d819 | 82 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 83 | |
aca6ff29 | 84 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 85 | unsigned int halt_poll_ns_grow = 2; |
039c5d1b | 86 | module_param(halt_poll_ns_grow, uint, 0644); |
ec76d819 | 87 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 | 88 | |
49113d36 NW |
89 | /* The start value to grow halt_poll_ns from */ |
90 | unsigned int halt_poll_ns_grow_start = 10000; /* 10us */ | |
91 | module_param(halt_poll_ns_grow_start, uint, 0644); | |
92 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start); | |
93 | ||
aca6ff29 | 94 | /* Default resets per-vcpu halt_poll_ns . */ |
ec76d819 | 95 | unsigned int halt_poll_ns_shrink; |
039c5d1b | 96 | module_param(halt_poll_ns_shrink, uint, 0644); |
ec76d819 | 97 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 98 | |
fa40a821 MT |
99 | /* |
100 | * Ordering of locks: | |
101 | * | |
b7d409de | 102 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
103 | */ |
104 | ||
0d9ce162 | 105 | DEFINE_MUTEX(kvm_lock); |
e9b11c17 | 106 | LIST_HEAD(vm_list); |
133de902 | 107 | |
aaba298c | 108 | static struct kmem_cache *kvm_vcpu_cache; |
1165f5fe | 109 | |
15ad7146 | 110 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
7495e22b | 111 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu); |
15ad7146 | 112 | |
76f7c879 | 113 | struct dentry *kvm_debugfs_dir; |
e23a808b | 114 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 115 | |
09cbcef6 | 116 | static const struct file_operations stat_fops_per_vm; |
536a6f88 | 117 | |
bccf2150 AK |
118 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
119 | unsigned long arg); | |
de8e5d74 | 120 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
121 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
122 | unsigned long arg); | |
7ddfd3e0 MZ |
123 | #define KVM_COMPAT(c) .compat_ioctl = (c) |
124 | #else | |
9cb09e7c MZ |
125 | /* |
126 | * For architectures that don't implement a compat infrastructure, | |
127 | * adopt a double line of defense: | |
128 | * - Prevent a compat task from opening /dev/kvm | |
129 | * - If the open has been done by a 64bit task, and the KVM fd | |
130 | * passed to a compat task, let the ioctls fail. | |
131 | */ | |
7ddfd3e0 MZ |
132 | static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl, |
133 | unsigned long arg) { return -EINVAL; } | |
b9876e6d MZ |
134 | |
135 | static int kvm_no_compat_open(struct inode *inode, struct file *file) | |
136 | { | |
137 | return is_compat_task() ? -ENODEV : 0; | |
138 | } | |
139 | #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \ | |
140 | .open = kvm_no_compat_open | |
1dda606c | 141 | #endif |
10474ae8 AG |
142 | static int hardware_enable_all(void); |
143 | static void hardware_disable_all(void); | |
bccf2150 | 144 | |
e93f8a0f | 145 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 146 | |
286de8f6 CI |
147 | #define KVM_EVENT_CREATE_VM 0 |
148 | #define KVM_EVENT_DESTROY_VM 1 | |
149 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); | |
150 | static unsigned long long kvm_createvm_count; | |
151 | static unsigned long long kvm_active_vms; | |
152 | ||
baff59cc VK |
153 | static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask); |
154 | ||
683412cc MZ |
155 | __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm) |
156 | { | |
157 | } | |
158 | ||
284dc493 | 159 | bool kvm_is_zone_device_page(struct page *page) |
a78986aa SC |
160 | { |
161 | /* | |
162 | * The metadata used by is_zone_device_page() to determine whether or | |
163 | * not a page is ZONE_DEVICE is guaranteed to be valid if and only if | |
164 | * the device has been pinned, e.g. by get_user_pages(). WARN if the | |
165 | * page_count() is zero to help detect bad usage of this helper. | |
166 | */ | |
284dc493 | 167 | if (WARN_ON_ONCE(!page_count(page))) |
a78986aa SC |
168 | return false; |
169 | ||
284dc493 | 170 | return is_zone_device_page(page); |
a78986aa SC |
171 | } |
172 | ||
b14b2690 SC |
173 | /* |
174 | * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted | |
175 | * page, NULL otherwise. Note, the list of refcounted PG_reserved page types | |
176 | * is likely incomplete, it has been compiled purely through people wanting to | |
177 | * back guest with a certain type of memory and encountering issues. | |
178 | */ | |
179 | struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn) | |
cbff90a7 | 180 | { |
b14b2690 SC |
181 | struct page *page; |
182 | ||
183 | if (!pfn_valid(pfn)) | |
184 | return NULL; | |
185 | ||
186 | page = pfn_to_page(pfn); | |
187 | if (!PageReserved(page)) | |
188 | return page; | |
189 | ||
190 | /* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */ | |
191 | if (is_zero_pfn(pfn)) | |
192 | return page; | |
193 | ||
a78986aa SC |
194 | /* |
195 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting | |
196 | * perspective they are "normal" pages, albeit with slightly different | |
197 | * usage rules. | |
198 | */ | |
b14b2690 SC |
199 | if (kvm_is_zone_device_page(page)) |
200 | return page; | |
cbff90a7 | 201 | |
b14b2690 | 202 | return NULL; |
cbff90a7 BAY |
203 | } |
204 | ||
bccf2150 AK |
205 | /* |
206 | * Switches to specified vcpu, until a matching vcpu_put() | |
207 | */ | |
ec7660cc | 208 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 209 | { |
ec7660cc | 210 | int cpu = get_cpu(); |
7495e22b PB |
211 | |
212 | __this_cpu_write(kvm_running_vcpu, vcpu); | |
15ad7146 | 213 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 214 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 215 | put_cpu(); |
6aa8b732 | 216 | } |
2f1fe811 | 217 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 218 | |
313a3dc7 | 219 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 220 | { |
15ad7146 | 221 | preempt_disable(); |
313a3dc7 | 222 | kvm_arch_vcpu_put(vcpu); |
15ad7146 | 223 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
7495e22b | 224 | __this_cpu_write(kvm_running_vcpu, NULL); |
15ad7146 | 225 | preempt_enable(); |
6aa8b732 | 226 | } |
2f1fe811 | 227 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 228 | |
7a97cec2 PB |
229 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
230 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
231 | { | |
232 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
233 | ||
234 | /* | |
235 | * We need to wait for the VCPU to reenable interrupts and get out of | |
236 | * READING_SHADOW_PAGE_TABLES mode. | |
237 | */ | |
238 | if (req & KVM_REQUEST_WAIT) | |
239 | return mode != OUTSIDE_GUEST_MODE; | |
240 | ||
241 | /* | |
242 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
243 | */ | |
244 | return mode == IN_GUEST_MODE; | |
245 | } | |
246 | ||
f24b44e4 | 247 | static void ack_kick(void *_completed) |
d9e368d6 | 248 | { |
d9e368d6 AK |
249 | } |
250 | ||
620b2438 | 251 | static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait) |
b49defe8 | 252 | { |
b49defe8 PB |
253 | if (cpumask_empty(cpus)) |
254 | return false; | |
255 | ||
f24b44e4 | 256 | smp_call_function_many(cpus, ack_kick, NULL, wait); |
b49defe8 PB |
257 | return true; |
258 | } | |
259 | ||
b56bd8e0 JL |
260 | static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req, |
261 | struct cpumask *tmp, int current_cpu) | |
ae0946cd VK |
262 | { |
263 | int cpu; | |
264 | ||
df06dae3 SC |
265 | if (likely(!(req & KVM_REQUEST_NO_ACTION))) |
266 | __kvm_make_request(req, vcpu); | |
ae0946cd VK |
267 | |
268 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) | |
269 | return; | |
270 | ||
ae0946cd VK |
271 | /* |
272 | * Note, the vCPU could get migrated to a different pCPU at any point | |
273 | * after kvm_request_needs_ipi(), which could result in sending an IPI | |
274 | * to the previous pCPU. But, that's OK because the purpose of the IPI | |
275 | * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is | |
276 | * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES | |
277 | * after this point is also OK, as the requirement is only that KVM wait | |
278 | * for vCPUs that were reading SPTEs _before_ any changes were | |
279 | * finalized. See kvm_vcpu_kick() for more details on handling requests. | |
280 | */ | |
281 | if (kvm_request_needs_ipi(vcpu, req)) { | |
282 | cpu = READ_ONCE(vcpu->cpu); | |
283 | if (cpu != -1 && cpu != current_cpu) | |
284 | __cpumask_set_cpu(cpu, tmp); | |
285 | } | |
286 | } | |
287 | ||
7053df4e | 288 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
620b2438 | 289 | unsigned long *vcpu_bitmap) |
d9e368d6 | 290 | { |
d9e368d6 | 291 | struct kvm_vcpu *vcpu; |
620b2438 | 292 | struct cpumask *cpus; |
ae0946cd | 293 | int i, me; |
7053df4e | 294 | bool called; |
6ef7a1bc | 295 | |
3cba4130 | 296 | me = get_cpu(); |
7053df4e | 297 | |
620b2438 VK |
298 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
299 | cpumask_clear(cpus); | |
300 | ||
ae0946cd VK |
301 | for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) { |
302 | vcpu = kvm_get_vcpu(kvm, i); | |
381cecc5 | 303 | if (!vcpu) |
7053df4e | 304 | continue; |
b56bd8e0 | 305 | kvm_make_vcpu_request(vcpu, req, cpus, me); |
49846896 | 306 | } |
7053df4e | 307 | |
620b2438 | 308 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); |
3cba4130 | 309 | put_cpu(); |
7053df4e VK |
310 | |
311 | return called; | |
312 | } | |
313 | ||
82e9c84d | 314 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
7053df4e | 315 | { |
ae0946cd | 316 | struct kvm_vcpu *vcpu; |
baff59cc | 317 | struct cpumask *cpus; |
46808a4c | 318 | unsigned long i; |
7053df4e | 319 | bool called; |
46808a4c | 320 | int me; |
7053df4e | 321 | |
ae0946cd VK |
322 | me = get_cpu(); |
323 | ||
baff59cc VK |
324 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
325 | cpumask_clear(cpus); | |
326 | ||
82e9c84d | 327 | kvm_for_each_vcpu(i, vcpu, kvm) |
b56bd8e0 | 328 | kvm_make_vcpu_request(vcpu, req, cpus, me); |
ae0946cd VK |
329 | |
330 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); | |
331 | put_cpu(); | |
7053df4e | 332 | |
49846896 | 333 | return called; |
d9e368d6 | 334 | } |
a2486020 | 335 | EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request); |
54163a34 | 336 | |
49846896 | 337 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 338 | { |
3cc4e148 | 339 | ++kvm->stat.generic.remote_tlb_flush_requests; |
6bc6db00 | 340 | |
4ae3cb3a LT |
341 | /* |
342 | * We want to publish modifications to the page tables before reading | |
343 | * mode. Pairs with a memory barrier in arch-specific code. | |
344 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
345 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
346 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
347 | * | |
348 | * There is already an smp_mb__after_atomic() before | |
349 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
350 | * barrier here. | |
351 | */ | |
a1342c80 | 352 | if (!kvm_arch_flush_remote_tlbs(kvm) |
b08660e5 | 353 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
0193cc90 | 354 | ++kvm->stat.generic.remote_tlb_flush; |
2e53d63a | 355 | } |
2ba9f0d8 | 356 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
2e53d63a | 357 | |
d4788996 DM |
358 | void kvm_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages) |
359 | { | |
360 | if (!kvm_arch_flush_remote_tlbs_range(kvm, gfn, nr_pages)) | |
361 | return; | |
362 | ||
363 | /* | |
364 | * Fall back to a flushing entire TLBs if the architecture range-based | |
365 | * TLB invalidation is unsupported or can't be performed for whatever | |
366 | * reason. | |
367 | */ | |
368 | kvm_flush_remote_tlbs(kvm); | |
369 | } | |
370 | ||
619b5072 DM |
371 | void kvm_flush_remote_tlbs_memslot(struct kvm *kvm, |
372 | const struct kvm_memory_slot *memslot) | |
373 | { | |
374 | /* | |
375 | * All current use cases for flushing the TLBs for a specific memslot | |
376 | * are related to dirty logging, and many do the TLB flush out of | |
377 | * mmu_lock. The interaction between the various operations on memslot | |
378 | * must be serialized by slots_locks to ensure the TLB flush from one | |
379 | * operation is observed by any other operation on the same memslot. | |
380 | */ | |
381 | lockdep_assert_held(&kvm->slots_lock); | |
382 | kvm_flush_remote_tlbs_range(kvm, memslot->base_gfn, memslot->npages); | |
383 | } | |
2e53d63a | 384 | |
683412cc MZ |
385 | static void kvm_flush_shadow_all(struct kvm *kvm) |
386 | { | |
387 | kvm_arch_flush_shadow_all(kvm); | |
388 | kvm_arch_guest_memory_reclaimed(kvm); | |
389 | } | |
390 | ||
6926f95a SC |
391 | #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE |
392 | static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc, | |
393 | gfp_t gfp_flags) | |
394 | { | |
395 | gfp_flags |= mc->gfp_zero; | |
396 | ||
397 | if (mc->kmem_cache) | |
398 | return kmem_cache_alloc(mc->kmem_cache, gfp_flags); | |
399 | else | |
400 | return (void *)__get_free_page(gfp_flags); | |
401 | } | |
402 | ||
837f66c7 | 403 | int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min) |
6926f95a | 404 | { |
63f4b210 | 405 | gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT; |
6926f95a SC |
406 | void *obj; |
407 | ||
408 | if (mc->nobjs >= min) | |
409 | return 0; | |
837f66c7 DM |
410 | |
411 | if (unlikely(!mc->objects)) { | |
412 | if (WARN_ON_ONCE(!capacity)) | |
413 | return -EIO; | |
414 | ||
ea3689d9 | 415 | mc->objects = kvmalloc_array(capacity, sizeof(void *), gfp); |
837f66c7 DM |
416 | if (!mc->objects) |
417 | return -ENOMEM; | |
418 | ||
419 | mc->capacity = capacity; | |
420 | } | |
421 | ||
422 | /* It is illegal to request a different capacity across topups. */ | |
423 | if (WARN_ON_ONCE(mc->capacity != capacity)) | |
424 | return -EIO; | |
425 | ||
426 | while (mc->nobjs < mc->capacity) { | |
427 | obj = mmu_memory_cache_alloc_obj(mc, gfp); | |
6926f95a SC |
428 | if (!obj) |
429 | return mc->nobjs >= min ? 0 : -ENOMEM; | |
430 | mc->objects[mc->nobjs++] = obj; | |
431 | } | |
432 | return 0; | |
433 | } | |
434 | ||
837f66c7 DM |
435 | int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min) |
436 | { | |
437 | return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min); | |
438 | } | |
439 | ||
6926f95a SC |
440 | int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc) |
441 | { | |
442 | return mc->nobjs; | |
443 | } | |
444 | ||
445 | void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
446 | { | |
447 | while (mc->nobjs) { | |
448 | if (mc->kmem_cache) | |
449 | kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]); | |
450 | else | |
451 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
452 | } | |
837f66c7 DM |
453 | |
454 | kvfree(mc->objects); | |
455 | ||
456 | mc->objects = NULL; | |
457 | mc->capacity = 0; | |
6926f95a SC |
458 | } |
459 | ||
460 | void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
461 | { | |
462 | void *p; | |
463 | ||
464 | if (WARN_ON(!mc->nobjs)) | |
465 | p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT); | |
466 | else | |
467 | p = mc->objects[--mc->nobjs]; | |
468 | BUG_ON(!p); | |
469 | return p; | |
470 | } | |
471 | #endif | |
472 | ||
8bd826d6 | 473 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 474 | { |
fb3f0f51 RR |
475 | mutex_init(&vcpu->mutex); |
476 | vcpu->cpu = -1; | |
fb3f0f51 RR |
477 | vcpu->kvm = kvm; |
478 | vcpu->vcpu_id = id; | |
34bb10b7 | 479 | vcpu->pid = NULL; |
510958e9 | 480 | #ifndef __KVM_HAVE_ARCH_WQP |
da4ad88c | 481 | rcuwait_init(&vcpu->wait); |
510958e9 | 482 | #endif |
af585b92 | 483 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 484 | |
4c088493 R |
485 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
486 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 487 | vcpu->preempted = false; |
d73eb57b | 488 | vcpu->ready = false; |
d5c48deb | 489 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
a54d8066 | 490 | vcpu->last_used_slot = NULL; |
58fc1166 OU |
491 | |
492 | /* Fill the stats id string for the vcpu */ | |
493 | snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d", | |
494 | task_pid_nr(current), id); | |
fb3f0f51 | 495 | } |
fb3f0f51 | 496 | |
27592ae8 | 497 | static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
4543bdc0 SC |
498 | { |
499 | kvm_arch_vcpu_destroy(vcpu); | |
5593473a | 500 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
e529ef66 | 501 | |
9941d224 SC |
502 | /* |
503 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
504 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
505 | * are already gone. | |
506 | */ | |
507 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
508 | ||
8bd826d6 | 509 | free_page((unsigned long)vcpu->run); |
e529ef66 | 510 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 | 511 | } |
27592ae8 MZ |
512 | |
513 | void kvm_destroy_vcpus(struct kvm *kvm) | |
514 | { | |
46808a4c | 515 | unsigned long i; |
27592ae8 MZ |
516 | struct kvm_vcpu *vcpu; |
517 | ||
518 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
519 | kvm_vcpu_destroy(vcpu); | |
c5b07754 | 520 | xa_erase(&kvm->vcpu_array, i); |
27592ae8 MZ |
521 | } |
522 | ||
523 | atomic_set(&kvm->online_vcpus, 0); | |
524 | } | |
525 | EXPORT_SYMBOL_GPL(kvm_destroy_vcpus); | |
4543bdc0 | 526 | |
f128cf8c | 527 | #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER |
e930bffe AA |
528 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) |
529 | { | |
530 | return container_of(mn, struct kvm, mmu_notifier); | |
531 | } | |
532 | ||
e97b39c5 | 533 | typedef bool (*gfn_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range); |
3039bcc7 | 534 | |
8569992d | 535 | typedef void (*on_lock_fn_t)(struct kvm *kvm); |
683412cc | 536 | |
e97b39c5 SC |
537 | struct kvm_mmu_notifier_range { |
538 | /* | |
539 | * 64-bit addresses, as KVM notifiers can operate on host virtual | |
540 | * addresses (unsigned long) and guest physical addresses (64-bit). | |
541 | */ | |
542 | u64 start; | |
543 | u64 end; | |
3e1efe2b | 544 | union kvm_mmu_notifier_arg arg; |
e97b39c5 | 545 | gfn_handler_t handler; |
f922bd9b | 546 | on_lock_fn_t on_lock; |
3039bcc7 SC |
547 | bool flush_on_ret; |
548 | bool may_block; | |
549 | }; | |
550 | ||
cec29eef SC |
551 | /* |
552 | * The inner-most helper returns a tuple containing the return value from the | |
553 | * arch- and action-specific handler, plus a flag indicating whether or not at | |
554 | * least one memslot was found, i.e. if the handler found guest memory. | |
555 | * | |
556 | * Note, most notifiers are averse to booleans, so even though KVM tracks the | |
557 | * return from arch code as a bool, outer helpers will cast it to an int. :-( | |
558 | */ | |
559 | typedef struct kvm_mmu_notifier_return { | |
560 | bool ret; | |
561 | bool found_memslot; | |
562 | } kvm_mn_ret_t; | |
563 | ||
f922bd9b SC |
564 | /* |
565 | * Use a dedicated stub instead of NULL to indicate that there is no callback | |
566 | * function/handler. The compiler technically can't guarantee that a real | |
567 | * function will have a non-zero address, and so it will generate code to | |
568 | * check for !NULL, whereas comparing against a stub will be elided at compile | |
569 | * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9). | |
570 | */ | |
571 | static void kvm_null_fn(void) | |
572 | { | |
573 | ||
574 | } | |
575 | #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn) | |
576 | ||
3e1efe2b SC |
577 | static const union kvm_mmu_notifier_arg KVM_MMU_NOTIFIER_NO_ARG; |
578 | ||
ed922739 MS |
579 | /* Iterate over each memslot intersecting [start, last] (inclusive) range */ |
580 | #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \ | |
581 | for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \ | |
582 | node; \ | |
583 | node = interval_tree_iter_next(node, start, last)) \ | |
584 | ||
cec29eef SC |
585 | static __always_inline kvm_mn_ret_t __kvm_handle_hva_range(struct kvm *kvm, |
586 | const struct kvm_mmu_notifier_range *range) | |
3039bcc7 | 587 | { |
cec29eef SC |
588 | struct kvm_mmu_notifier_return r = { |
589 | .ret = false, | |
590 | .found_memslot = false, | |
591 | }; | |
f922bd9b | 592 | struct kvm_gfn_range gfn_range; |
3039bcc7 SC |
593 | struct kvm_memory_slot *slot; |
594 | struct kvm_memslots *slots; | |
3039bcc7 SC |
595 | int i, idx; |
596 | ||
ed922739 | 597 | if (WARN_ON_ONCE(range->end <= range->start)) |
cec29eef | 598 | return r; |
ed922739 | 599 | |
f922bd9b SC |
600 | /* A null handler is allowed if and only if on_lock() is provided. */ |
601 | if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) && | |
602 | IS_KVM_NULL_FN(range->handler))) | |
cec29eef | 603 | return r; |
f922bd9b | 604 | |
3039bcc7 SC |
605 | idx = srcu_read_lock(&kvm->srcu); |
606 | ||
eed52e43 | 607 | for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { |
ed922739 MS |
608 | struct interval_tree_node *node; |
609 | ||
3039bcc7 | 610 | slots = __kvm_memslots(kvm, i); |
ed922739 MS |
611 | kvm_for_each_memslot_in_hva_range(node, slots, |
612 | range->start, range->end - 1) { | |
3039bcc7 SC |
613 | unsigned long hva_start, hva_end; |
614 | ||
a54d8066 | 615 | slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]); |
e97b39c5 SC |
616 | hva_start = max_t(unsigned long, range->start, slot->userspace_addr); |
617 | hva_end = min_t(unsigned long, range->end, | |
618 | slot->userspace_addr + (slot->npages << PAGE_SHIFT)); | |
3039bcc7 SC |
619 | |
620 | /* | |
621 | * To optimize for the likely case where the address | |
622 | * range is covered by zero or one memslots, don't | |
623 | * bother making these conditional (to avoid writes on | |
624 | * the second or later invocation of the handler). | |
625 | */ | |
3e1efe2b | 626 | gfn_range.arg = range->arg; |
3039bcc7 SC |
627 | gfn_range.may_block = range->may_block; |
628 | ||
629 | /* | |
630 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
631 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
632 | */ | |
633 | gfn_range.start = hva_to_gfn_memslot(hva_start, slot); | |
634 | gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot); | |
635 | gfn_range.slot = slot; | |
636 | ||
cec29eef SC |
637 | if (!r.found_memslot) { |
638 | r.found_memslot = true; | |
8931a454 | 639 | KVM_MMU_LOCK(kvm); |
071064f1 | 640 | if (!IS_KVM_NULL_FN(range->on_lock)) |
8569992d CP |
641 | range->on_lock(kvm); |
642 | ||
071064f1 PB |
643 | if (IS_KVM_NULL_FN(range->handler)) |
644 | break; | |
8931a454 | 645 | } |
cec29eef | 646 | r.ret |= range->handler(kvm, &gfn_range); |
3039bcc7 SC |
647 | } |
648 | } | |
649 | ||
cec29eef | 650 | if (range->flush_on_ret && r.ret) |
3039bcc7 SC |
651 | kvm_flush_remote_tlbs(kvm); |
652 | ||
193bbfaa | 653 | if (r.found_memslot) |
8931a454 | 654 | KVM_MMU_UNLOCK(kvm); |
f922bd9b | 655 | |
3039bcc7 SC |
656 | srcu_read_unlock(&kvm->srcu, idx); |
657 | ||
cec29eef | 658 | return r; |
3039bcc7 SC |
659 | } |
660 | ||
661 | static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn, | |
662 | unsigned long start, | |
663 | unsigned long end, | |
3e1efe2b | 664 | union kvm_mmu_notifier_arg arg, |
e97b39c5 | 665 | gfn_handler_t handler) |
3039bcc7 SC |
666 | { |
667 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
e97b39c5 | 668 | const struct kvm_mmu_notifier_range range = { |
3039bcc7 SC |
669 | .start = start, |
670 | .end = end, | |
3e1efe2b | 671 | .arg = arg, |
3039bcc7 | 672 | .handler = handler, |
f922bd9b | 673 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
674 | .flush_on_ret = true, |
675 | .may_block = false, | |
676 | }; | |
3039bcc7 | 677 | |
cec29eef | 678 | return __kvm_handle_hva_range(kvm, &range).ret; |
3039bcc7 SC |
679 | } |
680 | ||
681 | static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn, | |
682 | unsigned long start, | |
683 | unsigned long end, | |
e97b39c5 | 684 | gfn_handler_t handler) |
3039bcc7 SC |
685 | { |
686 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
e97b39c5 | 687 | const struct kvm_mmu_notifier_range range = { |
3039bcc7 SC |
688 | .start = start, |
689 | .end = end, | |
3039bcc7 | 690 | .handler = handler, |
f922bd9b | 691 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
692 | .flush_on_ret = false, |
693 | .may_block = false, | |
694 | }; | |
3039bcc7 | 695 | |
cec29eef | 696 | return __kvm_handle_hva_range(kvm, &range).ret; |
3039bcc7 | 697 | } |
2230f9e1 GS |
698 | |
699 | static bool kvm_change_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range) | |
700 | { | |
701 | /* | |
702 | * Skipping invalid memslots is correct if and only change_pte() is | |
703 | * surrounded by invalidate_range_{start,end}(), which is currently | |
704 | * guaranteed by the primary MMU. If that ever changes, KVM needs to | |
705 | * unmap the memslot instead of skipping the memslot to ensure that KVM | |
706 | * doesn't hold references to the old PFN. | |
707 | */ | |
708 | WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count)); | |
709 | ||
710 | if (range->slot->flags & KVM_MEMSLOT_INVALID) | |
711 | return false; | |
712 | ||
713 | return kvm_set_spte_gfn(kvm, range); | |
714 | } | |
715 | ||
3da0dd43 IE |
716 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
717 | struct mm_struct *mm, | |
718 | unsigned long address, | |
719 | pte_t pte) | |
720 | { | |
721 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
3e1efe2b | 722 | const union kvm_mmu_notifier_arg arg = { .pte = pte }; |
3da0dd43 | 723 | |
501b9185 SC |
724 | trace_kvm_set_spte_hva(address); |
725 | ||
c13fda23 | 726 | /* |
52ac8b35 | 727 | * .change_pte() must be surrounded by .invalidate_range_{start,end}(). |
20ec3ebd CP |
728 | * If mmu_invalidate_in_progress is zero, then no in-progress |
729 | * invalidations, including this one, found a relevant memslot at | |
730 | * start(); rechecking memslots here is unnecessary. Note, a false | |
731 | * positive (count elevated by a different invalidation) is sub-optimal | |
732 | * but functionally ok. | |
c13fda23 | 733 | */ |
52ac8b35 | 734 | WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count)); |
20ec3ebd | 735 | if (!READ_ONCE(kvm->mmu_invalidate_in_progress)) |
071064f1 | 736 | return; |
c13fda23 | 737 | |
3e1efe2b | 738 | kvm_handle_hva_range(mn, address, address + 1, arg, kvm_change_spte_gfn); |
3da0dd43 IE |
739 | } |
740 | ||
8569992d | 741 | void kvm_mmu_invalidate_begin(struct kvm *kvm) |
e930bffe | 742 | { |
8569992d | 743 | lockdep_assert_held_write(&kvm->mmu_lock); |
e930bffe AA |
744 | /* |
745 | * The count increase must become visible at unlock time as no | |
746 | * spte can be established without taking the mmu_lock and | |
747 | * count is also read inside the mmu_lock critical section. | |
748 | */ | |
20ec3ebd | 749 | kvm->mmu_invalidate_in_progress++; |
8569992d | 750 | |
20ec3ebd | 751 | if (likely(kvm->mmu_invalidate_in_progress == 1)) { |
8569992d CP |
752 | kvm->mmu_invalidate_range_start = INVALID_GPA; |
753 | kvm->mmu_invalidate_range_end = INVALID_GPA; | |
754 | } | |
755 | } | |
756 | ||
757 | void kvm_mmu_invalidate_range_add(struct kvm *kvm, gfn_t start, gfn_t end) | |
758 | { | |
759 | lockdep_assert_held_write(&kvm->mmu_lock); | |
760 | ||
761 | WARN_ON_ONCE(!kvm->mmu_invalidate_in_progress); | |
762 | ||
763 | if (likely(kvm->mmu_invalidate_range_start == INVALID_GPA)) { | |
20ec3ebd CP |
764 | kvm->mmu_invalidate_range_start = start; |
765 | kvm->mmu_invalidate_range_end = end; | |
4a42d848 DS |
766 | } else { |
767 | /* | |
a413a625 | 768 | * Fully tracking multiple concurrent ranges has diminishing |
4a42d848 DS |
769 | * returns. Keep things simple and just find the minimal range |
770 | * which includes the current and new ranges. As there won't be | |
771 | * enough information to subtract a range after its invalidate | |
772 | * completes, any ranges invalidated concurrently will | |
773 | * accumulate and persist until all outstanding invalidates | |
774 | * complete. | |
775 | */ | |
20ec3ebd CP |
776 | kvm->mmu_invalidate_range_start = |
777 | min(kvm->mmu_invalidate_range_start, start); | |
778 | kvm->mmu_invalidate_range_end = | |
779 | max(kvm->mmu_invalidate_range_end, end); | |
4a42d848 | 780 | } |
f922bd9b | 781 | } |
3039bcc7 | 782 | |
a7800aa8 | 783 | bool kvm_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range) |
8569992d CP |
784 | { |
785 | kvm_mmu_invalidate_range_add(kvm, range->start, range->end); | |
786 | return kvm_unmap_gfn_range(kvm, range); | |
787 | } | |
788 | ||
f922bd9b SC |
789 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
790 | const struct mmu_notifier_range *range) | |
791 | { | |
792 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
e97b39c5 | 793 | const struct kvm_mmu_notifier_range hva_range = { |
f922bd9b SC |
794 | .start = range->start, |
795 | .end = range->end, | |
8569992d | 796 | .handler = kvm_mmu_unmap_gfn_range, |
20ec3ebd | 797 | .on_lock = kvm_mmu_invalidate_begin, |
f922bd9b SC |
798 | .flush_on_ret = true, |
799 | .may_block = mmu_notifier_range_blockable(range), | |
800 | }; | |
565f3be2 | 801 | |
f922bd9b SC |
802 | trace_kvm_unmap_hva_range(range->start, range->end); |
803 | ||
52ac8b35 PB |
804 | /* |
805 | * Prevent memslot modification between range_start() and range_end() | |
806 | * so that conditionally locking provides the same result in both | |
20ec3ebd | 807 | * functions. Without that guarantee, the mmu_invalidate_in_progress |
52ac8b35 PB |
808 | * adjustments will be imbalanced. |
809 | * | |
810 | * Pairs with the decrement in range_end(). | |
811 | */ | |
812 | spin_lock(&kvm->mn_invalidate_lock); | |
813 | kvm->mn_active_invalidate_count++; | |
814 | spin_unlock(&kvm->mn_invalidate_lock); | |
815 | ||
58cd407c SC |
816 | /* |
817 | * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e. | |
818 | * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring | |
819 | * each cache's lock. There are relatively few caches in existence at | |
820 | * any given time, and the caches themselves can check for hva overlap, | |
821 | * i.e. don't need to rely on memslot overlap checks for performance. | |
822 | * Because this runs without holding mmu_lock, the pfn caches must use | |
20ec3ebd CP |
823 | * mn_active_invalidate_count (see above) instead of |
824 | * mmu_invalidate_in_progress. | |
58cd407c | 825 | */ |
982ed0de DW |
826 | gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end, |
827 | hva_range.may_block); | |
828 | ||
cec29eef SC |
829 | /* |
830 | * If one or more memslots were found and thus zapped, notify arch code | |
831 | * that guest memory has been reclaimed. This needs to be done *after* | |
832 | * dropping mmu_lock, as x86's reclaim path is slooooow. | |
833 | */ | |
834 | if (__kvm_handle_hva_range(kvm, &hva_range).found_memslot) | |
835 | kvm_arch_guest_memory_reclaimed(kvm); | |
93065ac7 | 836 | |
e649b3f0 | 837 | return 0; |
e930bffe AA |
838 | } |
839 | ||
8569992d | 840 | void kvm_mmu_invalidate_end(struct kvm *kvm) |
e930bffe | 841 | { |
8569992d CP |
842 | lockdep_assert_held_write(&kvm->mmu_lock); |
843 | ||
e930bffe AA |
844 | /* |
845 | * This sequence increase will notify the kvm page fault that | |
846 | * the page that is going to be mapped in the spte could have | |
847 | * been freed. | |
848 | */ | |
20ec3ebd | 849 | kvm->mmu_invalidate_seq++; |
a355aa54 | 850 | smp_wmb(); |
e930bffe AA |
851 | /* |
852 | * The above sequence increase must be visible before the | |
a355aa54 | 853 | * below count decrease, which is ensured by the smp_wmb above |
20ec3ebd | 854 | * in conjunction with the smp_rmb in mmu_invalidate_retry(). |
e930bffe | 855 | */ |
20ec3ebd | 856 | kvm->mmu_invalidate_in_progress--; |
c0db1923 | 857 | KVM_BUG_ON(kvm->mmu_invalidate_in_progress < 0, kvm); |
8569992d CP |
858 | |
859 | /* | |
860 | * Assert that at least one range was added between start() and end(). | |
861 | * Not adding a range isn't fatal, but it is a KVM bug. | |
862 | */ | |
863 | WARN_ON_ONCE(kvm->mmu_invalidate_range_start == INVALID_GPA); | |
f922bd9b SC |
864 | } |
865 | ||
866 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
867 | const struct mmu_notifier_range *range) | |
868 | { | |
869 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
e97b39c5 | 870 | const struct kvm_mmu_notifier_range hva_range = { |
f922bd9b SC |
871 | .start = range->start, |
872 | .end = range->end, | |
f922bd9b | 873 | .handler = (void *)kvm_null_fn, |
20ec3ebd | 874 | .on_lock = kvm_mmu_invalidate_end, |
f922bd9b SC |
875 | .flush_on_ret = false, |
876 | .may_block = mmu_notifier_range_blockable(range), | |
877 | }; | |
52ac8b35 | 878 | bool wake; |
f922bd9b SC |
879 | |
880 | __kvm_handle_hva_range(kvm, &hva_range); | |
e930bffe | 881 | |
52ac8b35 PB |
882 | /* Pairs with the increment in range_start(). */ |
883 | spin_lock(&kvm->mn_invalidate_lock); | |
d489ec95 SC |
884 | if (!WARN_ON_ONCE(!kvm->mn_active_invalidate_count)) |
885 | --kvm->mn_active_invalidate_count; | |
886 | wake = !kvm->mn_active_invalidate_count; | |
52ac8b35 PB |
887 | spin_unlock(&kvm->mn_invalidate_lock); |
888 | ||
889 | /* | |
890 | * There can only be one waiter, since the wait happens under | |
891 | * slots_lock. | |
892 | */ | |
893 | if (wake) | |
894 | rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait); | |
e930bffe AA |
895 | } |
896 | ||
897 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
898 | struct mm_struct *mm, | |
57128468 ALC |
899 | unsigned long start, |
900 | unsigned long end) | |
e930bffe | 901 | { |
501b9185 SC |
902 | trace_kvm_age_hva(start, end); |
903 | ||
3e1efe2b SC |
904 | return kvm_handle_hva_range(mn, start, end, KVM_MMU_NOTIFIER_NO_ARG, |
905 | kvm_age_gfn); | |
e930bffe AA |
906 | } |
907 | ||
1d7715c6 VD |
908 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
909 | struct mm_struct *mm, | |
910 | unsigned long start, | |
911 | unsigned long end) | |
912 | { | |
501b9185 SC |
913 | trace_kvm_age_hva(start, end); |
914 | ||
1d7715c6 VD |
915 | /* |
916 | * Even though we do not flush TLB, this will still adversely | |
917 | * affect performance on pre-Haswell Intel EPT, where there is | |
918 | * no EPT Access Bit to clear so that we have to tear down EPT | |
919 | * tables instead. If we find this unacceptable, we can always | |
920 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
921 | * do anything on clear_young. | |
922 | * | |
923 | * Also note that currently we never issue secondary TLB flushes | |
924 | * from clear_young, leaving this job up to the regular system | |
925 | * cadence. If we find this inaccurate, we might come up with a | |
926 | * more sophisticated heuristic later. | |
927 | */ | |
3039bcc7 | 928 | return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn); |
1d7715c6 VD |
929 | } |
930 | ||
8ee53820 AA |
931 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
932 | struct mm_struct *mm, | |
933 | unsigned long address) | |
934 | { | |
501b9185 SC |
935 | trace_kvm_test_age_hva(address); |
936 | ||
3039bcc7 SC |
937 | return kvm_handle_hva_range_no_flush(mn, address, address + 1, |
938 | kvm_test_age_gfn); | |
8ee53820 AA |
939 | } |
940 | ||
85db06e5 MT |
941 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
942 | struct mm_struct *mm) | |
943 | { | |
944 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
945 | int idx; |
946 | ||
947 | idx = srcu_read_lock(&kvm->srcu); | |
683412cc | 948 | kvm_flush_shadow_all(kvm); |
eda2beda | 949 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
950 | } |
951 | ||
e930bffe | 952 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e930bffe AA |
953 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
954 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
955 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 956 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 957 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 958 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 959 | .release = kvm_mmu_notifier_release, |
e930bffe | 960 | }; |
4c07b0a4 AK |
961 | |
962 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
963 | { | |
964 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
965 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
966 | } | |
967 | ||
f128cf8c | 968 | #else /* !CONFIG_KVM_GENERIC_MMU_NOTIFIER */ |
4c07b0a4 AK |
969 | |
970 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
971 | { | |
972 | return 0; | |
973 | } | |
974 | ||
f128cf8c | 975 | #endif /* CONFIG_KVM_GENERIC_MMU_NOTIFIER */ |
e930bffe | 976 | |
2fdef3a2 SS |
977 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
978 | static int kvm_pm_notifier_call(struct notifier_block *bl, | |
979 | unsigned long state, | |
980 | void *unused) | |
981 | { | |
982 | struct kvm *kvm = container_of(bl, struct kvm, pm_notifier); | |
983 | ||
984 | return kvm_arch_pm_notifier(kvm, state); | |
985 | } | |
986 | ||
987 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
988 | { | |
989 | kvm->pm_notifier.notifier_call = kvm_pm_notifier_call; | |
990 | /* Suspend KVM before we suspend ftrace, RCU, etc. */ | |
991 | kvm->pm_notifier.priority = INT_MAX; | |
992 | register_pm_notifier(&kvm->pm_notifier); | |
993 | } | |
994 | ||
995 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
996 | { | |
997 | unregister_pm_notifier(&kvm->pm_notifier); | |
998 | } | |
999 | #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
1000 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
1001 | { | |
1002 | } | |
1003 | ||
1004 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
1005 | { | |
1006 | } | |
1007 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
1008 | ||
a47d2b07 PB |
1009 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
1010 | { | |
1011 | if (!memslot->dirty_bitmap) | |
1012 | return; | |
1013 | ||
1014 | kvfree(memslot->dirty_bitmap); | |
1015 | memslot->dirty_bitmap = NULL; | |
1016 | } | |
1017 | ||
a54d8066 | 1018 | /* This does not remove the slot from struct kvm_memslots data structures */ |
e96c81ee | 1019 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 1020 | { |
a7800aa8 SC |
1021 | if (slot->flags & KVM_MEM_GUEST_MEMFD) |
1022 | kvm_gmem_unbind(slot); | |
1023 | ||
e96c81ee | 1024 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 1025 | |
e96c81ee | 1026 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 1027 | |
a54d8066 | 1028 | kfree(slot); |
a47d2b07 PB |
1029 | } |
1030 | ||
1031 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
1032 | { | |
a54d8066 | 1033 | struct hlist_node *idnode; |
a47d2b07 | 1034 | struct kvm_memory_slot *memslot; |
a54d8066 | 1035 | int bkt; |
a47d2b07 | 1036 | |
a54d8066 MS |
1037 | /* |
1038 | * The same memslot objects live in both active and inactive sets, | |
1039 | * arbitrarily free using index '1' so the second invocation of this | |
1040 | * function isn't operating over a structure with dangling pointers | |
1041 | * (even though this function isn't actually touching them). | |
1042 | */ | |
1043 | if (!slots->node_idx) | |
a47d2b07 PB |
1044 | return; |
1045 | ||
a54d8066 | 1046 | hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1]) |
e96c81ee | 1047 | kvm_free_memslot(kvm, memslot); |
bf3e05bc XG |
1048 | } |
1049 | ||
bc9e9e67 JZ |
1050 | static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc) |
1051 | { | |
1052 | switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) { | |
1053 | case KVM_STATS_TYPE_INSTANT: | |
1054 | return 0444; | |
1055 | case KVM_STATS_TYPE_CUMULATIVE: | |
1056 | case KVM_STATS_TYPE_PEAK: | |
1057 | default: | |
1058 | return 0644; | |
1059 | } | |
1060 | } | |
1061 | ||
1062 | ||
536a6f88 JF |
1063 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
1064 | { | |
1065 | int i; | |
bc9e9e67 JZ |
1066 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
1067 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 | 1068 | |
a44a4cc1 | 1069 | if (IS_ERR(kvm->debugfs_dentry)) |
536a6f88 JF |
1070 | return; |
1071 | ||
1072 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
1073 | ||
9d5a1dce LC |
1074 | if (kvm->debugfs_stat_data) { |
1075 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
1076 | kfree(kvm->debugfs_stat_data[i]); | |
1077 | kfree(kvm->debugfs_stat_data); | |
1078 | } | |
536a6f88 JF |
1079 | } |
1080 | ||
59f82aad | 1081 | static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname) |
536a6f88 | 1082 | { |
85cd39af PB |
1083 | static DEFINE_MUTEX(kvm_debugfs_lock); |
1084 | struct dentry *dent; | |
536a6f88 JF |
1085 | char dir_name[ITOA_MAX_LEN * 2]; |
1086 | struct kvm_stat_data *stat_data; | |
bc9e9e67 | 1087 | const struct _kvm_stats_desc *pdesc; |
b74ed7a6 | 1088 | int i, ret = -ENOMEM; |
bc9e9e67 JZ |
1089 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
1090 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
1091 | |
1092 | if (!debugfs_initialized()) | |
1093 | return 0; | |
1094 | ||
59f82aad | 1095 | snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname); |
85cd39af PB |
1096 | mutex_lock(&kvm_debugfs_lock); |
1097 | dent = debugfs_lookup(dir_name, kvm_debugfs_dir); | |
1098 | if (dent) { | |
1099 | pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name); | |
1100 | dput(dent); | |
1101 | mutex_unlock(&kvm_debugfs_lock); | |
1102 | return 0; | |
1103 | } | |
1104 | dent = debugfs_create_dir(dir_name, kvm_debugfs_dir); | |
1105 | mutex_unlock(&kvm_debugfs_lock); | |
1106 | if (IS_ERR(dent)) | |
1107 | return 0; | |
536a6f88 | 1108 | |
85cd39af | 1109 | kvm->debugfs_dentry = dent; |
536a6f88 JF |
1110 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
1111 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 1112 | GFP_KERNEL_ACCOUNT); |
536a6f88 | 1113 | if (!kvm->debugfs_stat_data) |
b74ed7a6 | 1114 | goto out_err; |
536a6f88 | 1115 | |
bc9e9e67 JZ |
1116 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
1117 | pdesc = &kvm_vm_stats_desc[i]; | |
b12ce36a | 1118 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 | 1119 | if (!stat_data) |
b74ed7a6 | 1120 | goto out_err; |
536a6f88 JF |
1121 | |
1122 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1123 | stat_data->desc = pdesc; |
1124 | stat_data->kind = KVM_STAT_VM; | |
1125 | kvm->debugfs_stat_data[i] = stat_data; | |
1126 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
1127 | kvm->debugfs_dentry, stat_data, | |
1128 | &stat_fops_per_vm); | |
1129 | } | |
1130 | ||
1131 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
1132 | pdesc = &kvm_vcpu_stats_desc[i]; | |
b12ce36a | 1133 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 | 1134 | if (!stat_data) |
b74ed7a6 | 1135 | goto out_err; |
536a6f88 JF |
1136 | |
1137 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1138 | stat_data->desc = pdesc; |
1139 | stat_data->kind = KVM_STAT_VCPU; | |
004d62eb | 1140 | kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data; |
bc9e9e67 | 1141 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), |
09cbcef6 MP |
1142 | kvm->debugfs_dentry, stat_data, |
1143 | &stat_fops_per_vm); | |
536a6f88 | 1144 | } |
3165af73 | 1145 | |
284851ee | 1146 | kvm_arch_create_vm_debugfs(kvm); |
536a6f88 | 1147 | return 0; |
b74ed7a6 OU |
1148 | out_err: |
1149 | kvm_destroy_vm_debugfs(kvm); | |
1150 | return ret; | |
536a6f88 JF |
1151 | } |
1152 | ||
1aa9b957 JS |
1153 | /* |
1154 | * Called after the VM is otherwise initialized, but just before adding it to | |
1155 | * the vm_list. | |
1156 | */ | |
1157 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
1158 | { | |
1159 | return 0; | |
1160 | } | |
1161 | ||
1162 | /* | |
1163 | * Called just after removing the VM from the vm_list, but before doing any | |
1164 | * other destruction. | |
1165 | */ | |
1166 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
1167 | { | |
1168 | } | |
1169 | ||
3165af73 PX |
1170 | /* |
1171 | * Called after per-vm debugfs created. When called kvm->debugfs_dentry should | |
1172 | * be setup already, so we can create arch-specific debugfs entries under it. | |
1173 | * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so | |
1174 | * a per-arch destroy interface is not needed. | |
1175 | */ | |
284851ee | 1176 | void __weak kvm_arch_create_vm_debugfs(struct kvm *kvm) |
3165af73 | 1177 | { |
3165af73 PX |
1178 | } |
1179 | ||
b74ed7a6 | 1180 | static struct kvm *kvm_create_vm(unsigned long type, const char *fdname) |
6aa8b732 | 1181 | { |
d89f5eff | 1182 | struct kvm *kvm = kvm_arch_alloc_vm(); |
a54d8066 | 1183 | struct kvm_memslots *slots; |
9121923c | 1184 | int r = -ENOMEM; |
a54d8066 | 1185 | int i, j; |
6aa8b732 | 1186 | |
d89f5eff JK |
1187 | if (!kvm) |
1188 | return ERR_PTR(-ENOMEM); | |
1189 | ||
531810ca | 1190 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 1191 | mmgrab(current->mm); |
e9ad4ec8 PB |
1192 | kvm->mm = current->mm; |
1193 | kvm_eventfd_init(kvm); | |
1194 | mutex_init(&kvm->lock); | |
1195 | mutex_init(&kvm->irq_lock); | |
1196 | mutex_init(&kvm->slots_lock); | |
b10a038e | 1197 | mutex_init(&kvm->slots_arch_lock); |
52ac8b35 PB |
1198 | spin_lock_init(&kvm->mn_invalidate_lock); |
1199 | rcuwait_init(&kvm->mn_memslots_update_rcuwait); | |
c5b07754 | 1200 | xa_init(&kvm->vcpu_array); |
5a475554 CP |
1201 | #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES |
1202 | xa_init(&kvm->mem_attr_array); | |
1203 | #endif | |
52ac8b35 | 1204 | |
982ed0de DW |
1205 | INIT_LIST_HEAD(&kvm->gpc_list); |
1206 | spin_lock_init(&kvm->gpc_lock); | |
52ac8b35 | 1207 | |
e9ad4ec8 | 1208 | INIT_LIST_HEAD(&kvm->devices); |
f502cc56 | 1209 | kvm->max_vcpus = KVM_MAX_VCPUS; |
e9ad4ec8 | 1210 | |
1e702d9a AW |
1211 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
1212 | ||
5c697c36 SC |
1213 | /* |
1214 | * Force subsequent debugfs file creations to fail if the VM directory | |
1215 | * is not created (by kvm_create_vm_debugfs()). | |
1216 | */ | |
1217 | kvm->debugfs_dentry = ERR_PTR(-ENOENT); | |
1218 | ||
f2759c08 OU |
1219 | snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d", |
1220 | task_pid_nr(current)); | |
1221 | ||
8a44119a PB |
1222 | if (init_srcu_struct(&kvm->srcu)) |
1223 | goto out_err_no_srcu; | |
1224 | if (init_srcu_struct(&kvm->irq_srcu)) | |
1225 | goto out_err_no_irq_srcu; | |
1226 | ||
e2d3fcaf | 1227 | refcount_set(&kvm->users_count, 1); |
eed52e43 | 1228 | for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { |
a54d8066 MS |
1229 | for (j = 0; j < 2; j++) { |
1230 | slots = &kvm->__memslots[i][j]; | |
9121923c | 1231 | |
a54d8066 MS |
1232 | atomic_long_set(&slots->last_used_slot, (unsigned long)NULL); |
1233 | slots->hva_tree = RB_ROOT_CACHED; | |
1234 | slots->gfn_tree = RB_ROOT; | |
1235 | hash_init(slots->id_hash); | |
1236 | slots->node_idx = j; | |
1237 | ||
1238 | /* Generations must be different for each address space. */ | |
1239 | slots->generation = i; | |
1240 | } | |
1241 | ||
1242 | rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]); | |
f481b069 | 1243 | } |
00f034a1 | 1244 | |
e93f8a0f | 1245 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 1246 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 1247 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 1248 | if (!kvm->buses[i]) |
a97b0e77 | 1249 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 1250 | } |
e930bffe | 1251 | |
e08b9637 | 1252 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 1253 | if (r) |
a97b0e77 | 1254 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
1255 | |
1256 | r = hardware_enable_all(); | |
1257 | if (r) | |
719d93cd | 1258 | goto out_err_no_disable; |
10474ae8 | 1259 | |
c5b31cc2 | 1260 | #ifdef CONFIG_HAVE_KVM_IRQCHIP |
136bdfee | 1261 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 1262 | #endif |
6aa8b732 | 1263 | |
74b5c5bf | 1264 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
1265 | if (r) |
1266 | goto out_err_no_mmu_notifier; | |
1267 | ||
c2b82397 SC |
1268 | r = kvm_coalesced_mmio_init(kvm); |
1269 | if (r < 0) | |
1270 | goto out_no_coalesced_mmio; | |
1271 | ||
4ba4f419 SC |
1272 | r = kvm_create_vm_debugfs(kvm, fdname); |
1273 | if (r) | |
1274 | goto out_err_no_debugfs; | |
1275 | ||
1aa9b957 | 1276 | r = kvm_arch_post_init_vm(kvm); |
74b5c5bf | 1277 | if (r) |
4ba4f419 | 1278 | goto out_err; |
74b5c5bf | 1279 | |
0d9ce162 | 1280 | mutex_lock(&kvm_lock); |
5e58cfe4 | 1281 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 1282 | mutex_unlock(&kvm_lock); |
d89f5eff | 1283 | |
2ecd9d29 | 1284 | preempt_notifier_inc(); |
2fdef3a2 | 1285 | kvm_init_pm_notifier(kvm); |
2ecd9d29 | 1286 | |
f17abe9a | 1287 | return kvm; |
10474ae8 AG |
1288 | |
1289 | out_err: | |
4ba4f419 SC |
1290 | kvm_destroy_vm_debugfs(kvm); |
1291 | out_err_no_debugfs: | |
c2b82397 SC |
1292 | kvm_coalesced_mmio_free(kvm); |
1293 | out_no_coalesced_mmio: | |
f128cf8c | 1294 | #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER |
1aa9b957 JS |
1295 | if (kvm->mmu_notifier.ops) |
1296 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
1297 | #endif | |
1298 | out_err_no_mmu_notifier: | |
10474ae8 | 1299 | hardware_disable_all(); |
719d93cd | 1300 | out_err_no_disable: |
a97b0e77 | 1301 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 1302 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 1303 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 1304 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 1305 | kfree(kvm_get_bus(kvm, i)); |
8a44119a PB |
1306 | cleanup_srcu_struct(&kvm->irq_srcu); |
1307 | out_err_no_irq_srcu: | |
1308 | cleanup_srcu_struct(&kvm->srcu); | |
1309 | out_err_no_srcu: | |
d89f5eff | 1310 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 1311 | mmdrop(current->mm); |
10474ae8 | 1312 | return ERR_PTR(r); |
f17abe9a AK |
1313 | } |
1314 | ||
07f0a7bd SW |
1315 | static void kvm_destroy_devices(struct kvm *kvm) |
1316 | { | |
e6e3b5a6 | 1317 | struct kvm_device *dev, *tmp; |
07f0a7bd | 1318 | |
a28ebea2 CD |
1319 | /* |
1320 | * We do not need to take the kvm->lock here, because nobody else | |
1321 | * has a reference to the struct kvm at this point and therefore | |
1322 | * cannot access the devices list anyhow. | |
1323 | */ | |
e6e3b5a6 GT |
1324 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
1325 | list_del(&dev->vm_node); | |
07f0a7bd SW |
1326 | dev->ops->destroy(dev); |
1327 | } | |
1328 | } | |
1329 | ||
f17abe9a AK |
1330 | static void kvm_destroy_vm(struct kvm *kvm) |
1331 | { | |
e93f8a0f | 1332 | int i; |
6d4e4c4f AK |
1333 | struct mm_struct *mm = kvm->mm; |
1334 | ||
2fdef3a2 | 1335 | kvm_destroy_pm_notifier(kvm); |
286de8f6 | 1336 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 1337 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 1338 | kvm_arch_sync_events(kvm); |
0d9ce162 | 1339 | mutex_lock(&kvm_lock); |
133de902 | 1340 | list_del(&kvm->vm_list); |
0d9ce162 | 1341 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
1342 | kvm_arch_pre_destroy_vm(kvm); |
1343 | ||
399ec807 | 1344 | kvm_free_irq_routing(kvm); |
df630b8c | 1345 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 1346 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 1347 | |
4a12f951 CB |
1348 | if (bus) |
1349 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
1350 | kvm->buses[i] = NULL; |
1351 | } | |
980da6ce | 1352 | kvm_coalesced_mmio_free(kvm); |
f128cf8c | 1353 | #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER |
e930bffe | 1354 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); |
52ac8b35 PB |
1355 | /* |
1356 | * At this point, pending calls to invalidate_range_start() | |
1357 | * have completed but no more MMU notifiers will run, so | |
1358 | * mn_active_invalidate_count may remain unbalanced. | |
b0d23708 | 1359 | * No threads can be waiting in kvm_swap_active_memslots() as the |
52ac8b35 PB |
1360 | * last reference on KVM has been dropped, but freeing |
1361 | * memslots would deadlock without this manual intervention. | |
d497a0fa SC |
1362 | * |
1363 | * If the count isn't unbalanced, i.e. KVM did NOT unregister its MMU | |
1364 | * notifier between a start() and end(), then there shouldn't be any | |
1365 | * in-progress invalidations. | |
52ac8b35 PB |
1366 | */ |
1367 | WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait)); | |
d497a0fa SC |
1368 | if (kvm->mn_active_invalidate_count) |
1369 | kvm->mn_active_invalidate_count = 0; | |
1370 | else | |
1371 | WARN_ON(kvm->mmu_invalidate_in_progress); | |
f00be0ca | 1372 | #else |
683412cc | 1373 | kvm_flush_shadow_all(kvm); |
5f94c174 | 1374 | #endif |
d19a9cd2 | 1375 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 1376 | kvm_destroy_devices(kvm); |
eed52e43 | 1377 | for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { |
a54d8066 MS |
1378 | kvm_free_memslots(kvm, &kvm->__memslots[i][0]); |
1379 | kvm_free_memslots(kvm, &kvm->__memslots[i][1]); | |
1380 | } | |
820b3fcd | 1381 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff | 1382 | cleanup_srcu_struct(&kvm->srcu); |
5a475554 CP |
1383 | #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES |
1384 | xa_destroy(&kvm->mem_attr_array); | |
1385 | #endif | |
d89f5eff | 1386 | kvm_arch_free_vm(kvm); |
2ecd9d29 | 1387 | preempt_notifier_dec(); |
10474ae8 | 1388 | hardware_disable_all(); |
6d4e4c4f | 1389 | mmdrop(mm); |
f17abe9a AK |
1390 | } |
1391 | ||
d39f13b0 IE |
1392 | void kvm_get_kvm(struct kvm *kvm) |
1393 | { | |
e3736c3e | 1394 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
1395 | } |
1396 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
1397 | ||
605c7130 PX |
1398 | /* |
1399 | * Make sure the vm is not during destruction, which is a safe version of | |
1400 | * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise. | |
1401 | */ | |
1402 | bool kvm_get_kvm_safe(struct kvm *kvm) | |
1403 | { | |
1404 | return refcount_inc_not_zero(&kvm->users_count); | |
1405 | } | |
1406 | EXPORT_SYMBOL_GPL(kvm_get_kvm_safe); | |
1407 | ||
d39f13b0 IE |
1408 | void kvm_put_kvm(struct kvm *kvm) |
1409 | { | |
e3736c3e | 1410 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
1411 | kvm_destroy_vm(kvm); |
1412 | } | |
1413 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
1414 | ||
149487bd SC |
1415 | /* |
1416 | * Used to put a reference that was taken on behalf of an object associated | |
1417 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
1418 | * of the new file descriptor fails and the reference cannot be transferred to | |
1419 | * its final owner. In such cases, the caller is still actively using @kvm and | |
1420 | * will fail miserably if the refcount unexpectedly hits zero. | |
1421 | */ | |
1422 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
1423 | { | |
1424 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
1425 | } | |
1426 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 1427 | |
f17abe9a AK |
1428 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
1429 | { | |
1430 | struct kvm *kvm = filp->private_data; | |
1431 | ||
721eecbf GH |
1432 | kvm_irqfd_release(kvm); |
1433 | ||
d39f13b0 | 1434 | kvm_put_kvm(kvm); |
6aa8b732 AK |
1435 | return 0; |
1436 | } | |
1437 | ||
515a0127 TY |
1438 | /* |
1439 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 1440 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 1441 | */ |
3c9bd400 | 1442 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 1443 | { |
37b2a651 | 1444 | unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 1445 | |
37b2a651 | 1446 | memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
1447 | if (!memslot->dirty_bitmap) |
1448 | return -ENOMEM; | |
1449 | ||
a36a57b1 TY |
1450 | return 0; |
1451 | } | |
1452 | ||
a54d8066 | 1453 | static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id) |
bf3e05bc | 1454 | { |
a54d8066 MS |
1455 | struct kvm_memslots *active = __kvm_memslots(kvm, as_id); |
1456 | int node_idx_inactive = active->node_idx ^ 1; | |
0e60b079 | 1457 | |
a54d8066 | 1458 | return &kvm->__memslots[as_id][node_idx_inactive]; |
0577d1ab SC |
1459 | } |
1460 | ||
1461 | /* | |
a54d8066 MS |
1462 | * Helper to get the address space ID when one of memslot pointers may be NULL. |
1463 | * This also serves as a sanity that at least one of the pointers is non-NULL, | |
1464 | * and that their address space IDs don't diverge. | |
0577d1ab | 1465 | */ |
a54d8066 MS |
1466 | static int kvm_memslots_get_as_id(struct kvm_memory_slot *a, |
1467 | struct kvm_memory_slot *b) | |
0577d1ab | 1468 | { |
a54d8066 MS |
1469 | if (WARN_ON_ONCE(!a && !b)) |
1470 | return 0; | |
0577d1ab | 1471 | |
a54d8066 MS |
1472 | if (!a) |
1473 | return b->as_id; | |
1474 | if (!b) | |
1475 | return a->as_id; | |
0577d1ab | 1476 | |
a54d8066 MS |
1477 | WARN_ON_ONCE(a->as_id != b->as_id); |
1478 | return a->as_id; | |
0577d1ab | 1479 | } |
efbeec70 | 1480 | |
a54d8066 MS |
1481 | static void kvm_insert_gfn_node(struct kvm_memslots *slots, |
1482 | struct kvm_memory_slot *slot) | |
0577d1ab | 1483 | { |
a54d8066 MS |
1484 | struct rb_root *gfn_tree = &slots->gfn_tree; |
1485 | struct rb_node **node, *parent; | |
1486 | int idx = slots->node_idx; | |
0577d1ab | 1487 | |
a54d8066 MS |
1488 | parent = NULL; |
1489 | for (node = &gfn_tree->rb_node; *node; ) { | |
1490 | struct kvm_memory_slot *tmp; | |
f85e2cb5 | 1491 | |
a54d8066 MS |
1492 | tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]); |
1493 | parent = *node; | |
1494 | if (slot->base_gfn < tmp->base_gfn) | |
1495 | node = &(*node)->rb_left; | |
1496 | else if (slot->base_gfn > tmp->base_gfn) | |
1497 | node = &(*node)->rb_right; | |
1498 | else | |
1499 | BUG(); | |
0577d1ab | 1500 | } |
a54d8066 MS |
1501 | |
1502 | rb_link_node(&slot->gfn_node[idx], parent, node); | |
1503 | rb_insert_color(&slot->gfn_node[idx], gfn_tree); | |
0577d1ab SC |
1504 | } |
1505 | ||
a54d8066 MS |
1506 | static void kvm_erase_gfn_node(struct kvm_memslots *slots, |
1507 | struct kvm_memory_slot *slot) | |
0577d1ab | 1508 | { |
a54d8066 MS |
1509 | rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree); |
1510 | } | |
0577d1ab | 1511 | |
a54d8066 MS |
1512 | static void kvm_replace_gfn_node(struct kvm_memslots *slots, |
1513 | struct kvm_memory_slot *old, | |
1514 | struct kvm_memory_slot *new) | |
1515 | { | |
1516 | int idx = slots->node_idx; | |
0577d1ab | 1517 | |
a54d8066 | 1518 | WARN_ON_ONCE(old->base_gfn != new->base_gfn); |
0577d1ab | 1519 | |
a54d8066 MS |
1520 | rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx], |
1521 | &slots->gfn_tree); | |
0577d1ab SC |
1522 | } |
1523 | ||
1524 | /* | |
a54d8066 | 1525 | * Replace @old with @new in the inactive memslots. |
0577d1ab | 1526 | * |
a54d8066 MS |
1527 | * With NULL @old this simply adds @new. |
1528 | * With NULL @new this simply removes @old. | |
0577d1ab | 1529 | * |
a54d8066 MS |
1530 | * If @new is non-NULL its hva_node[slots_idx] range has to be set |
1531 | * appropriately. | |
0577d1ab | 1532 | */ |
a54d8066 MS |
1533 | static void kvm_replace_memslot(struct kvm *kvm, |
1534 | struct kvm_memory_slot *old, | |
1535 | struct kvm_memory_slot *new) | |
0577d1ab | 1536 | { |
a54d8066 MS |
1537 | int as_id = kvm_memslots_get_as_id(old, new); |
1538 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); | |
1539 | int idx = slots->node_idx; | |
0577d1ab | 1540 | |
a54d8066 MS |
1541 | if (old) { |
1542 | hash_del(&old->id_node[idx]); | |
1543 | interval_tree_remove(&old->hva_node[idx], &slots->hva_tree); | |
0577d1ab | 1544 | |
a54d8066 MS |
1545 | if ((long)old == atomic_long_read(&slots->last_used_slot)) |
1546 | atomic_long_set(&slots->last_used_slot, (long)new); | |
0577d1ab | 1547 | |
a54d8066 MS |
1548 | if (!new) { |
1549 | kvm_erase_gfn_node(slots, old); | |
1e8617d3 | 1550 | return; |
a54d8066 MS |
1551 | } |
1552 | } | |
1e8617d3 | 1553 | |
a54d8066 MS |
1554 | /* |
1555 | * Initialize @new's hva range. Do this even when replacing an @old | |
1556 | * slot, kvm_copy_memslot() deliberately does not touch node data. | |
1557 | */ | |
1558 | new->hva_node[idx].start = new->userspace_addr; | |
1559 | new->hva_node[idx].last = new->userspace_addr + | |
1560 | (new->npages << PAGE_SHIFT) - 1; | |
1561 | ||
1562 | /* | |
1563 | * (Re)Add the new memslot. There is no O(1) interval_tree_replace(), | |
1564 | * hva_node needs to be swapped with remove+insert even though hva can't | |
1565 | * change when replacing an existing slot. | |
1566 | */ | |
1567 | hash_add(slots->id_hash, &new->id_node[idx], new->id); | |
1568 | interval_tree_insert(&new->hva_node[idx], &slots->hva_tree); | |
1569 | ||
1570 | /* | |
1571 | * If the memslot gfn is unchanged, rb_replace_node() can be used to | |
1572 | * switch the node in the gfn tree instead of removing the old and | |
1573 | * inserting the new as two separate operations. Replacement is a | |
1574 | * single O(1) operation versus two O(log(n)) operations for | |
1575 | * remove+insert. | |
1576 | */ | |
1577 | if (old && old->base_gfn == new->base_gfn) { | |
1578 | kvm_replace_gfn_node(slots, old, new); | |
1579 | } else { | |
1580 | if (old) | |
1581 | kvm_erase_gfn_node(slots, old); | |
1582 | kvm_insert_gfn_node(slots, new); | |
0577d1ab | 1583 | } |
bf3e05bc XG |
1584 | } |
1585 | ||
bb58b90b SC |
1586 | /* |
1587 | * Flags that do not access any of the extra space of struct | |
1588 | * kvm_userspace_memory_region2. KVM_SET_USER_MEMORY_REGION_V1_FLAGS | |
1589 | * only allows these. | |
1590 | */ | |
1591 | #define KVM_SET_USER_MEMORY_REGION_V1_FLAGS \ | |
1592 | (KVM_MEM_LOG_DIRTY_PAGES | KVM_MEM_READONLY) | |
1593 | ||
a7800aa8 SC |
1594 | static int check_memory_region_flags(struct kvm *kvm, |
1595 | const struct kvm_userspace_memory_region2 *mem) | |
a50d64d6 | 1596 | { |
4d8b81ab XG |
1597 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1598 | ||
a7800aa8 SC |
1599 | if (kvm_arch_has_private_mem(kvm)) |
1600 | valid_flags |= KVM_MEM_GUEST_MEMFD; | |
1601 | ||
1602 | /* Dirty logging private memory is not currently supported. */ | |
1603 | if (mem->flags & KVM_MEM_GUEST_MEMFD) | |
1604 | valid_flags &= ~KVM_MEM_LOG_DIRTY_PAGES; | |
1605 | ||
8886640d | 1606 | #ifdef CONFIG_HAVE_KVM_READONLY_MEM |
e5635922 SC |
1607 | /* |
1608 | * GUEST_MEMFD is incompatible with read-only memslots, as writes to | |
1609 | * read-only memslots have emulated MMIO, not page fault, semantics, | |
1610 | * and KVM doesn't allow emulated MMIO for private memory. | |
1611 | */ | |
1612 | if (!(mem->flags & KVM_MEM_GUEST_MEMFD)) | |
1613 | valid_flags |= KVM_MEM_READONLY; | |
4d8b81ab XG |
1614 | #endif |
1615 | ||
1616 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1617 | return -EINVAL; |
1618 | ||
1619 | return 0; | |
1620 | } | |
1621 | ||
a54d8066 | 1622 | static void kvm_swap_active_memslots(struct kvm *kvm, int as_id) |
7ec4fb44 | 1623 | { |
a54d8066 MS |
1624 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); |
1625 | ||
1626 | /* Grab the generation from the activate memslots. */ | |
1627 | u64 gen = __kvm_memslots(kvm, as_id)->generation; | |
7ec4fb44 | 1628 | |
361209e0 SC |
1629 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1630 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1631 | |
52ac8b35 PB |
1632 | /* |
1633 | * Do not store the new memslots while there are invalidations in | |
071064f1 PB |
1634 | * progress, otherwise the locking in invalidate_range_start and |
1635 | * invalidate_range_end will be unbalanced. | |
52ac8b35 PB |
1636 | */ |
1637 | spin_lock(&kvm->mn_invalidate_lock); | |
1638 | prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
1639 | while (kvm->mn_active_invalidate_count) { | |
1640 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1641 | spin_unlock(&kvm->mn_invalidate_lock); | |
1642 | schedule(); | |
1643 | spin_lock(&kvm->mn_invalidate_lock); | |
1644 | } | |
1645 | finish_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
f481b069 | 1646 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
52ac8b35 | 1647 | spin_unlock(&kvm->mn_invalidate_lock); |
b10a038e BG |
1648 | |
1649 | /* | |
1650 | * Acquired in kvm_set_memslot. Must be released before synchronize | |
1651 | * SRCU below in order to avoid deadlock with another thread | |
1652 | * acquiring the slots_arch_lock in an srcu critical section. | |
1653 | */ | |
1654 | mutex_unlock(&kvm->slots_arch_lock); | |
1655 | ||
7ec4fb44 | 1656 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1657 | |
ee3d1570 | 1658 | /* |
361209e0 | 1659 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1660 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1661 | * the number of address spaces. This provides a unique and easily |
1662 | * identifiable generation number while the memslots are in flux. | |
1663 | */ | |
1664 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1665 | ||
1666 | /* | |
4bd518f1 PB |
1667 | * Generations must be unique even across address spaces. We do not need |
1668 | * a global counter for that, instead the generation space is evenly split | |
1669 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1670 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1671 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1672 | */ |
eed52e43 | 1673 | gen += kvm_arch_nr_memslot_as_ids(kvm); |
ee3d1570 | 1674 | |
15248258 | 1675 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1676 | |
15248258 | 1677 | slots->generation = gen; |
7ec4fb44 GN |
1678 | } |
1679 | ||
07921665 SC |
1680 | static int kvm_prepare_memory_region(struct kvm *kvm, |
1681 | const struct kvm_memory_slot *old, | |
1682 | struct kvm_memory_slot *new, | |
1683 | enum kvm_mr_change change) | |
ddc12f2a | 1684 | { |
07921665 SC |
1685 | int r; |
1686 | ||
1687 | /* | |
1688 | * If dirty logging is disabled, nullify the bitmap; the old bitmap | |
1689 | * will be freed on "commit". If logging is enabled in both old and | |
1690 | * new, reuse the existing bitmap. If logging is enabled only in the | |
1691 | * new and KVM isn't using a ring buffer, allocate and initialize a | |
1692 | * new bitmap. | |
1693 | */ | |
244893fa SC |
1694 | if (change != KVM_MR_DELETE) { |
1695 | if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1696 | new->dirty_bitmap = NULL; | |
1697 | else if (old && old->dirty_bitmap) | |
1698 | new->dirty_bitmap = old->dirty_bitmap; | |
86bdf3eb | 1699 | else if (kvm_use_dirty_bitmap(kvm)) { |
244893fa SC |
1700 | r = kvm_alloc_dirty_bitmap(new); |
1701 | if (r) | |
1702 | return r; | |
1703 | ||
1704 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1705 | bitmap_set(new->dirty_bitmap, 0, new->npages); | |
1706 | } | |
07921665 SC |
1707 | } |
1708 | ||
1709 | r = kvm_arch_prepare_memory_region(kvm, old, new, change); | |
1710 | ||
1711 | /* Free the bitmap on failure if it was allocated above. */ | |
c87661f8 | 1712 | if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap)) |
07921665 SC |
1713 | kvm_destroy_dirty_bitmap(new); |
1714 | ||
1715 | return r; | |
ddc12f2a BG |
1716 | } |
1717 | ||
07921665 SC |
1718 | static void kvm_commit_memory_region(struct kvm *kvm, |
1719 | struct kvm_memory_slot *old, | |
1720 | const struct kvm_memory_slot *new, | |
1721 | enum kvm_mr_change change) | |
ddc12f2a | 1722 | { |
6c7b2202 PB |
1723 | int old_flags = old ? old->flags : 0; |
1724 | int new_flags = new ? new->flags : 0; | |
07921665 SC |
1725 | /* |
1726 | * Update the total number of memslot pages before calling the arch | |
1727 | * hook so that architectures can consume the result directly. | |
1728 | */ | |
1729 | if (change == KVM_MR_DELETE) | |
1730 | kvm->nr_memslot_pages -= old->npages; | |
1731 | else if (change == KVM_MR_CREATE) | |
1732 | kvm->nr_memslot_pages += new->npages; | |
1733 | ||
6c7b2202 PB |
1734 | if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES) { |
1735 | int change = (new_flags & KVM_MEM_LOG_DIRTY_PAGES) ? 1 : -1; | |
1736 | atomic_set(&kvm->nr_memslots_dirty_logging, | |
1737 | atomic_read(&kvm->nr_memslots_dirty_logging) + change); | |
1738 | } | |
1739 | ||
07921665 SC |
1740 | kvm_arch_commit_memory_region(kvm, old, new, change); |
1741 | ||
a54d8066 MS |
1742 | switch (change) { |
1743 | case KVM_MR_CREATE: | |
1744 | /* Nothing more to do. */ | |
1745 | break; | |
1746 | case KVM_MR_DELETE: | |
1747 | /* Free the old memslot and all its metadata. */ | |
1748 | kvm_free_memslot(kvm, old); | |
1749 | break; | |
1750 | case KVM_MR_MOVE: | |
1751 | case KVM_MR_FLAGS_ONLY: | |
1752 | /* | |
1753 | * Free the dirty bitmap as needed; the below check encompasses | |
1754 | * both the flags and whether a ring buffer is being used) | |
1755 | */ | |
1756 | if (old->dirty_bitmap && !new->dirty_bitmap) | |
1757 | kvm_destroy_dirty_bitmap(old); | |
1758 | ||
1759 | /* | |
1760 | * The final quirk. Free the detached, old slot, but only its | |
1761 | * memory, not any metadata. Metadata, including arch specific | |
1762 | * data, may be reused by @new. | |
1763 | */ | |
1764 | kfree(old); | |
1765 | break; | |
1766 | default: | |
1767 | BUG(); | |
1768 | } | |
ddc12f2a BG |
1769 | } |
1770 | ||
36947254 | 1771 | /* |
a54d8066 MS |
1772 | * Activate @new, which must be installed in the inactive slots by the caller, |
1773 | * by swapping the active slots and then propagating @new to @old once @old is | |
1774 | * unreachable and can be safely modified. | |
1775 | * | |
1776 | * With NULL @old this simply adds @new to @active (while swapping the sets). | |
1777 | * With NULL @new this simply removes @old from @active and frees it | |
1778 | * (while also swapping the sets). | |
36947254 | 1779 | */ |
a54d8066 MS |
1780 | static void kvm_activate_memslot(struct kvm *kvm, |
1781 | struct kvm_memory_slot *old, | |
1782 | struct kvm_memory_slot *new) | |
36947254 | 1783 | { |
a54d8066 | 1784 | int as_id = kvm_memslots_get_as_id(old, new); |
36947254 | 1785 | |
a54d8066 MS |
1786 | kvm_swap_active_memslots(kvm, as_id); |
1787 | ||
1788 | /* Propagate the new memslot to the now inactive memslots. */ | |
1789 | kvm_replace_memslot(kvm, old, new); | |
1790 | } | |
1791 | ||
1792 | static void kvm_copy_memslot(struct kvm_memory_slot *dest, | |
1793 | const struct kvm_memory_slot *src) | |
1794 | { | |
1795 | dest->base_gfn = src->base_gfn; | |
1796 | dest->npages = src->npages; | |
1797 | dest->dirty_bitmap = src->dirty_bitmap; | |
1798 | dest->arch = src->arch; | |
1799 | dest->userspace_addr = src->userspace_addr; | |
1800 | dest->flags = src->flags; | |
1801 | dest->id = src->id; | |
1802 | dest->as_id = src->as_id; | |
1803 | } | |
1804 | ||
1805 | static void kvm_invalidate_memslot(struct kvm *kvm, | |
1806 | struct kvm_memory_slot *old, | |
244893fa | 1807 | struct kvm_memory_slot *invalid_slot) |
a54d8066 | 1808 | { |
07921665 | 1809 | /* |
a54d8066 MS |
1810 | * Mark the current slot INVALID. As with all memslot modifications, |
1811 | * this must be done on an unreachable slot to avoid modifying the | |
1812 | * current slot in the active tree. | |
07921665 | 1813 | */ |
244893fa SC |
1814 | kvm_copy_memslot(invalid_slot, old); |
1815 | invalid_slot->flags |= KVM_MEMSLOT_INVALID; | |
1816 | kvm_replace_memslot(kvm, old, invalid_slot); | |
a54d8066 MS |
1817 | |
1818 | /* | |
1819 | * Activate the slot that is now marked INVALID, but don't propagate | |
1820 | * the slot to the now inactive slots. The slot is either going to be | |
1821 | * deleted or recreated as a new slot. | |
1822 | */ | |
1823 | kvm_swap_active_memslots(kvm, old->as_id); | |
1824 | ||
1825 | /* | |
1826 | * From this point no new shadow pages pointing to a deleted, or moved, | |
1827 | * memslot will be created. Validation of sp->gfn happens in: | |
1828 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1829 | * - kvm_is_visible_gfn (mmu_check_root) | |
1830 | */ | |
bcb63dcd | 1831 | kvm_arch_flush_shadow_memslot(kvm, old); |
683412cc | 1832 | kvm_arch_guest_memory_reclaimed(kvm); |
a54d8066 | 1833 | |
b0d23708 | 1834 | /* Was released by kvm_swap_active_memslots(), reacquire. */ |
a54d8066 MS |
1835 | mutex_lock(&kvm->slots_arch_lock); |
1836 | ||
1837 | /* | |
1838 | * Copy the arch-specific field of the newly-installed slot back to the | |
1839 | * old slot as the arch data could have changed between releasing | |
b0d23708 | 1840 | * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock |
a54d8066 MS |
1841 | * above. Writers are required to retrieve memslots *after* acquiring |
1842 | * slots_arch_lock, thus the active slot's data is guaranteed to be fresh. | |
1843 | */ | |
244893fa | 1844 | old->arch = invalid_slot->arch; |
a54d8066 MS |
1845 | } |
1846 | ||
1847 | static void kvm_create_memslot(struct kvm *kvm, | |
244893fa | 1848 | struct kvm_memory_slot *new) |
a54d8066 | 1849 | { |
244893fa SC |
1850 | /* Add the new memslot to the inactive set and activate. */ |
1851 | kvm_replace_memslot(kvm, NULL, new); | |
1852 | kvm_activate_memslot(kvm, NULL, new); | |
a54d8066 MS |
1853 | } |
1854 | ||
1855 | static void kvm_delete_memslot(struct kvm *kvm, | |
1856 | struct kvm_memory_slot *old, | |
1857 | struct kvm_memory_slot *invalid_slot) | |
1858 | { | |
1859 | /* | |
1860 | * Remove the old memslot (in the inactive memslots) by passing NULL as | |
244893fa | 1861 | * the "new" slot, and for the invalid version in the active slots. |
a54d8066 MS |
1862 | */ |
1863 | kvm_replace_memslot(kvm, old, NULL); | |
a54d8066 | 1864 | kvm_activate_memslot(kvm, invalid_slot, NULL); |
a54d8066 | 1865 | } |
36947254 | 1866 | |
244893fa SC |
1867 | static void kvm_move_memslot(struct kvm *kvm, |
1868 | struct kvm_memory_slot *old, | |
1869 | struct kvm_memory_slot *new, | |
1870 | struct kvm_memory_slot *invalid_slot) | |
a54d8066 | 1871 | { |
a54d8066 | 1872 | /* |
244893fa SC |
1873 | * Replace the old memslot in the inactive slots, and then swap slots |
1874 | * and replace the current INVALID with the new as well. | |
a54d8066 | 1875 | */ |
244893fa SC |
1876 | kvm_replace_memslot(kvm, old, new); |
1877 | kvm_activate_memslot(kvm, invalid_slot, new); | |
a54d8066 | 1878 | } |
36947254 | 1879 | |
a54d8066 MS |
1880 | static void kvm_update_flags_memslot(struct kvm *kvm, |
1881 | struct kvm_memory_slot *old, | |
244893fa | 1882 | struct kvm_memory_slot *new) |
a54d8066 MS |
1883 | { |
1884 | /* | |
1885 | * Similar to the MOVE case, but the slot doesn't need to be zapped as | |
1886 | * an intermediate step. Instead, the old memslot is simply replaced | |
1887 | * with a new, updated copy in both memslot sets. | |
1888 | */ | |
244893fa SC |
1889 | kvm_replace_memslot(kvm, old, new); |
1890 | kvm_activate_memslot(kvm, old, new); | |
36947254 SC |
1891 | } |
1892 | ||
cf47f50b | 1893 | static int kvm_set_memslot(struct kvm *kvm, |
a54d8066 | 1894 | struct kvm_memory_slot *old, |
ce5f0215 | 1895 | struct kvm_memory_slot *new, |
cf47f50b SC |
1896 | enum kvm_mr_change change) |
1897 | { | |
244893fa | 1898 | struct kvm_memory_slot *invalid_slot; |
cf47f50b SC |
1899 | int r; |
1900 | ||
b10a038e | 1901 | /* |
b0d23708 | 1902 | * Released in kvm_swap_active_memslots(). |
b10a038e | 1903 | * |
b0d23708 JM |
1904 | * Must be held from before the current memslots are copied until after |
1905 | * the new memslots are installed with rcu_assign_pointer, then | |
1906 | * released before the synchronize srcu in kvm_swap_active_memslots(). | |
b10a038e BG |
1907 | * |
1908 | * When modifying memslots outside of the slots_lock, must be held | |
1909 | * before reading the pointer to the current memslots until after all | |
1910 | * changes to those memslots are complete. | |
1911 | * | |
1912 | * These rules ensure that installing new memslots does not lose | |
1913 | * changes made to the previous memslots. | |
1914 | */ | |
1915 | mutex_lock(&kvm->slots_arch_lock); | |
1916 | ||
a54d8066 MS |
1917 | /* |
1918 | * Invalidate the old slot if it's being deleted or moved. This is | |
1919 | * done prior to actually deleting/moving the memslot to allow vCPUs to | |
1920 | * continue running by ensuring there are no mappings or shadow pages | |
1921 | * for the memslot when it is deleted/moved. Without pre-invalidation | |
1922 | * (and without a lock), a window would exist between effecting the | |
1923 | * delete/move and committing the changes in arch code where KVM or a | |
1924 | * guest could access a non-existent memslot. | |
244893fa SC |
1925 | * |
1926 | * Modifications are done on a temporary, unreachable slot. The old | |
1927 | * slot needs to be preserved in case a later step fails and the | |
1928 | * invalidation needs to be reverted. | |
a54d8066 | 1929 | */ |
cf47f50b | 1930 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
244893fa SC |
1931 | invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT); |
1932 | if (!invalid_slot) { | |
1933 | mutex_unlock(&kvm->slots_arch_lock); | |
1934 | return -ENOMEM; | |
1935 | } | |
1936 | kvm_invalidate_memslot(kvm, old, invalid_slot); | |
1937 | } | |
b10a038e | 1938 | |
a54d8066 MS |
1939 | r = kvm_prepare_memory_region(kvm, old, new, change); |
1940 | if (r) { | |
b10a038e | 1941 | /* |
a54d8066 MS |
1942 | * For DELETE/MOVE, revert the above INVALID change. No |
1943 | * modifications required since the original slot was preserved | |
1944 | * in the inactive slots. Changing the active memslots also | |
1945 | * release slots_arch_lock. | |
b10a038e | 1946 | */ |
244893fa SC |
1947 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
1948 | kvm_activate_memslot(kvm, invalid_slot, old); | |
1949 | kfree(invalid_slot); | |
1950 | } else { | |
a54d8066 | 1951 | mutex_unlock(&kvm->slots_arch_lock); |
244893fa | 1952 | } |
a54d8066 | 1953 | return r; |
cf47f50b SC |
1954 | } |
1955 | ||
bda44d84 | 1956 | /* |
a54d8066 MS |
1957 | * For DELETE and MOVE, the working slot is now active as the INVALID |
1958 | * version of the old slot. MOVE is particularly special as it reuses | |
1959 | * the old slot and returns a copy of the old slot (in working_slot). | |
1960 | * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the | |
1961 | * old slot is detached but otherwise preserved. | |
bda44d84 | 1962 | */ |
a54d8066 | 1963 | if (change == KVM_MR_CREATE) |
244893fa | 1964 | kvm_create_memslot(kvm, new); |
a54d8066 | 1965 | else if (change == KVM_MR_DELETE) |
244893fa | 1966 | kvm_delete_memslot(kvm, old, invalid_slot); |
a54d8066 | 1967 | else if (change == KVM_MR_MOVE) |
244893fa | 1968 | kvm_move_memslot(kvm, old, new, invalid_slot); |
a54d8066 | 1969 | else if (change == KVM_MR_FLAGS_ONLY) |
244893fa | 1970 | kvm_update_flags_memslot(kvm, old, new); |
a54d8066 MS |
1971 | else |
1972 | BUG(); | |
cf47f50b | 1973 | |
244893fa SC |
1974 | /* Free the temporary INVALID slot used for DELETE and MOVE. */ |
1975 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) | |
1976 | kfree(invalid_slot); | |
bda44d84 | 1977 | |
a54d8066 MS |
1978 | /* |
1979 | * No need to refresh new->arch, changes after dropping slots_arch_lock | |
a413a625 | 1980 | * will directly hit the final, active memslot. Architectures are |
a54d8066 MS |
1981 | * responsible for knowing that new->arch may be stale. |
1982 | */ | |
1983 | kvm_commit_memory_region(kvm, old, new, change); | |
cf47f50b | 1984 | |
cf47f50b | 1985 | return 0; |
cf47f50b SC |
1986 | } |
1987 | ||
44401a20 MS |
1988 | static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id, |
1989 | gfn_t start, gfn_t end) | |
5c0b4f3d | 1990 | { |
44401a20 | 1991 | struct kvm_memslot_iter iter; |
5c0b4f3d | 1992 | |
44401a20 MS |
1993 | kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) { |
1994 | if (iter.slot->id != id) | |
1995 | return true; | |
1996 | } | |
5c0b4f3d | 1997 | |
44401a20 | 1998 | return false; |
5c0b4f3d SC |
1999 | } |
2000 | ||
6aa8b732 AK |
2001 | /* |
2002 | * Allocate some memory and give it an address in the guest physical address | |
2003 | * space. | |
2004 | * | |
2005 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 2006 | * |
02d5d55b | 2007 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 2008 | */ |
f78e0e2e | 2009 | int __kvm_set_memory_region(struct kvm *kvm, |
bb58b90b | 2010 | const struct kvm_userspace_memory_region2 *mem) |
6aa8b732 | 2011 | { |
244893fa | 2012 | struct kvm_memory_slot *old, *new; |
44401a20 | 2013 | struct kvm_memslots *slots; |
f64c0398 | 2014 | enum kvm_mr_change change; |
0f9bdef3 SC |
2015 | unsigned long npages; |
2016 | gfn_t base_gfn; | |
163da372 SC |
2017 | int as_id, id; |
2018 | int r; | |
6aa8b732 | 2019 | |
a7800aa8 | 2020 | r = check_memory_region_flags(kvm, mem); |
a50d64d6 | 2021 | if (r) |
71a4c30b | 2022 | return r; |
a50d64d6 | 2023 | |
f481b069 PB |
2024 | as_id = mem->slot >> 16; |
2025 | id = (u16)mem->slot; | |
2026 | ||
6aa8b732 | 2027 | /* General sanity checks */ |
6b285a55 SC |
2028 | if ((mem->memory_size & (PAGE_SIZE - 1)) || |
2029 | (mem->memory_size != (unsigned long)mem->memory_size)) | |
71a4c30b | 2030 | return -EINVAL; |
6aa8b732 | 2031 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 2032 | return -EINVAL; |
fa3d315a | 2033 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 2034 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 2035 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 2036 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 2037 | mem->memory_size)) |
71a4c30b | 2038 | return -EINVAL; |
a7800aa8 SC |
2039 | if (mem->flags & KVM_MEM_GUEST_MEMFD && |
2040 | (mem->guest_memfd_offset & (PAGE_SIZE - 1) || | |
2041 | mem->guest_memfd_offset + mem->memory_size < mem->guest_memfd_offset)) | |
2042 | return -EINVAL; | |
eed52e43 | 2043 | if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 2044 | return -EINVAL; |
6aa8b732 | 2045 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 2046 | return -EINVAL; |
0f9bdef3 SC |
2047 | if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES) |
2048 | return -EINVAL; | |
6aa8b732 | 2049 | |
44401a20 | 2050 | slots = __kvm_memslots(kvm, as_id); |
6aa8b732 | 2051 | |
5c0b4f3d | 2052 | /* |
7cd08553 SC |
2053 | * Note, the old memslot (and the pointer itself!) may be invalidated |
2054 | * and/or destroyed by kvm_set_memslot(). | |
5c0b4f3d | 2055 | */ |
44401a20 | 2056 | old = id_to_memslot(slots, id); |
163da372 | 2057 | |
47ea7d90 | 2058 | if (!mem->memory_size) { |
7cd08553 | 2059 | if (!old || !old->npages) |
47ea7d90 | 2060 | return -EINVAL; |
5c0b4f3d | 2061 | |
7cd08553 | 2062 | if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages)) |
47ea7d90 | 2063 | return -EIO; |
6aa8b732 | 2064 | |
244893fa | 2065 | return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE); |
47ea7d90 | 2066 | } |
5c0b4f3d | 2067 | |
0f9bdef3 SC |
2068 | base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT); |
2069 | npages = (mem->memory_size >> PAGE_SHIFT); | |
163da372 | 2070 | |
7cd08553 | 2071 | if (!old || !old->npages) { |
5c0b4f3d | 2072 | change = KVM_MR_CREATE; |
afa319a5 SC |
2073 | |
2074 | /* | |
2075 | * To simplify KVM internals, the total number of pages across | |
2076 | * all memslots must fit in an unsigned long. | |
2077 | */ | |
0f9bdef3 | 2078 | if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages) |
afa319a5 | 2079 | return -EINVAL; |
5c0b4f3d | 2080 | } else { /* Modify an existing slot. */ |
a7800aa8 SC |
2081 | /* Private memslots are immutable, they can only be deleted. */ |
2082 | if (mem->flags & KVM_MEM_GUEST_MEMFD) | |
2083 | return -EINVAL; | |
0f9bdef3 SC |
2084 | if ((mem->userspace_addr != old->userspace_addr) || |
2085 | (npages != old->npages) || | |
2086 | ((mem->flags ^ old->flags) & KVM_MEM_READONLY)) | |
71a4c30b | 2087 | return -EINVAL; |
09170a49 | 2088 | |
0f9bdef3 | 2089 | if (base_gfn != old->base_gfn) |
5c0b4f3d | 2090 | change = KVM_MR_MOVE; |
0f9bdef3 | 2091 | else if (mem->flags != old->flags) |
5c0b4f3d SC |
2092 | change = KVM_MR_FLAGS_ONLY; |
2093 | else /* Nothing to change. */ | |
2094 | return 0; | |
09170a49 | 2095 | } |
6aa8b732 | 2096 | |
44401a20 | 2097 | if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) && |
0f9bdef3 | 2098 | kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages)) |
44401a20 | 2099 | return -EEXIST; |
6aa8b732 | 2100 | |
244893fa SC |
2101 | /* Allocate a slot that will persist in the memslot. */ |
2102 | new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT); | |
2103 | if (!new) | |
2104 | return -ENOMEM; | |
3c9bd400 | 2105 | |
244893fa SC |
2106 | new->as_id = as_id; |
2107 | new->id = id; | |
2108 | new->base_gfn = base_gfn; | |
2109 | new->npages = npages; | |
2110 | new->flags = mem->flags; | |
2111 | new->userspace_addr = mem->userspace_addr; | |
a7800aa8 SC |
2112 | if (mem->flags & KVM_MEM_GUEST_MEMFD) { |
2113 | r = kvm_gmem_bind(kvm, new, mem->guest_memfd, mem->guest_memfd_offset); | |
2114 | if (r) | |
2115 | goto out; | |
2116 | } | |
6aa8b732 | 2117 | |
244893fa | 2118 | r = kvm_set_memslot(kvm, old, new, change); |
cf47f50b | 2119 | if (r) |
a7800aa8 SC |
2120 | goto out_unbind; |
2121 | ||
2122 | return 0; | |
2123 | ||
2124 | out_unbind: | |
2125 | if (mem->flags & KVM_MEM_GUEST_MEMFD) | |
2126 | kvm_gmem_unbind(new); | |
2127 | out: | |
2128 | kfree(new); | |
6aa8b732 | 2129 | return r; |
210c7c4d | 2130 | } |
f78e0e2e SY |
2131 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
2132 | ||
2133 | int kvm_set_memory_region(struct kvm *kvm, | |
bb58b90b | 2134 | const struct kvm_userspace_memory_region2 *mem) |
f78e0e2e SY |
2135 | { |
2136 | int r; | |
2137 | ||
79fac95e | 2138 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 2139 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 2140 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
2141 | return r; |
2142 | } | |
210c7c4d IE |
2143 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
2144 | ||
7940876e | 2145 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
bb58b90b | 2146 | struct kvm_userspace_memory_region2 *mem) |
210c7c4d | 2147 | { |
f481b069 | 2148 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 2149 | return -EINVAL; |
09170a49 | 2150 | |
47ae31e2 | 2151 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
2152 | } |
2153 | ||
0dff0846 | 2154 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
2155 | /** |
2156 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
2157 | * @kvm: pointer to kvm instance | |
2158 | * @log: slot id and address to which we copy the log | |
2159 | * @is_dirty: set to '1' if any dirty pages were found | |
2160 | * @memslot: set to the associated memslot, always valid on success | |
2161 | */ | |
2162 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
2163 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 2164 | { |
9f6b8029 | 2165 | struct kvm_memslots *slots; |
843574a3 | 2166 | int i, as_id, id; |
87bf6e7d | 2167 | unsigned long n; |
6aa8b732 AK |
2168 | unsigned long any = 0; |
2169 | ||
86bdf3eb GS |
2170 | /* Dirty ring tracking may be exclusive to dirty log tracking */ |
2171 | if (!kvm_use_dirty_bitmap(kvm)) | |
b2cc64c4 PX |
2172 | return -ENXIO; |
2173 | ||
2a49f61d SC |
2174 | *memslot = NULL; |
2175 | *is_dirty = 0; | |
2176 | ||
f481b069 PB |
2177 | as_id = log->slot >> 16; |
2178 | id = (u16)log->slot; | |
eed52e43 | 2179 | if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS) |
843574a3 | 2180 | return -EINVAL; |
6aa8b732 | 2181 | |
f481b069 | 2182 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 2183 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 2184 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 2185 | return -ENOENT; |
6aa8b732 | 2186 | |
2a49f61d SC |
2187 | kvm_arch_sync_dirty_log(kvm, *memslot); |
2188 | ||
2189 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 2190 | |
cd1a4a98 | 2191 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 2192 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 2193 | |
2a49f61d | 2194 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 2195 | return -EFAULT; |
6aa8b732 | 2196 | |
5bb064dc ZX |
2197 | if (any) |
2198 | *is_dirty = 1; | |
843574a3 | 2199 | return 0; |
6aa8b732 | 2200 | } |
2ba9f0d8 | 2201 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 2202 | |
0dff0846 | 2203 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 2204 | /** |
b8b00220 | 2205 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 2206 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
2207 | * @kvm: pointer to kvm instance |
2208 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
2209 | * |
2210 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
2211 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
2212 | * following order: | |
2213 | * | |
2214 | * 1. Take a snapshot of the bit and clear it if needed. | |
2215 | * 2. Write protect the corresponding page. | |
2216 | * 3. Copy the snapshot to the userspace. | |
2217 | * 4. Upon return caller flushes TLB's if needed. | |
2218 | * | |
2219 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
2220 | * entry. This is not a problem because the page is reported dirty using | |
2221 | * the snapshot taken before and step 4 ensures that writes done after | |
2222 | * exiting to userspace will be logged for the next call. | |
2223 | * | |
2224 | */ | |
0dff0846 | 2225 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 2226 | { |
9f6b8029 | 2227 | struct kvm_memslots *slots; |
ba0513b5 | 2228 | struct kvm_memory_slot *memslot; |
58d6db34 | 2229 | int i, as_id, id; |
ba0513b5 MS |
2230 | unsigned long n; |
2231 | unsigned long *dirty_bitmap; | |
2232 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2233 | bool flush; |
ba0513b5 | 2234 | |
86bdf3eb GS |
2235 | /* Dirty ring tracking may be exclusive to dirty log tracking */ |
2236 | if (!kvm_use_dirty_bitmap(kvm)) | |
b2cc64c4 PX |
2237 | return -ENXIO; |
2238 | ||
f481b069 PB |
2239 | as_id = log->slot >> 16; |
2240 | id = (u16)log->slot; | |
eed52e43 | 2241 | if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS) |
58d6db34 | 2242 | return -EINVAL; |
ba0513b5 | 2243 | |
f481b069 PB |
2244 | slots = __kvm_memslots(kvm, as_id); |
2245 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2246 | if (!memslot || !memslot->dirty_bitmap) |
2247 | return -ENOENT; | |
ba0513b5 MS |
2248 | |
2249 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 2250 | |
0dff0846 SC |
2251 | kvm_arch_sync_dirty_log(kvm, memslot); |
2252 | ||
ba0513b5 | 2253 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 2254 | flush = false; |
2a31b9db PB |
2255 | if (kvm->manual_dirty_log_protect) { |
2256 | /* | |
2257 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
2258 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
2259 | * is some code duplication between this function and | |
2260 | * kvm_get_dirty_log, but hopefully all architecture | |
2261 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
2262 | * can be eliminated. | |
2263 | */ | |
2264 | dirty_bitmap_buffer = dirty_bitmap; | |
2265 | } else { | |
2266 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
2267 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 2268 | |
531810ca | 2269 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
2270 | for (i = 0; i < n / sizeof(long); i++) { |
2271 | unsigned long mask; | |
2272 | gfn_t offset; | |
ba0513b5 | 2273 | |
2a31b9db PB |
2274 | if (!dirty_bitmap[i]) |
2275 | continue; | |
2276 | ||
0dff0846 | 2277 | flush = true; |
2a31b9db PB |
2278 | mask = xchg(&dirty_bitmap[i], 0); |
2279 | dirty_bitmap_buffer[i] = mask; | |
2280 | ||
a67794ca LT |
2281 | offset = i * BITS_PER_LONG; |
2282 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
2283 | offset, mask); | |
2a31b9db | 2284 | } |
531810ca | 2285 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
2286 | } |
2287 | ||
0dff0846 | 2288 | if (flush) |
619b5072 | 2289 | kvm_flush_remote_tlbs_memslot(kvm, memslot); |
0dff0846 | 2290 | |
2a31b9db PB |
2291 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
2292 | return -EFAULT; | |
2293 | return 0; | |
2294 | } | |
0dff0846 SC |
2295 | |
2296 | ||
2297 | /** | |
2298 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
2299 | * @kvm: kvm instance | |
2300 | * @log: slot id and address to which we copy the log | |
2301 | * | |
2302 | * Steps 1-4 below provide general overview of dirty page logging. See | |
2303 | * kvm_get_dirty_log_protect() function description for additional details. | |
2304 | * | |
2305 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
2306 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
2307 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
2308 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
2309 | * writes will be marked dirty for next log read. | |
2310 | * | |
2311 | * 1. Take a snapshot of the bit and clear it if needed. | |
2312 | * 2. Write protect the corresponding page. | |
2313 | * 3. Copy the snapshot to the userspace. | |
2314 | * 4. Flush TLB's if needed. | |
2315 | */ | |
2316 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
2317 | struct kvm_dirty_log *log) | |
2318 | { | |
2319 | int r; | |
2320 | ||
2321 | mutex_lock(&kvm->slots_lock); | |
2322 | ||
2323 | r = kvm_get_dirty_log_protect(kvm, log); | |
2324 | ||
2325 | mutex_unlock(&kvm->slots_lock); | |
2326 | return r; | |
2327 | } | |
2a31b9db PB |
2328 | |
2329 | /** | |
2330 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
2331 | * and reenable dirty page tracking for the corresponding pages. | |
2332 | * @kvm: pointer to kvm instance | |
2333 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
2334 | */ | |
0dff0846 SC |
2335 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
2336 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
2337 | { |
2338 | struct kvm_memslots *slots; | |
2339 | struct kvm_memory_slot *memslot; | |
98938aa8 | 2340 | int as_id, id; |
2a31b9db | 2341 | gfn_t offset; |
98938aa8 | 2342 | unsigned long i, n; |
2a31b9db PB |
2343 | unsigned long *dirty_bitmap; |
2344 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2345 | bool flush; |
2a31b9db | 2346 | |
86bdf3eb GS |
2347 | /* Dirty ring tracking may be exclusive to dirty log tracking */ |
2348 | if (!kvm_use_dirty_bitmap(kvm)) | |
b2cc64c4 PX |
2349 | return -ENXIO; |
2350 | ||
2a31b9db PB |
2351 | as_id = log->slot >> 16; |
2352 | id = (u16)log->slot; | |
eed52e43 | 2353 | if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS) |
2a31b9db PB |
2354 | return -EINVAL; |
2355 | ||
76d58e0f | 2356 | if (log->first_page & 63) |
2a31b9db PB |
2357 | return -EINVAL; |
2358 | ||
2359 | slots = __kvm_memslots(kvm, as_id); | |
2360 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2361 | if (!memslot || !memslot->dirty_bitmap) |
2362 | return -ENOENT; | |
2a31b9db PB |
2363 | |
2364 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 2365 | |
4ddc9204 | 2366 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
2367 | |
2368 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
2369 | log->num_pages > memslot->npages - log->first_page || |
2370 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
2371 | return -EINVAL; | |
98938aa8 | 2372 | |
0dff0846 SC |
2373 | kvm_arch_sync_dirty_log(kvm, memslot); |
2374 | ||
2375 | flush = false; | |
2a31b9db PB |
2376 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
2377 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
2378 | return -EFAULT; | |
ba0513b5 | 2379 | |
531810ca | 2380 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
2381 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
2382 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
2383 | i++, offset += BITS_PER_LONG) { |
2384 | unsigned long mask = *dirty_bitmap_buffer++; | |
2385 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
2386 | if (!mask) | |
ba0513b5 MS |
2387 | continue; |
2388 | ||
2a31b9db | 2389 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 2390 | |
2a31b9db PB |
2391 | /* |
2392 | * mask contains the bits that really have been cleared. This | |
2393 | * never includes any bits beyond the length of the memslot (if | |
2394 | * the length is not aligned to 64 pages), therefore it is not | |
2395 | * a problem if userspace sets them in log->dirty_bitmap. | |
2396 | */ | |
58d2930f | 2397 | if (mask) { |
0dff0846 | 2398 | flush = true; |
58d2930f TY |
2399 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
2400 | offset, mask); | |
2401 | } | |
ba0513b5 | 2402 | } |
531810ca | 2403 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 2404 | |
0dff0846 | 2405 | if (flush) |
619b5072 | 2406 | kvm_flush_remote_tlbs_memslot(kvm, memslot); |
0dff0846 | 2407 | |
58d6db34 | 2408 | return 0; |
ba0513b5 | 2409 | } |
0dff0846 SC |
2410 | |
2411 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
2412 | struct kvm_clear_dirty_log *log) | |
2413 | { | |
2414 | int r; | |
2415 | ||
2416 | mutex_lock(&kvm->slots_lock); | |
2417 | ||
2418 | r = kvm_clear_dirty_log_protect(kvm, log); | |
2419 | ||
2420 | mutex_unlock(&kvm->slots_lock); | |
2421 | return r; | |
2422 | } | |
2423 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 2424 | |
5a475554 CP |
2425 | #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES |
2426 | /* | |
2427 | * Returns true if _all_ gfns in the range [@start, @end) have attributes | |
2428 | * matching @attrs. | |
2429 | */ | |
2430 | bool kvm_range_has_memory_attributes(struct kvm *kvm, gfn_t start, gfn_t end, | |
2431 | unsigned long attrs) | |
2432 | { | |
2433 | XA_STATE(xas, &kvm->mem_attr_array, start); | |
2434 | unsigned long index; | |
2435 | bool has_attrs; | |
2436 | void *entry; | |
2437 | ||
2438 | rcu_read_lock(); | |
2439 | ||
2440 | if (!attrs) { | |
2441 | has_attrs = !xas_find(&xas, end - 1); | |
2442 | goto out; | |
2443 | } | |
2444 | ||
2445 | has_attrs = true; | |
2446 | for (index = start; index < end; index++) { | |
2447 | do { | |
2448 | entry = xas_next(&xas); | |
2449 | } while (xas_retry(&xas, entry)); | |
2450 | ||
2451 | if (xas.xa_index != index || xa_to_value(entry) != attrs) { | |
2452 | has_attrs = false; | |
2453 | break; | |
2454 | } | |
2455 | } | |
2456 | ||
2457 | out: | |
2458 | rcu_read_unlock(); | |
2459 | return has_attrs; | |
2460 | } | |
2461 | ||
2462 | static u64 kvm_supported_mem_attributes(struct kvm *kvm) | |
2463 | { | |
a7800aa8 | 2464 | if (!kvm || kvm_arch_has_private_mem(kvm)) |
5a475554 CP |
2465 | return KVM_MEMORY_ATTRIBUTE_PRIVATE; |
2466 | ||
2467 | return 0; | |
2468 | } | |
2469 | ||
2470 | static __always_inline void kvm_handle_gfn_range(struct kvm *kvm, | |
2471 | struct kvm_mmu_notifier_range *range) | |
2472 | { | |
2473 | struct kvm_gfn_range gfn_range; | |
2474 | struct kvm_memory_slot *slot; | |
2475 | struct kvm_memslots *slots; | |
2476 | struct kvm_memslot_iter iter; | |
2477 | bool found_memslot = false; | |
2478 | bool ret = false; | |
2479 | int i; | |
2480 | ||
2481 | gfn_range.arg = range->arg; | |
2482 | gfn_range.may_block = range->may_block; | |
2483 | ||
eed52e43 | 2484 | for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { |
5a475554 CP |
2485 | slots = __kvm_memslots(kvm, i); |
2486 | ||
2487 | kvm_for_each_memslot_in_gfn_range(&iter, slots, range->start, range->end) { | |
2488 | slot = iter.slot; | |
2489 | gfn_range.slot = slot; | |
2490 | ||
2491 | gfn_range.start = max(range->start, slot->base_gfn); | |
2492 | gfn_range.end = min(range->end, slot->base_gfn + slot->npages); | |
2493 | if (gfn_range.start >= gfn_range.end) | |
2494 | continue; | |
2495 | ||
2496 | if (!found_memslot) { | |
2497 | found_memslot = true; | |
2498 | KVM_MMU_LOCK(kvm); | |
2499 | if (!IS_KVM_NULL_FN(range->on_lock)) | |
2500 | range->on_lock(kvm); | |
2501 | } | |
2502 | ||
2503 | ret |= range->handler(kvm, &gfn_range); | |
2504 | } | |
2505 | } | |
2506 | ||
2507 | if (range->flush_on_ret && ret) | |
2508 | kvm_flush_remote_tlbs(kvm); | |
2509 | ||
2510 | if (found_memslot) | |
2511 | KVM_MMU_UNLOCK(kvm); | |
2512 | } | |
2513 | ||
2514 | static bool kvm_pre_set_memory_attributes(struct kvm *kvm, | |
2515 | struct kvm_gfn_range *range) | |
2516 | { | |
2517 | /* | |
2518 | * Unconditionally add the range to the invalidation set, regardless of | |
2519 | * whether or not the arch callback actually needs to zap SPTEs. E.g. | |
2520 | * if KVM supports RWX attributes in the future and the attributes are | |
2521 | * going from R=>RW, zapping isn't strictly necessary. Unconditionally | |
2522 | * adding the range allows KVM to require that MMU invalidations add at | |
2523 | * least one range between begin() and end(), e.g. allows KVM to detect | |
2524 | * bugs where the add() is missed. Relaxing the rule *might* be safe, | |
2525 | * but it's not obvious that allowing new mappings while the attributes | |
2526 | * are in flux is desirable or worth the complexity. | |
2527 | */ | |
2528 | kvm_mmu_invalidate_range_add(kvm, range->start, range->end); | |
2529 | ||
2530 | return kvm_arch_pre_set_memory_attributes(kvm, range); | |
2531 | } | |
2532 | ||
2533 | /* Set @attributes for the gfn range [@start, @end). */ | |
2534 | static int kvm_vm_set_mem_attributes(struct kvm *kvm, gfn_t start, gfn_t end, | |
2535 | unsigned long attributes) | |
2536 | { | |
2537 | struct kvm_mmu_notifier_range pre_set_range = { | |
2538 | .start = start, | |
2539 | .end = end, | |
2540 | .handler = kvm_pre_set_memory_attributes, | |
2541 | .on_lock = kvm_mmu_invalidate_begin, | |
2542 | .flush_on_ret = true, | |
2543 | .may_block = true, | |
2544 | }; | |
2545 | struct kvm_mmu_notifier_range post_set_range = { | |
2546 | .start = start, | |
2547 | .end = end, | |
2548 | .arg.attributes = attributes, | |
2549 | .handler = kvm_arch_post_set_memory_attributes, | |
2550 | .on_lock = kvm_mmu_invalidate_end, | |
2551 | .may_block = true, | |
2552 | }; | |
2553 | unsigned long i; | |
2554 | void *entry; | |
2555 | int r = 0; | |
2556 | ||
2557 | entry = attributes ? xa_mk_value(attributes) : NULL; | |
2558 | ||
2559 | mutex_lock(&kvm->slots_lock); | |
2560 | ||
2561 | /* Nothing to do if the entire range as the desired attributes. */ | |
2562 | if (kvm_range_has_memory_attributes(kvm, start, end, attributes)) | |
2563 | goto out_unlock; | |
2564 | ||
2565 | /* | |
2566 | * Reserve memory ahead of time to avoid having to deal with failures | |
2567 | * partway through setting the new attributes. | |
2568 | */ | |
2569 | for (i = start; i < end; i++) { | |
2570 | r = xa_reserve(&kvm->mem_attr_array, i, GFP_KERNEL_ACCOUNT); | |
2571 | if (r) | |
2572 | goto out_unlock; | |
2573 | } | |
2574 | ||
2575 | kvm_handle_gfn_range(kvm, &pre_set_range); | |
2576 | ||
2577 | for (i = start; i < end; i++) { | |
2578 | r = xa_err(xa_store(&kvm->mem_attr_array, i, entry, | |
2579 | GFP_KERNEL_ACCOUNT)); | |
2580 | KVM_BUG_ON(r, kvm); | |
2581 | } | |
2582 | ||
2583 | kvm_handle_gfn_range(kvm, &post_set_range); | |
2584 | ||
2585 | out_unlock: | |
2586 | mutex_unlock(&kvm->slots_lock); | |
2587 | ||
2588 | return r; | |
2589 | } | |
2590 | static int kvm_vm_ioctl_set_mem_attributes(struct kvm *kvm, | |
2591 | struct kvm_memory_attributes *attrs) | |
2592 | { | |
2593 | gfn_t start, end; | |
2594 | ||
2595 | /* flags is currently not used. */ | |
2596 | if (attrs->flags) | |
2597 | return -EINVAL; | |
2598 | if (attrs->attributes & ~kvm_supported_mem_attributes(kvm)) | |
2599 | return -EINVAL; | |
2600 | if (attrs->size == 0 || attrs->address + attrs->size < attrs->address) | |
2601 | return -EINVAL; | |
2602 | if (!PAGE_ALIGNED(attrs->address) || !PAGE_ALIGNED(attrs->size)) | |
2603 | return -EINVAL; | |
2604 | ||
2605 | start = attrs->address >> PAGE_SHIFT; | |
2606 | end = (attrs->address + attrs->size) >> PAGE_SHIFT; | |
2607 | ||
2608 | /* | |
2609 | * xarray tracks data using "unsigned long", and as a result so does | |
2610 | * KVM. For simplicity, supports generic attributes only on 64-bit | |
2611 | * architectures. | |
2612 | */ | |
2613 | BUILD_BUG_ON(sizeof(attrs->attributes) != sizeof(unsigned long)); | |
2614 | ||
2615 | return kvm_vm_set_mem_attributes(kvm, start, end, attrs->attributes); | |
2616 | } | |
2617 | #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */ | |
2618 | ||
49c7754c GN |
2619 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
2620 | { | |
2621 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
2622 | } | |
a1f4d395 | 2623 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 2624 | |
8e73485c PB |
2625 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
2626 | { | |
fe22ed82 | 2627 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
a54d8066 | 2628 | u64 gen = slots->generation; |
fe22ed82 | 2629 | struct kvm_memory_slot *slot; |
fe22ed82 | 2630 | |
a54d8066 MS |
2631 | /* |
2632 | * This also protects against using a memslot from a different address space, | |
2633 | * since different address spaces have different generation numbers. | |
2634 | */ | |
2635 | if (unlikely(gen != vcpu->last_used_slot_gen)) { | |
2636 | vcpu->last_used_slot = NULL; | |
2637 | vcpu->last_used_slot_gen = gen; | |
2638 | } | |
2639 | ||
2640 | slot = try_get_memslot(vcpu->last_used_slot, gfn); | |
fe22ed82 DM |
2641 | if (slot) |
2642 | return slot; | |
2643 | ||
2644 | /* | |
2645 | * Fall back to searching all memslots. We purposely use | |
2646 | * search_memslots() instead of __gfn_to_memslot() to avoid | |
a54d8066 | 2647 | * thrashing the VM-wide last_used_slot in kvm_memslots. |
fe22ed82 | 2648 | */ |
a54d8066 | 2649 | slot = search_memslots(slots, gfn, false); |
fe22ed82 | 2650 | if (slot) { |
a54d8066 | 2651 | vcpu->last_used_slot = slot; |
fe22ed82 DM |
2652 | return slot; |
2653 | } | |
2654 | ||
2655 | return NULL; | |
8e73485c PB |
2656 | } |
2657 | ||
33e94154 | 2658 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 2659 | { |
bf3e05bc | 2660 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 2661 | |
c36b7150 | 2662 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
2663 | } |
2664 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
2665 | ||
995decb6 VK |
2666 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
2667 | { | |
2668 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2669 | ||
2670 | return kvm_is_visible_memslot(memslot); | |
2671 | } | |
2672 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
2673 | ||
f9b84e19 | 2674 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
2675 | { |
2676 | struct vm_area_struct *vma; | |
2677 | unsigned long addr, size; | |
2678 | ||
2679 | size = PAGE_SIZE; | |
2680 | ||
42cde48b | 2681 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
2682 | if (kvm_is_error_hva(addr)) |
2683 | return PAGE_SIZE; | |
2684 | ||
d8ed45c5 | 2685 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
2686 | vma = find_vma(current->mm, addr); |
2687 | if (!vma) | |
2688 | goto out; | |
2689 | ||
2690 | size = vma_kernel_pagesize(vma); | |
2691 | ||
2692 | out: | |
d8ed45c5 | 2693 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
2694 | |
2695 | return size; | |
2696 | } | |
2697 | ||
8283e36a | 2698 | static bool memslot_is_readonly(const struct kvm_memory_slot *slot) |
4d8b81ab XG |
2699 | { |
2700 | return slot->flags & KVM_MEM_READONLY; | |
2701 | } | |
2702 | ||
8283e36a | 2703 | static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn, |
4d8b81ab | 2704 | gfn_t *nr_pages, bool write) |
539cb660 | 2705 | { |
bc6678a3 | 2706 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 2707 | return KVM_HVA_ERR_BAD; |
48987781 | 2708 | |
4d8b81ab XG |
2709 | if (memslot_is_readonly(slot) && write) |
2710 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
2711 | |
2712 | if (nr_pages) | |
2713 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
2714 | ||
4d8b81ab | 2715 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 2716 | } |
48987781 | 2717 | |
4d8b81ab XG |
2718 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2719 | gfn_t *nr_pages) | |
2720 | { | |
2721 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 2722 | } |
48987781 | 2723 | |
4d8b81ab | 2724 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 2725 | gfn_t gfn) |
4d8b81ab XG |
2726 | { |
2727 | return gfn_to_hva_many(slot, gfn, NULL); | |
2728 | } | |
2729 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
2730 | ||
48987781 XG |
2731 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
2732 | { | |
49c7754c | 2733 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 2734 | } |
0d150298 | 2735 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 2736 | |
8e73485c PB |
2737 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
2738 | { | |
2739 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
2740 | } | |
2741 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
2742 | ||
86ab8cff | 2743 | /* |
970c0d4b WY |
2744 | * Return the hva of a @gfn and the R/W attribute if possible. |
2745 | * | |
2746 | * @slot: the kvm_memory_slot which contains @gfn | |
2747 | * @gfn: the gfn to be translated | |
2748 | * @writable: used to return the read/write attribute of the @slot if the hva | |
2749 | * is valid and @writable is not NULL | |
86ab8cff | 2750 | */ |
64d83126 CD |
2751 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
2752 | gfn_t gfn, bool *writable) | |
86ab8cff | 2753 | { |
a2ac07fe GN |
2754 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
2755 | ||
2756 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
2757 | *writable = !memslot_is_readonly(slot); |
2758 | ||
a2ac07fe | 2759 | return hva; |
86ab8cff XG |
2760 | } |
2761 | ||
64d83126 CD |
2762 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
2763 | { | |
2764 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2765 | ||
2766 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2767 | } | |
2768 | ||
8e73485c PB |
2769 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
2770 | { | |
2771 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2772 | ||
2773 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2774 | } | |
2775 | ||
fafc3dba HY |
2776 | static inline int check_user_page_hwpoison(unsigned long addr) |
2777 | { | |
0d731759 | 2778 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 2779 | |
54d02069 | 2780 | rc = get_user_pages(addr, 1, flags, NULL); |
fafc3dba HY |
2781 | return rc == -EHWPOISON; |
2782 | } | |
2783 | ||
2fc84311 | 2784 | /* |
b9b33da2 PB |
2785 | * The fast path to get the writable pfn which will be stored in @pfn, |
2786 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 2787 | * only part that runs if we can in atomic context. |
2fc84311 | 2788 | */ |
b9b33da2 PB |
2789 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
2790 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 2791 | { |
8d4e1288 | 2792 | struct page *page[1]; |
954bbbc2 | 2793 | |
12ce13fe XG |
2794 | /* |
2795 | * Fast pin a writable pfn only if it is a write fault request | |
2796 | * or the caller allows to map a writable pfn for a read fault | |
2797 | * request. | |
2798 | */ | |
2799 | if (!(write_fault || writable)) | |
2800 | return false; | |
612819c3 | 2801 | |
dadbb612 | 2802 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 2803 | *pfn = page_to_pfn(page[0]); |
612819c3 | 2804 | |
2fc84311 XG |
2805 | if (writable) |
2806 | *writable = true; | |
2807 | return true; | |
2808 | } | |
af585b92 | 2809 | |
2fc84311 XG |
2810 | return false; |
2811 | } | |
612819c3 | 2812 | |
2fc84311 XG |
2813 | /* |
2814 | * The slow path to get the pfn of the specified host virtual address, | |
2815 | * 1 indicates success, -errno is returned if error is detected. | |
2816 | */ | |
2817 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
c8b88b33 | 2818 | bool interruptible, bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 2819 | { |
b1e1296d DH |
2820 | /* |
2821 | * When a VCPU accesses a page that is not mapped into the secondary | |
2822 | * MMU, we lookup the page using GUP to map it, so the guest VCPU can | |
2823 | * make progress. We always want to honor NUMA hinting faults in that | |
2824 | * case, because GUP usage corresponds to memory accesses from the VCPU. | |
2825 | * Otherwise, we'd not trigger NUMA hinting faults once a page is | |
2826 | * mapped into the secondary MMU and gets accessed by a VCPU. | |
2827 | * | |
2828 | * Note that get_user_page_fast_only() and FOLL_WRITE for now | |
2829 | * implicitly honor NUMA hinting faults and don't need this flag. | |
2830 | */ | |
2831 | unsigned int flags = FOLL_HWPOISON | FOLL_HONOR_NUMA_FAULT; | |
ce53053c | 2832 | struct page *page; |
28249139 | 2833 | int npages; |
612819c3 | 2834 | |
2fc84311 XG |
2835 | might_sleep(); |
2836 | ||
2837 | if (writable) | |
2838 | *writable = write_fault; | |
2839 | ||
ce53053c AV |
2840 | if (write_fault) |
2841 | flags |= FOLL_WRITE; | |
2842 | if (async) | |
2843 | flags |= FOLL_NOWAIT; | |
c8b88b33 PX |
2844 | if (interruptible) |
2845 | flags |= FOLL_INTERRUPTIBLE; | |
d4944b0e | 2846 | |
ce53053c | 2847 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
2848 | if (npages != 1) |
2849 | return npages; | |
2850 | ||
2851 | /* map read fault as writable if possible */ | |
12ce13fe | 2852 | if (unlikely(!write_fault) && writable) { |
ce53053c | 2853 | struct page *wpage; |
2fc84311 | 2854 | |
dadbb612 | 2855 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 2856 | *writable = true; |
ce53053c AV |
2857 | put_page(page); |
2858 | page = wpage; | |
612819c3 | 2859 | } |
887c08ac | 2860 | } |
ce53053c | 2861 | *pfn = page_to_pfn(page); |
2fc84311 XG |
2862 | return npages; |
2863 | } | |
539cb660 | 2864 | |
4d8b81ab XG |
2865 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
2866 | { | |
2867 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
2868 | return false; | |
2e2e3738 | 2869 | |
4d8b81ab XG |
2870 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
2871 | return false; | |
887c08ac | 2872 | |
4d8b81ab XG |
2873 | return true; |
2874 | } | |
bf998156 | 2875 | |
f8be156b NP |
2876 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
2877 | { | |
b14b2690 SC |
2878 | struct page *page = kvm_pfn_to_refcounted_page(pfn); |
2879 | ||
2880 | if (!page) | |
f8be156b | 2881 | return 1; |
b14b2690 SC |
2882 | |
2883 | return get_page_unless_zero(page); | |
f8be156b NP |
2884 | } |
2885 | ||
92176a8e | 2886 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1625566e XT |
2887 | unsigned long addr, bool write_fault, |
2888 | bool *writable, kvm_pfn_t *p_pfn) | |
92176a8e | 2889 | { |
a9545779 | 2890 | kvm_pfn_t pfn; |
bd2fae8d | 2891 | pte_t *ptep; |
c33c7948 | 2892 | pte_t pte; |
bd2fae8d | 2893 | spinlock_t *ptl; |
add6a0cd PB |
2894 | int r; |
2895 | ||
9fd6dad1 | 2896 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2897 | if (r) { |
2898 | /* | |
2899 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
2900 | * not call the fault handler, so do it here. | |
2901 | */ | |
2902 | bool unlocked = false; | |
64019a2e | 2903 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
2904 | (write_fault ? FAULT_FLAG_WRITE : 0), |
2905 | &unlocked); | |
a8387d0b PB |
2906 | if (unlocked) |
2907 | return -EAGAIN; | |
add6a0cd PB |
2908 | if (r) |
2909 | return r; | |
2910 | ||
9fd6dad1 | 2911 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2912 | if (r) |
2913 | return r; | |
bd2fae8d | 2914 | } |
add6a0cd | 2915 | |
c33c7948 RR |
2916 | pte = ptep_get(ptep); |
2917 | ||
2918 | if (write_fault && !pte_write(pte)) { | |
bd2fae8d PB |
2919 | pfn = KVM_PFN_ERR_RO_FAULT; |
2920 | goto out; | |
add6a0cd PB |
2921 | } |
2922 | ||
a340b3e2 | 2923 | if (writable) |
c33c7948 RR |
2924 | *writable = pte_write(pte); |
2925 | pfn = pte_pfn(pte); | |
add6a0cd PB |
2926 | |
2927 | /* | |
2928 | * Get a reference here because callers of *hva_to_pfn* and | |
2929 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
2930 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
36c3ce6c | 2931 | * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will |
add6a0cd PB |
2932 | * simply do nothing for reserved pfns. |
2933 | * | |
2934 | * Whoever called remap_pfn_range is also going to call e.g. | |
2935 | * unmap_mapping_range before the underlying pages are freed, | |
2936 | * causing a call to our MMU notifier. | |
f8be156b NP |
2937 | * |
2938 | * Certain IO or PFNMAP mappings can be backed with valid | |
2939 | * struct pages, but be allocated without refcounting e.g., | |
2940 | * tail pages of non-compound higher order allocations, which | |
2941 | * would then underflow the refcount when the caller does the | |
2942 | * required put_page. Don't allow those pages here. | |
c33c7948 | 2943 | */ |
f8be156b NP |
2944 | if (!kvm_try_get_pfn(pfn)) |
2945 | r = -EFAULT; | |
add6a0cd | 2946 | |
bd2fae8d PB |
2947 | out: |
2948 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 2949 | *p_pfn = pfn; |
f8be156b NP |
2950 | |
2951 | return r; | |
92176a8e PB |
2952 | } |
2953 | ||
12ce13fe XG |
2954 | /* |
2955 | * Pin guest page in memory and return its pfn. | |
2956 | * @addr: host virtual address which maps memory to the guest | |
ed2f049f | 2957 | * @atomic: whether this function is forbidden from sleeping |
c8b88b33 | 2958 | * @interruptible: whether the process can be interrupted by non-fatal signals |
12ce13fe XG |
2959 | * @async: whether this function need to wait IO complete if the |
2960 | * host page is not in the memory | |
2961 | * @write_fault: whether we should get a writable host page | |
2962 | * @writable: whether it allows to map a writable host page for !@write_fault | |
2963 | * | |
2964 | * The function will map a writable host page for these two cases: | |
2965 | * 1): @write_fault = true | |
2966 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
2967 | * whether the mapping is writable. | |
2968 | */ | |
c8b88b33 PX |
2969 | kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool interruptible, |
2970 | bool *async, bool write_fault, bool *writable) | |
2fc84311 XG |
2971 | { |
2972 | struct vm_area_struct *vma; | |
943dfea8 | 2973 | kvm_pfn_t pfn; |
92176a8e | 2974 | int npages, r; |
2e2e3738 | 2975 | |
2fc84311 XG |
2976 | /* we can do it either atomically or asynchronously, not both */ |
2977 | BUG_ON(atomic && async); | |
8d4e1288 | 2978 | |
b9b33da2 | 2979 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
2980 | return pfn; |
2981 | ||
2982 | if (atomic) | |
2983 | return KVM_PFN_ERR_FAULT; | |
2984 | ||
c8b88b33 PX |
2985 | npages = hva_to_pfn_slow(addr, async, write_fault, interruptible, |
2986 | writable, &pfn); | |
2fc84311 XG |
2987 | if (npages == 1) |
2988 | return pfn; | |
fe5ed56c PX |
2989 | if (npages == -EINTR) |
2990 | return KVM_PFN_ERR_SIGPENDING; | |
8d4e1288 | 2991 | |
d8ed45c5 | 2992 | mmap_read_lock(current->mm); |
2fc84311 XG |
2993 | if (npages == -EHWPOISON || |
2994 | (!async && check_user_page_hwpoison(addr))) { | |
2995 | pfn = KVM_PFN_ERR_HWPOISON; | |
2996 | goto exit; | |
2997 | } | |
2998 | ||
a8387d0b | 2999 | retry: |
fc98c03b | 3000 | vma = vma_lookup(current->mm, addr); |
2fc84311 XG |
3001 | |
3002 | if (vma == NULL) | |
3003 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 3004 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
1625566e | 3005 | r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn); |
a8387d0b PB |
3006 | if (r == -EAGAIN) |
3007 | goto retry; | |
92176a8e PB |
3008 | if (r < 0) |
3009 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 3010 | } else { |
4d8b81ab | 3011 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
3012 | *async = true; |
3013 | pfn = KVM_PFN_ERR_FAULT; | |
3014 | } | |
3015 | exit: | |
d8ed45c5 | 3016 | mmap_read_unlock(current->mm); |
2e2e3738 | 3017 | return pfn; |
35149e21 AL |
3018 | } |
3019 | ||
8283e36a | 3020 | kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn, |
c8b88b33 PX |
3021 | bool atomic, bool interruptible, bool *async, |
3022 | bool write_fault, bool *writable, hva_t *hva) | |
887c08ac | 3023 | { |
4d8b81ab XG |
3024 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
3025 | ||
4a42d848 DS |
3026 | if (hva) |
3027 | *hva = addr; | |
3028 | ||
b2740d35 PB |
3029 | if (kvm_is_error_hva(addr)) { |
3030 | if (writable) | |
3031 | *writable = false; | |
f588557a AM |
3032 | |
3033 | return addr == KVM_HVA_ERR_RO_BAD ? KVM_PFN_ERR_RO_FAULT : | |
3034 | KVM_PFN_NOSLOT; | |
b2740d35 | 3035 | } |
4d8b81ab XG |
3036 | |
3037 | /* Do not map writable pfn in the readonly memslot. */ | |
3038 | if (writable && memslot_is_readonly(slot)) { | |
3039 | *writable = false; | |
3040 | writable = NULL; | |
3041 | } | |
3042 | ||
c8b88b33 | 3043 | return hva_to_pfn(addr, atomic, interruptible, async, write_fault, |
4d8b81ab | 3044 | writable); |
887c08ac | 3045 | } |
3520469d | 3046 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 3047 | |
ba049e93 | 3048 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
3049 | bool *writable) |
3050 | { | |
c8b88b33 PX |
3051 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, false, |
3052 | NULL, write_fault, writable, NULL); | |
612819c3 MT |
3053 | } |
3054 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
3055 | ||
8283e36a | 3056 | kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 3057 | { |
c8b88b33 PX |
3058 | return __gfn_to_pfn_memslot(slot, gfn, false, false, NULL, true, |
3059 | NULL, NULL); | |
506f0d6f | 3060 | } |
e37afc6e | 3061 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 3062 | |
8283e36a | 3063 | kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 3064 | { |
c8b88b33 PX |
3065 | return __gfn_to_pfn_memslot(slot, gfn, true, false, NULL, true, |
3066 | NULL, NULL); | |
506f0d6f | 3067 | } |
037d92dc | 3068 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 3069 | |
ba049e93 | 3070 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
3071 | { |
3072 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
3073 | } | |
3074 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
3075 | ||
ba049e93 | 3076 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
3077 | { |
3078 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
3079 | } | |
3080 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
3081 | ||
ba049e93 | 3082 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
3083 | { |
3084 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
3085 | } | |
3086 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
3087 | ||
d9ef13c2 PB |
3088 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
3089 | struct page **pages, int nr_pages) | |
48987781 XG |
3090 | { |
3091 | unsigned long addr; | |
076b925d | 3092 | gfn_t entry = 0; |
48987781 | 3093 | |
d9ef13c2 | 3094 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
3095 | if (kvm_is_error_hva(addr)) |
3096 | return -1; | |
3097 | ||
3098 | if (entry < nr_pages) | |
3099 | return 0; | |
3100 | ||
dadbb612 | 3101 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
3102 | } |
3103 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
3104 | ||
b1624f99 SC |
3105 | /* |
3106 | * Do not use this helper unless you are absolutely certain the gfn _must_ be | |
3107 | * backed by 'struct page'. A valid example is if the backing memslot is | |
3108 | * controlled by KVM. Note, if the returned page is valid, it's refcount has | |
3109 | * been elevated by gfn_to_pfn(). | |
3110 | */ | |
35149e21 AL |
3111 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
3112 | { | |
b14b2690 | 3113 | struct page *page; |
ba049e93 | 3114 | kvm_pfn_t pfn; |
2e2e3738 AL |
3115 | |
3116 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 3117 | |
81c52c56 | 3118 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 3119 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 3120 | |
b14b2690 SC |
3121 | page = kvm_pfn_to_refcounted_page(pfn); |
3122 | if (!page) | |
6cede2e6 | 3123 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 3124 | |
b14b2690 | 3125 | return page; |
954bbbc2 AK |
3126 | } |
3127 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
3128 | ||
357a18ad | 3129 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty) |
91724814 | 3130 | { |
91724814 BO |
3131 | if (dirty) |
3132 | kvm_release_pfn_dirty(pfn); | |
3133 | else | |
3134 | kvm_release_pfn_clean(pfn); | |
3135 | } | |
3136 | ||
357a18ad | 3137 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
e45adf66 KA |
3138 | { |
3139 | kvm_pfn_t pfn; | |
3140 | void *hva = NULL; | |
3141 | struct page *page = KVM_UNMAPPED_PAGE; | |
3142 | ||
3143 | if (!map) | |
3144 | return -EINVAL; | |
3145 | ||
357a18ad | 3146 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
e45adf66 KA |
3147 | if (is_error_noslot_pfn(pfn)) |
3148 | return -EINVAL; | |
3149 | ||
3150 | if (pfn_valid(pfn)) { | |
3151 | page = pfn_to_page(pfn); | |
357a18ad | 3152 | hva = kmap(page); |
d30b214d | 3153 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 3154 | } else { |
357a18ad | 3155 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
d30b214d | 3156 | #endif |
e45adf66 KA |
3157 | } |
3158 | ||
3159 | if (!hva) | |
3160 | return -EFAULT; | |
3161 | ||
3162 | map->page = page; | |
3163 | map->hva = hva; | |
3164 | map->pfn = pfn; | |
3165 | map->gfn = gfn; | |
3166 | ||
3167 | return 0; | |
3168 | } | |
e45adf66 KA |
3169 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); |
3170 | ||
357a18ad | 3171 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) |
e45adf66 KA |
3172 | { |
3173 | if (!map) | |
3174 | return; | |
3175 | ||
3176 | if (!map->hva) | |
3177 | return; | |
3178 | ||
357a18ad DW |
3179 | if (map->page != KVM_UNMAPPED_PAGE) |
3180 | kunmap(map->page); | |
eb1f2f38 | 3181 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 3182 | else |
357a18ad | 3183 | memunmap(map->hva); |
eb1f2f38 | 3184 | #endif |
e45adf66 | 3185 | |
91724814 | 3186 | if (dirty) |
357a18ad | 3187 | kvm_vcpu_mark_page_dirty(vcpu, map->gfn); |
91724814 | 3188 | |
357a18ad | 3189 | kvm_release_pfn(map->pfn, dirty); |
e45adf66 KA |
3190 | |
3191 | map->hva = NULL; | |
3192 | map->page = NULL; | |
3193 | } | |
3194 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); | |
3195 | ||
8e1c6914 | 3196 | static bool kvm_is_ad_tracked_page(struct page *page) |
8e73485c | 3197 | { |
8e1c6914 SC |
3198 | /* |
3199 | * Per page-flags.h, pages tagged PG_reserved "should in general not be | |
3200 | * touched (e.g. set dirty) except by its owner". | |
3201 | */ | |
3202 | return !PageReserved(page); | |
3203 | } | |
8e73485c | 3204 | |
8e1c6914 SC |
3205 | static void kvm_set_page_dirty(struct page *page) |
3206 | { | |
3207 | if (kvm_is_ad_tracked_page(page)) | |
3208 | SetPageDirty(page); | |
3209 | } | |
8e73485c | 3210 | |
8e1c6914 SC |
3211 | static void kvm_set_page_accessed(struct page *page) |
3212 | { | |
3213 | if (kvm_is_ad_tracked_page(page)) | |
3214 | mark_page_accessed(page); | |
8e73485c | 3215 | } |
8e73485c | 3216 | |
b4231d61 IE |
3217 | void kvm_release_page_clean(struct page *page) |
3218 | { | |
32cad84f XG |
3219 | WARN_ON(is_error_page(page)); |
3220 | ||
8e1c6914 SC |
3221 | kvm_set_page_accessed(page); |
3222 | put_page(page); | |
b4231d61 IE |
3223 | } |
3224 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
3225 | ||
ba049e93 | 3226 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 3227 | { |
b14b2690 SC |
3228 | struct page *page; |
3229 | ||
3230 | if (is_error_noslot_pfn(pfn)) | |
3231 | return; | |
3232 | ||
3233 | page = kvm_pfn_to_refcounted_page(pfn); | |
3234 | if (!page) | |
3235 | return; | |
3236 | ||
3237 | kvm_release_page_clean(page); | |
35149e21 AL |
3238 | } |
3239 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
3240 | ||
b4231d61 | 3241 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 3242 | { |
a2766325 XG |
3243 | WARN_ON(is_error_page(page)); |
3244 | ||
8e1c6914 SC |
3245 | kvm_set_page_dirty(page); |
3246 | kvm_release_page_clean(page); | |
35149e21 AL |
3247 | } |
3248 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
3249 | ||
f7a6509f | 3250 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 3251 | { |
b14b2690 SC |
3252 | struct page *page; |
3253 | ||
3254 | if (is_error_noslot_pfn(pfn)) | |
3255 | return; | |
3256 | ||
3257 | page = kvm_pfn_to_refcounted_page(pfn); | |
3258 | if (!page) | |
3259 | return; | |
3260 | ||
3261 | kvm_release_page_dirty(page); | |
35149e21 | 3262 | } |
f7a6509f | 3263 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 3264 | |
8e1c6914 SC |
3265 | /* |
3266 | * Note, checking for an error/noslot pfn is the caller's responsibility when | |
3267 | * directly marking a page dirty/accessed. Unlike the "release" helpers, the | |
3268 | * "set" helpers are not to be used when the pfn might point at garbage. | |
3269 | */ | |
ba049e93 | 3270 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 3271 | { |
8e1c6914 SC |
3272 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
3273 | return; | |
3274 | ||
3275 | if (pfn_valid(pfn)) | |
3276 | kvm_set_page_dirty(pfn_to_page(pfn)); | |
8a7ae055 | 3277 | } |
35149e21 AL |
3278 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
3279 | ||
ba049e93 | 3280 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 3281 | { |
8e1c6914 SC |
3282 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
3283 | return; | |
3284 | ||
3285 | if (pfn_valid(pfn)) | |
3286 | kvm_set_page_accessed(pfn_to_page(pfn)); | |
35149e21 AL |
3287 | } |
3288 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
3289 | ||
195aefde IE |
3290 | static int next_segment(unsigned long len, int offset) |
3291 | { | |
3292 | if (len > PAGE_SIZE - offset) | |
3293 | return PAGE_SIZE - offset; | |
3294 | else | |
3295 | return len; | |
3296 | } | |
3297 | ||
a3bd2f7e | 3298 | /* Copy @len bytes from guest memory at '(@gfn * PAGE_SIZE) + @offset' to @data */ |
8e73485c PB |
3299 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
3300 | void *data, int offset, int len) | |
195aefde | 3301 | { |
e0506bcb IE |
3302 | int r; |
3303 | unsigned long addr; | |
195aefde | 3304 | |
8e73485c | 3305 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
3306 | if (kvm_is_error_hva(addr)) |
3307 | return -EFAULT; | |
3180a7fc | 3308 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 3309 | if (r) |
195aefde | 3310 | return -EFAULT; |
195aefde IE |
3311 | return 0; |
3312 | } | |
8e73485c PB |
3313 | |
3314 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
3315 | int len) | |
3316 | { | |
3317 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
3318 | ||
3319 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
3320 | } | |
195aefde IE |
3321 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
3322 | ||
8e73485c PB |
3323 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
3324 | int offset, int len) | |
3325 | { | |
3326 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3327 | ||
3328 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
3329 | } | |
3330 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
3331 | ||
195aefde IE |
3332 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
3333 | { | |
3334 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3335 | int seg; | |
3336 | int offset = offset_in_page(gpa); | |
3337 | int ret; | |
3338 | ||
3339 | while ((seg = next_segment(len, offset)) != 0) { | |
3340 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
3341 | if (ret < 0) | |
3342 | return ret; | |
3343 | offset = 0; | |
3344 | len -= seg; | |
3345 | data += seg; | |
3346 | ++gfn; | |
3347 | } | |
3348 | return 0; | |
3349 | } | |
3350 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
3351 | ||
8e73485c | 3352 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 3353 | { |
7ec54588 | 3354 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 3355 | int seg; |
7ec54588 | 3356 | int offset = offset_in_page(gpa); |
8e73485c PB |
3357 | int ret; |
3358 | ||
3359 | while ((seg = next_segment(len, offset)) != 0) { | |
3360 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
3361 | if (ret < 0) | |
3362 | return ret; | |
3363 | offset = 0; | |
3364 | len -= seg; | |
3365 | data += seg; | |
3366 | ++gfn; | |
3367 | } | |
3368 | return 0; | |
3369 | } | |
3370 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 3371 | |
8e73485c PB |
3372 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
3373 | void *data, int offset, unsigned long len) | |
3374 | { | |
3375 | int r; | |
3376 | unsigned long addr; | |
3377 | ||
3378 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
3379 | if (kvm_is_error_hva(addr)) |
3380 | return -EFAULT; | |
0aac03f0 | 3381 | pagefault_disable(); |
3180a7fc | 3382 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 3383 | pagefault_enable(); |
7ec54588 MT |
3384 | if (r) |
3385 | return -EFAULT; | |
3386 | return 0; | |
3387 | } | |
7ec54588 | 3388 | |
8e73485c PB |
3389 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
3390 | void *data, unsigned long len) | |
3391 | { | |
3392 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3393 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3394 | int offset = offset_in_page(gpa); | |
3395 | ||
3396 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
3397 | } | |
3398 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
3399 | ||
a3bd2f7e | 3400 | /* Copy @len bytes from @data into guest memory at '(@gfn * PAGE_SIZE) + @offset' */ |
28bd726a PX |
3401 | static int __kvm_write_guest_page(struct kvm *kvm, |
3402 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 3403 | const void *data, int offset, int len) |
195aefde | 3404 | { |
e0506bcb IE |
3405 | int r; |
3406 | unsigned long addr; | |
195aefde | 3407 | |
251eb841 | 3408 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
3409 | if (kvm_is_error_hva(addr)) |
3410 | return -EFAULT; | |
8b0cedff | 3411 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 3412 | if (r) |
195aefde | 3413 | return -EFAULT; |
28bd726a | 3414 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
3415 | return 0; |
3416 | } | |
8e73485c PB |
3417 | |
3418 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
3419 | const void *data, int offset, int len) | |
3420 | { | |
3421 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
3422 | ||
28bd726a | 3423 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 3424 | } |
195aefde IE |
3425 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
3426 | ||
8e73485c PB |
3427 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3428 | const void *data, int offset, int len) | |
3429 | { | |
3430 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3431 | ||
28bd726a | 3432 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
3433 | } |
3434 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
3435 | ||
195aefde IE |
3436 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
3437 | unsigned long len) | |
3438 | { | |
3439 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3440 | int seg; | |
3441 | int offset = offset_in_page(gpa); | |
3442 | int ret; | |
3443 | ||
3444 | while ((seg = next_segment(len, offset)) != 0) { | |
3445 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
3446 | if (ret < 0) | |
3447 | return ret; | |
3448 | offset = 0; | |
3449 | len -= seg; | |
3450 | data += seg; | |
3451 | ++gfn; | |
3452 | } | |
3453 | return 0; | |
3454 | } | |
ff651cb6 | 3455 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 3456 | |
8e73485c PB |
3457 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
3458 | unsigned long len) | |
3459 | { | |
3460 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3461 | int seg; | |
3462 | int offset = offset_in_page(gpa); | |
3463 | int ret; | |
3464 | ||
3465 | while ((seg = next_segment(len, offset)) != 0) { | |
3466 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
3467 | if (ret < 0) | |
3468 | return ret; | |
3469 | offset = 0; | |
3470 | len -= seg; | |
3471 | data += seg; | |
3472 | ++gfn; | |
3473 | } | |
3474 | return 0; | |
3475 | } | |
3476 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
3477 | ||
5a2d4365 PB |
3478 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
3479 | struct gfn_to_hva_cache *ghc, | |
3480 | gpa_t gpa, unsigned long len) | |
49c7754c | 3481 | { |
49c7754c | 3482 | int offset = offset_in_page(gpa); |
8f964525 AH |
3483 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
3484 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
3485 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
3486 | gfn_t nr_pages_avail; | |
49c7754c | 3487 | |
6ad1e29f | 3488 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 3489 | ghc->generation = slots->generation; |
6ad1e29f SC |
3490 | |
3491 | if (start_gfn > end_gfn) { | |
3492 | ghc->hva = KVM_HVA_ERR_BAD; | |
3493 | return -EINVAL; | |
3494 | } | |
f1b9dd5e JM |
3495 | |
3496 | /* | |
3497 | * If the requested region crosses two memslots, we still | |
3498 | * verify that the entire region is valid here. | |
3499 | */ | |
6ad1e29f | 3500 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
3501 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
3502 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
3503 | &nr_pages_avail); | |
3504 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 3505 | return -EFAULT; |
f1b9dd5e JM |
3506 | } |
3507 | ||
3508 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 3509 | if (nr_pages_needed == 1) |
49c7754c | 3510 | ghc->hva += offset; |
f1b9dd5e | 3511 | else |
8f964525 | 3512 | ghc->memslot = NULL; |
f1b9dd5e | 3513 | |
6ad1e29f SC |
3514 | ghc->gpa = gpa; |
3515 | ghc->len = len; | |
3516 | return 0; | |
49c7754c | 3517 | } |
5a2d4365 | 3518 | |
4e335d9e | 3519 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
3520 | gpa_t gpa, unsigned long len) |
3521 | { | |
4e335d9e | 3522 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
3523 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
3524 | } | |
4e335d9e | 3525 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 3526 | |
4e335d9e | 3527 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
3528 | void *data, unsigned int offset, |
3529 | unsigned long len) | |
49c7754c | 3530 | { |
4e335d9e | 3531 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 3532 | int r; |
4ec6e863 | 3533 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 3534 | |
5f25e71e PB |
3535 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3536 | return -EINVAL; | |
8f964525 | 3537 | |
dc9ce71e SC |
3538 | if (slots->generation != ghc->generation) { |
3539 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3540 | return -EFAULT; | |
3541 | } | |
8f964525 | 3542 | |
49c7754c GN |
3543 | if (kvm_is_error_hva(ghc->hva)) |
3544 | return -EFAULT; | |
3545 | ||
fcfbc617 SC |
3546 | if (unlikely(!ghc->memslot)) |
3547 | return kvm_write_guest(kvm, gpa, data, len); | |
3548 | ||
4ec6e863 | 3549 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
3550 | if (r) |
3551 | return -EFAULT; | |
28bd726a | 3552 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
3553 | |
3554 | return 0; | |
3555 | } | |
4e335d9e | 3556 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 3557 | |
4e335d9e PB |
3558 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3559 | void *data, unsigned long len) | |
4ec6e863 | 3560 | { |
4e335d9e | 3561 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 3562 | } |
4e335d9e | 3563 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 3564 | |
0958f0ce VK |
3565 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3566 | void *data, unsigned int offset, | |
3567 | unsigned long len) | |
e03b644f | 3568 | { |
4e335d9e | 3569 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 3570 | int r; |
0958f0ce | 3571 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 3572 | |
5f25e71e PB |
3573 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3574 | return -EINVAL; | |
8f964525 | 3575 | |
dc9ce71e SC |
3576 | if (slots->generation != ghc->generation) { |
3577 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3578 | return -EFAULT; | |
3579 | } | |
8f964525 | 3580 | |
e03b644f GN |
3581 | if (kvm_is_error_hva(ghc->hva)) |
3582 | return -EFAULT; | |
3583 | ||
fcfbc617 | 3584 | if (unlikely(!ghc->memslot)) |
0958f0ce | 3585 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 3586 | |
0958f0ce | 3587 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
3588 | if (r) |
3589 | return -EFAULT; | |
3590 | ||
3591 | return 0; | |
3592 | } | |
0958f0ce VK |
3593 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
3594 | ||
3595 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
3596 | void *data, unsigned long len) | |
3597 | { | |
3598 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
3599 | } | |
4e335d9e | 3600 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 3601 | |
195aefde IE |
3602 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
3603 | { | |
2f541442 | 3604 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
3605 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3606 | int seg; | |
3607 | int offset = offset_in_page(gpa); | |
3608 | int ret; | |
3609 | ||
bfda0e84 | 3610 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 3611 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
3612 | if (ret < 0) |
3613 | return ret; | |
3614 | offset = 0; | |
3615 | len -= seg; | |
3616 | ++gfn; | |
3617 | } | |
3618 | return 0; | |
3619 | } | |
3620 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
3621 | ||
28bd726a | 3622 | void mark_page_dirty_in_slot(struct kvm *kvm, |
8283e36a | 3623 | const struct kvm_memory_slot *memslot, |
28bd726a | 3624 | gfn_t gfn) |
6aa8b732 | 3625 | { |
2efd61a6 DW |
3626 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); |
3627 | ||
e09fccb5 | 3628 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
86bdf3eb | 3629 | if (WARN_ON_ONCE(vcpu && vcpu->kvm != kvm)) |
2efd61a6 | 3630 | return; |
86bdf3eb | 3631 | |
c57351a7 | 3632 | WARN_ON_ONCE(!vcpu && !kvm_arch_allow_write_without_running_vcpu(kvm)); |
e09fccb5 | 3633 | #endif |
2efd61a6 | 3634 | |
044c59c4 | 3635 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 3636 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 3637 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 3638 | |
86bdf3eb | 3639 | if (kvm->dirty_ring_size && vcpu) |
cf87ac73 | 3640 | kvm_dirty_ring_push(vcpu, slot, rel_gfn); |
c57351a7 | 3641 | else if (memslot->dirty_bitmap) |
fb04a1ed | 3642 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
3643 | } |
3644 | } | |
a6a0b05d | 3645 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 3646 | |
49c7754c GN |
3647 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
3648 | { | |
3649 | struct kvm_memory_slot *memslot; | |
3650 | ||
3651 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 3652 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 3653 | } |
2ba9f0d8 | 3654 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 3655 | |
8e73485c PB |
3656 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
3657 | { | |
3658 | struct kvm_memory_slot *memslot; | |
3659 | ||
3660 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 3661 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
3662 | } |
3663 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
3664 | ||
20b7035c JS |
3665 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
3666 | { | |
3667 | if (!vcpu->sigset_active) | |
3668 | return; | |
3669 | ||
3670 | /* | |
3671 | * This does a lockless modification of ->real_blocked, which is fine | |
3672 | * because, only current can change ->real_blocked and all readers of | |
3673 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
3674 | * of ->blocked. | |
3675 | */ | |
3676 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
3677 | } | |
3678 | ||
3679 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
3680 | { | |
3681 | if (!vcpu->sigset_active) | |
3682 | return; | |
3683 | ||
3684 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
3685 | sigemptyset(¤t->real_blocked); | |
3686 | } | |
3687 | ||
aca6ff29 WL |
3688 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
3689 | { | |
dee339b5 | 3690 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 3691 | |
2cbd7824 | 3692 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 3693 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3694 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
3695 | if (!grow) |
3696 | goto out; | |
3697 | ||
dee339b5 NW |
3698 | val *= grow; |
3699 | if (val < grow_start) | |
3700 | val = grow_start; | |
aca6ff29 WL |
3701 | |
3702 | vcpu->halt_poll_ns = val; | |
7fa08e71 | 3703 | out: |
2cbd7824 | 3704 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3705 | } |
3706 | ||
3707 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
3708 | { | |
ae232ea4 | 3709 | unsigned int old, val, shrink, grow_start; |
aca6ff29 | 3710 | |
2cbd7824 | 3711 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 3712 | shrink = READ_ONCE(halt_poll_ns_shrink); |
ae232ea4 | 3713 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3714 | if (shrink == 0) |
aca6ff29 WL |
3715 | val = 0; |
3716 | else | |
6b6de68c | 3717 | val /= shrink; |
aca6ff29 | 3718 | |
ae232ea4 SS |
3719 | if (val < grow_start) |
3720 | val = 0; | |
3721 | ||
aca6ff29 | 3722 | vcpu->halt_poll_ns = val; |
2cbd7824 | 3723 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3724 | } |
3725 | ||
f7819512 PB |
3726 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
3727 | { | |
50c28f21 JS |
3728 | int ret = -EINTR; |
3729 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
3730 | ||
c59fb127 | 3731 | if (kvm_arch_vcpu_runnable(vcpu)) |
50c28f21 | 3732 | goto out; |
f7819512 | 3733 | if (kvm_cpu_has_pending_timer(vcpu)) |
50c28f21 | 3734 | goto out; |
f7819512 | 3735 | if (signal_pending(current)) |
50c28f21 | 3736 | goto out; |
084071d5 MT |
3737 | if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu)) |
3738 | goto out; | |
f7819512 | 3739 | |
50c28f21 JS |
3740 | ret = 0; |
3741 | out: | |
3742 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
3743 | return ret; | |
f7819512 PB |
3744 | } |
3745 | ||
fac42688 SC |
3746 | /* |
3747 | * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is | |
3748 | * pending. This is mostly used when halting a vCPU, but may also be used | |
3749 | * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI. | |
3750 | */ | |
3751 | bool kvm_vcpu_block(struct kvm_vcpu *vcpu) | |
cb953129 | 3752 | { |
fac42688 SC |
3753 | struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); |
3754 | bool waited = false; | |
3755 | ||
c3858335 JZ |
3756 | vcpu->stat.generic.blocking = 1; |
3757 | ||
18869f26 | 3758 | preempt_disable(); |
fac42688 | 3759 | kvm_arch_vcpu_blocking(vcpu); |
fac42688 | 3760 | prepare_to_rcuwait(wait); |
18869f26 ML |
3761 | preempt_enable(); |
3762 | ||
fac42688 SC |
3763 | for (;;) { |
3764 | set_current_state(TASK_INTERRUPTIBLE); | |
3765 | ||
3766 | if (kvm_vcpu_check_block(vcpu) < 0) | |
3767 | break; | |
3768 | ||
3769 | waited = true; | |
3770 | schedule(); | |
3771 | } | |
fac42688 | 3772 | |
18869f26 ML |
3773 | preempt_disable(); |
3774 | finish_rcuwait(wait); | |
fac42688 | 3775 | kvm_arch_vcpu_unblocking(vcpu); |
18869f26 | 3776 | preempt_enable(); |
fac42688 | 3777 | |
c3858335 JZ |
3778 | vcpu->stat.generic.blocking = 0; |
3779 | ||
fac42688 SC |
3780 | return waited; |
3781 | } | |
3782 | ||
29e72893 SC |
3783 | static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start, |
3784 | ktime_t end, bool success) | |
cb953129 | 3785 | { |
30c94347 | 3786 | struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic; |
29e72893 SC |
3787 | u64 poll_ns = ktime_to_ns(ktime_sub(end, start)); |
3788 | ||
30c94347 SC |
3789 | ++vcpu->stat.generic.halt_attempted_poll; |
3790 | ||
3791 | if (success) { | |
3792 | ++vcpu->stat.generic.halt_successful_poll; | |
3793 | ||
3794 | if (!vcpu_valid_wakeup(vcpu)) | |
3795 | ++vcpu->stat.generic.halt_poll_invalid; | |
3796 | ||
3797 | stats->halt_poll_success_ns += poll_ns; | |
3798 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns); | |
3799 | } else { | |
3800 | stats->halt_poll_fail_ns += poll_ns; | |
3801 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns); | |
3802 | } | |
cb953129 DM |
3803 | } |
3804 | ||
175d5dc7 DM |
3805 | static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu *vcpu) |
3806 | { | |
9eb8ca04 DM |
3807 | struct kvm *kvm = vcpu->kvm; |
3808 | ||
3809 | if (kvm->override_halt_poll_ns) { | |
3810 | /* | |
3811 | * Ensure kvm->max_halt_poll_ns is not read before | |
3812 | * kvm->override_halt_poll_ns. | |
3813 | * | |
3814 | * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL. | |
3815 | */ | |
3816 | smp_rmb(); | |
3817 | return READ_ONCE(kvm->max_halt_poll_ns); | |
3818 | } | |
3819 | ||
3820 | return READ_ONCE(halt_poll_ns); | |
175d5dc7 DM |
3821 | } |
3822 | ||
b6958ce4 | 3823 | /* |
fac42688 SC |
3824 | * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt |
3825 | * polling is enabled, busy wait for a short time before blocking to avoid the | |
3826 | * expensive block+unblock sequence if a wake event arrives soon after the vCPU | |
3827 | * is halted. | |
b6958ce4 | 3828 | */ |
91b99ea7 | 3829 | void kvm_vcpu_halt(struct kvm_vcpu *vcpu) |
d3bef15f | 3830 | { |
175d5dc7 | 3831 | unsigned int max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu); |
6f390916 | 3832 | bool halt_poll_allowed = !kvm_arch_no_poll(vcpu); |
cb953129 | 3833 | ktime_t start, cur, poll_end; |
f7819512 | 3834 | bool waited = false; |
97b6847a | 3835 | bool do_halt_poll; |
91b99ea7 | 3836 | u64 halt_ns; |
07ab0f8d | 3837 | |
175d5dc7 DM |
3838 | if (vcpu->halt_poll_ns > max_halt_poll_ns) |
3839 | vcpu->halt_poll_ns = max_halt_poll_ns; | |
97b6847a DM |
3840 | |
3841 | do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns; | |
3842 | ||
cb953129 | 3843 | start = cur = poll_end = ktime_get(); |
8df6a61c | 3844 | if (do_halt_poll) { |
109a9826 | 3845 | ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns); |
f95ef0cd | 3846 | |
f7819512 | 3847 | do { |
30c94347 | 3848 | if (kvm_vcpu_check_block(vcpu) < 0) |
f7819512 | 3849 | goto out; |
74775654 | 3850 | cpu_relax(); |
cb953129 | 3851 | poll_end = cur = ktime_get(); |
6bd5b743 | 3852 | } while (kvm_vcpu_can_poll(cur, stop)); |
f7819512 | 3853 | } |
e5c239cf | 3854 | |
fac42688 | 3855 | waited = kvm_vcpu_block(vcpu); |
8ccba534 | 3856 | |
f7819512 | 3857 | cur = ktime_get(); |
87bcc5fa JZ |
3858 | if (waited) { |
3859 | vcpu->stat.generic.halt_wait_ns += | |
3860 | ktime_to_ns(cur) - ktime_to_ns(poll_end); | |
8ccba534 JZ |
3861 | KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist, |
3862 | ktime_to_ns(cur) - ktime_to_ns(poll_end)); | |
87bcc5fa | 3863 | } |
f7819512 | 3864 | out: |
91b99ea7 SC |
3865 | /* The total time the vCPU was "halted", including polling time. */ |
3866 | halt_ns = ktime_to_ns(cur) - ktime_to_ns(start); | |
aca6ff29 | 3867 | |
29e72893 SC |
3868 | /* |
3869 | * Note, halt-polling is considered successful so long as the vCPU was | |
3870 | * never actually scheduled out, i.e. even if the wake event arrived | |
3871 | * after of the halt-polling loop itself, but before the full wait. | |
3872 | */ | |
8df6a61c | 3873 | if (do_halt_poll) |
29e72893 | 3874 | update_halt_poll_stats(vcpu, start, poll_end, !waited); |
cb953129 | 3875 | |
6f390916 | 3876 | if (halt_poll_allowed) { |
175d5dc7 DM |
3877 | /* Recompute the max halt poll time in case it changed. */ |
3878 | max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu); | |
3879 | ||
44551b2f | 3880 | if (!vcpu_valid_wakeup(vcpu)) { |
aca6ff29 | 3881 | shrink_halt_poll_ns(vcpu); |
175d5dc7 | 3882 | } else if (max_halt_poll_ns) { |
91b99ea7 | 3883 | if (halt_ns <= vcpu->halt_poll_ns) |
44551b2f WL |
3884 | ; |
3885 | /* we had a long block, shrink polling */ | |
acd05785 | 3886 | else if (vcpu->halt_poll_ns && |
175d5dc7 | 3887 | halt_ns > max_halt_poll_ns) |
44551b2f WL |
3888 | shrink_halt_poll_ns(vcpu); |
3889 | /* we had a short halt and our poll time is too small */ | |
175d5dc7 DM |
3890 | else if (vcpu->halt_poll_ns < max_halt_poll_ns && |
3891 | halt_ns < max_halt_poll_ns) | |
44551b2f WL |
3892 | grow_halt_poll_ns(vcpu); |
3893 | } else { | |
3894 | vcpu->halt_poll_ns = 0; | |
3895 | } | |
3896 | } | |
aca6ff29 | 3897 | |
91b99ea7 | 3898 | trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu)); |
b6958ce4 | 3899 | } |
91b99ea7 | 3900 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
b6958ce4 | 3901 | |
178f02ff | 3902 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 3903 | { |
d92a5d1c | 3904 | if (__kvm_vcpu_wake_up(vcpu)) { |
d73eb57b | 3905 | WRITE_ONCE(vcpu->ready, true); |
0193cc90 | 3906 | ++vcpu->stat.generic.halt_wakeup; |
178f02ff | 3907 | return true; |
b6d33834 CD |
3908 | } |
3909 | ||
178f02ff | 3910 | return false; |
dd1a4cc1 RK |
3911 | } |
3912 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
3913 | ||
0266c894 | 3914 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
3915 | /* |
3916 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
3917 | */ | |
3918 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
3919 | { | |
85b64045 | 3920 | int me, cpu; |
dd1a4cc1 | 3921 | |
178f02ff RK |
3922 | if (kvm_vcpu_wake_up(vcpu)) |
3923 | return; | |
3924 | ||
aefdc2ed PB |
3925 | me = get_cpu(); |
3926 | /* | |
3927 | * The only state change done outside the vcpu mutex is IN_GUEST_MODE | |
3928 | * to EXITING_GUEST_MODE. Therefore the moderately expensive "should | |
3929 | * kick" check does not need atomic operations if kvm_vcpu_kick is used | |
3930 | * within the vCPU thread itself. | |
3931 | */ | |
3932 | if (vcpu == __this_cpu_read(kvm_running_vcpu)) { | |
3933 | if (vcpu->mode == IN_GUEST_MODE) | |
3934 | WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE); | |
3935 | goto out; | |
3936 | } | |
3937 | ||
85b64045 SC |
3938 | /* |
3939 | * Note, the vCPU could get migrated to a different pCPU at any point | |
3940 | * after kvm_arch_vcpu_should_kick(), which could result in sending an | |
3941 | * IPI to the previous pCPU. But, that's ok because the purpose of the | |
3942 | * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the | |
3943 | * vCPU also requires it to leave IN_GUEST_MODE. | |
3944 | */ | |
85b64045 SC |
3945 | if (kvm_arch_vcpu_should_kick(vcpu)) { |
3946 | cpu = READ_ONCE(vcpu->cpu); | |
3947 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
b6d33834 | 3948 | smp_send_reschedule(cpu); |
85b64045 | 3949 | } |
aefdc2ed | 3950 | out: |
b6d33834 CD |
3951 | put_cpu(); |
3952 | } | |
a20ed54d | 3953 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 3954 | #endif /* !CONFIG_S390 */ |
b6d33834 | 3955 | |
fa93384f | 3956 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
3957 | { |
3958 | struct pid *pid; | |
3959 | struct task_struct *task = NULL; | |
fa93384f | 3960 | int ret = 0; |
41628d33 KW |
3961 | |
3962 | rcu_read_lock(); | |
3963 | pid = rcu_dereference(target->pid); | |
3964 | if (pid) | |
27fbe64b | 3965 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
3966 | rcu_read_unlock(); |
3967 | if (!task) | |
c45c528e | 3968 | return ret; |
c45c528e | 3969 | ret = yield_to(task, 1); |
41628d33 | 3970 | put_task_struct(task); |
c45c528e R |
3971 | |
3972 | return ret; | |
41628d33 KW |
3973 | } |
3974 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
3975 | ||
06e48c51 R |
3976 | /* |
3977 | * Helper that checks whether a VCPU is eligible for directed yield. | |
3978 | * Most eligible candidate to yield is decided by following heuristics: | |
3979 | * | |
3980 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
3981 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 3982 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
3983 | * |
3984 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
3985 | * chance last time (mostly it has become eligible now since we have probably | |
3986 | * yielded to lockholder in last iteration. This is done by toggling | |
3987 | * @dy_eligible each time a VCPU checked for eligibility.) | |
3988 | * | |
3989 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
3990 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
3991 | * burning. Giving priority for a potential lock-holder increases lock | |
3992 | * progress. | |
3993 | * | |
3994 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
3995 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
3996 | * and continue with next VCPU and so on. | |
3997 | */ | |
7940876e | 3998 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 3999 | { |
4a55dd72 | 4000 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
4001 | bool eligible; |
4002 | ||
4003 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 4004 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
4005 | |
4006 | if (vcpu->spin_loop.in_spin_loop) | |
4007 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
4008 | ||
4009 | return eligible; | |
4a55dd72 SW |
4010 | #else |
4011 | return true; | |
06e48c51 | 4012 | #endif |
4a55dd72 | 4013 | } |
c45c528e | 4014 | |
17e433b5 WL |
4015 | /* |
4016 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
4017 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
4018 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
4019 | */ | |
4020 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
4021 | { | |
4022 | return kvm_arch_vcpu_runnable(vcpu); | |
4023 | } | |
4024 | ||
4025 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
4026 | { | |
4027 | if (kvm_arch_dy_runnable(vcpu)) | |
4028 | return true; | |
4029 | ||
4030 | #ifdef CONFIG_KVM_ASYNC_PF | |
4031 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
4032 | return true; | |
4033 | #endif | |
4034 | ||
4035 | return false; | |
4036 | } | |
4037 | ||
77bcd9e6 SC |
4038 | /* |
4039 | * By default, simply query the target vCPU's current mode when checking if a | |
4040 | * vCPU was preempted in kernel mode. All architectures except x86 (or more | |
4041 | * specifical, except VMX) allow querying whether or not a vCPU is in kernel | |
4042 | * mode even if the vCPU is NOT loaded, i.e. using kvm_arch_vcpu_in_kernel() | |
4043 | * directly for cross-vCPU checks is functionally correct and accurate. | |
4044 | */ | |
4045 | bool __weak kvm_arch_vcpu_preempted_in_kernel(struct kvm_vcpu *vcpu) | |
4046 | { | |
4047 | return kvm_arch_vcpu_in_kernel(vcpu); | |
4048 | } | |
4049 | ||
52acd22f WL |
4050 | bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
4051 | { | |
4052 | return false; | |
4053 | } | |
4054 | ||
199b5763 | 4055 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 4056 | { |
217ece61 RR |
4057 | struct kvm *kvm = me->kvm; |
4058 | struct kvm_vcpu *vcpu; | |
4059 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
46808a4c | 4060 | unsigned long i; |
217ece61 | 4061 | int yielded = 0; |
c45c528e | 4062 | int try = 3; |
217ece61 | 4063 | int pass; |
d255f4f2 | 4064 | |
4c088493 | 4065 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
4066 | /* |
4067 | * We boost the priority of a VCPU that is runnable but not | |
4068 | * currently running, because it got preempted by something | |
4069 | * else and called schedule in __vcpu_run. Hopefully that | |
4070 | * VCPU is holding the lock that we need and will release it. | |
4071 | * We approximate round-robin by starting at the last boosted VCPU. | |
4072 | */ | |
c45c528e | 4073 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 4074 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 4075 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
4076 | i = last_boosted_vcpu; |
4077 | continue; | |
4078 | } else if (pass && i > last_boosted_vcpu) | |
4079 | break; | |
d73eb57b | 4080 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 4081 | continue; |
217ece61 RR |
4082 | if (vcpu == me) |
4083 | continue; | |
d92a5d1c | 4084 | if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu)) |
217ece61 | 4085 | continue; |
dafc17dd SC |
4086 | |
4087 | /* | |
4088 | * Treat the target vCPU as being in-kernel if it has a | |
4089 | * pending interrupt, as the vCPU trying to yield may | |
4090 | * be spinning waiting on IPI delivery, i.e. the target | |
4091 | * vCPU is in-kernel for the purposes of directed yield. | |
4092 | */ | |
046ddeed | 4093 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
52acd22f | 4094 | !kvm_arch_dy_has_pending_interrupt(vcpu) && |
77bcd9e6 | 4095 | !kvm_arch_vcpu_preempted_in_kernel(vcpu)) |
199b5763 | 4096 | continue; |
06e48c51 R |
4097 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
4098 | continue; | |
c45c528e R |
4099 | |
4100 | yielded = kvm_vcpu_yield_to(vcpu); | |
4101 | if (yielded > 0) { | |
217ece61 | 4102 | kvm->last_boosted_vcpu = i; |
217ece61 | 4103 | break; |
c45c528e R |
4104 | } else if (yielded < 0) { |
4105 | try--; | |
4106 | if (!try) | |
4107 | break; | |
217ece61 | 4108 | } |
217ece61 RR |
4109 | } |
4110 | } | |
4c088493 | 4111 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
4112 | |
4113 | /* Ensure vcpu is not eligible during next spinloop */ | |
4114 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
4115 | } |
4116 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
4117 | ||
fb04a1ed PX |
4118 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
4119 | { | |
dc70ec21 | 4120 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
fb04a1ed PX |
4121 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && |
4122 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
4123 | kvm->dirty_ring_size / PAGE_SIZE); | |
4124 | #else | |
4125 | return false; | |
4126 | #endif | |
4127 | } | |
4128 | ||
1499fa80 | 4129 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 4130 | { |
11bac800 | 4131 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
4132 | struct page *page; |
4133 | ||
e4a533a4 | 4134 | if (vmf->pgoff == 0) |
039576c0 | 4135 | page = virt_to_page(vcpu->run); |
09566765 | 4136 | #ifdef CONFIG_X86 |
e4a533a4 | 4137 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 4138 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 4139 | #endif |
4b4357e0 | 4140 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
4141 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
4142 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 4143 | #endif |
fb04a1ed PX |
4144 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
4145 | page = kvm_dirty_ring_get_page( | |
4146 | &vcpu->dirty_ring, | |
4147 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 4148 | else |
5b1c1493 | 4149 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 4150 | get_page(page); |
e4a533a4 | 4151 | vmf->page = page; |
4152 | return 0; | |
9a2bb7f4 AK |
4153 | } |
4154 | ||
f0f37e2f | 4155 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 4156 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
4157 | }; |
4158 | ||
4159 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
4160 | { | |
fb04a1ed | 4161 | struct kvm_vcpu *vcpu = file->private_data; |
11476d27 | 4162 | unsigned long pages = vma_pages(vma); |
fb04a1ed PX |
4163 | |
4164 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
4165 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
4166 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
4167 | return -EINVAL; | |
4168 | ||
9a2bb7f4 AK |
4169 | vma->vm_ops = &kvm_vcpu_vm_ops; |
4170 | return 0; | |
4171 | } | |
4172 | ||
bccf2150 AK |
4173 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
4174 | { | |
4175 | struct kvm_vcpu *vcpu = filp->private_data; | |
4176 | ||
66c0b394 | 4177 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
4178 | return 0; |
4179 | } | |
4180 | ||
087e1520 | 4181 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
4182 | .release = kvm_vcpu_release, |
4183 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 4184 | .mmap = kvm_vcpu_mmap, |
6038f373 | 4185 | .llseek = noop_llseek, |
7ddfd3e0 | 4186 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
4187 | }; |
4188 | ||
4189 | /* | |
4190 | * Allocates an inode for the vcpu. | |
4191 | */ | |
4192 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
4193 | { | |
e46b4692 MY |
4194 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
4195 | ||
4196 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
4197 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
4198 | } |
4199 | ||
e36de87d VP |
4200 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
4201 | static int vcpu_get_pid(void *data, u64 *val) | |
4202 | { | |
14aa40a1 | 4203 | struct kvm_vcpu *vcpu = data; |
76021e96 SC |
4204 | |
4205 | rcu_read_lock(); | |
4206 | *val = pid_nr(rcu_dereference(vcpu->pid)); | |
4207 | rcu_read_unlock(); | |
e36de87d VP |
4208 | return 0; |
4209 | } | |
4210 | ||
4211 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n"); | |
4212 | ||
3e7093d0 | 4213 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 4214 | { |
d56f5136 | 4215 | struct dentry *debugfs_dentry; |
45b5939e | 4216 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 4217 | |
45b5939e | 4218 | if (!debugfs_initialized()) |
3e7093d0 | 4219 | return; |
45b5939e LC |
4220 | |
4221 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
4222 | debugfs_dentry = debugfs_create_dir(dir_name, |
4223 | vcpu->kvm->debugfs_dentry); | |
e36de87d VP |
4224 | debugfs_create_file("pid", 0444, debugfs_dentry, vcpu, |
4225 | &vcpu_get_pid_fops); | |
45b5939e | 4226 | |
d56f5136 | 4227 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
45b5939e | 4228 | } |
e36de87d | 4229 | #endif |
45b5939e | 4230 | |
c5ea7660 AK |
4231 | /* |
4232 | * Creates some virtual cpus. Good luck creating more than one. | |
4233 | */ | |
73880c80 | 4234 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
4235 | { |
4236 | int r; | |
e09fefde | 4237 | struct kvm_vcpu *vcpu; |
8bd826d6 | 4238 | struct page *page; |
c5ea7660 | 4239 | |
a1c42dde | 4240 | if (id >= KVM_MAX_VCPU_IDS) |
338c7dba AH |
4241 | return -EINVAL; |
4242 | ||
6c7caebc | 4243 | mutex_lock(&kvm->lock); |
f502cc56 | 4244 | if (kvm->created_vcpus >= kvm->max_vcpus) { |
6c7caebc PB |
4245 | mutex_unlock(&kvm->lock); |
4246 | return -EINVAL; | |
4247 | } | |
4248 | ||
1d5e740d ZG |
4249 | r = kvm_arch_vcpu_precreate(kvm, id); |
4250 | if (r) { | |
4251 | mutex_unlock(&kvm->lock); | |
4252 | return r; | |
4253 | } | |
4254 | ||
6c7caebc PB |
4255 | kvm->created_vcpus++; |
4256 | mutex_unlock(&kvm->lock); | |
4257 | ||
85f47930 | 4258 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
e529ef66 SC |
4259 | if (!vcpu) { |
4260 | r = -ENOMEM; | |
6c7caebc PB |
4261 | goto vcpu_decrement; |
4262 | } | |
c5ea7660 | 4263 | |
fcd97ad5 | 4264 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 4265 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
4266 | if (!page) { |
4267 | r = -ENOMEM; | |
e529ef66 | 4268 | goto vcpu_free; |
8bd826d6 SC |
4269 | } |
4270 | vcpu->run = page_address(page); | |
4271 | ||
4272 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
4273 | |
4274 | r = kvm_arch_vcpu_create(vcpu); | |
4275 | if (r) | |
8bd826d6 | 4276 | goto vcpu_free_run_page; |
e529ef66 | 4277 | |
fb04a1ed PX |
4278 | if (kvm->dirty_ring_size) { |
4279 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
4280 | id, kvm->dirty_ring_size); | |
4281 | if (r) | |
4282 | goto arch_vcpu_destroy; | |
4283 | } | |
4284 | ||
11ec2804 | 4285 | mutex_lock(&kvm->lock); |
42a90008 DW |
4286 | |
4287 | #ifdef CONFIG_LOCKDEP | |
4288 | /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */ | |
4289 | mutex_lock(&vcpu->mutex); | |
4290 | mutex_unlock(&vcpu->mutex); | |
4291 | #endif | |
4292 | ||
e09fefde DH |
4293 | if (kvm_get_vcpu_by_id(kvm, id)) { |
4294 | r = -EEXIST; | |
4295 | goto unlock_vcpu_destroy; | |
4296 | } | |
73880c80 | 4297 | |
8750e72a | 4298 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
afb2acb2 | 4299 | r = xa_reserve(&kvm->vcpu_array, vcpu->vcpu_idx, GFP_KERNEL_ACCOUNT); |
c5b07754 MZ |
4300 | if (r) |
4301 | goto unlock_vcpu_destroy; | |
c5ea7660 | 4302 | |
fb3f0f51 | 4303 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 4304 | kvm_get_kvm(kvm); |
bccf2150 | 4305 | r = create_vcpu_fd(vcpu); |
afb2acb2 ML |
4306 | if (r < 0) |
4307 | goto kvm_put_xa_release; | |
4308 | ||
5f643e46 | 4309 | if (KVM_BUG_ON(xa_store(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, 0), kvm)) { |
afb2acb2 ML |
4310 | r = -EINVAL; |
4311 | goto kvm_put_xa_release; | |
73880c80 GN |
4312 | } |
4313 | ||
dd489240 | 4314 | /* |
c5b07754 MZ |
4315 | * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu |
4316 | * pointer before kvm->online_vcpu's incremented value. | |
dd489240 | 4317 | */ |
73880c80 GN |
4318 | smp_wmb(); |
4319 | atomic_inc(&kvm->online_vcpus); | |
4320 | ||
73880c80 | 4321 | mutex_unlock(&kvm->lock); |
42897d86 | 4322 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 4323 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 4324 | return r; |
39c3b86e | 4325 | |
afb2acb2 ML |
4326 | kvm_put_xa_release: |
4327 | kvm_put_kvm_no_destroy(kvm); | |
4328 | xa_release(&kvm->vcpu_array, vcpu->vcpu_idx); | |
d780592b | 4329 | unlock_vcpu_destroy: |
7d8fece6 | 4330 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
4331 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
4332 | arch_vcpu_destroy: | |
d40ccc62 | 4333 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
4334 | vcpu_free_run_page: |
4335 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
4336 | vcpu_free: |
4337 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
4338 | vcpu_decrement: |
4339 | mutex_lock(&kvm->lock); | |
4340 | kvm->created_vcpus--; | |
4341 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
4342 | return r; |
4343 | } | |
4344 | ||
1961d276 AK |
4345 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
4346 | { | |
4347 | if (sigset) { | |
4348 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
4349 | vcpu->sigset_active = 1; | |
4350 | vcpu->sigset = *sigset; | |
4351 | } else | |
4352 | vcpu->sigset_active = 0; | |
4353 | return 0; | |
4354 | } | |
4355 | ||
ce55c049 JZ |
4356 | static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer, |
4357 | size_t size, loff_t *offset) | |
4358 | { | |
4359 | struct kvm_vcpu *vcpu = file->private_data; | |
4360 | ||
4361 | return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header, | |
4362 | &kvm_vcpu_stats_desc[0], &vcpu->stat, | |
4363 | sizeof(vcpu->stat), user_buffer, size, offset); | |
4364 | } | |
4365 | ||
eed3013f SC |
4366 | static int kvm_vcpu_stats_release(struct inode *inode, struct file *file) |
4367 | { | |
4368 | struct kvm_vcpu *vcpu = file->private_data; | |
4369 | ||
4370 | kvm_put_kvm(vcpu->kvm); | |
4371 | return 0; | |
4372 | } | |
4373 | ||
ce55c049 | 4374 | static const struct file_operations kvm_vcpu_stats_fops = { |
087e1520 | 4375 | .owner = THIS_MODULE, |
ce55c049 | 4376 | .read = kvm_vcpu_stats_read, |
eed3013f | 4377 | .release = kvm_vcpu_stats_release, |
ce55c049 JZ |
4378 | .llseek = noop_llseek, |
4379 | }; | |
4380 | ||
4381 | static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu) | |
4382 | { | |
4383 | int fd; | |
4384 | struct file *file; | |
4385 | char name[15 + ITOA_MAX_LEN + 1]; | |
4386 | ||
4387 | snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id); | |
4388 | ||
4389 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4390 | if (fd < 0) | |
4391 | return fd; | |
4392 | ||
4393 | file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY); | |
4394 | if (IS_ERR(file)) { | |
4395 | put_unused_fd(fd); | |
4396 | return PTR_ERR(file); | |
4397 | } | |
eed3013f SC |
4398 | |
4399 | kvm_get_kvm(vcpu->kvm); | |
4400 | ||
ce55c049 JZ |
4401 | file->f_mode |= FMODE_PREAD; |
4402 | fd_install(fd, file); | |
4403 | ||
4404 | return fd; | |
4405 | } | |
4406 | ||
bccf2150 AK |
4407 | static long kvm_vcpu_ioctl(struct file *filp, |
4408 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 4409 | { |
bccf2150 | 4410 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 4411 | void __user *argp = (void __user *)arg; |
313a3dc7 | 4412 | int r; |
fa3795a7 DH |
4413 | struct kvm_fpu *fpu = NULL; |
4414 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 4415 | |
f4d31653 | 4416 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
6d4e4c4f | 4417 | return -EIO; |
2122ff5e | 4418 | |
2ea75be3 DM |
4419 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
4420 | return -EINVAL; | |
4421 | ||
2122ff5e | 4422 | /* |
5cb0944c PB |
4423 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
4424 | * execution; mutex_lock() would break them. | |
2122ff5e | 4425 | */ |
5cb0944c PB |
4426 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
4427 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 4428 | return r; |
2122ff5e | 4429 | |
ec7660cc CD |
4430 | if (mutex_lock_killable(&vcpu->mutex)) |
4431 | return -EINTR; | |
6aa8b732 | 4432 | switch (ioctl) { |
0e4524a5 CB |
4433 | case KVM_RUN: { |
4434 | struct pid *oldpid; | |
f0fe5108 AK |
4435 | r = -EINVAL; |
4436 | if (arg) | |
4437 | goto out; | |
0e4524a5 | 4438 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 4439 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 4440 | /* The thread running this VCPU changed. */ |
bd2a6394 | 4441 | struct pid *newpid; |
f95ef0cd | 4442 | |
bd2a6394 CD |
4443 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
4444 | if (r) | |
4445 | break; | |
4446 | ||
4447 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
4448 | rcu_assign_pointer(vcpu->pid, newpid); |
4449 | if (oldpid) | |
4450 | synchronize_rcu(); | |
4451 | put_pid(oldpid); | |
4452 | } | |
1b94f6f8 | 4453 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 4454 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 4455 | break; |
0e4524a5 | 4456 | } |
6aa8b732 | 4457 | case KVM_GET_REGS: { |
3e4bb3ac | 4458 | struct kvm_regs *kvm_regs; |
6aa8b732 | 4459 | |
3e4bb3ac | 4460 | r = -ENOMEM; |
b12ce36a | 4461 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 4462 | if (!kvm_regs) |
6aa8b732 | 4463 | goto out; |
3e4bb3ac XZ |
4464 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
4465 | if (r) | |
4466 | goto out_free1; | |
6aa8b732 | 4467 | r = -EFAULT; |
3e4bb3ac XZ |
4468 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
4469 | goto out_free1; | |
6aa8b732 | 4470 | r = 0; |
3e4bb3ac XZ |
4471 | out_free1: |
4472 | kfree(kvm_regs); | |
6aa8b732 AK |
4473 | break; |
4474 | } | |
4475 | case KVM_SET_REGS: { | |
3e4bb3ac | 4476 | struct kvm_regs *kvm_regs; |
6aa8b732 | 4477 | |
ff5c2c03 SL |
4478 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
4479 | if (IS_ERR(kvm_regs)) { | |
4480 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 4481 | goto out; |
ff5c2c03 | 4482 | } |
3e4bb3ac | 4483 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 4484 | kfree(kvm_regs); |
6aa8b732 AK |
4485 | break; |
4486 | } | |
4487 | case KVM_GET_SREGS: { | |
b12ce36a BG |
4488 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
4489 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
4490 | r = -ENOMEM; |
4491 | if (!kvm_sregs) | |
4492 | goto out; | |
4493 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
4494 | if (r) |
4495 | goto out; | |
4496 | r = -EFAULT; | |
fa3795a7 | 4497 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
4498 | goto out; |
4499 | r = 0; | |
4500 | break; | |
4501 | } | |
4502 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
4503 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
4504 | if (IS_ERR(kvm_sregs)) { | |
4505 | r = PTR_ERR(kvm_sregs); | |
18595411 | 4506 | kvm_sregs = NULL; |
6aa8b732 | 4507 | goto out; |
ff5c2c03 | 4508 | } |
fa3795a7 | 4509 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
4510 | break; |
4511 | } | |
62d9f0db MT |
4512 | case KVM_GET_MP_STATE: { |
4513 | struct kvm_mp_state mp_state; | |
4514 | ||
4515 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
4516 | if (r) | |
4517 | goto out; | |
4518 | r = -EFAULT; | |
893bdbf1 | 4519 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
4520 | goto out; |
4521 | r = 0; | |
4522 | break; | |
4523 | } | |
4524 | case KVM_SET_MP_STATE: { | |
4525 | struct kvm_mp_state mp_state; | |
4526 | ||
4527 | r = -EFAULT; | |
893bdbf1 | 4528 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
4529 | goto out; |
4530 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
4531 | break; |
4532 | } | |
6aa8b732 AK |
4533 | case KVM_TRANSLATE: { |
4534 | struct kvm_translation tr; | |
4535 | ||
4536 | r = -EFAULT; | |
893bdbf1 | 4537 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 4538 | goto out; |
8b006791 | 4539 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
4540 | if (r) |
4541 | goto out; | |
4542 | r = -EFAULT; | |
893bdbf1 | 4543 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
4544 | goto out; |
4545 | r = 0; | |
4546 | break; | |
4547 | } | |
d0bfb940 JK |
4548 | case KVM_SET_GUEST_DEBUG: { |
4549 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
4550 | |
4551 | r = -EFAULT; | |
893bdbf1 | 4552 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 4553 | goto out; |
d0bfb940 | 4554 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
4555 | break; |
4556 | } | |
1961d276 AK |
4557 | case KVM_SET_SIGNAL_MASK: { |
4558 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4559 | struct kvm_signal_mask kvm_sigmask; | |
4560 | sigset_t sigset, *p; | |
4561 | ||
4562 | p = NULL; | |
4563 | if (argp) { | |
4564 | r = -EFAULT; | |
4565 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4566 | sizeof(kvm_sigmask))) |
1961d276 AK |
4567 | goto out; |
4568 | r = -EINVAL; | |
893bdbf1 | 4569 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
4570 | goto out; |
4571 | r = -EFAULT; | |
4572 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 4573 | sizeof(sigset))) |
1961d276 AK |
4574 | goto out; |
4575 | p = &sigset; | |
4576 | } | |
376d41ff | 4577 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
4578 | break; |
4579 | } | |
b8836737 | 4580 | case KVM_GET_FPU: { |
b12ce36a | 4581 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
4582 | r = -ENOMEM; |
4583 | if (!fpu) | |
4584 | goto out; | |
4585 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
4586 | if (r) |
4587 | goto out; | |
4588 | r = -EFAULT; | |
fa3795a7 | 4589 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
4590 | goto out; |
4591 | r = 0; | |
4592 | break; | |
4593 | } | |
4594 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
4595 | fpu = memdup_user(argp, sizeof(*fpu)); |
4596 | if (IS_ERR(fpu)) { | |
4597 | r = PTR_ERR(fpu); | |
18595411 | 4598 | fpu = NULL; |
b8836737 | 4599 | goto out; |
ff5c2c03 | 4600 | } |
fa3795a7 | 4601 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
4602 | break; |
4603 | } | |
ce55c049 JZ |
4604 | case KVM_GET_STATS_FD: { |
4605 | r = kvm_vcpu_ioctl_get_stats_fd(vcpu); | |
4606 | break; | |
4607 | } | |
bccf2150 | 4608 | default: |
313a3dc7 | 4609 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
4610 | } |
4611 | out: | |
ec7660cc | 4612 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
4613 | kfree(fpu); |
4614 | kfree(kvm_sregs); | |
bccf2150 AK |
4615 | return r; |
4616 | } | |
4617 | ||
de8e5d74 | 4618 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
4619 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
4620 | unsigned int ioctl, unsigned long arg) | |
4621 | { | |
4622 | struct kvm_vcpu *vcpu = filp->private_data; | |
4623 | void __user *argp = compat_ptr(arg); | |
4624 | int r; | |
4625 | ||
f4d31653 | 4626 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
1dda606c AG |
4627 | return -EIO; |
4628 | ||
4629 | switch (ioctl) { | |
4630 | case KVM_SET_SIGNAL_MASK: { | |
4631 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4632 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
4633 | sigset_t sigset; |
4634 | ||
4635 | if (argp) { | |
4636 | r = -EFAULT; | |
4637 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4638 | sizeof(kvm_sigmask))) |
1dda606c AG |
4639 | goto out; |
4640 | r = -EINVAL; | |
3968cf62 | 4641 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
4642 | goto out; |
4643 | r = -EFAULT; | |
1393b4aa PB |
4644 | if (get_compat_sigset(&sigset, |
4645 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 4646 | goto out; |
760a9a30 AC |
4647 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
4648 | } else | |
4649 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
4650 | break; |
4651 | } | |
4652 | default: | |
4653 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
4654 | } | |
4655 | ||
4656 | out: | |
4657 | return r; | |
4658 | } | |
4659 | #endif | |
4660 | ||
a1cd3f08 CLG |
4661 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
4662 | { | |
4663 | struct kvm_device *dev = filp->private_data; | |
4664 | ||
4665 | if (dev->ops->mmap) | |
4666 | return dev->ops->mmap(dev, vma); | |
4667 | ||
4668 | return -ENODEV; | |
4669 | } | |
4670 | ||
852b6d57 SW |
4671 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
4672 | int (*accessor)(struct kvm_device *dev, | |
4673 | struct kvm_device_attr *attr), | |
4674 | unsigned long arg) | |
4675 | { | |
4676 | struct kvm_device_attr attr; | |
4677 | ||
4678 | if (!accessor) | |
4679 | return -EPERM; | |
4680 | ||
4681 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
4682 | return -EFAULT; | |
4683 | ||
4684 | return accessor(dev, &attr); | |
4685 | } | |
4686 | ||
4687 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
4688 | unsigned long arg) | |
4689 | { | |
4690 | struct kvm_device *dev = filp->private_data; | |
4691 | ||
f4d31653 | 4692 | if (dev->kvm->mm != current->mm || dev->kvm->vm_dead) |
ddba9180 SC |
4693 | return -EIO; |
4694 | ||
852b6d57 SW |
4695 | switch (ioctl) { |
4696 | case KVM_SET_DEVICE_ATTR: | |
4697 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
4698 | case KVM_GET_DEVICE_ATTR: | |
4699 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
4700 | case KVM_HAS_DEVICE_ATTR: | |
4701 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
4702 | default: | |
4703 | if (dev->ops->ioctl) | |
4704 | return dev->ops->ioctl(dev, ioctl, arg); | |
4705 | ||
4706 | return -ENOTTY; | |
4707 | } | |
4708 | } | |
4709 | ||
852b6d57 SW |
4710 | static int kvm_device_release(struct inode *inode, struct file *filp) |
4711 | { | |
4712 | struct kvm_device *dev = filp->private_data; | |
4713 | struct kvm *kvm = dev->kvm; | |
4714 | ||
2bde9b3e CLG |
4715 | if (dev->ops->release) { |
4716 | mutex_lock(&kvm->lock); | |
4717 | list_del(&dev->vm_node); | |
4718 | dev->ops->release(dev); | |
4719 | mutex_unlock(&kvm->lock); | |
4720 | } | |
4721 | ||
852b6d57 SW |
4722 | kvm_put_kvm(kvm); |
4723 | return 0; | |
4724 | } | |
4725 | ||
087e1520 | 4726 | static struct file_operations kvm_device_fops = { |
852b6d57 SW |
4727 | .unlocked_ioctl = kvm_device_ioctl, |
4728 | .release = kvm_device_release, | |
7ddfd3e0 | 4729 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 4730 | .mmap = kvm_device_mmap, |
852b6d57 SW |
4731 | }; |
4732 | ||
4733 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
4734 | { | |
4735 | if (filp->f_op != &kvm_device_fops) | |
4736 | return NULL; | |
4737 | ||
4738 | return filp->private_data; | |
4739 | } | |
4740 | ||
8538cb22 | 4741 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 4742 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
4743 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
4744 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 4745 | #endif |
d60eacb0 WD |
4746 | }; |
4747 | ||
8538cb22 | 4748 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
4749 | { |
4750 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
4751 | return -ENOSPC; | |
4752 | ||
4753 | if (kvm_device_ops_table[type] != NULL) | |
4754 | return -EEXIST; | |
4755 | ||
4756 | kvm_device_ops_table[type] = ops; | |
4757 | return 0; | |
4758 | } | |
4759 | ||
571ee1b6 WL |
4760 | void kvm_unregister_device_ops(u32 type) |
4761 | { | |
4762 | if (kvm_device_ops_table[type] != NULL) | |
4763 | kvm_device_ops_table[type] = NULL; | |
4764 | } | |
4765 | ||
852b6d57 SW |
4766 | static int kvm_ioctl_create_device(struct kvm *kvm, |
4767 | struct kvm_create_device *cd) | |
4768 | { | |
eceb6e1d | 4769 | const struct kvm_device_ops *ops; |
852b6d57 SW |
4770 | struct kvm_device *dev; |
4771 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 4772 | int type; |
852b6d57 SW |
4773 | int ret; |
4774 | ||
d60eacb0 WD |
4775 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
4776 | return -ENODEV; | |
4777 | ||
1d487e9b PB |
4778 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
4779 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 4780 | if (ops == NULL) |
852b6d57 | 4781 | return -ENODEV; |
852b6d57 SW |
4782 | |
4783 | if (test) | |
4784 | return 0; | |
4785 | ||
b12ce36a | 4786 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
4787 | if (!dev) |
4788 | return -ENOMEM; | |
4789 | ||
4790 | dev->ops = ops; | |
4791 | dev->kvm = kvm; | |
852b6d57 | 4792 | |
a28ebea2 | 4793 | mutex_lock(&kvm->lock); |
1d487e9b | 4794 | ret = ops->create(dev, type); |
852b6d57 | 4795 | if (ret < 0) { |
a28ebea2 | 4796 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
4797 | kfree(dev); |
4798 | return ret; | |
4799 | } | |
a28ebea2 CD |
4800 | list_add(&dev->vm_node, &kvm->devices); |
4801 | mutex_unlock(&kvm->lock); | |
852b6d57 | 4802 | |
023e9fdd CD |
4803 | if (ops->init) |
4804 | ops->init(dev); | |
4805 | ||
cfa39381 | 4806 | kvm_get_kvm(kvm); |
24009b05 | 4807 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 4808 | if (ret < 0) { |
149487bd | 4809 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
4810 | mutex_lock(&kvm->lock); |
4811 | list_del(&dev->vm_node); | |
e8bc2427 AK |
4812 | if (ops->release) |
4813 | ops->release(dev); | |
a28ebea2 | 4814 | mutex_unlock(&kvm->lock); |
e8bc2427 AK |
4815 | if (ops->destroy) |
4816 | ops->destroy(dev); | |
852b6d57 SW |
4817 | return ret; |
4818 | } | |
4819 | ||
852b6d57 SW |
4820 | cd->fd = ret; |
4821 | return 0; | |
4822 | } | |
4823 | ||
f15ba52b | 4824 | static int kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
92b591a4 AG |
4825 | { |
4826 | switch (arg) { | |
4827 | case KVM_CAP_USER_MEMORY: | |
bb58b90b | 4828 | case KVM_CAP_USER_MEMORY2: |
92b591a4 AG |
4829 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: |
4830 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
4831 | case KVM_CAP_INTERNAL_ERROR_DATA: |
4832 | #ifdef CONFIG_HAVE_KVM_MSI | |
4833 | case KVM_CAP_SIGNAL_MSI: | |
4834 | #endif | |
c5b31cc2 | 4835 | #ifdef CONFIG_HAVE_KVM_IRQCHIP |
dc9be0fa | 4836 | case KVM_CAP_IRQFD: |
92b591a4 | 4837 | #endif |
e9ea5069 | 4838 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 4839 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 4840 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 4841 | case KVM_CAP_HALT_POLL: |
92b591a4 | 4842 | return 1; |
4b4357e0 | 4843 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
4844 | case KVM_CAP_COALESCED_MMIO: |
4845 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
4846 | case KVM_CAP_COALESCED_PIO: |
4847 | return 1; | |
30422558 | 4848 | #endif |
3c9bd400 JZ |
4849 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4850 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
4851 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
4852 | #endif | |
92b591a4 AG |
4853 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4854 | case KVM_CAP_IRQ_ROUTING: | |
4855 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 | 4856 | #endif |
eed52e43 | 4857 | #if KVM_MAX_NR_ADDRESS_SPACES > 1 |
f481b069 | 4858 | case KVM_CAP_MULTI_ADDRESS_SPACE: |
eed52e43 SC |
4859 | if (kvm) |
4860 | return kvm_arch_nr_memslot_as_ids(kvm); | |
4861 | return KVM_MAX_NR_ADDRESS_SPACES; | |
92b591a4 | 4862 | #endif |
c110ae57 PB |
4863 | case KVM_CAP_NR_MEMSLOTS: |
4864 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed | 4865 | case KVM_CAP_DIRTY_LOG_RING: |
17601bfe MZ |
4866 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO |
4867 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); | |
4868 | #else | |
4869 | return 0; | |
4870 | #endif | |
4871 | case KVM_CAP_DIRTY_LOG_RING_ACQ_REL: | |
4872 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL | |
fb04a1ed PX |
4873 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); |
4874 | #else | |
4875 | return 0; | |
86bdf3eb GS |
4876 | #endif |
4877 | #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP | |
4878 | case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: | |
fb04a1ed | 4879 | #endif |
ce55c049 | 4880 | case KVM_CAP_BINARY_STATS_FD: |
d495f942 | 4881 | case KVM_CAP_SYSTEM_EVENT_DATA: |
63912245 | 4882 | case KVM_CAP_DEVICE_CTRL: |
ce55c049 | 4883 | return 1; |
5a475554 CP |
4884 | #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES |
4885 | case KVM_CAP_MEMORY_ATTRIBUTES: | |
4886 | return kvm_supported_mem_attributes(kvm); | |
a7800aa8 SC |
4887 | #endif |
4888 | #ifdef CONFIG_KVM_PRIVATE_MEM | |
4889 | case KVM_CAP_GUEST_MEMFD: | |
4890 | return !kvm || kvm_arch_has_private_mem(kvm); | |
5a475554 | 4891 | #endif |
92b591a4 AG |
4892 | default: |
4893 | break; | |
4894 | } | |
4895 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
4896 | } | |
4897 | ||
fb04a1ed PX |
4898 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
4899 | { | |
4900 | int r; | |
4901 | ||
4902 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
4903 | return -EINVAL; | |
4904 | ||
4905 | /* the size should be power of 2 */ | |
4906 | if (!size || (size & (size - 1))) | |
4907 | return -EINVAL; | |
4908 | ||
4909 | /* Should be bigger to keep the reserved entries, or a page */ | |
4910 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
4911 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
4912 | return -EINVAL; | |
4913 | ||
4914 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
4915 | sizeof(struct kvm_dirty_gfn)) | |
4916 | return -E2BIG; | |
4917 | ||
4918 | /* We only allow it to set once */ | |
4919 | if (kvm->dirty_ring_size) | |
4920 | return -EINVAL; | |
4921 | ||
4922 | mutex_lock(&kvm->lock); | |
4923 | ||
4924 | if (kvm->created_vcpus) { | |
4925 | /* We don't allow to change this value after vcpu created */ | |
4926 | r = -EINVAL; | |
4927 | } else { | |
4928 | kvm->dirty_ring_size = size; | |
4929 | r = 0; | |
4930 | } | |
4931 | ||
4932 | mutex_unlock(&kvm->lock); | |
4933 | return r; | |
4934 | } | |
4935 | ||
4936 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
4937 | { | |
46808a4c | 4938 | unsigned long i; |
fb04a1ed PX |
4939 | struct kvm_vcpu *vcpu; |
4940 | int cleared = 0; | |
4941 | ||
4942 | if (!kvm->dirty_ring_size) | |
4943 | return -EINVAL; | |
4944 | ||
4945 | mutex_lock(&kvm->slots_lock); | |
4946 | ||
4947 | kvm_for_each_vcpu(i, vcpu, kvm) | |
4948 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
4949 | ||
4950 | mutex_unlock(&kvm->slots_lock); | |
4951 | ||
4952 | if (cleared) | |
4953 | kvm_flush_remote_tlbs(kvm); | |
4954 | ||
4955 | return cleared; | |
4956 | } | |
4957 | ||
e5d83c74 PB |
4958 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4959 | struct kvm_enable_cap *cap) | |
4960 | { | |
4961 | return -EINVAL; | |
4962 | } | |
4963 | ||
26f45714 | 4964 | bool kvm_are_all_memslots_empty(struct kvm *kvm) |
86bdf3eb GS |
4965 | { |
4966 | int i; | |
4967 | ||
4968 | lockdep_assert_held(&kvm->slots_lock); | |
4969 | ||
eed52e43 | 4970 | for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { |
86bdf3eb GS |
4971 | if (!kvm_memslots_empty(__kvm_memslots(kvm, i))) |
4972 | return false; | |
4973 | } | |
4974 | ||
4975 | return true; | |
4976 | } | |
26f45714 | 4977 | EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty); |
86bdf3eb | 4978 | |
e5d83c74 PB |
4979 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, |
4980 | struct kvm_enable_cap *cap) | |
4981 | { | |
4982 | switch (cap->cap) { | |
2a31b9db | 4983 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
4984 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
4985 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
4986 | ||
4987 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
4988 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
4989 | ||
4990 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
4991 | return -EINVAL; |
4992 | kvm->manual_dirty_log_protect = cap->args[0]; | |
4993 | return 0; | |
3c9bd400 | 4994 | } |
2a31b9db | 4995 | #endif |
acd05785 DM |
4996 | case KVM_CAP_HALT_POLL: { |
4997 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
4998 | return -EINVAL; | |
4999 | ||
5000 | kvm->max_halt_poll_ns = cap->args[0]; | |
9eb8ca04 DM |
5001 | |
5002 | /* | |
5003 | * Ensure kvm->override_halt_poll_ns does not become visible | |
5004 | * before kvm->max_halt_poll_ns. | |
5005 | * | |
5006 | * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns(). | |
5007 | */ | |
5008 | smp_wmb(); | |
5009 | kvm->override_halt_poll_ns = true; | |
5010 | ||
acd05785 DM |
5011 | return 0; |
5012 | } | |
fb04a1ed | 5013 | case KVM_CAP_DIRTY_LOG_RING: |
17601bfe | 5014 | case KVM_CAP_DIRTY_LOG_RING_ACQ_REL: |
7a2726ec GS |
5015 | if (!kvm_vm_ioctl_check_extension_generic(kvm, cap->cap)) |
5016 | return -EINVAL; | |
5017 | ||
fb04a1ed | 5018 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); |
86bdf3eb GS |
5019 | case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: { |
5020 | int r = -EINVAL; | |
5021 | ||
5022 | if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP) || | |
5023 | !kvm->dirty_ring_size || cap->flags) | |
5024 | return r; | |
5025 | ||
5026 | mutex_lock(&kvm->slots_lock); | |
5027 | ||
5028 | /* | |
5029 | * For simplicity, allow enabling ring+bitmap if and only if | |
5030 | * there are no memslots, e.g. to ensure all memslots allocate | |
5031 | * a bitmap after the capability is enabled. | |
5032 | */ | |
5033 | if (kvm_are_all_memslots_empty(kvm)) { | |
5034 | kvm->dirty_ring_with_bitmap = true; | |
5035 | r = 0; | |
5036 | } | |
5037 | ||
5038 | mutex_unlock(&kvm->slots_lock); | |
5039 | ||
5040 | return r; | |
5041 | } | |
e5d83c74 PB |
5042 | default: |
5043 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
5044 | } | |
5045 | } | |
5046 | ||
fcfe1bae JZ |
5047 | static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer, |
5048 | size_t size, loff_t *offset) | |
5049 | { | |
5050 | struct kvm *kvm = file->private_data; | |
5051 | ||
5052 | return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header, | |
5053 | &kvm_vm_stats_desc[0], &kvm->stat, | |
5054 | sizeof(kvm->stat), user_buffer, size, offset); | |
5055 | } | |
5056 | ||
eed3013f SC |
5057 | static int kvm_vm_stats_release(struct inode *inode, struct file *file) |
5058 | { | |
5059 | struct kvm *kvm = file->private_data; | |
5060 | ||
5061 | kvm_put_kvm(kvm); | |
5062 | return 0; | |
5063 | } | |
5064 | ||
fcfe1bae | 5065 | static const struct file_operations kvm_vm_stats_fops = { |
087e1520 | 5066 | .owner = THIS_MODULE, |
fcfe1bae | 5067 | .read = kvm_vm_stats_read, |
eed3013f | 5068 | .release = kvm_vm_stats_release, |
fcfe1bae JZ |
5069 | .llseek = noop_llseek, |
5070 | }; | |
5071 | ||
5072 | static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm) | |
5073 | { | |
5074 | int fd; | |
5075 | struct file *file; | |
5076 | ||
5077 | fd = get_unused_fd_flags(O_CLOEXEC); | |
5078 | if (fd < 0) | |
5079 | return fd; | |
5080 | ||
5081 | file = anon_inode_getfile("kvm-vm-stats", | |
5082 | &kvm_vm_stats_fops, kvm, O_RDONLY); | |
5083 | if (IS_ERR(file)) { | |
5084 | put_unused_fd(fd); | |
5085 | return PTR_ERR(file); | |
5086 | } | |
eed3013f SC |
5087 | |
5088 | kvm_get_kvm(kvm); | |
5089 | ||
fcfe1bae JZ |
5090 | file->f_mode |= FMODE_PREAD; |
5091 | fd_install(fd, file); | |
5092 | ||
5093 | return fd; | |
5094 | } | |
5095 | ||
bb58b90b SC |
5096 | #define SANITY_CHECK_MEM_REGION_FIELD(field) \ |
5097 | do { \ | |
5098 | BUILD_BUG_ON(offsetof(struct kvm_userspace_memory_region, field) != \ | |
5099 | offsetof(struct kvm_userspace_memory_region2, field)); \ | |
5100 | BUILD_BUG_ON(sizeof_field(struct kvm_userspace_memory_region, field) != \ | |
5101 | sizeof_field(struct kvm_userspace_memory_region2, field)); \ | |
5102 | } while (0) | |
5103 | ||
bccf2150 AK |
5104 | static long kvm_vm_ioctl(struct file *filp, |
5105 | unsigned int ioctl, unsigned long arg) | |
5106 | { | |
5107 | struct kvm *kvm = filp->private_data; | |
5108 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 5109 | int r; |
bccf2150 | 5110 | |
f4d31653 | 5111 | if (kvm->mm != current->mm || kvm->vm_dead) |
6d4e4c4f | 5112 | return -EIO; |
bccf2150 AK |
5113 | switch (ioctl) { |
5114 | case KVM_CREATE_VCPU: | |
5115 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 5116 | break; |
e5d83c74 PB |
5117 | case KVM_ENABLE_CAP: { |
5118 | struct kvm_enable_cap cap; | |
5119 | ||
5120 | r = -EFAULT; | |
5121 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
5122 | goto out; | |
5123 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
5124 | break; | |
5125 | } | |
bb58b90b | 5126 | case KVM_SET_USER_MEMORY_REGION2: |
6fc138d2 | 5127 | case KVM_SET_USER_MEMORY_REGION: { |
bb58b90b SC |
5128 | struct kvm_userspace_memory_region2 mem; |
5129 | unsigned long size; | |
5130 | ||
5131 | if (ioctl == KVM_SET_USER_MEMORY_REGION) { | |
5132 | /* | |
5133 | * Fields beyond struct kvm_userspace_memory_region shouldn't be | |
5134 | * accessed, but avoid leaking kernel memory in case of a bug. | |
5135 | */ | |
5136 | memset(&mem, 0, sizeof(mem)); | |
5137 | size = sizeof(struct kvm_userspace_memory_region); | |
5138 | } else { | |
5139 | size = sizeof(struct kvm_userspace_memory_region2); | |
5140 | } | |
5141 | ||
5142 | /* Ensure the common parts of the two structs are identical. */ | |
5143 | SANITY_CHECK_MEM_REGION_FIELD(slot); | |
5144 | SANITY_CHECK_MEM_REGION_FIELD(flags); | |
5145 | SANITY_CHECK_MEM_REGION_FIELD(guest_phys_addr); | |
5146 | SANITY_CHECK_MEM_REGION_FIELD(memory_size); | |
5147 | SANITY_CHECK_MEM_REGION_FIELD(userspace_addr); | |
6fc138d2 IE |
5148 | |
5149 | r = -EFAULT; | |
bb58b90b SC |
5150 | if (copy_from_user(&mem, argp, size)) |
5151 | goto out; | |
5152 | ||
5153 | r = -EINVAL; | |
5154 | if (ioctl == KVM_SET_USER_MEMORY_REGION && | |
5155 | (mem.flags & ~KVM_SET_USER_MEMORY_REGION_V1_FLAGS)) | |
6fc138d2 IE |
5156 | goto out; |
5157 | ||
bb58b90b | 5158 | r = kvm_vm_ioctl_set_memory_region(kvm, &mem); |
6aa8b732 AK |
5159 | break; |
5160 | } | |
5161 | case KVM_GET_DIRTY_LOG: { | |
5162 | struct kvm_dirty_log log; | |
5163 | ||
5164 | r = -EFAULT; | |
893bdbf1 | 5165 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 5166 | goto out; |
2c6f5df9 | 5167 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
5168 | break; |
5169 | } | |
2a31b9db PB |
5170 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
5171 | case KVM_CLEAR_DIRTY_LOG: { | |
5172 | struct kvm_clear_dirty_log log; | |
5173 | ||
5174 | r = -EFAULT; | |
5175 | if (copy_from_user(&log, argp, sizeof(log))) | |
5176 | goto out; | |
5177 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
5178 | break; | |
5179 | } | |
5180 | #endif | |
4b4357e0 | 5181 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
5182 | case KVM_REGISTER_COALESCED_MMIO: { |
5183 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 5184 | |
5f94c174 | 5185 | r = -EFAULT; |
893bdbf1 | 5186 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 5187 | goto out; |
5f94c174 | 5188 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
5189 | break; |
5190 | } | |
5191 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
5192 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 5193 | |
5f94c174 | 5194 | r = -EFAULT; |
893bdbf1 | 5195 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 5196 | goto out; |
5f94c174 | 5197 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
5198 | break; |
5199 | } | |
5200 | #endif | |
721eecbf GH |
5201 | case KVM_IRQFD: { |
5202 | struct kvm_irqfd data; | |
5203 | ||
5204 | r = -EFAULT; | |
893bdbf1 | 5205 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 5206 | goto out; |
d4db2935 | 5207 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
5208 | break; |
5209 | } | |
d34e6b17 GH |
5210 | case KVM_IOEVENTFD: { |
5211 | struct kvm_ioeventfd data; | |
5212 | ||
5213 | r = -EFAULT; | |
893bdbf1 | 5214 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
5215 | goto out; |
5216 | r = kvm_ioeventfd(kvm, &data); | |
5217 | break; | |
5218 | } | |
07975ad3 JK |
5219 | #ifdef CONFIG_HAVE_KVM_MSI |
5220 | case KVM_SIGNAL_MSI: { | |
5221 | struct kvm_msi msi; | |
5222 | ||
5223 | r = -EFAULT; | |
893bdbf1 | 5224 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
5225 | goto out; |
5226 | r = kvm_send_userspace_msi(kvm, &msi); | |
5227 | break; | |
5228 | } | |
23d43cf9 CD |
5229 | #endif |
5230 | #ifdef __KVM_HAVE_IRQ_LINE | |
5231 | case KVM_IRQ_LINE_STATUS: | |
5232 | case KVM_IRQ_LINE: { | |
5233 | struct kvm_irq_level irq_event; | |
5234 | ||
5235 | r = -EFAULT; | |
893bdbf1 | 5236 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
5237 | goto out; |
5238 | ||
aa2fbe6d YZ |
5239 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
5240 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
5241 | if (r) |
5242 | goto out; | |
5243 | ||
5244 | r = -EFAULT; | |
5245 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 5246 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
5247 | goto out; |
5248 | } | |
5249 | ||
5250 | r = 0; | |
5251 | break; | |
5252 | } | |
73880c80 | 5253 | #endif |
aa8d5944 AG |
5254 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
5255 | case KVM_SET_GSI_ROUTING: { | |
5256 | struct kvm_irq_routing routing; | |
5257 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 5258 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
5259 | |
5260 | r = -EFAULT; | |
5261 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
5262 | goto out; | |
5263 | r = -EINVAL; | |
5c0aea0e DH |
5264 | if (!kvm_arch_can_set_irq_routing(kvm)) |
5265 | goto out; | |
caf1ff26 | 5266 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
5267 | goto out; |
5268 | if (routing.flags) | |
5269 | goto out; | |
f8c1b85b | 5270 | if (routing.nr) { |
f8c1b85b | 5271 | urouting = argp; |
1f829359 PS |
5272 | entries = vmemdup_array_user(urouting->entries, |
5273 | routing.nr, sizeof(*entries)); | |
7ec28e26 DE |
5274 | if (IS_ERR(entries)) { |
5275 | r = PTR_ERR(entries); | |
5276 | goto out; | |
5277 | } | |
f8c1b85b | 5278 | } |
aa8d5944 AG |
5279 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
5280 | routing.flags); | |
7ec28e26 | 5281 | kvfree(entries); |
aa8d5944 AG |
5282 | break; |
5283 | } | |
5284 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
5a475554 CP |
5285 | #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES |
5286 | case KVM_SET_MEMORY_ATTRIBUTES: { | |
5287 | struct kvm_memory_attributes attrs; | |
5288 | ||
5289 | r = -EFAULT; | |
5290 | if (copy_from_user(&attrs, argp, sizeof(attrs))) | |
5291 | goto out; | |
5292 | ||
5293 | r = kvm_vm_ioctl_set_mem_attributes(kvm, &attrs); | |
5294 | break; | |
5295 | } | |
5296 | #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */ | |
852b6d57 SW |
5297 | case KVM_CREATE_DEVICE: { |
5298 | struct kvm_create_device cd; | |
5299 | ||
5300 | r = -EFAULT; | |
5301 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
5302 | goto out; | |
5303 | ||
5304 | r = kvm_ioctl_create_device(kvm, &cd); | |
5305 | if (r) | |
5306 | goto out; | |
5307 | ||
5308 | r = -EFAULT; | |
5309 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
5310 | goto out; | |
5311 | ||
5312 | r = 0; | |
5313 | break; | |
5314 | } | |
92b591a4 AG |
5315 | case KVM_CHECK_EXTENSION: |
5316 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
5317 | break; | |
fb04a1ed PX |
5318 | case KVM_RESET_DIRTY_RINGS: |
5319 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
5320 | break; | |
fcfe1bae JZ |
5321 | case KVM_GET_STATS_FD: |
5322 | r = kvm_vm_ioctl_get_stats_fd(kvm); | |
5323 | break; | |
a7800aa8 SC |
5324 | #ifdef CONFIG_KVM_PRIVATE_MEM |
5325 | case KVM_CREATE_GUEST_MEMFD: { | |
5326 | struct kvm_create_guest_memfd guest_memfd; | |
5327 | ||
5328 | r = -EFAULT; | |
5329 | if (copy_from_user(&guest_memfd, argp, sizeof(guest_memfd))) | |
5330 | goto out; | |
5331 | ||
5332 | r = kvm_gmem_create(kvm, &guest_memfd); | |
5333 | break; | |
5334 | } | |
5335 | #endif | |
f17abe9a | 5336 | default: |
1fe779f8 | 5337 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
5338 | } |
5339 | out: | |
5340 | return r; | |
5341 | } | |
5342 | ||
de8e5d74 | 5343 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
5344 | struct compat_kvm_dirty_log { |
5345 | __u32 slot; | |
5346 | __u32 padding1; | |
5347 | union { | |
5348 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
5349 | __u64 padding2; | |
5350 | }; | |
5351 | }; | |
5352 | ||
8750f9bb PB |
5353 | struct compat_kvm_clear_dirty_log { |
5354 | __u32 slot; | |
5355 | __u32 num_pages; | |
5356 | __u64 first_page; | |
5357 | union { | |
5358 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
5359 | __u64 padding2; | |
5360 | }; | |
5361 | }; | |
5362 | ||
ed51862f AG |
5363 | long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl, |
5364 | unsigned long arg) | |
5365 | { | |
5366 | return -ENOTTY; | |
5367 | } | |
5368 | ||
6ff5894c AB |
5369 | static long kvm_vm_compat_ioctl(struct file *filp, |
5370 | unsigned int ioctl, unsigned long arg) | |
5371 | { | |
5372 | struct kvm *kvm = filp->private_data; | |
5373 | int r; | |
5374 | ||
f4d31653 | 5375 | if (kvm->mm != current->mm || kvm->vm_dead) |
6ff5894c | 5376 | return -EIO; |
ed51862f AG |
5377 | |
5378 | r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg); | |
5379 | if (r != -ENOTTY) | |
5380 | return r; | |
5381 | ||
6ff5894c | 5382 | switch (ioctl) { |
8750f9bb PB |
5383 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
5384 | case KVM_CLEAR_DIRTY_LOG: { | |
5385 | struct compat_kvm_clear_dirty_log compat_log; | |
5386 | struct kvm_clear_dirty_log log; | |
5387 | ||
5388 | if (copy_from_user(&compat_log, (void __user *)arg, | |
5389 | sizeof(compat_log))) | |
5390 | return -EFAULT; | |
5391 | log.slot = compat_log.slot; | |
5392 | log.num_pages = compat_log.num_pages; | |
5393 | log.first_page = compat_log.first_page; | |
5394 | log.padding2 = compat_log.padding2; | |
5395 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
5396 | ||
5397 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
5398 | break; | |
5399 | } | |
5400 | #endif | |
6ff5894c AB |
5401 | case KVM_GET_DIRTY_LOG: { |
5402 | struct compat_kvm_dirty_log compat_log; | |
5403 | struct kvm_dirty_log log; | |
5404 | ||
6ff5894c AB |
5405 | if (copy_from_user(&compat_log, (void __user *)arg, |
5406 | sizeof(compat_log))) | |
f6a3b168 | 5407 | return -EFAULT; |
6ff5894c AB |
5408 | log.slot = compat_log.slot; |
5409 | log.padding1 = compat_log.padding1; | |
5410 | log.padding2 = compat_log.padding2; | |
5411 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
5412 | ||
5413 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
5414 | break; |
5415 | } | |
5416 | default: | |
5417 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
5418 | } | |
6ff5894c AB |
5419 | return r; |
5420 | } | |
5421 | #endif | |
5422 | ||
087e1520 | 5423 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
5424 | .release = kvm_vm_release, |
5425 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 5426 | .llseek = noop_llseek, |
7ddfd3e0 | 5427 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
5428 | }; |
5429 | ||
54526d1f NT |
5430 | bool file_is_kvm(struct file *file) |
5431 | { | |
5432 | return file && file->f_op == &kvm_vm_fops; | |
5433 | } | |
5434 | EXPORT_SYMBOL_GPL(file_is_kvm); | |
5435 | ||
e08b9637 | 5436 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 5437 | { |
59f82aad | 5438 | char fdname[ITOA_MAX_LEN + 1]; |
20020f4c | 5439 | int r, fd; |
f17abe9a | 5440 | struct kvm *kvm; |
506cfba9 | 5441 | struct file *file; |
f17abe9a | 5442 | |
20020f4c OU |
5443 | fd = get_unused_fd_flags(O_CLOEXEC); |
5444 | if (fd < 0) | |
5445 | return fd; | |
5446 | ||
59f82aad OU |
5447 | snprintf(fdname, sizeof(fdname), "%d", fd); |
5448 | ||
b74ed7a6 | 5449 | kvm = kvm_create_vm(type, fdname); |
20020f4c OU |
5450 | if (IS_ERR(kvm)) { |
5451 | r = PTR_ERR(kvm); | |
5452 | goto put_fd; | |
5453 | } | |
5454 | ||
506cfba9 AV |
5455 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
5456 | if (IS_ERR(file)) { | |
78588335 ME |
5457 | r = PTR_ERR(file); |
5458 | goto put_kvm; | |
506cfba9 | 5459 | } |
536a6f88 | 5460 | |
525df861 PB |
5461 | /* |
5462 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
5463 | * already set, with ->release() being kvm_vm_release(). In error | |
5464 | * cases it will be called by the final fput(file) and will take | |
5465 | * care of doing kvm_put_kvm(kvm). | |
5466 | */ | |
286de8f6 | 5467 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 5468 | |
20020f4c OU |
5469 | fd_install(fd, file); |
5470 | return fd; | |
78588335 ME |
5471 | |
5472 | put_kvm: | |
5473 | kvm_put_kvm(kvm); | |
20020f4c OU |
5474 | put_fd: |
5475 | put_unused_fd(fd); | |
78588335 | 5476 | return r; |
f17abe9a AK |
5477 | } |
5478 | ||
5479 | static long kvm_dev_ioctl(struct file *filp, | |
5480 | unsigned int ioctl, unsigned long arg) | |
5481 | { | |
f15ba52b | 5482 | int r = -EINVAL; |
f17abe9a AK |
5483 | |
5484 | switch (ioctl) { | |
5485 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
5486 | if (arg) |
5487 | goto out; | |
f17abe9a AK |
5488 | r = KVM_API_VERSION; |
5489 | break; | |
5490 | case KVM_CREATE_VM: | |
e08b9637 | 5491 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 5492 | break; |
018d00d2 | 5493 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 5494 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 5495 | break; |
07c45a36 | 5496 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
5497 | if (arg) |
5498 | goto out; | |
adb1ff46 AK |
5499 | r = PAGE_SIZE; /* struct kvm_run */ |
5500 | #ifdef CONFIG_X86 | |
5501 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 5502 | #endif |
4b4357e0 | 5503 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 5504 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 5505 | #endif |
07c45a36 | 5506 | break; |
6aa8b732 | 5507 | default: |
043405e1 | 5508 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
5509 | } |
5510 | out: | |
5511 | return r; | |
5512 | } | |
5513 | ||
6aa8b732 | 5514 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 5515 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 5516 | .llseek = noop_llseek, |
7ddfd3e0 | 5517 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
5518 | }; |
5519 | ||
5520 | static struct miscdevice kvm_dev = { | |
bbe4432e | 5521 | KVM_MINOR, |
6aa8b732 AK |
5522 | "kvm", |
5523 | &kvm_chardev_ops, | |
5524 | }; | |
5525 | ||
441f7bfa SC |
5526 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
5527 | __visible bool kvm_rebooting; | |
5528 | EXPORT_SYMBOL_GPL(kvm_rebooting); | |
5529 | ||
5530 | static DEFINE_PER_CPU(bool, hardware_enabled); | |
5531 | static int kvm_usage_count; | |
5532 | ||
e6fb7d6e | 5533 | static int __hardware_enable_nolock(void) |
1b6c0168 | 5534 | { |
37d25881 | 5535 | if (__this_cpu_read(hardware_enabled)) |
e6fb7d6e | 5536 | return 0; |
10474ae8 | 5537 | |
37d25881 | 5538 | if (kvm_arch_hardware_enable()) { |
37d25881 SC |
5539 | pr_info("kvm: enabling virtualization on CPU%d failed\n", |
5540 | raw_smp_processor_id()); | |
e6fb7d6e | 5541 | return -EIO; |
10474ae8 | 5542 | } |
37d25881 SC |
5543 | |
5544 | __this_cpu_write(hardware_enabled, true); | |
e6fb7d6e IY |
5545 | return 0; |
5546 | } | |
5547 | ||
5548 | static void hardware_enable_nolock(void *failed) | |
5549 | { | |
5550 | if (__hardware_enable_nolock()) | |
5551 | atomic_inc(failed); | |
1b6c0168 AK |
5552 | } |
5553 | ||
aaf12a7b | 5554 | static int kvm_online_cpu(unsigned int cpu) |
75b7127c | 5555 | { |
aaf12a7b CG |
5556 | int ret = 0; |
5557 | ||
5558 | /* | |
5559 | * Abort the CPU online process if hardware virtualization cannot | |
5560 | * be enabled. Otherwise running VMs would encounter unrecoverable | |
5561 | * errors when scheduled to this CPU. | |
5562 | */ | |
0bf50497 | 5563 | mutex_lock(&kvm_lock); |
e6fb7d6e IY |
5564 | if (kvm_usage_count) |
5565 | ret = __hardware_enable_nolock(); | |
0bf50497 | 5566 | mutex_unlock(&kvm_lock); |
aaf12a7b | 5567 | return ret; |
75b7127c TY |
5568 | } |
5569 | ||
5570 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 | 5571 | { |
37d25881 SC |
5572 | /* |
5573 | * Note, hardware_disable_all_nolock() tells all online CPUs to disable | |
5574 | * hardware, not just CPUs that successfully enabled hardware! | |
5575 | */ | |
5576 | if (!__this_cpu_read(hardware_enabled)) | |
1b6c0168 | 5577 | return; |
37d25881 | 5578 | |
13a34e06 | 5579 | kvm_arch_hardware_disable(); |
37d25881 SC |
5580 | |
5581 | __this_cpu_write(hardware_enabled, false); | |
1b6c0168 AK |
5582 | } |
5583 | ||
aaf12a7b | 5584 | static int kvm_offline_cpu(unsigned int cpu) |
75b7127c | 5585 | { |
0bf50497 | 5586 | mutex_lock(&kvm_lock); |
4fa92fb2 PB |
5587 | if (kvm_usage_count) |
5588 | hardware_disable_nolock(NULL); | |
0bf50497 | 5589 | mutex_unlock(&kvm_lock); |
8c18b2d2 | 5590 | return 0; |
75b7127c TY |
5591 | } |
5592 | ||
10474ae8 AG |
5593 | static void hardware_disable_all_nolock(void) |
5594 | { | |
5595 | BUG_ON(!kvm_usage_count); | |
5596 | ||
5597 | kvm_usage_count--; | |
5598 | if (!kvm_usage_count) | |
75b7127c | 5599 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
5600 | } |
5601 | ||
5602 | static void hardware_disable_all(void) | |
5603 | { | |
e4aa7f88 | 5604 | cpus_read_lock(); |
0bf50497 | 5605 | mutex_lock(&kvm_lock); |
10474ae8 | 5606 | hardware_disable_all_nolock(); |
0bf50497 | 5607 | mutex_unlock(&kvm_lock); |
e4aa7f88 | 5608 | cpus_read_unlock(); |
10474ae8 AG |
5609 | } |
5610 | ||
5611 | static int hardware_enable_all(void) | |
5612 | { | |
e6fb7d6e | 5613 | atomic_t failed = ATOMIC_INIT(0); |
e0ceec22 SC |
5614 | int r; |
5615 | ||
5616 | /* | |
5617 | * Do not enable hardware virtualization if the system is going down. | |
5618 | * If userspace initiated a forced reboot, e.g. reboot -f, then it's | |
5619 | * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling | |
5620 | * after kvm_reboot() is called. Note, this relies on system_state | |
5621 | * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops | |
5622 | * hook instead of registering a dedicated reboot notifier (the latter | |
5623 | * runs before system_state is updated). | |
5624 | */ | |
5625 | if (system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF || | |
5626 | system_state == SYSTEM_RESTART) | |
5627 | return -EBUSY; | |
10474ae8 | 5628 | |
e4aa7f88 CG |
5629 | /* |
5630 | * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu() | |
5631 | * is called, and so on_each_cpu() between them includes the CPU that | |
5632 | * is being onlined. As a result, hardware_enable_nolock() may get | |
5633 | * invoked before kvm_online_cpu(), which also enables hardware if the | |
5634 | * usage count is non-zero. Disable CPU hotplug to avoid attempting to | |
5635 | * enable hardware multiple times. | |
5636 | */ | |
5637 | cpus_read_lock(); | |
0bf50497 | 5638 | mutex_lock(&kvm_lock); |
10474ae8 | 5639 | |
e0ceec22 SC |
5640 | r = 0; |
5641 | ||
10474ae8 AG |
5642 | kvm_usage_count++; |
5643 | if (kvm_usage_count == 1) { | |
e6fb7d6e | 5644 | on_each_cpu(hardware_enable_nolock, &failed, 1); |
10474ae8 | 5645 | |
e6fb7d6e | 5646 | if (atomic_read(&failed)) { |
10474ae8 AG |
5647 | hardware_disable_all_nolock(); |
5648 | r = -EBUSY; | |
5649 | } | |
5650 | } | |
5651 | ||
0bf50497 | 5652 | mutex_unlock(&kvm_lock); |
e4aa7f88 | 5653 | cpus_read_unlock(); |
10474ae8 AG |
5654 | |
5655 | return r; | |
5656 | } | |
5657 | ||
6735150b | 5658 | static void kvm_shutdown(void) |
9a2b85c6 | 5659 | { |
8e1c1815 | 5660 | /* |
6735150b SC |
5661 | * Disable hardware virtualization and set kvm_rebooting to indicate |
5662 | * that KVM has asynchronously disabled hardware virtualization, i.e. | |
5663 | * that relevant errors and exceptions aren't entirely unexpected. | |
5664 | * Some flavors of hardware virtualization need to be disabled before | |
5665 | * transferring control to firmware (to perform shutdown/reboot), e.g. | |
5666 | * on x86, virtualization can block INIT interrupts, which are used by | |
5667 | * firmware to pull APs back under firmware control. Note, this path | |
5668 | * is used for both shutdown and reboot scenarios, i.e. neither name is | |
5669 | * 100% comprehensive. | |
8e1c1815 | 5670 | */ |
1170adc6 | 5671 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 5672 | kvm_rebooting = true; |
75b7127c | 5673 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
5674 | } |
5675 | ||
35774a9f SC |
5676 | static int kvm_suspend(void) |
5677 | { | |
5678 | /* | |
5679 | * Secondary CPUs and CPU hotplug are disabled across the suspend/resume | |
5680 | * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count | |
5681 | * is stable. Assert that kvm_lock is not held to ensure the system | |
5682 | * isn't suspended while KVM is enabling hardware. Hardware enabling | |
5683 | * can be preempted, but the task cannot be frozen until it has dropped | |
5684 | * all locks (userspace tasks are frozen via a fake signal). | |
5685 | */ | |
5686 | lockdep_assert_not_held(&kvm_lock); | |
5687 | lockdep_assert_irqs_disabled(); | |
5688 | ||
5689 | if (kvm_usage_count) | |
5690 | hardware_disable_nolock(NULL); | |
5691 | return 0; | |
5692 | } | |
5693 | ||
5694 | static void kvm_resume(void) | |
5695 | { | |
5696 | lockdep_assert_not_held(&kvm_lock); | |
5697 | lockdep_assert_irqs_disabled(); | |
5698 | ||
5699 | if (kvm_usage_count) | |
5700 | WARN_ON_ONCE(__hardware_enable_nolock()); | |
5701 | } | |
5702 | ||
5703 | static struct syscore_ops kvm_syscore_ops = { | |
5704 | .suspend = kvm_suspend, | |
5705 | .resume = kvm_resume, | |
6735150b | 5706 | .shutdown = kvm_shutdown, |
35774a9f | 5707 | }; |
441f7bfa SC |
5708 | #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */ |
5709 | static int hardware_enable_all(void) | |
5710 | { | |
5711 | return 0; | |
5712 | } | |
5713 | ||
5714 | static void hardware_disable_all(void) | |
5715 | { | |
5716 | ||
5717 | } | |
5718 | #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */ | |
35774a9f | 5719 | |
5ea5ca3c WW |
5720 | static void kvm_iodevice_destructor(struct kvm_io_device *dev) |
5721 | { | |
5722 | if (dev->ops->destructor) | |
5723 | dev->ops->destructor(dev); | |
5724 | } | |
5725 | ||
e93f8a0f | 5726 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
5727 | { |
5728 | int i; | |
5729 | ||
5730 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 5731 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
5732 | |
5733 | kvm_iodevice_destructor(pos); | |
5734 | } | |
e93f8a0f | 5735 | kfree(bus); |
2eeb2e94 GH |
5736 | } |
5737 | ||
c21fbff1 | 5738 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 5739 | const struct kvm_io_range *r2) |
743eeb0b | 5740 | { |
8f4216c7 JW |
5741 | gpa_t addr1 = r1->addr; |
5742 | gpa_t addr2 = r2->addr; | |
5743 | ||
5744 | if (addr1 < addr2) | |
743eeb0b | 5745 | return -1; |
8f4216c7 JW |
5746 | |
5747 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
5748 | * accept any overlapping write. Any order is acceptable for | |
5749 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
5750 | * we process all of them. | |
5751 | */ | |
5752 | if (r2->len) { | |
5753 | addr1 += r1->len; | |
5754 | addr2 += r2->len; | |
5755 | } | |
5756 | ||
5757 | if (addr1 > addr2) | |
743eeb0b | 5758 | return 1; |
8f4216c7 | 5759 | |
743eeb0b SL |
5760 | return 0; |
5761 | } | |
5762 | ||
a343c9b7 PB |
5763 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
5764 | { | |
c21fbff1 | 5765 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
5766 | } |
5767 | ||
39369f7a | 5768 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
5769 | gpa_t addr, int len) |
5770 | { | |
5771 | struct kvm_io_range *range, key; | |
5772 | int off; | |
5773 | ||
5774 | key = (struct kvm_io_range) { | |
5775 | .addr = addr, | |
5776 | .len = len, | |
5777 | }; | |
5778 | ||
5779 | range = bsearch(&key, bus->range, bus->dev_count, | |
5780 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
5781 | if (range == NULL) | |
5782 | return -ENOENT; | |
5783 | ||
5784 | off = range - bus->range; | |
5785 | ||
c21fbff1 | 5786 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
5787 | off--; |
5788 | ||
5789 | return off; | |
5790 | } | |
5791 | ||
e32edf4f | 5792 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
5793 | struct kvm_io_range *range, const void *val) |
5794 | { | |
5795 | int idx; | |
5796 | ||
5797 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
5798 | if (idx < 0) | |
5799 | return -EOPNOTSUPP; | |
5800 | ||
5801 | while (idx < bus->dev_count && | |
c21fbff1 | 5802 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5803 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5804 | range->len, val)) |
5805 | return idx; | |
5806 | idx++; | |
5807 | } | |
5808 | ||
5809 | return -EOPNOTSUPP; | |
5810 | } | |
5811 | ||
bda9020e | 5812 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 5813 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 5814 | int len, const void *val) |
2eeb2e94 | 5815 | { |
90d83dc3 | 5816 | struct kvm_io_bus *bus; |
743eeb0b | 5817 | struct kvm_io_range range; |
126a5af5 | 5818 | int r; |
743eeb0b SL |
5819 | |
5820 | range = (struct kvm_io_range) { | |
5821 | .addr = addr, | |
5822 | .len = len, | |
5823 | }; | |
90d83dc3 | 5824 | |
e32edf4f | 5825 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5826 | if (!bus) |
5827 | return -ENOMEM; | |
e32edf4f | 5828 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5829 | return r < 0 ? r : 0; |
5830 | } | |
a2420107 | 5831 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
5832 | |
5833 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
5834 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
5835 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
5836 | { |
5837 | struct kvm_io_bus *bus; | |
5838 | struct kvm_io_range range; | |
5839 | ||
5840 | range = (struct kvm_io_range) { | |
5841 | .addr = addr, | |
5842 | .len = len, | |
5843 | }; | |
5844 | ||
e32edf4f | 5845 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5846 | if (!bus) |
5847 | return -ENOMEM; | |
126a5af5 CH |
5848 | |
5849 | /* First try the device referenced by cookie. */ | |
5850 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 5851 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 5852 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
5853 | val)) |
5854 | return cookie; | |
5855 | ||
5856 | /* | |
5857 | * cookie contained garbage; fall back to search and return the | |
5858 | * correct cookie value. | |
5859 | */ | |
e32edf4f | 5860 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5861 | } |
5862 | ||
e32edf4f NN |
5863 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
5864 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
5865 | { |
5866 | int idx; | |
5867 | ||
5868 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
5869 | if (idx < 0) |
5870 | return -EOPNOTSUPP; | |
5871 | ||
5872 | while (idx < bus->dev_count && | |
c21fbff1 | 5873 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5874 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5875 | range->len, val)) |
5876 | return idx; | |
743eeb0b SL |
5877 | idx++; |
5878 | } | |
5879 | ||
bda9020e MT |
5880 | return -EOPNOTSUPP; |
5881 | } | |
2eeb2e94 | 5882 | |
bda9020e | 5883 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 5884 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 5885 | int len, void *val) |
bda9020e | 5886 | { |
90d83dc3 | 5887 | struct kvm_io_bus *bus; |
743eeb0b | 5888 | struct kvm_io_range range; |
126a5af5 | 5889 | int r; |
743eeb0b SL |
5890 | |
5891 | range = (struct kvm_io_range) { | |
5892 | .addr = addr, | |
5893 | .len = len, | |
5894 | }; | |
e93f8a0f | 5895 | |
e32edf4f | 5896 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5897 | if (!bus) |
5898 | return -ENOMEM; | |
e32edf4f | 5899 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
5900 | return r < 0 ? r : 0; |
5901 | } | |
743eeb0b | 5902 | |
743eeb0b SL |
5903 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
5904 | int len, struct kvm_io_device *dev) | |
6c474694 | 5905 | { |
d4c67a7a | 5906 | int i; |
e93f8a0f | 5907 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 5908 | struct kvm_io_range range; |
090b7aff | 5909 | |
b1a39a71 MZ |
5910 | lockdep_assert_held(&kvm->slots_lock); |
5911 | ||
4a12f951 | 5912 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
5913 | if (!bus) |
5914 | return -ENOMEM; | |
5915 | ||
6ea34c9b AK |
5916 | /* exclude ioeventfd which is limited by maximum fd */ |
5917 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 5918 | return -ENOSPC; |
2eeb2e94 | 5919 | |
90952cd3 | 5920 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 5921 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
5922 | if (!new_bus) |
5923 | return -ENOMEM; | |
d4c67a7a GH |
5924 | |
5925 | range = (struct kvm_io_range) { | |
5926 | .addr = addr, | |
5927 | .len = len, | |
5928 | .dev = dev, | |
5929 | }; | |
5930 | ||
5931 | for (i = 0; i < bus->dev_count; i++) | |
5932 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
5933 | break; | |
5934 | ||
5935 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
5936 | new_bus->dev_count++; | |
5937 | new_bus->range[i] = range; | |
5938 | memcpy(new_bus->range + i + 1, bus->range + i, | |
5939 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
5940 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
5941 | synchronize_srcu_expedited(&kvm->srcu); | |
5942 | kfree(bus); | |
090b7aff GH |
5943 | |
5944 | return 0; | |
5945 | } | |
5946 | ||
5d3c4c79 SC |
5947 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5948 | struct kvm_io_device *dev) | |
090b7aff | 5949 | { |
5ea5ca3c | 5950 | int i; |
e93f8a0f | 5951 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 5952 | |
7c896d37 SC |
5953 | lockdep_assert_held(&kvm->slots_lock); |
5954 | ||
4a12f951 | 5955 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 5956 | if (!bus) |
5d3c4c79 | 5957 | return 0; |
df630b8c | 5958 | |
7c896d37 | 5959 | for (i = 0; i < bus->dev_count; i++) { |
a1300716 | 5960 | if (bus->range[i].dev == dev) { |
090b7aff GH |
5961 | break; |
5962 | } | |
7c896d37 | 5963 | } |
e93f8a0f | 5964 | |
90db1043 | 5965 | if (i == bus->dev_count) |
5d3c4c79 | 5966 | return 0; |
a1300716 | 5967 | |
90952cd3 | 5968 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 5969 | GFP_KERNEL_ACCOUNT); |
f6588660 | 5970 | if (new_bus) { |
871c433b | 5971 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
5972 | new_bus->dev_count--; |
5973 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 5974 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
2ee37574 SC |
5975 | } |
5976 | ||
5977 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
5978 | synchronize_srcu_expedited(&kvm->srcu); | |
5979 | ||
5ea5ca3c WW |
5980 | /* |
5981 | * If NULL bus is installed, destroy the old bus, including all the | |
5982 | * attached devices. Otherwise, destroy the caller's device only. | |
5983 | */ | |
2ee37574 | 5984 | if (!new_bus) { |
90db1043 | 5985 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
5ea5ca3c WW |
5986 | kvm_io_bus_destroy(bus); |
5987 | return -ENOMEM; | |
90db1043 | 5988 | } |
a1300716 | 5989 | |
5ea5ca3c | 5990 | kvm_iodevice_destructor(dev); |
e93f8a0f | 5991 | kfree(bus); |
5ea5ca3c | 5992 | return 0; |
2eeb2e94 GH |
5993 | } |
5994 | ||
8a39d006 AP |
5995 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5996 | gpa_t addr) | |
5997 | { | |
5998 | struct kvm_io_bus *bus; | |
5999 | int dev_idx, srcu_idx; | |
6000 | struct kvm_io_device *iodev = NULL; | |
6001 | ||
6002 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
6003 | ||
6004 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
6005 | if (!bus) |
6006 | goto out_unlock; | |
8a39d006 AP |
6007 | |
6008 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
6009 | if (dev_idx < 0) | |
6010 | goto out_unlock; | |
6011 | ||
6012 | iodev = bus->range[dev_idx].dev; | |
6013 | ||
6014 | out_unlock: | |
6015 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
6016 | ||
6017 | return iodev; | |
6018 | } | |
6019 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
6020 | ||
536a6f88 JF |
6021 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
6022 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
6023 | const char *fmt) | |
6024 | { | |
180418e2 | 6025 | int ret; |
14aa40a1 | 6026 | struct kvm_stat_data *stat_data = inode->i_private; |
536a6f88 | 6027 | |
605c7130 PX |
6028 | /* |
6029 | * The debugfs files are a reference to the kvm struct which | |
6030 | * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe | |
6031 | * avoids the race between open and the removal of the debugfs directory. | |
536a6f88 | 6032 | */ |
605c7130 | 6033 | if (!kvm_get_kvm_safe(stat_data->kvm)) |
536a6f88 JF |
6034 | return -ENOENT; |
6035 | ||
180418e2 HW |
6036 | ret = simple_attr_open(inode, file, get, |
6037 | kvm_stats_debugfs_mode(stat_data->desc) & 0222 | |
6038 | ? set : NULL, fmt); | |
6039 | if (ret) | |
536a6f88 | 6040 | kvm_put_kvm(stat_data->kvm); |
536a6f88 | 6041 | |
180418e2 | 6042 | return ret; |
536a6f88 JF |
6043 | } |
6044 | ||
6045 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
6046 | { | |
14aa40a1 | 6047 | struct kvm_stat_data *stat_data = inode->i_private; |
536a6f88 JF |
6048 | |
6049 | simple_attr_release(inode, file); | |
6050 | kvm_put_kvm(stat_data->kvm); | |
6051 | ||
6052 | return 0; | |
6053 | } | |
6054 | ||
09cbcef6 | 6055 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 6056 | { |
bc9e9e67 | 6057 | *val = *(u64 *)((void *)(&kvm->stat) + offset); |
536a6f88 | 6058 | |
09cbcef6 MP |
6059 | return 0; |
6060 | } | |
6061 | ||
6062 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
6063 | { | |
bc9e9e67 | 6064 | *(u64 *)((void *)(&kvm->stat) + offset) = 0; |
536a6f88 JF |
6065 | |
6066 | return 0; | |
6067 | } | |
6068 | ||
09cbcef6 | 6069 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 6070 | { |
46808a4c | 6071 | unsigned long i; |
09cbcef6 | 6072 | struct kvm_vcpu *vcpu; |
ce35ef27 | 6073 | |
09cbcef6 | 6074 | *val = 0; |
ce35ef27 | 6075 | |
09cbcef6 | 6076 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 6077 | *val += *(u64 *)((void *)(&vcpu->stat) + offset); |
ce35ef27 SJS |
6078 | |
6079 | return 0; | |
6080 | } | |
6081 | ||
09cbcef6 | 6082 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 6083 | { |
46808a4c | 6084 | unsigned long i; |
09cbcef6 | 6085 | struct kvm_vcpu *vcpu; |
536a6f88 | 6086 | |
09cbcef6 | 6087 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 6088 | *(u64 *)((void *)(&vcpu->stat) + offset) = 0; |
09cbcef6 MP |
6089 | |
6090 | return 0; | |
6091 | } | |
536a6f88 | 6092 | |
09cbcef6 | 6093 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 6094 | { |
09cbcef6 | 6095 | int r = -EFAULT; |
14aa40a1 | 6096 | struct kvm_stat_data *stat_data = data; |
536a6f88 | 6097 | |
bc9e9e67 | 6098 | switch (stat_data->kind) { |
09cbcef6 MP |
6099 | case KVM_STAT_VM: |
6100 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 6101 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
6102 | break; |
6103 | case KVM_STAT_VCPU: | |
6104 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 6105 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
6106 | break; |
6107 | } | |
536a6f88 | 6108 | |
09cbcef6 | 6109 | return r; |
536a6f88 JF |
6110 | } |
6111 | ||
09cbcef6 | 6112 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 6113 | { |
09cbcef6 | 6114 | int r = -EFAULT; |
14aa40a1 | 6115 | struct kvm_stat_data *stat_data = data; |
ce35ef27 SJS |
6116 | |
6117 | if (val) | |
6118 | return -EINVAL; | |
6119 | ||
bc9e9e67 | 6120 | switch (stat_data->kind) { |
09cbcef6 MP |
6121 | case KVM_STAT_VM: |
6122 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 6123 | stat_data->desc->desc.offset); |
09cbcef6 MP |
6124 | break; |
6125 | case KVM_STAT_VCPU: | |
6126 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 6127 | stat_data->desc->desc.offset); |
09cbcef6 MP |
6128 | break; |
6129 | } | |
ce35ef27 | 6130 | |
09cbcef6 | 6131 | return r; |
ce35ef27 SJS |
6132 | } |
6133 | ||
09cbcef6 | 6134 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
6135 | { |
6136 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
6137 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
6138 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
6139 | } |
6140 | ||
09cbcef6 MP |
6141 | static const struct file_operations stat_fops_per_vm = { |
6142 | .owner = THIS_MODULE, | |
6143 | .open = kvm_stat_data_open, | |
536a6f88 | 6144 | .release = kvm_debugfs_release, |
09cbcef6 MP |
6145 | .read = simple_attr_read, |
6146 | .write = simple_attr_write, | |
6147 | .llseek = no_llseek, | |
536a6f88 JF |
6148 | }; |
6149 | ||
8b88b099 | 6150 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
6151 | { |
6152 | unsigned offset = (long)_offset; | |
ba1389b7 | 6153 | struct kvm *kvm; |
536a6f88 | 6154 | u64 tmp_val; |
ba1389b7 | 6155 | |
8b88b099 | 6156 | *val = 0; |
0d9ce162 | 6157 | mutex_lock(&kvm_lock); |
536a6f88 | 6158 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 6159 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
6160 | *val += tmp_val; |
6161 | } | |
0d9ce162 | 6162 | mutex_unlock(&kvm_lock); |
8b88b099 | 6163 | return 0; |
ba1389b7 AK |
6164 | } |
6165 | ||
ce35ef27 SJS |
6166 | static int vm_stat_clear(void *_offset, u64 val) |
6167 | { | |
6168 | unsigned offset = (long)_offset; | |
6169 | struct kvm *kvm; | |
ce35ef27 SJS |
6170 | |
6171 | if (val) | |
6172 | return -EINVAL; | |
6173 | ||
0d9ce162 | 6174 | mutex_lock(&kvm_lock); |
ce35ef27 | 6175 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 6176 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 6177 | } |
0d9ce162 | 6178 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
6179 | |
6180 | return 0; | |
6181 | } | |
6182 | ||
6183 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
bc9e9e67 | 6184 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n"); |
ba1389b7 | 6185 | |
8b88b099 | 6186 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
6187 | { |
6188 | unsigned offset = (long)_offset; | |
1165f5fe | 6189 | struct kvm *kvm; |
536a6f88 | 6190 | u64 tmp_val; |
1165f5fe | 6191 | |
8b88b099 | 6192 | *val = 0; |
0d9ce162 | 6193 | mutex_lock(&kvm_lock); |
536a6f88 | 6194 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 6195 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
6196 | *val += tmp_val; |
6197 | } | |
0d9ce162 | 6198 | mutex_unlock(&kvm_lock); |
8b88b099 | 6199 | return 0; |
1165f5fe AK |
6200 | } |
6201 | ||
ce35ef27 SJS |
6202 | static int vcpu_stat_clear(void *_offset, u64 val) |
6203 | { | |
6204 | unsigned offset = (long)_offset; | |
6205 | struct kvm *kvm; | |
ce35ef27 SJS |
6206 | |
6207 | if (val) | |
6208 | return -EINVAL; | |
6209 | ||
0d9ce162 | 6210 | mutex_lock(&kvm_lock); |
ce35ef27 | 6211 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 6212 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 6213 | } |
0d9ce162 | 6214 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
6215 | |
6216 | return 0; | |
6217 | } | |
6218 | ||
6219 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
6220 | "%llu\n"); | |
bc9e9e67 | 6221 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n"); |
1165f5fe | 6222 | |
286de8f6 CI |
6223 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
6224 | { | |
6225 | struct kobj_uevent_env *env; | |
286de8f6 CI |
6226 | unsigned long long created, active; |
6227 | ||
6228 | if (!kvm_dev.this_device || !kvm) | |
6229 | return; | |
6230 | ||
0d9ce162 | 6231 | mutex_lock(&kvm_lock); |
286de8f6 CI |
6232 | if (type == KVM_EVENT_CREATE_VM) { |
6233 | kvm_createvm_count++; | |
6234 | kvm_active_vms++; | |
6235 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
6236 | kvm_active_vms--; | |
6237 | } | |
6238 | created = kvm_createvm_count; | |
6239 | active = kvm_active_vms; | |
0d9ce162 | 6240 | mutex_unlock(&kvm_lock); |
286de8f6 | 6241 | |
b12ce36a | 6242 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
6243 | if (!env) |
6244 | return; | |
6245 | ||
6246 | add_uevent_var(env, "CREATED=%llu", created); | |
6247 | add_uevent_var(env, "COUNT=%llu", active); | |
6248 | ||
fdeaf7e3 | 6249 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 6250 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
6251 | kvm->userspace_pid = task_pid_nr(current); |
6252 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 6253 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
6254 | } |
6255 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 6256 | |
a44a4cc1 | 6257 | if (!IS_ERR(kvm->debugfs_dentry)) { |
b12ce36a | 6258 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
6259 | |
6260 | if (p) { | |
6261 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
6262 | if (!IS_ERR(tmp)) | |
6263 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
6264 | kfree(p); | |
286de8f6 CI |
6265 | } |
6266 | } | |
6267 | /* no need for checks, since we are adding at most only 5 keys */ | |
6268 | env->envp[env->envp_idx++] = NULL; | |
6269 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
6270 | kfree(env); | |
286de8f6 CI |
6271 | } |
6272 | ||
929f45e3 | 6273 | static void kvm_init_debug(void) |
6aa8b732 | 6274 | { |
bc9e9e67 JZ |
6275 | const struct file_operations *fops; |
6276 | const struct _kvm_stats_desc *pdesc; | |
6277 | int i; | |
6aa8b732 | 6278 | |
76f7c879 | 6279 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 6280 | |
bc9e9e67 JZ |
6281 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
6282 | pdesc = &kvm_vm_stats_desc[i]; | |
6283 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
6284 | fops = &vm_stat_fops; | |
6285 | else | |
6286 | fops = &vm_stat_readonly_fops; | |
6287 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
6288 | kvm_debugfs_dir, | |
6289 | (void *)(long)pdesc->desc.offset, fops); | |
6290 | } | |
6291 | ||
6292 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
6293 | pdesc = &kvm_vcpu_stats_desc[i]; | |
6294 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
6295 | fops = &vcpu_stat_fops; | |
6296 | else | |
6297 | fops = &vcpu_stat_readonly_fops; | |
6298 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
6299 | kvm_debugfs_dir, | |
6300 | (void *)(long)pdesc->desc.offset, fops); | |
4f69b680 | 6301 | } |
6aa8b732 AK |
6302 | } |
6303 | ||
15ad7146 AK |
6304 | static inline |
6305 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
6306 | { | |
6307 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
6308 | } | |
6309 | ||
6310 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
6311 | { | |
6312 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 6313 | |
046ddeed | 6314 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 6315 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 6316 | |
7495e22b | 6317 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 6318 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 6319 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
6320 | } |
6321 | ||
6322 | static void kvm_sched_out(struct preempt_notifier *pn, | |
6323 | struct task_struct *next) | |
6324 | { | |
6325 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
6326 | ||
3ba9f93b | 6327 | if (current->on_rq) { |
046ddeed | 6328 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
6329 | WRITE_ONCE(vcpu->ready, true); |
6330 | } | |
e9b11c17 | 6331 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
6332 | __this_cpu_write(kvm_running_vcpu, NULL); |
6333 | } | |
6334 | ||
6335 | /** | |
6336 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
6337 | * |
6338 | * We can disable preemption locally around accessing the per-CPU variable, | |
6339 | * and use the resolved vcpu pointer after enabling preemption again, | |
6340 | * because even if the current thread is migrated to another CPU, reading | |
6341 | * the per-CPU value later will give us the same value as we update the | |
6342 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
6343 | */ |
6344 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
6345 | { | |
1f03b2bc MZ |
6346 | struct kvm_vcpu *vcpu; |
6347 | ||
6348 | preempt_disable(); | |
6349 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
6350 | preempt_enable(); | |
6351 | ||
6352 | return vcpu; | |
7495e22b | 6353 | } |
379a3c8e | 6354 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
6355 | |
6356 | /** | |
6357 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
6358 | */ | |
6359 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
6360 | { | |
6361 | return &kvm_running_vcpu; | |
15ad7146 AK |
6362 | } |
6363 | ||
e1bfc245 SC |
6364 | #ifdef CONFIG_GUEST_PERF_EVENTS |
6365 | static unsigned int kvm_guest_state(void) | |
6366 | { | |
6367 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
6368 | unsigned int state; | |
6369 | ||
6370 | if (!kvm_arch_pmi_in_guest(vcpu)) | |
6371 | return 0; | |
6372 | ||
6373 | state = PERF_GUEST_ACTIVE; | |
6374 | if (!kvm_arch_vcpu_in_kernel(vcpu)) | |
6375 | state |= PERF_GUEST_USER; | |
6376 | ||
6377 | return state; | |
6378 | } | |
6379 | ||
6380 | static unsigned long kvm_guest_get_ip(void) | |
6381 | { | |
6382 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
6383 | ||
6384 | /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */ | |
6385 | if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu))) | |
6386 | return 0; | |
6387 | ||
6388 | return kvm_arch_vcpu_get_ip(vcpu); | |
6389 | } | |
6390 | ||
6391 | static struct perf_guest_info_callbacks kvm_guest_cbs = { | |
6392 | .state = kvm_guest_state, | |
6393 | .get_ip = kvm_guest_get_ip, | |
6394 | .handle_intel_pt_intr = NULL, | |
6395 | }; | |
6396 | ||
6397 | void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void)) | |
6398 | { | |
6399 | kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler; | |
6400 | perf_register_guest_info_callbacks(&kvm_guest_cbs); | |
6401 | } | |
6402 | void kvm_unregister_perf_callbacks(void) | |
6403 | { | |
6404 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); | |
6405 | } | |
6406 | #endif | |
6407 | ||
81a1cf9f | 6408 | int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module) |
f257d6dc | 6409 | { |
6aa8b732 | 6410 | int r; |
002c7f7c | 6411 | int cpu; |
6aa8b732 | 6412 | |
441f7bfa | 6413 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
aaf12a7b CG |
6414 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE, "kvm/cpu:online", |
6415 | kvm_online_cpu, kvm_offline_cpu); | |
774c47f1 | 6416 | if (r) |
37d25881 SC |
6417 | return r; |
6418 | ||
35774a9f | 6419 | register_syscore_ops(&kvm_syscore_ops); |
441f7bfa | 6420 | #endif |
6aa8b732 | 6421 | |
c16f862d | 6422 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
6423 | if (!vcpu_align) |
6424 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
6425 | kvm_vcpu_cache = |
6426 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
6427 | SLAB_ACCOUNT, | |
6428 | offsetof(struct kvm_vcpu, arch), | |
ce55c049 JZ |
6429 | offsetofend(struct kvm_vcpu, stats_id) |
6430 | - offsetof(struct kvm_vcpu, arch), | |
46515736 | 6431 | NULL); |
c16f862d RR |
6432 | if (!kvm_vcpu_cache) { |
6433 | r = -ENOMEM; | |
9f1a4c00 | 6434 | goto err_vcpu_cache; |
c16f862d RR |
6435 | } |
6436 | ||
baff59cc VK |
6437 | for_each_possible_cpu(cpu) { |
6438 | if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu), | |
6439 | GFP_KERNEL, cpu_to_node(cpu))) { | |
6440 | r = -ENOMEM; | |
9f1a4c00 | 6441 | goto err_cpu_kick_mask; |
baff59cc VK |
6442 | } |
6443 | } | |
6444 | ||
5910ccf0 SC |
6445 | r = kvm_irqfd_init(); |
6446 | if (r) | |
6447 | goto err_irqfd; | |
6448 | ||
af585b92 GN |
6449 | r = kvm_async_pf_init(); |
6450 | if (r) | |
5910ccf0 | 6451 | goto err_async_pf; |
af585b92 | 6452 | |
6aa8b732 | 6453 | kvm_chardev_ops.owner = module; |
087e1520 SC |
6454 | kvm_vm_fops.owner = module; |
6455 | kvm_vcpu_fops.owner = module; | |
6456 | kvm_device_fops.owner = module; | |
6aa8b732 | 6457 | |
15ad7146 AK |
6458 | kvm_preempt_ops.sched_in = kvm_sched_in; |
6459 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
6460 | ||
929f45e3 | 6461 | kvm_init_debug(); |
0ea4ed8e | 6462 | |
3c3c29fd | 6463 | r = kvm_vfio_ops_init(); |
2b012812 SC |
6464 | if (WARN_ON_ONCE(r)) |
6465 | goto err_vfio; | |
6466 | ||
a7800aa8 SC |
6467 | kvm_gmem_init(module); |
6468 | ||
2b012812 SC |
6469 | /* |
6470 | * Registration _must_ be the very last thing done, as this exposes | |
6471 | * /dev/kvm to userspace, i.e. all infrastructure must be setup! | |
6472 | */ | |
6473 | r = misc_register(&kvm_dev); | |
6474 | if (r) { | |
6475 | pr_err("kvm: misc device register failed\n"); | |
6476 | goto err_register; | |
6477 | } | |
3c3c29fd | 6478 | |
c7addb90 | 6479 | return 0; |
6aa8b732 | 6480 | |
2b012812 SC |
6481 | err_register: |
6482 | kvm_vfio_ops_exit(); | |
6483 | err_vfio: | |
af585b92 | 6484 | kvm_async_pf_deinit(); |
5910ccf0 SC |
6485 | err_async_pf: |
6486 | kvm_irqfd_exit(); | |
6487 | err_irqfd: | |
9f1a4c00 | 6488 | err_cpu_kick_mask: |
baff59cc VK |
6489 | for_each_possible_cpu(cpu) |
6490 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
c16f862d | 6491 | kmem_cache_destroy(kvm_vcpu_cache); |
9f1a4c00 | 6492 | err_vcpu_cache: |
441f7bfa | 6493 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
35774a9f | 6494 | unregister_syscore_ops(&kvm_syscore_ops); |
aaf12a7b | 6495 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE); |
441f7bfa | 6496 | #endif |
6aa8b732 AK |
6497 | return r; |
6498 | } | |
cb498ea2 | 6499 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 6500 | |
cb498ea2 | 6501 | void kvm_exit(void) |
6aa8b732 | 6502 | { |
baff59cc VK |
6503 | int cpu; |
6504 | ||
2b012812 SC |
6505 | /* |
6506 | * Note, unregistering /dev/kvm doesn't strictly need to come first, | |
6507 | * fops_get(), a.k.a. try_module_get(), prevents acquiring references | |
6508 | * to KVM while the module is being stopped. | |
6509 | */ | |
6aa8b732 | 6510 | misc_deregister(&kvm_dev); |
2b012812 SC |
6511 | |
6512 | debugfs_remove_recursive(kvm_debugfs_dir); | |
baff59cc VK |
6513 | for_each_possible_cpu(cpu) |
6514 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
c16f862d | 6515 | kmem_cache_destroy(kvm_vcpu_cache); |
73b8dc04 | 6516 | kvm_vfio_ops_exit(); |
af585b92 | 6517 | kvm_async_pf_deinit(); |
441f7bfa | 6518 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
fb3600cc | 6519 | unregister_syscore_ops(&kvm_syscore_ops); |
aaf12a7b | 6520 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE); |
441f7bfa | 6521 | #endif |
5910ccf0 | 6522 | kvm_irqfd_exit(); |
6aa8b732 | 6523 | } |
cb498ea2 | 6524 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
6525 | |
6526 | struct kvm_vm_worker_thread_context { | |
6527 | struct kvm *kvm; | |
6528 | struct task_struct *parent; | |
6529 | struct completion init_done; | |
6530 | kvm_vm_thread_fn_t thread_fn; | |
6531 | uintptr_t data; | |
6532 | int err; | |
6533 | }; | |
6534 | ||
6535 | static int kvm_vm_worker_thread(void *context) | |
6536 | { | |
6537 | /* | |
6538 | * The init_context is allocated on the stack of the parent thread, so | |
6539 | * we have to locally copy anything that is needed beyond initialization | |
6540 | */ | |
6541 | struct kvm_vm_worker_thread_context *init_context = context; | |
e45cce30 | 6542 | struct task_struct *parent; |
c57c8046 JS |
6543 | struct kvm *kvm = init_context->kvm; |
6544 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
6545 | uintptr_t data = init_context->data; | |
6546 | int err; | |
6547 | ||
6548 | err = kthread_park(current); | |
6549 | /* kthread_park(current) is never supposed to return an error */ | |
6550 | WARN_ON(err != 0); | |
6551 | if (err) | |
6552 | goto init_complete; | |
6553 | ||
6554 | err = cgroup_attach_task_all(init_context->parent, current); | |
6555 | if (err) { | |
6556 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
6557 | __func__, err); | |
6558 | goto init_complete; | |
6559 | } | |
6560 | ||
6561 | set_user_nice(current, task_nice(init_context->parent)); | |
6562 | ||
6563 | init_complete: | |
6564 | init_context->err = err; | |
6565 | complete(&init_context->init_done); | |
6566 | init_context = NULL; | |
6567 | ||
6568 | if (err) | |
e45cce30 | 6569 | goto out; |
c57c8046 JS |
6570 | |
6571 | /* Wait to be woken up by the spawner before proceeding. */ | |
6572 | kthread_parkme(); | |
6573 | ||
6574 | if (!kthread_should_stop()) | |
6575 | err = thread_fn(kvm, data); | |
6576 | ||
e45cce30 VS |
6577 | out: |
6578 | /* | |
6579 | * Move kthread back to its original cgroup to prevent it lingering in | |
6580 | * the cgroup of the VM process, after the latter finishes its | |
6581 | * execution. | |
6582 | * | |
6583 | * kthread_stop() waits on the 'exited' completion condition which is | |
6584 | * set in exit_mm(), via mm_release(), in do_exit(). However, the | |
6585 | * kthread is removed from the cgroup in the cgroup_exit() which is | |
6586 | * called after the exit_mm(). This causes the kthread_stop() to return | |
6587 | * before the kthread actually quits the cgroup. | |
6588 | */ | |
6589 | rcu_read_lock(); | |
6590 | parent = rcu_dereference(current->real_parent); | |
6591 | get_task_struct(parent); | |
6592 | rcu_read_unlock(); | |
6593 | cgroup_attach_task_all(parent, current); | |
6594 | put_task_struct(parent); | |
6595 | ||
c57c8046 JS |
6596 | return err; |
6597 | } | |
6598 | ||
6599 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
6600 | uintptr_t data, const char *name, | |
6601 | struct task_struct **thread_ptr) | |
6602 | { | |
6603 | struct kvm_vm_worker_thread_context init_context = {}; | |
6604 | struct task_struct *thread; | |
6605 | ||
6606 | *thread_ptr = NULL; | |
6607 | init_context.kvm = kvm; | |
6608 | init_context.parent = current; | |
6609 | init_context.thread_fn = thread_fn; | |
6610 | init_context.data = data; | |
6611 | init_completion(&init_context.init_done); | |
6612 | ||
6613 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
6614 | "%s-%d", name, task_pid_nr(current)); | |
6615 | if (IS_ERR(thread)) | |
6616 | return PTR_ERR(thread); | |
6617 | ||
6618 | /* kthread_run is never supposed to return NULL */ | |
6619 | WARN_ON(thread == NULL); | |
6620 | ||
6621 | wait_for_completion(&init_context.init_done); | |
6622 | ||
6623 | if (!init_context.err) | |
6624 | *thread_ptr = thread; | |
6625 | ||
6626 | return init_context.err; | |
6627 | } |