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 | ||
54163a34 SS |
314 | bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req, |
315 | struct kvm_vcpu *except) | |
7053df4e | 316 | { |
ae0946cd | 317 | struct kvm_vcpu *vcpu; |
baff59cc | 318 | struct cpumask *cpus; |
46808a4c | 319 | unsigned long i; |
7053df4e | 320 | bool called; |
46808a4c | 321 | int me; |
7053df4e | 322 | |
ae0946cd VK |
323 | me = get_cpu(); |
324 | ||
baff59cc VK |
325 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
326 | cpumask_clear(cpus); | |
327 | ||
ae0946cd VK |
328 | kvm_for_each_vcpu(i, vcpu, kvm) { |
329 | if (vcpu == except) | |
330 | continue; | |
b56bd8e0 | 331 | kvm_make_vcpu_request(vcpu, req, cpus, me); |
ae0946cd VK |
332 | } |
333 | ||
334 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); | |
335 | put_cpu(); | |
7053df4e | 336 | |
49846896 | 337 | return called; |
d9e368d6 AK |
338 | } |
339 | ||
54163a34 SS |
340 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
341 | { | |
342 | return kvm_make_all_cpus_request_except(kvm, req, NULL); | |
343 | } | |
a2486020 | 344 | EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request); |
54163a34 | 345 | |
49846896 | 346 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 347 | { |
3cc4e148 | 348 | ++kvm->stat.generic.remote_tlb_flush_requests; |
6bc6db00 | 349 | |
4ae3cb3a LT |
350 | /* |
351 | * We want to publish modifications to the page tables before reading | |
352 | * mode. Pairs with a memory barrier in arch-specific code. | |
353 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
354 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
355 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
356 | * | |
357 | * There is already an smp_mb__after_atomic() before | |
358 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
359 | * barrier here. | |
360 | */ | |
a1342c80 | 361 | if (!kvm_arch_flush_remote_tlbs(kvm) |
b08660e5 | 362 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
0193cc90 | 363 | ++kvm->stat.generic.remote_tlb_flush; |
2e53d63a | 364 | } |
2ba9f0d8 | 365 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
2e53d63a | 366 | |
d4788996 DM |
367 | void kvm_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages) |
368 | { | |
369 | if (!kvm_arch_flush_remote_tlbs_range(kvm, gfn, nr_pages)) | |
370 | return; | |
371 | ||
372 | /* | |
373 | * Fall back to a flushing entire TLBs if the architecture range-based | |
374 | * TLB invalidation is unsupported or can't be performed for whatever | |
375 | * reason. | |
376 | */ | |
377 | kvm_flush_remote_tlbs(kvm); | |
378 | } | |
379 | ||
619b5072 DM |
380 | void kvm_flush_remote_tlbs_memslot(struct kvm *kvm, |
381 | const struct kvm_memory_slot *memslot) | |
382 | { | |
383 | /* | |
384 | * All current use cases for flushing the TLBs for a specific memslot | |
385 | * are related to dirty logging, and many do the TLB flush out of | |
386 | * mmu_lock. The interaction between the various operations on memslot | |
387 | * must be serialized by slots_locks to ensure the TLB flush from one | |
388 | * operation is observed by any other operation on the same memslot. | |
389 | */ | |
390 | lockdep_assert_held(&kvm->slots_lock); | |
391 | kvm_flush_remote_tlbs_range(kvm, memslot->base_gfn, memslot->npages); | |
392 | } | |
2e53d63a | 393 | |
683412cc MZ |
394 | static void kvm_flush_shadow_all(struct kvm *kvm) |
395 | { | |
396 | kvm_arch_flush_shadow_all(kvm); | |
397 | kvm_arch_guest_memory_reclaimed(kvm); | |
398 | } | |
399 | ||
6926f95a SC |
400 | #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE |
401 | static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc, | |
402 | gfp_t gfp_flags) | |
403 | { | |
c23e2b71 SC |
404 | void *page; |
405 | ||
6926f95a SC |
406 | gfp_flags |= mc->gfp_zero; |
407 | ||
408 | if (mc->kmem_cache) | |
409 | return kmem_cache_alloc(mc->kmem_cache, gfp_flags); | |
c23e2b71 SC |
410 | |
411 | page = (void *)__get_free_page(gfp_flags); | |
412 | if (page && mc->init_value) | |
413 | memset64(page, mc->init_value, PAGE_SIZE / sizeof(u64)); | |
414 | return page; | |
6926f95a SC |
415 | } |
416 | ||
837f66c7 | 417 | int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min) |
6926f95a | 418 | { |
63f4b210 | 419 | gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT; |
6926f95a SC |
420 | void *obj; |
421 | ||
422 | if (mc->nobjs >= min) | |
423 | return 0; | |
837f66c7 DM |
424 | |
425 | if (unlikely(!mc->objects)) { | |
426 | if (WARN_ON_ONCE(!capacity)) | |
427 | return -EIO; | |
428 | ||
c23e2b71 SC |
429 | /* |
430 | * Custom init values can be used only for page allocations, | |
431 | * and obviously conflict with __GFP_ZERO. | |
432 | */ | |
433 | if (WARN_ON_ONCE(mc->init_value && (mc->kmem_cache || mc->gfp_zero))) | |
434 | return -EIO; | |
435 | ||
ea3689d9 | 436 | mc->objects = kvmalloc_array(capacity, sizeof(void *), gfp); |
837f66c7 DM |
437 | if (!mc->objects) |
438 | return -ENOMEM; | |
439 | ||
440 | mc->capacity = capacity; | |
441 | } | |
442 | ||
443 | /* It is illegal to request a different capacity across topups. */ | |
444 | if (WARN_ON_ONCE(mc->capacity != capacity)) | |
445 | return -EIO; | |
446 | ||
447 | while (mc->nobjs < mc->capacity) { | |
448 | obj = mmu_memory_cache_alloc_obj(mc, gfp); | |
6926f95a SC |
449 | if (!obj) |
450 | return mc->nobjs >= min ? 0 : -ENOMEM; | |
451 | mc->objects[mc->nobjs++] = obj; | |
452 | } | |
453 | return 0; | |
454 | } | |
455 | ||
837f66c7 DM |
456 | int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min) |
457 | { | |
458 | return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min); | |
459 | } | |
460 | ||
6926f95a SC |
461 | int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc) |
462 | { | |
463 | return mc->nobjs; | |
464 | } | |
465 | ||
466 | void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
467 | { | |
468 | while (mc->nobjs) { | |
469 | if (mc->kmem_cache) | |
470 | kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]); | |
471 | else | |
472 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
473 | } | |
837f66c7 DM |
474 | |
475 | kvfree(mc->objects); | |
476 | ||
477 | mc->objects = NULL; | |
478 | mc->capacity = 0; | |
6926f95a SC |
479 | } |
480 | ||
481 | void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
482 | { | |
483 | void *p; | |
484 | ||
485 | if (WARN_ON(!mc->nobjs)) | |
486 | p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT); | |
487 | else | |
488 | p = mc->objects[--mc->nobjs]; | |
489 | BUG_ON(!p); | |
490 | return p; | |
491 | } | |
492 | #endif | |
493 | ||
8bd826d6 | 494 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 495 | { |
fb3f0f51 RR |
496 | mutex_init(&vcpu->mutex); |
497 | vcpu->cpu = -1; | |
fb3f0f51 RR |
498 | vcpu->kvm = kvm; |
499 | vcpu->vcpu_id = id; | |
34bb10b7 | 500 | vcpu->pid = NULL; |
510958e9 | 501 | #ifndef __KVM_HAVE_ARCH_WQP |
da4ad88c | 502 | rcuwait_init(&vcpu->wait); |
510958e9 | 503 | #endif |
af585b92 | 504 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 505 | |
4c088493 R |
506 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
507 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 508 | vcpu->preempted = false; |
d73eb57b | 509 | vcpu->ready = false; |
d5c48deb | 510 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
a54d8066 | 511 | vcpu->last_used_slot = NULL; |
58fc1166 OU |
512 | |
513 | /* Fill the stats id string for the vcpu */ | |
514 | snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d", | |
515 | task_pid_nr(current), id); | |
fb3f0f51 | 516 | } |
fb3f0f51 | 517 | |
27592ae8 | 518 | static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
4543bdc0 SC |
519 | { |
520 | kvm_arch_vcpu_destroy(vcpu); | |
5593473a | 521 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
e529ef66 | 522 | |
9941d224 SC |
523 | /* |
524 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
525 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
526 | * are already gone. | |
527 | */ | |
528 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
529 | ||
8bd826d6 | 530 | free_page((unsigned long)vcpu->run); |
e529ef66 | 531 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 | 532 | } |
27592ae8 MZ |
533 | |
534 | void kvm_destroy_vcpus(struct kvm *kvm) | |
535 | { | |
46808a4c | 536 | unsigned long i; |
27592ae8 MZ |
537 | struct kvm_vcpu *vcpu; |
538 | ||
539 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
540 | kvm_vcpu_destroy(vcpu); | |
c5b07754 | 541 | xa_erase(&kvm->vcpu_array, i); |
27592ae8 MZ |
542 | } |
543 | ||
544 | atomic_set(&kvm->online_vcpus, 0); | |
545 | } | |
546 | EXPORT_SYMBOL_GPL(kvm_destroy_vcpus); | |
4543bdc0 | 547 | |
f128cf8c | 548 | #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER |
e930bffe AA |
549 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) |
550 | { | |
551 | return container_of(mn, struct kvm, mmu_notifier); | |
552 | } | |
553 | ||
e97b39c5 | 554 | typedef bool (*gfn_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range); |
3039bcc7 | 555 | |
8569992d | 556 | typedef void (*on_lock_fn_t)(struct kvm *kvm); |
683412cc | 557 | |
e97b39c5 SC |
558 | struct kvm_mmu_notifier_range { |
559 | /* | |
560 | * 64-bit addresses, as KVM notifiers can operate on host virtual | |
561 | * addresses (unsigned long) and guest physical addresses (64-bit). | |
562 | */ | |
563 | u64 start; | |
564 | u64 end; | |
3e1efe2b | 565 | union kvm_mmu_notifier_arg arg; |
e97b39c5 | 566 | gfn_handler_t handler; |
f922bd9b | 567 | on_lock_fn_t on_lock; |
3039bcc7 SC |
568 | bool flush_on_ret; |
569 | bool may_block; | |
570 | }; | |
571 | ||
cec29eef SC |
572 | /* |
573 | * The inner-most helper returns a tuple containing the return value from the | |
574 | * arch- and action-specific handler, plus a flag indicating whether or not at | |
575 | * least one memslot was found, i.e. if the handler found guest memory. | |
576 | * | |
577 | * Note, most notifiers are averse to booleans, so even though KVM tracks the | |
578 | * return from arch code as a bool, outer helpers will cast it to an int. :-( | |
579 | */ | |
580 | typedef struct kvm_mmu_notifier_return { | |
581 | bool ret; | |
582 | bool found_memslot; | |
583 | } kvm_mn_ret_t; | |
584 | ||
f922bd9b SC |
585 | /* |
586 | * Use a dedicated stub instead of NULL to indicate that there is no callback | |
587 | * function/handler. The compiler technically can't guarantee that a real | |
588 | * function will have a non-zero address, and so it will generate code to | |
589 | * check for !NULL, whereas comparing against a stub will be elided at compile | |
590 | * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9). | |
591 | */ | |
592 | static void kvm_null_fn(void) | |
593 | { | |
594 | ||
595 | } | |
596 | #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn) | |
597 | ||
ed922739 MS |
598 | /* Iterate over each memslot intersecting [start, last] (inclusive) range */ |
599 | #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \ | |
600 | for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \ | |
601 | node; \ | |
602 | node = interval_tree_iter_next(node, start, last)) \ | |
603 | ||
cec29eef SC |
604 | static __always_inline kvm_mn_ret_t __kvm_handle_hva_range(struct kvm *kvm, |
605 | const struct kvm_mmu_notifier_range *range) | |
3039bcc7 | 606 | { |
cec29eef SC |
607 | struct kvm_mmu_notifier_return r = { |
608 | .ret = false, | |
609 | .found_memslot = false, | |
610 | }; | |
f922bd9b | 611 | struct kvm_gfn_range gfn_range; |
3039bcc7 SC |
612 | struct kvm_memory_slot *slot; |
613 | struct kvm_memslots *slots; | |
3039bcc7 SC |
614 | int i, idx; |
615 | ||
ed922739 | 616 | if (WARN_ON_ONCE(range->end <= range->start)) |
cec29eef | 617 | return r; |
ed922739 | 618 | |
f922bd9b SC |
619 | /* A null handler is allowed if and only if on_lock() is provided. */ |
620 | if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) && | |
621 | IS_KVM_NULL_FN(range->handler))) | |
cec29eef | 622 | return r; |
f922bd9b | 623 | |
3039bcc7 SC |
624 | idx = srcu_read_lock(&kvm->srcu); |
625 | ||
eed52e43 | 626 | for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { |
ed922739 MS |
627 | struct interval_tree_node *node; |
628 | ||
3039bcc7 | 629 | slots = __kvm_memslots(kvm, i); |
ed922739 MS |
630 | kvm_for_each_memslot_in_hva_range(node, slots, |
631 | range->start, range->end - 1) { | |
3039bcc7 SC |
632 | unsigned long hva_start, hva_end; |
633 | ||
a54d8066 | 634 | slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]); |
e97b39c5 SC |
635 | hva_start = max_t(unsigned long, range->start, slot->userspace_addr); |
636 | hva_end = min_t(unsigned long, range->end, | |
637 | slot->userspace_addr + (slot->npages << PAGE_SHIFT)); | |
3039bcc7 SC |
638 | |
639 | /* | |
640 | * To optimize for the likely case where the address | |
641 | * range is covered by zero or one memslots, don't | |
642 | * bother making these conditional (to avoid writes on | |
643 | * the second or later invocation of the handler). | |
644 | */ | |
3e1efe2b | 645 | gfn_range.arg = range->arg; |
3039bcc7 SC |
646 | gfn_range.may_block = range->may_block; |
647 | ||
648 | /* | |
649 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
650 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
651 | */ | |
652 | gfn_range.start = hva_to_gfn_memslot(hva_start, slot); | |
653 | gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot); | |
654 | gfn_range.slot = slot; | |
655 | ||
cec29eef SC |
656 | if (!r.found_memslot) { |
657 | r.found_memslot = true; | |
8931a454 | 658 | KVM_MMU_LOCK(kvm); |
071064f1 | 659 | if (!IS_KVM_NULL_FN(range->on_lock)) |
8569992d CP |
660 | range->on_lock(kvm); |
661 | ||
071064f1 PB |
662 | if (IS_KVM_NULL_FN(range->handler)) |
663 | break; | |
8931a454 | 664 | } |
cec29eef | 665 | r.ret |= range->handler(kvm, &gfn_range); |
3039bcc7 SC |
666 | } |
667 | } | |
668 | ||
cec29eef | 669 | if (range->flush_on_ret && r.ret) |
3039bcc7 SC |
670 | kvm_flush_remote_tlbs(kvm); |
671 | ||
193bbfaa | 672 | if (r.found_memslot) |
8931a454 | 673 | KVM_MMU_UNLOCK(kvm); |
f922bd9b | 674 | |
3039bcc7 SC |
675 | srcu_read_unlock(&kvm->srcu, idx); |
676 | ||
cec29eef | 677 | return r; |
3039bcc7 SC |
678 | } |
679 | ||
680 | static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn, | |
681 | unsigned long start, | |
682 | unsigned long end, | |
e97b39c5 | 683 | gfn_handler_t handler) |
3039bcc7 SC |
684 | { |
685 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
e97b39c5 | 686 | const struct kvm_mmu_notifier_range range = { |
3039bcc7 SC |
687 | .start = start, |
688 | .end = end, | |
3039bcc7 | 689 | .handler = handler, |
f922bd9b | 690 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
691 | .flush_on_ret = true, |
692 | .may_block = false, | |
693 | }; | |
3039bcc7 | 694 | |
cec29eef | 695 | return __kvm_handle_hva_range(kvm, &range).ret; |
3039bcc7 SC |
696 | } |
697 | ||
698 | static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn, | |
699 | unsigned long start, | |
700 | unsigned long end, | |
e97b39c5 | 701 | gfn_handler_t handler) |
3039bcc7 SC |
702 | { |
703 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
e97b39c5 | 704 | const struct kvm_mmu_notifier_range range = { |
3039bcc7 SC |
705 | .start = start, |
706 | .end = end, | |
3039bcc7 | 707 | .handler = handler, |
f922bd9b | 708 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
709 | .flush_on_ret = false, |
710 | .may_block = false, | |
711 | }; | |
3039bcc7 | 712 | |
cec29eef | 713 | return __kvm_handle_hva_range(kvm, &range).ret; |
3039bcc7 | 714 | } |
2230f9e1 | 715 | |
8569992d | 716 | void kvm_mmu_invalidate_begin(struct kvm *kvm) |
e930bffe | 717 | { |
8569992d | 718 | lockdep_assert_held_write(&kvm->mmu_lock); |
e930bffe AA |
719 | /* |
720 | * The count increase must become visible at unlock time as no | |
721 | * spte can be established without taking the mmu_lock and | |
722 | * count is also read inside the mmu_lock critical section. | |
723 | */ | |
20ec3ebd | 724 | kvm->mmu_invalidate_in_progress++; |
8569992d | 725 | |
20ec3ebd | 726 | if (likely(kvm->mmu_invalidate_in_progress == 1)) { |
8569992d CP |
727 | kvm->mmu_invalidate_range_start = INVALID_GPA; |
728 | kvm->mmu_invalidate_range_end = INVALID_GPA; | |
729 | } | |
730 | } | |
731 | ||
732 | void kvm_mmu_invalidate_range_add(struct kvm *kvm, gfn_t start, gfn_t end) | |
733 | { | |
734 | lockdep_assert_held_write(&kvm->mmu_lock); | |
735 | ||
736 | WARN_ON_ONCE(!kvm->mmu_invalidate_in_progress); | |
737 | ||
738 | if (likely(kvm->mmu_invalidate_range_start == INVALID_GPA)) { | |
20ec3ebd CP |
739 | kvm->mmu_invalidate_range_start = start; |
740 | kvm->mmu_invalidate_range_end = end; | |
4a42d848 DS |
741 | } else { |
742 | /* | |
a413a625 | 743 | * Fully tracking multiple concurrent ranges has diminishing |
4a42d848 DS |
744 | * returns. Keep things simple and just find the minimal range |
745 | * which includes the current and new ranges. As there won't be | |
746 | * enough information to subtract a range after its invalidate | |
747 | * completes, any ranges invalidated concurrently will | |
748 | * accumulate and persist until all outstanding invalidates | |
749 | * complete. | |
750 | */ | |
20ec3ebd CP |
751 | kvm->mmu_invalidate_range_start = |
752 | min(kvm->mmu_invalidate_range_start, start); | |
753 | kvm->mmu_invalidate_range_end = | |
754 | max(kvm->mmu_invalidate_range_end, end); | |
4a42d848 | 755 | } |
f922bd9b | 756 | } |
3039bcc7 | 757 | |
a7800aa8 | 758 | bool kvm_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range) |
8569992d CP |
759 | { |
760 | kvm_mmu_invalidate_range_add(kvm, range->start, range->end); | |
761 | return kvm_unmap_gfn_range(kvm, range); | |
762 | } | |
763 | ||
f922bd9b SC |
764 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
765 | const struct mmu_notifier_range *range) | |
766 | { | |
767 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
e97b39c5 | 768 | const struct kvm_mmu_notifier_range hva_range = { |
f922bd9b SC |
769 | .start = range->start, |
770 | .end = range->end, | |
8569992d | 771 | .handler = kvm_mmu_unmap_gfn_range, |
20ec3ebd | 772 | .on_lock = kvm_mmu_invalidate_begin, |
f922bd9b SC |
773 | .flush_on_ret = true, |
774 | .may_block = mmu_notifier_range_blockable(range), | |
775 | }; | |
565f3be2 | 776 | |
f922bd9b SC |
777 | trace_kvm_unmap_hva_range(range->start, range->end); |
778 | ||
52ac8b35 PB |
779 | /* |
780 | * Prevent memslot modification between range_start() and range_end() | |
781 | * so that conditionally locking provides the same result in both | |
20ec3ebd | 782 | * functions. Without that guarantee, the mmu_invalidate_in_progress |
52ac8b35 PB |
783 | * adjustments will be imbalanced. |
784 | * | |
785 | * Pairs with the decrement in range_end(). | |
786 | */ | |
787 | spin_lock(&kvm->mn_invalidate_lock); | |
788 | kvm->mn_active_invalidate_count++; | |
789 | spin_unlock(&kvm->mn_invalidate_lock); | |
790 | ||
58cd407c SC |
791 | /* |
792 | * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e. | |
793 | * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring | |
794 | * each cache's lock. There are relatively few caches in existence at | |
795 | * any given time, and the caches themselves can check for hva overlap, | |
796 | * i.e. don't need to rely on memslot overlap checks for performance. | |
797 | * Because this runs without holding mmu_lock, the pfn caches must use | |
20ec3ebd CP |
798 | * mn_active_invalidate_count (see above) instead of |
799 | * mmu_invalidate_in_progress. | |
58cd407c | 800 | */ |
982ed0de DW |
801 | gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end, |
802 | hva_range.may_block); | |
803 | ||
cec29eef SC |
804 | /* |
805 | * If one or more memslots were found and thus zapped, notify arch code | |
806 | * that guest memory has been reclaimed. This needs to be done *after* | |
807 | * dropping mmu_lock, as x86's reclaim path is slooooow. | |
808 | */ | |
809 | if (__kvm_handle_hva_range(kvm, &hva_range).found_memslot) | |
810 | kvm_arch_guest_memory_reclaimed(kvm); | |
93065ac7 | 811 | |
e649b3f0 | 812 | return 0; |
e930bffe AA |
813 | } |
814 | ||
8569992d | 815 | void kvm_mmu_invalidate_end(struct kvm *kvm) |
e930bffe | 816 | { |
8569992d CP |
817 | lockdep_assert_held_write(&kvm->mmu_lock); |
818 | ||
e930bffe AA |
819 | /* |
820 | * This sequence increase will notify the kvm page fault that | |
821 | * the page that is going to be mapped in the spte could have | |
822 | * been freed. | |
823 | */ | |
20ec3ebd | 824 | kvm->mmu_invalidate_seq++; |
a355aa54 | 825 | smp_wmb(); |
e930bffe AA |
826 | /* |
827 | * The above sequence increase must be visible before the | |
a355aa54 | 828 | * below count decrease, which is ensured by the smp_wmb above |
20ec3ebd | 829 | * in conjunction with the smp_rmb in mmu_invalidate_retry(). |
e930bffe | 830 | */ |
20ec3ebd | 831 | kvm->mmu_invalidate_in_progress--; |
c0db1923 | 832 | KVM_BUG_ON(kvm->mmu_invalidate_in_progress < 0, kvm); |
8569992d CP |
833 | |
834 | /* | |
835 | * Assert that at least one range was added between start() and end(). | |
836 | * Not adding a range isn't fatal, but it is a KVM bug. | |
837 | */ | |
838 | WARN_ON_ONCE(kvm->mmu_invalidate_range_start == INVALID_GPA); | |
f922bd9b SC |
839 | } |
840 | ||
841 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
842 | const struct mmu_notifier_range *range) | |
843 | { | |
844 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
e97b39c5 | 845 | const struct kvm_mmu_notifier_range hva_range = { |
f922bd9b SC |
846 | .start = range->start, |
847 | .end = range->end, | |
f922bd9b | 848 | .handler = (void *)kvm_null_fn, |
20ec3ebd | 849 | .on_lock = kvm_mmu_invalidate_end, |
f922bd9b SC |
850 | .flush_on_ret = false, |
851 | .may_block = mmu_notifier_range_blockable(range), | |
852 | }; | |
52ac8b35 | 853 | bool wake; |
f922bd9b SC |
854 | |
855 | __kvm_handle_hva_range(kvm, &hva_range); | |
e930bffe | 856 | |
52ac8b35 PB |
857 | /* Pairs with the increment in range_start(). */ |
858 | spin_lock(&kvm->mn_invalidate_lock); | |
d489ec95 SC |
859 | if (!WARN_ON_ONCE(!kvm->mn_active_invalidate_count)) |
860 | --kvm->mn_active_invalidate_count; | |
861 | wake = !kvm->mn_active_invalidate_count; | |
52ac8b35 PB |
862 | spin_unlock(&kvm->mn_invalidate_lock); |
863 | ||
864 | /* | |
865 | * There can only be one waiter, since the wait happens under | |
866 | * slots_lock. | |
867 | */ | |
868 | if (wake) | |
869 | rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait); | |
e930bffe AA |
870 | } |
871 | ||
872 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
873 | struct mm_struct *mm, | |
57128468 ALC |
874 | unsigned long start, |
875 | unsigned long end) | |
e930bffe | 876 | { |
501b9185 SC |
877 | trace_kvm_age_hva(start, end); |
878 | ||
5257de95 | 879 | return kvm_handle_hva_range(mn, start, end, kvm_age_gfn); |
e930bffe AA |
880 | } |
881 | ||
1d7715c6 VD |
882 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
883 | struct mm_struct *mm, | |
884 | unsigned long start, | |
885 | unsigned long end) | |
886 | { | |
501b9185 SC |
887 | trace_kvm_age_hva(start, end); |
888 | ||
1d7715c6 VD |
889 | /* |
890 | * Even though we do not flush TLB, this will still adversely | |
891 | * affect performance on pre-Haswell Intel EPT, where there is | |
892 | * no EPT Access Bit to clear so that we have to tear down EPT | |
893 | * tables instead. If we find this unacceptable, we can always | |
894 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
895 | * do anything on clear_young. | |
896 | * | |
897 | * Also note that currently we never issue secondary TLB flushes | |
898 | * from clear_young, leaving this job up to the regular system | |
899 | * cadence. If we find this inaccurate, we might come up with a | |
900 | * more sophisticated heuristic later. | |
901 | */ | |
3039bcc7 | 902 | return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn); |
1d7715c6 VD |
903 | } |
904 | ||
8ee53820 AA |
905 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
906 | struct mm_struct *mm, | |
907 | unsigned long address) | |
908 | { | |
501b9185 SC |
909 | trace_kvm_test_age_hva(address); |
910 | ||
3039bcc7 SC |
911 | return kvm_handle_hva_range_no_flush(mn, address, address + 1, |
912 | kvm_test_age_gfn); | |
8ee53820 AA |
913 | } |
914 | ||
85db06e5 MT |
915 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
916 | struct mm_struct *mm) | |
917 | { | |
918 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
919 | int idx; |
920 | ||
921 | idx = srcu_read_lock(&kvm->srcu); | |
683412cc | 922 | kvm_flush_shadow_all(kvm); |
eda2beda | 923 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
924 | } |
925 | ||
e930bffe | 926 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e930bffe AA |
927 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
928 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
929 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 930 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 931 | .test_young = kvm_mmu_notifier_test_young, |
85db06e5 | 932 | .release = kvm_mmu_notifier_release, |
e930bffe | 933 | }; |
4c07b0a4 AK |
934 | |
935 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
936 | { | |
937 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
938 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
939 | } | |
940 | ||
f128cf8c | 941 | #else /* !CONFIG_KVM_GENERIC_MMU_NOTIFIER */ |
4c07b0a4 AK |
942 | |
943 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
944 | { | |
945 | return 0; | |
946 | } | |
947 | ||
f128cf8c | 948 | #endif /* CONFIG_KVM_GENERIC_MMU_NOTIFIER */ |
e930bffe | 949 | |
2fdef3a2 SS |
950 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
951 | static int kvm_pm_notifier_call(struct notifier_block *bl, | |
952 | unsigned long state, | |
953 | void *unused) | |
954 | { | |
955 | struct kvm *kvm = container_of(bl, struct kvm, pm_notifier); | |
956 | ||
957 | return kvm_arch_pm_notifier(kvm, state); | |
958 | } | |
959 | ||
960 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
961 | { | |
962 | kvm->pm_notifier.notifier_call = kvm_pm_notifier_call; | |
963 | /* Suspend KVM before we suspend ftrace, RCU, etc. */ | |
964 | kvm->pm_notifier.priority = INT_MAX; | |
965 | register_pm_notifier(&kvm->pm_notifier); | |
966 | } | |
967 | ||
968 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
969 | { | |
970 | unregister_pm_notifier(&kvm->pm_notifier); | |
971 | } | |
972 | #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
973 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
974 | { | |
975 | } | |
976 | ||
977 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
978 | { | |
979 | } | |
980 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
981 | ||
a47d2b07 PB |
982 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
983 | { | |
984 | if (!memslot->dirty_bitmap) | |
985 | return; | |
986 | ||
987 | kvfree(memslot->dirty_bitmap); | |
988 | memslot->dirty_bitmap = NULL; | |
989 | } | |
990 | ||
a54d8066 | 991 | /* This does not remove the slot from struct kvm_memslots data structures */ |
e96c81ee | 992 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 993 | { |
a7800aa8 SC |
994 | if (slot->flags & KVM_MEM_GUEST_MEMFD) |
995 | kvm_gmem_unbind(slot); | |
996 | ||
e96c81ee | 997 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 998 | |
e96c81ee | 999 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 1000 | |
a54d8066 | 1001 | kfree(slot); |
a47d2b07 PB |
1002 | } |
1003 | ||
1004 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
1005 | { | |
a54d8066 | 1006 | struct hlist_node *idnode; |
a47d2b07 | 1007 | struct kvm_memory_slot *memslot; |
a54d8066 | 1008 | int bkt; |
a47d2b07 | 1009 | |
a54d8066 MS |
1010 | /* |
1011 | * The same memslot objects live in both active and inactive sets, | |
1012 | * arbitrarily free using index '1' so the second invocation of this | |
1013 | * function isn't operating over a structure with dangling pointers | |
1014 | * (even though this function isn't actually touching them). | |
1015 | */ | |
1016 | if (!slots->node_idx) | |
a47d2b07 PB |
1017 | return; |
1018 | ||
a54d8066 | 1019 | hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1]) |
e96c81ee | 1020 | kvm_free_memslot(kvm, memslot); |
bf3e05bc XG |
1021 | } |
1022 | ||
bc9e9e67 JZ |
1023 | static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc) |
1024 | { | |
1025 | switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) { | |
1026 | case KVM_STATS_TYPE_INSTANT: | |
1027 | return 0444; | |
1028 | case KVM_STATS_TYPE_CUMULATIVE: | |
1029 | case KVM_STATS_TYPE_PEAK: | |
1030 | default: | |
1031 | return 0644; | |
1032 | } | |
1033 | } | |
1034 | ||
1035 | ||
536a6f88 JF |
1036 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
1037 | { | |
1038 | int i; | |
bc9e9e67 JZ |
1039 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
1040 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 | 1041 | |
a44a4cc1 | 1042 | if (IS_ERR(kvm->debugfs_dentry)) |
536a6f88 JF |
1043 | return; |
1044 | ||
1045 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
1046 | ||
9d5a1dce LC |
1047 | if (kvm->debugfs_stat_data) { |
1048 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
1049 | kfree(kvm->debugfs_stat_data[i]); | |
1050 | kfree(kvm->debugfs_stat_data); | |
1051 | } | |
536a6f88 JF |
1052 | } |
1053 | ||
59f82aad | 1054 | static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname) |
536a6f88 | 1055 | { |
85cd39af PB |
1056 | static DEFINE_MUTEX(kvm_debugfs_lock); |
1057 | struct dentry *dent; | |
536a6f88 JF |
1058 | char dir_name[ITOA_MAX_LEN * 2]; |
1059 | struct kvm_stat_data *stat_data; | |
bc9e9e67 | 1060 | const struct _kvm_stats_desc *pdesc; |
b74ed7a6 | 1061 | int i, ret = -ENOMEM; |
bc9e9e67 JZ |
1062 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
1063 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
1064 | |
1065 | if (!debugfs_initialized()) | |
1066 | return 0; | |
1067 | ||
59f82aad | 1068 | snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname); |
85cd39af PB |
1069 | mutex_lock(&kvm_debugfs_lock); |
1070 | dent = debugfs_lookup(dir_name, kvm_debugfs_dir); | |
1071 | if (dent) { | |
1072 | pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name); | |
1073 | dput(dent); | |
1074 | mutex_unlock(&kvm_debugfs_lock); | |
1075 | return 0; | |
1076 | } | |
1077 | dent = debugfs_create_dir(dir_name, kvm_debugfs_dir); | |
1078 | mutex_unlock(&kvm_debugfs_lock); | |
1079 | if (IS_ERR(dent)) | |
1080 | return 0; | |
536a6f88 | 1081 | |
85cd39af | 1082 | kvm->debugfs_dentry = dent; |
536a6f88 JF |
1083 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
1084 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 1085 | GFP_KERNEL_ACCOUNT); |
536a6f88 | 1086 | if (!kvm->debugfs_stat_data) |
b74ed7a6 | 1087 | goto out_err; |
536a6f88 | 1088 | |
bc9e9e67 JZ |
1089 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
1090 | pdesc = &kvm_vm_stats_desc[i]; | |
b12ce36a | 1091 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 | 1092 | if (!stat_data) |
b74ed7a6 | 1093 | goto out_err; |
536a6f88 JF |
1094 | |
1095 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1096 | stat_data->desc = pdesc; |
1097 | stat_data->kind = KVM_STAT_VM; | |
1098 | kvm->debugfs_stat_data[i] = stat_data; | |
1099 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
1100 | kvm->debugfs_dentry, stat_data, | |
1101 | &stat_fops_per_vm); | |
1102 | } | |
1103 | ||
1104 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
1105 | pdesc = &kvm_vcpu_stats_desc[i]; | |
b12ce36a | 1106 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 | 1107 | if (!stat_data) |
b74ed7a6 | 1108 | goto out_err; |
536a6f88 JF |
1109 | |
1110 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1111 | stat_data->desc = pdesc; |
1112 | stat_data->kind = KVM_STAT_VCPU; | |
004d62eb | 1113 | kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data; |
bc9e9e67 | 1114 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), |
09cbcef6 MP |
1115 | kvm->debugfs_dentry, stat_data, |
1116 | &stat_fops_per_vm); | |
536a6f88 | 1117 | } |
3165af73 | 1118 | |
284851ee | 1119 | kvm_arch_create_vm_debugfs(kvm); |
536a6f88 | 1120 | return 0; |
b74ed7a6 OU |
1121 | out_err: |
1122 | kvm_destroy_vm_debugfs(kvm); | |
1123 | return ret; | |
536a6f88 JF |
1124 | } |
1125 | ||
1aa9b957 JS |
1126 | /* |
1127 | * Called after the VM is otherwise initialized, but just before adding it to | |
1128 | * the vm_list. | |
1129 | */ | |
1130 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
1131 | { | |
1132 | return 0; | |
1133 | } | |
1134 | ||
1135 | /* | |
1136 | * Called just after removing the VM from the vm_list, but before doing any | |
1137 | * other destruction. | |
1138 | */ | |
1139 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
1140 | { | |
1141 | } | |
1142 | ||
3165af73 PX |
1143 | /* |
1144 | * Called after per-vm debugfs created. When called kvm->debugfs_dentry should | |
1145 | * be setup already, so we can create arch-specific debugfs entries under it. | |
1146 | * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so | |
1147 | * a per-arch destroy interface is not needed. | |
1148 | */ | |
284851ee | 1149 | void __weak kvm_arch_create_vm_debugfs(struct kvm *kvm) |
3165af73 | 1150 | { |
3165af73 PX |
1151 | } |
1152 | ||
b74ed7a6 | 1153 | static struct kvm *kvm_create_vm(unsigned long type, const char *fdname) |
6aa8b732 | 1154 | { |
d89f5eff | 1155 | struct kvm *kvm = kvm_arch_alloc_vm(); |
a54d8066 | 1156 | struct kvm_memslots *slots; |
9121923c | 1157 | int r = -ENOMEM; |
a54d8066 | 1158 | int i, j; |
6aa8b732 | 1159 | |
d89f5eff JK |
1160 | if (!kvm) |
1161 | return ERR_PTR(-ENOMEM); | |
1162 | ||
531810ca | 1163 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 1164 | mmgrab(current->mm); |
e9ad4ec8 PB |
1165 | kvm->mm = current->mm; |
1166 | kvm_eventfd_init(kvm); | |
1167 | mutex_init(&kvm->lock); | |
1168 | mutex_init(&kvm->irq_lock); | |
1169 | mutex_init(&kvm->slots_lock); | |
b10a038e | 1170 | mutex_init(&kvm->slots_arch_lock); |
52ac8b35 PB |
1171 | spin_lock_init(&kvm->mn_invalidate_lock); |
1172 | rcuwait_init(&kvm->mn_memslots_update_rcuwait); | |
c5b07754 | 1173 | xa_init(&kvm->vcpu_array); |
5a475554 CP |
1174 | #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES |
1175 | xa_init(&kvm->mem_attr_array); | |
1176 | #endif | |
52ac8b35 | 1177 | |
982ed0de DW |
1178 | INIT_LIST_HEAD(&kvm->gpc_list); |
1179 | spin_lock_init(&kvm->gpc_lock); | |
52ac8b35 | 1180 | |
e9ad4ec8 | 1181 | INIT_LIST_HEAD(&kvm->devices); |
f502cc56 | 1182 | kvm->max_vcpus = KVM_MAX_VCPUS; |
e9ad4ec8 | 1183 | |
1e702d9a AW |
1184 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
1185 | ||
5c697c36 SC |
1186 | /* |
1187 | * Force subsequent debugfs file creations to fail if the VM directory | |
1188 | * is not created (by kvm_create_vm_debugfs()). | |
1189 | */ | |
1190 | kvm->debugfs_dentry = ERR_PTR(-ENOENT); | |
1191 | ||
f2759c08 OU |
1192 | snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d", |
1193 | task_pid_nr(current)); | |
1194 | ||
8a44119a PB |
1195 | if (init_srcu_struct(&kvm->srcu)) |
1196 | goto out_err_no_srcu; | |
1197 | if (init_srcu_struct(&kvm->irq_srcu)) | |
1198 | goto out_err_no_irq_srcu; | |
1199 | ||
e2d3fcaf | 1200 | refcount_set(&kvm->users_count, 1); |
eed52e43 | 1201 | for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { |
a54d8066 MS |
1202 | for (j = 0; j < 2; j++) { |
1203 | slots = &kvm->__memslots[i][j]; | |
9121923c | 1204 | |
a54d8066 MS |
1205 | atomic_long_set(&slots->last_used_slot, (unsigned long)NULL); |
1206 | slots->hva_tree = RB_ROOT_CACHED; | |
1207 | slots->gfn_tree = RB_ROOT; | |
1208 | hash_init(slots->id_hash); | |
1209 | slots->node_idx = j; | |
1210 | ||
1211 | /* Generations must be different for each address space. */ | |
1212 | slots->generation = i; | |
1213 | } | |
1214 | ||
1215 | rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]); | |
f481b069 | 1216 | } |
00f034a1 | 1217 | |
e93f8a0f | 1218 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 1219 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 1220 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 1221 | if (!kvm->buses[i]) |
a97b0e77 | 1222 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 1223 | } |
e930bffe | 1224 | |
e08b9637 | 1225 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 1226 | if (r) |
a97b0e77 | 1227 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
1228 | |
1229 | r = hardware_enable_all(); | |
1230 | if (r) | |
719d93cd | 1231 | goto out_err_no_disable; |
10474ae8 | 1232 | |
c5b31cc2 | 1233 | #ifdef CONFIG_HAVE_KVM_IRQCHIP |
136bdfee | 1234 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 1235 | #endif |
6aa8b732 | 1236 | |
74b5c5bf | 1237 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
1238 | if (r) |
1239 | goto out_err_no_mmu_notifier; | |
1240 | ||
c2b82397 SC |
1241 | r = kvm_coalesced_mmio_init(kvm); |
1242 | if (r < 0) | |
1243 | goto out_no_coalesced_mmio; | |
1244 | ||
4ba4f419 SC |
1245 | r = kvm_create_vm_debugfs(kvm, fdname); |
1246 | if (r) | |
1247 | goto out_err_no_debugfs; | |
1248 | ||
1aa9b957 | 1249 | r = kvm_arch_post_init_vm(kvm); |
74b5c5bf | 1250 | if (r) |
4ba4f419 | 1251 | goto out_err; |
74b5c5bf | 1252 | |
0d9ce162 | 1253 | mutex_lock(&kvm_lock); |
5e58cfe4 | 1254 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 1255 | mutex_unlock(&kvm_lock); |
d89f5eff | 1256 | |
2ecd9d29 | 1257 | preempt_notifier_inc(); |
2fdef3a2 | 1258 | kvm_init_pm_notifier(kvm); |
2ecd9d29 | 1259 | |
f17abe9a | 1260 | return kvm; |
10474ae8 AG |
1261 | |
1262 | out_err: | |
4ba4f419 SC |
1263 | kvm_destroy_vm_debugfs(kvm); |
1264 | out_err_no_debugfs: | |
c2b82397 SC |
1265 | kvm_coalesced_mmio_free(kvm); |
1266 | out_no_coalesced_mmio: | |
f128cf8c | 1267 | #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER |
1aa9b957 JS |
1268 | if (kvm->mmu_notifier.ops) |
1269 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
1270 | #endif | |
1271 | out_err_no_mmu_notifier: | |
10474ae8 | 1272 | hardware_disable_all(); |
719d93cd | 1273 | out_err_no_disable: |
a97b0e77 | 1274 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 1275 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 1276 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 1277 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 1278 | kfree(kvm_get_bus(kvm, i)); |
8a44119a PB |
1279 | cleanup_srcu_struct(&kvm->irq_srcu); |
1280 | out_err_no_irq_srcu: | |
1281 | cleanup_srcu_struct(&kvm->srcu); | |
1282 | out_err_no_srcu: | |
d89f5eff | 1283 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 1284 | mmdrop(current->mm); |
10474ae8 | 1285 | return ERR_PTR(r); |
f17abe9a AK |
1286 | } |
1287 | ||
07f0a7bd SW |
1288 | static void kvm_destroy_devices(struct kvm *kvm) |
1289 | { | |
e6e3b5a6 | 1290 | struct kvm_device *dev, *tmp; |
07f0a7bd | 1291 | |
a28ebea2 CD |
1292 | /* |
1293 | * We do not need to take the kvm->lock here, because nobody else | |
1294 | * has a reference to the struct kvm at this point and therefore | |
1295 | * cannot access the devices list anyhow. | |
1296 | */ | |
e6e3b5a6 GT |
1297 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
1298 | list_del(&dev->vm_node); | |
07f0a7bd SW |
1299 | dev->ops->destroy(dev); |
1300 | } | |
1301 | } | |
1302 | ||
f17abe9a AK |
1303 | static void kvm_destroy_vm(struct kvm *kvm) |
1304 | { | |
e93f8a0f | 1305 | int i; |
6d4e4c4f AK |
1306 | struct mm_struct *mm = kvm->mm; |
1307 | ||
2fdef3a2 | 1308 | kvm_destroy_pm_notifier(kvm); |
286de8f6 | 1309 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 1310 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 1311 | kvm_arch_sync_events(kvm); |
0d9ce162 | 1312 | mutex_lock(&kvm_lock); |
133de902 | 1313 | list_del(&kvm->vm_list); |
0d9ce162 | 1314 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
1315 | kvm_arch_pre_destroy_vm(kvm); |
1316 | ||
399ec807 | 1317 | kvm_free_irq_routing(kvm); |
df630b8c | 1318 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 1319 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 1320 | |
4a12f951 CB |
1321 | if (bus) |
1322 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
1323 | kvm->buses[i] = NULL; |
1324 | } | |
980da6ce | 1325 | kvm_coalesced_mmio_free(kvm); |
f128cf8c | 1326 | #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER |
e930bffe | 1327 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); |
52ac8b35 PB |
1328 | /* |
1329 | * At this point, pending calls to invalidate_range_start() | |
1330 | * have completed but no more MMU notifiers will run, so | |
1331 | * mn_active_invalidate_count may remain unbalanced. | |
b0d23708 | 1332 | * No threads can be waiting in kvm_swap_active_memslots() as the |
52ac8b35 PB |
1333 | * last reference on KVM has been dropped, but freeing |
1334 | * memslots would deadlock without this manual intervention. | |
d497a0fa SC |
1335 | * |
1336 | * If the count isn't unbalanced, i.e. KVM did NOT unregister its MMU | |
1337 | * notifier between a start() and end(), then there shouldn't be any | |
1338 | * in-progress invalidations. | |
52ac8b35 PB |
1339 | */ |
1340 | WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait)); | |
d497a0fa SC |
1341 | if (kvm->mn_active_invalidate_count) |
1342 | kvm->mn_active_invalidate_count = 0; | |
1343 | else | |
1344 | WARN_ON(kvm->mmu_invalidate_in_progress); | |
f00be0ca | 1345 | #else |
683412cc | 1346 | kvm_flush_shadow_all(kvm); |
5f94c174 | 1347 | #endif |
d19a9cd2 | 1348 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 1349 | kvm_destroy_devices(kvm); |
eed52e43 | 1350 | for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { |
a54d8066 MS |
1351 | kvm_free_memslots(kvm, &kvm->__memslots[i][0]); |
1352 | kvm_free_memslots(kvm, &kvm->__memslots[i][1]); | |
1353 | } | |
820b3fcd | 1354 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff | 1355 | cleanup_srcu_struct(&kvm->srcu); |
5a475554 CP |
1356 | #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES |
1357 | xa_destroy(&kvm->mem_attr_array); | |
1358 | #endif | |
d89f5eff | 1359 | kvm_arch_free_vm(kvm); |
2ecd9d29 | 1360 | preempt_notifier_dec(); |
10474ae8 | 1361 | hardware_disable_all(); |
6d4e4c4f | 1362 | mmdrop(mm); |
f17abe9a AK |
1363 | } |
1364 | ||
d39f13b0 IE |
1365 | void kvm_get_kvm(struct kvm *kvm) |
1366 | { | |
e3736c3e | 1367 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
1368 | } |
1369 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
1370 | ||
605c7130 PX |
1371 | /* |
1372 | * Make sure the vm is not during destruction, which is a safe version of | |
1373 | * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise. | |
1374 | */ | |
1375 | bool kvm_get_kvm_safe(struct kvm *kvm) | |
1376 | { | |
1377 | return refcount_inc_not_zero(&kvm->users_count); | |
1378 | } | |
1379 | EXPORT_SYMBOL_GPL(kvm_get_kvm_safe); | |
1380 | ||
d39f13b0 IE |
1381 | void kvm_put_kvm(struct kvm *kvm) |
1382 | { | |
e3736c3e | 1383 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
1384 | kvm_destroy_vm(kvm); |
1385 | } | |
1386 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
1387 | ||
149487bd SC |
1388 | /* |
1389 | * Used to put a reference that was taken on behalf of an object associated | |
1390 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
1391 | * of the new file descriptor fails and the reference cannot be transferred to | |
1392 | * its final owner. In such cases, the caller is still actively using @kvm and | |
1393 | * will fail miserably if the refcount unexpectedly hits zero. | |
1394 | */ | |
1395 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
1396 | { | |
1397 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
1398 | } | |
1399 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 1400 | |
f17abe9a AK |
1401 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
1402 | { | |
1403 | struct kvm *kvm = filp->private_data; | |
1404 | ||
721eecbf GH |
1405 | kvm_irqfd_release(kvm); |
1406 | ||
d39f13b0 | 1407 | kvm_put_kvm(kvm); |
6aa8b732 AK |
1408 | return 0; |
1409 | } | |
1410 | ||
515a0127 TY |
1411 | /* |
1412 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 1413 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 1414 | */ |
3c9bd400 | 1415 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 1416 | { |
37b2a651 | 1417 | unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 1418 | |
37b2a651 | 1419 | memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
1420 | if (!memslot->dirty_bitmap) |
1421 | return -ENOMEM; | |
1422 | ||
a36a57b1 TY |
1423 | return 0; |
1424 | } | |
1425 | ||
a54d8066 | 1426 | static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id) |
bf3e05bc | 1427 | { |
a54d8066 MS |
1428 | struct kvm_memslots *active = __kvm_memslots(kvm, as_id); |
1429 | int node_idx_inactive = active->node_idx ^ 1; | |
0e60b079 | 1430 | |
a54d8066 | 1431 | return &kvm->__memslots[as_id][node_idx_inactive]; |
0577d1ab SC |
1432 | } |
1433 | ||
1434 | /* | |
a54d8066 MS |
1435 | * Helper to get the address space ID when one of memslot pointers may be NULL. |
1436 | * This also serves as a sanity that at least one of the pointers is non-NULL, | |
1437 | * and that their address space IDs don't diverge. | |
0577d1ab | 1438 | */ |
a54d8066 MS |
1439 | static int kvm_memslots_get_as_id(struct kvm_memory_slot *a, |
1440 | struct kvm_memory_slot *b) | |
0577d1ab | 1441 | { |
a54d8066 MS |
1442 | if (WARN_ON_ONCE(!a && !b)) |
1443 | return 0; | |
0577d1ab | 1444 | |
a54d8066 MS |
1445 | if (!a) |
1446 | return b->as_id; | |
1447 | if (!b) | |
1448 | return a->as_id; | |
0577d1ab | 1449 | |
a54d8066 MS |
1450 | WARN_ON_ONCE(a->as_id != b->as_id); |
1451 | return a->as_id; | |
0577d1ab | 1452 | } |
efbeec70 | 1453 | |
a54d8066 MS |
1454 | static void kvm_insert_gfn_node(struct kvm_memslots *slots, |
1455 | struct kvm_memory_slot *slot) | |
0577d1ab | 1456 | { |
a54d8066 MS |
1457 | struct rb_root *gfn_tree = &slots->gfn_tree; |
1458 | struct rb_node **node, *parent; | |
1459 | int idx = slots->node_idx; | |
0577d1ab | 1460 | |
a54d8066 MS |
1461 | parent = NULL; |
1462 | for (node = &gfn_tree->rb_node; *node; ) { | |
1463 | struct kvm_memory_slot *tmp; | |
f85e2cb5 | 1464 | |
a54d8066 MS |
1465 | tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]); |
1466 | parent = *node; | |
1467 | if (slot->base_gfn < tmp->base_gfn) | |
1468 | node = &(*node)->rb_left; | |
1469 | else if (slot->base_gfn > tmp->base_gfn) | |
1470 | node = &(*node)->rb_right; | |
1471 | else | |
1472 | BUG(); | |
0577d1ab | 1473 | } |
a54d8066 MS |
1474 | |
1475 | rb_link_node(&slot->gfn_node[idx], parent, node); | |
1476 | rb_insert_color(&slot->gfn_node[idx], gfn_tree); | |
0577d1ab SC |
1477 | } |
1478 | ||
a54d8066 MS |
1479 | static void kvm_erase_gfn_node(struct kvm_memslots *slots, |
1480 | struct kvm_memory_slot *slot) | |
0577d1ab | 1481 | { |
a54d8066 MS |
1482 | rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree); |
1483 | } | |
0577d1ab | 1484 | |
a54d8066 MS |
1485 | static void kvm_replace_gfn_node(struct kvm_memslots *slots, |
1486 | struct kvm_memory_slot *old, | |
1487 | struct kvm_memory_slot *new) | |
1488 | { | |
1489 | int idx = slots->node_idx; | |
0577d1ab | 1490 | |
a54d8066 | 1491 | WARN_ON_ONCE(old->base_gfn != new->base_gfn); |
0577d1ab | 1492 | |
a54d8066 MS |
1493 | rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx], |
1494 | &slots->gfn_tree); | |
0577d1ab SC |
1495 | } |
1496 | ||
1497 | /* | |
a54d8066 | 1498 | * Replace @old with @new in the inactive memslots. |
0577d1ab | 1499 | * |
a54d8066 MS |
1500 | * With NULL @old this simply adds @new. |
1501 | * With NULL @new this simply removes @old. | |
0577d1ab | 1502 | * |
a54d8066 MS |
1503 | * If @new is non-NULL its hva_node[slots_idx] range has to be set |
1504 | * appropriately. | |
0577d1ab | 1505 | */ |
a54d8066 MS |
1506 | static void kvm_replace_memslot(struct kvm *kvm, |
1507 | struct kvm_memory_slot *old, | |
1508 | struct kvm_memory_slot *new) | |
0577d1ab | 1509 | { |
a54d8066 MS |
1510 | int as_id = kvm_memslots_get_as_id(old, new); |
1511 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); | |
1512 | int idx = slots->node_idx; | |
0577d1ab | 1513 | |
a54d8066 MS |
1514 | if (old) { |
1515 | hash_del(&old->id_node[idx]); | |
1516 | interval_tree_remove(&old->hva_node[idx], &slots->hva_tree); | |
0577d1ab | 1517 | |
a54d8066 MS |
1518 | if ((long)old == atomic_long_read(&slots->last_used_slot)) |
1519 | atomic_long_set(&slots->last_used_slot, (long)new); | |
0577d1ab | 1520 | |
a54d8066 MS |
1521 | if (!new) { |
1522 | kvm_erase_gfn_node(slots, old); | |
1e8617d3 | 1523 | return; |
a54d8066 MS |
1524 | } |
1525 | } | |
1e8617d3 | 1526 | |
a54d8066 MS |
1527 | /* |
1528 | * Initialize @new's hva range. Do this even when replacing an @old | |
1529 | * slot, kvm_copy_memslot() deliberately does not touch node data. | |
1530 | */ | |
1531 | new->hva_node[idx].start = new->userspace_addr; | |
1532 | new->hva_node[idx].last = new->userspace_addr + | |
1533 | (new->npages << PAGE_SHIFT) - 1; | |
1534 | ||
1535 | /* | |
1536 | * (Re)Add the new memslot. There is no O(1) interval_tree_replace(), | |
1537 | * hva_node needs to be swapped with remove+insert even though hva can't | |
1538 | * change when replacing an existing slot. | |
1539 | */ | |
1540 | hash_add(slots->id_hash, &new->id_node[idx], new->id); | |
1541 | interval_tree_insert(&new->hva_node[idx], &slots->hva_tree); | |
1542 | ||
1543 | /* | |
1544 | * If the memslot gfn is unchanged, rb_replace_node() can be used to | |
1545 | * switch the node in the gfn tree instead of removing the old and | |
1546 | * inserting the new as two separate operations. Replacement is a | |
1547 | * single O(1) operation versus two O(log(n)) operations for | |
1548 | * remove+insert. | |
1549 | */ | |
1550 | if (old && old->base_gfn == new->base_gfn) { | |
1551 | kvm_replace_gfn_node(slots, old, new); | |
1552 | } else { | |
1553 | if (old) | |
1554 | kvm_erase_gfn_node(slots, old); | |
1555 | kvm_insert_gfn_node(slots, new); | |
0577d1ab | 1556 | } |
bf3e05bc XG |
1557 | } |
1558 | ||
bb58b90b SC |
1559 | /* |
1560 | * Flags that do not access any of the extra space of struct | |
1561 | * kvm_userspace_memory_region2. KVM_SET_USER_MEMORY_REGION_V1_FLAGS | |
1562 | * only allows these. | |
1563 | */ | |
1564 | #define KVM_SET_USER_MEMORY_REGION_V1_FLAGS \ | |
1565 | (KVM_MEM_LOG_DIRTY_PAGES | KVM_MEM_READONLY) | |
1566 | ||
a7800aa8 SC |
1567 | static int check_memory_region_flags(struct kvm *kvm, |
1568 | const struct kvm_userspace_memory_region2 *mem) | |
a50d64d6 | 1569 | { |
4d8b81ab XG |
1570 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1571 | ||
a7800aa8 SC |
1572 | if (kvm_arch_has_private_mem(kvm)) |
1573 | valid_flags |= KVM_MEM_GUEST_MEMFD; | |
1574 | ||
1575 | /* Dirty logging private memory is not currently supported. */ | |
1576 | if (mem->flags & KVM_MEM_GUEST_MEMFD) | |
1577 | valid_flags &= ~KVM_MEM_LOG_DIRTY_PAGES; | |
1578 | ||
8886640d | 1579 | #ifdef CONFIG_HAVE_KVM_READONLY_MEM |
e5635922 SC |
1580 | /* |
1581 | * GUEST_MEMFD is incompatible with read-only memslots, as writes to | |
1582 | * read-only memslots have emulated MMIO, not page fault, semantics, | |
1583 | * and KVM doesn't allow emulated MMIO for private memory. | |
1584 | */ | |
1585 | if (!(mem->flags & KVM_MEM_GUEST_MEMFD)) | |
1586 | valid_flags |= KVM_MEM_READONLY; | |
4d8b81ab XG |
1587 | #endif |
1588 | ||
1589 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1590 | return -EINVAL; |
1591 | ||
1592 | return 0; | |
1593 | } | |
1594 | ||
a54d8066 | 1595 | static void kvm_swap_active_memslots(struct kvm *kvm, int as_id) |
7ec4fb44 | 1596 | { |
a54d8066 MS |
1597 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); |
1598 | ||
1599 | /* Grab the generation from the activate memslots. */ | |
1600 | u64 gen = __kvm_memslots(kvm, as_id)->generation; | |
7ec4fb44 | 1601 | |
361209e0 SC |
1602 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1603 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1604 | |
52ac8b35 PB |
1605 | /* |
1606 | * Do not store the new memslots while there are invalidations in | |
071064f1 PB |
1607 | * progress, otherwise the locking in invalidate_range_start and |
1608 | * invalidate_range_end will be unbalanced. | |
52ac8b35 PB |
1609 | */ |
1610 | spin_lock(&kvm->mn_invalidate_lock); | |
1611 | prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
1612 | while (kvm->mn_active_invalidate_count) { | |
1613 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1614 | spin_unlock(&kvm->mn_invalidate_lock); | |
1615 | schedule(); | |
1616 | spin_lock(&kvm->mn_invalidate_lock); | |
1617 | } | |
1618 | finish_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
f481b069 | 1619 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
52ac8b35 | 1620 | spin_unlock(&kvm->mn_invalidate_lock); |
b10a038e BG |
1621 | |
1622 | /* | |
1623 | * Acquired in kvm_set_memslot. Must be released before synchronize | |
1624 | * SRCU below in order to avoid deadlock with another thread | |
1625 | * acquiring the slots_arch_lock in an srcu critical section. | |
1626 | */ | |
1627 | mutex_unlock(&kvm->slots_arch_lock); | |
1628 | ||
7ec4fb44 | 1629 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1630 | |
ee3d1570 | 1631 | /* |
361209e0 | 1632 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1633 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1634 | * the number of address spaces. This provides a unique and easily |
1635 | * identifiable generation number while the memslots are in flux. | |
1636 | */ | |
1637 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1638 | ||
1639 | /* | |
4bd518f1 PB |
1640 | * Generations must be unique even across address spaces. We do not need |
1641 | * a global counter for that, instead the generation space is evenly split | |
1642 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1643 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1644 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1645 | */ |
eed52e43 | 1646 | gen += kvm_arch_nr_memslot_as_ids(kvm); |
ee3d1570 | 1647 | |
15248258 | 1648 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1649 | |
15248258 | 1650 | slots->generation = gen; |
7ec4fb44 GN |
1651 | } |
1652 | ||
07921665 SC |
1653 | static int kvm_prepare_memory_region(struct kvm *kvm, |
1654 | const struct kvm_memory_slot *old, | |
1655 | struct kvm_memory_slot *new, | |
1656 | enum kvm_mr_change change) | |
ddc12f2a | 1657 | { |
07921665 SC |
1658 | int r; |
1659 | ||
1660 | /* | |
1661 | * If dirty logging is disabled, nullify the bitmap; the old bitmap | |
1662 | * will be freed on "commit". If logging is enabled in both old and | |
1663 | * new, reuse the existing bitmap. If logging is enabled only in the | |
1664 | * new and KVM isn't using a ring buffer, allocate and initialize a | |
1665 | * new bitmap. | |
1666 | */ | |
244893fa SC |
1667 | if (change != KVM_MR_DELETE) { |
1668 | if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1669 | new->dirty_bitmap = NULL; | |
1670 | else if (old && old->dirty_bitmap) | |
1671 | new->dirty_bitmap = old->dirty_bitmap; | |
86bdf3eb | 1672 | else if (kvm_use_dirty_bitmap(kvm)) { |
244893fa SC |
1673 | r = kvm_alloc_dirty_bitmap(new); |
1674 | if (r) | |
1675 | return r; | |
1676 | ||
1677 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1678 | bitmap_set(new->dirty_bitmap, 0, new->npages); | |
1679 | } | |
07921665 SC |
1680 | } |
1681 | ||
1682 | r = kvm_arch_prepare_memory_region(kvm, old, new, change); | |
1683 | ||
1684 | /* Free the bitmap on failure if it was allocated above. */ | |
c87661f8 | 1685 | if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap)) |
07921665 SC |
1686 | kvm_destroy_dirty_bitmap(new); |
1687 | ||
1688 | return r; | |
ddc12f2a BG |
1689 | } |
1690 | ||
07921665 SC |
1691 | static void kvm_commit_memory_region(struct kvm *kvm, |
1692 | struct kvm_memory_slot *old, | |
1693 | const struct kvm_memory_slot *new, | |
1694 | enum kvm_mr_change change) | |
ddc12f2a | 1695 | { |
6c7b2202 PB |
1696 | int old_flags = old ? old->flags : 0; |
1697 | int new_flags = new ? new->flags : 0; | |
07921665 SC |
1698 | /* |
1699 | * Update the total number of memslot pages before calling the arch | |
1700 | * hook so that architectures can consume the result directly. | |
1701 | */ | |
1702 | if (change == KVM_MR_DELETE) | |
1703 | kvm->nr_memslot_pages -= old->npages; | |
1704 | else if (change == KVM_MR_CREATE) | |
1705 | kvm->nr_memslot_pages += new->npages; | |
1706 | ||
6c7b2202 PB |
1707 | if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES) { |
1708 | int change = (new_flags & KVM_MEM_LOG_DIRTY_PAGES) ? 1 : -1; | |
1709 | atomic_set(&kvm->nr_memslots_dirty_logging, | |
1710 | atomic_read(&kvm->nr_memslots_dirty_logging) + change); | |
1711 | } | |
1712 | ||
07921665 SC |
1713 | kvm_arch_commit_memory_region(kvm, old, new, change); |
1714 | ||
a54d8066 MS |
1715 | switch (change) { |
1716 | case KVM_MR_CREATE: | |
1717 | /* Nothing more to do. */ | |
1718 | break; | |
1719 | case KVM_MR_DELETE: | |
1720 | /* Free the old memslot and all its metadata. */ | |
1721 | kvm_free_memslot(kvm, old); | |
1722 | break; | |
1723 | case KVM_MR_MOVE: | |
1724 | case KVM_MR_FLAGS_ONLY: | |
1725 | /* | |
1726 | * Free the dirty bitmap as needed; the below check encompasses | |
1727 | * both the flags and whether a ring buffer is being used) | |
1728 | */ | |
1729 | if (old->dirty_bitmap && !new->dirty_bitmap) | |
1730 | kvm_destroy_dirty_bitmap(old); | |
1731 | ||
1732 | /* | |
1733 | * The final quirk. Free the detached, old slot, but only its | |
1734 | * memory, not any metadata. Metadata, including arch specific | |
1735 | * data, may be reused by @new. | |
1736 | */ | |
1737 | kfree(old); | |
1738 | break; | |
1739 | default: | |
1740 | BUG(); | |
1741 | } | |
ddc12f2a BG |
1742 | } |
1743 | ||
36947254 | 1744 | /* |
a54d8066 MS |
1745 | * Activate @new, which must be installed in the inactive slots by the caller, |
1746 | * by swapping the active slots and then propagating @new to @old once @old is | |
1747 | * unreachable and can be safely modified. | |
1748 | * | |
1749 | * With NULL @old this simply adds @new to @active (while swapping the sets). | |
1750 | * With NULL @new this simply removes @old from @active and frees it | |
1751 | * (while also swapping the sets). | |
36947254 | 1752 | */ |
a54d8066 MS |
1753 | static void kvm_activate_memslot(struct kvm *kvm, |
1754 | struct kvm_memory_slot *old, | |
1755 | struct kvm_memory_slot *new) | |
36947254 | 1756 | { |
a54d8066 | 1757 | int as_id = kvm_memslots_get_as_id(old, new); |
36947254 | 1758 | |
a54d8066 MS |
1759 | kvm_swap_active_memslots(kvm, as_id); |
1760 | ||
1761 | /* Propagate the new memslot to the now inactive memslots. */ | |
1762 | kvm_replace_memslot(kvm, old, new); | |
1763 | } | |
1764 | ||
1765 | static void kvm_copy_memslot(struct kvm_memory_slot *dest, | |
1766 | const struct kvm_memory_slot *src) | |
1767 | { | |
1768 | dest->base_gfn = src->base_gfn; | |
1769 | dest->npages = src->npages; | |
1770 | dest->dirty_bitmap = src->dirty_bitmap; | |
1771 | dest->arch = src->arch; | |
1772 | dest->userspace_addr = src->userspace_addr; | |
1773 | dest->flags = src->flags; | |
1774 | dest->id = src->id; | |
1775 | dest->as_id = src->as_id; | |
1776 | } | |
1777 | ||
1778 | static void kvm_invalidate_memslot(struct kvm *kvm, | |
1779 | struct kvm_memory_slot *old, | |
244893fa | 1780 | struct kvm_memory_slot *invalid_slot) |
a54d8066 | 1781 | { |
07921665 | 1782 | /* |
a54d8066 MS |
1783 | * Mark the current slot INVALID. As with all memslot modifications, |
1784 | * this must be done on an unreachable slot to avoid modifying the | |
1785 | * current slot in the active tree. | |
07921665 | 1786 | */ |
244893fa SC |
1787 | kvm_copy_memslot(invalid_slot, old); |
1788 | invalid_slot->flags |= KVM_MEMSLOT_INVALID; | |
1789 | kvm_replace_memslot(kvm, old, invalid_slot); | |
a54d8066 MS |
1790 | |
1791 | /* | |
1792 | * Activate the slot that is now marked INVALID, but don't propagate | |
1793 | * the slot to the now inactive slots. The slot is either going to be | |
1794 | * deleted or recreated as a new slot. | |
1795 | */ | |
1796 | kvm_swap_active_memslots(kvm, old->as_id); | |
1797 | ||
1798 | /* | |
1799 | * From this point no new shadow pages pointing to a deleted, or moved, | |
1800 | * memslot will be created. Validation of sp->gfn happens in: | |
1801 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1802 | * - kvm_is_visible_gfn (mmu_check_root) | |
1803 | */ | |
bcb63dcd | 1804 | kvm_arch_flush_shadow_memslot(kvm, old); |
683412cc | 1805 | kvm_arch_guest_memory_reclaimed(kvm); |
a54d8066 | 1806 | |
b0d23708 | 1807 | /* Was released by kvm_swap_active_memslots(), reacquire. */ |
a54d8066 MS |
1808 | mutex_lock(&kvm->slots_arch_lock); |
1809 | ||
1810 | /* | |
1811 | * Copy the arch-specific field of the newly-installed slot back to the | |
1812 | * old slot as the arch data could have changed between releasing | |
b0d23708 | 1813 | * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock |
a54d8066 MS |
1814 | * above. Writers are required to retrieve memslots *after* acquiring |
1815 | * slots_arch_lock, thus the active slot's data is guaranteed to be fresh. | |
1816 | */ | |
244893fa | 1817 | old->arch = invalid_slot->arch; |
a54d8066 MS |
1818 | } |
1819 | ||
1820 | static void kvm_create_memslot(struct kvm *kvm, | |
244893fa | 1821 | struct kvm_memory_slot *new) |
a54d8066 | 1822 | { |
244893fa SC |
1823 | /* Add the new memslot to the inactive set and activate. */ |
1824 | kvm_replace_memslot(kvm, NULL, new); | |
1825 | kvm_activate_memslot(kvm, NULL, new); | |
a54d8066 MS |
1826 | } |
1827 | ||
1828 | static void kvm_delete_memslot(struct kvm *kvm, | |
1829 | struct kvm_memory_slot *old, | |
1830 | struct kvm_memory_slot *invalid_slot) | |
1831 | { | |
1832 | /* | |
1833 | * Remove the old memslot (in the inactive memslots) by passing NULL as | |
244893fa | 1834 | * the "new" slot, and for the invalid version in the active slots. |
a54d8066 MS |
1835 | */ |
1836 | kvm_replace_memslot(kvm, old, NULL); | |
a54d8066 | 1837 | kvm_activate_memslot(kvm, invalid_slot, NULL); |
a54d8066 | 1838 | } |
36947254 | 1839 | |
244893fa SC |
1840 | static void kvm_move_memslot(struct kvm *kvm, |
1841 | struct kvm_memory_slot *old, | |
1842 | struct kvm_memory_slot *new, | |
1843 | struct kvm_memory_slot *invalid_slot) | |
a54d8066 | 1844 | { |
a54d8066 | 1845 | /* |
244893fa SC |
1846 | * Replace the old memslot in the inactive slots, and then swap slots |
1847 | * and replace the current INVALID with the new as well. | |
a54d8066 | 1848 | */ |
244893fa SC |
1849 | kvm_replace_memslot(kvm, old, new); |
1850 | kvm_activate_memslot(kvm, invalid_slot, new); | |
a54d8066 | 1851 | } |
36947254 | 1852 | |
a54d8066 MS |
1853 | static void kvm_update_flags_memslot(struct kvm *kvm, |
1854 | struct kvm_memory_slot *old, | |
244893fa | 1855 | struct kvm_memory_slot *new) |
a54d8066 MS |
1856 | { |
1857 | /* | |
1858 | * Similar to the MOVE case, but the slot doesn't need to be zapped as | |
1859 | * an intermediate step. Instead, the old memslot is simply replaced | |
1860 | * with a new, updated copy in both memslot sets. | |
1861 | */ | |
244893fa SC |
1862 | kvm_replace_memslot(kvm, old, new); |
1863 | kvm_activate_memslot(kvm, old, new); | |
36947254 SC |
1864 | } |
1865 | ||
cf47f50b | 1866 | static int kvm_set_memslot(struct kvm *kvm, |
a54d8066 | 1867 | struct kvm_memory_slot *old, |
ce5f0215 | 1868 | struct kvm_memory_slot *new, |
cf47f50b SC |
1869 | enum kvm_mr_change change) |
1870 | { | |
244893fa | 1871 | struct kvm_memory_slot *invalid_slot; |
cf47f50b SC |
1872 | int r; |
1873 | ||
b10a038e | 1874 | /* |
b0d23708 | 1875 | * Released in kvm_swap_active_memslots(). |
b10a038e | 1876 | * |
b0d23708 JM |
1877 | * Must be held from before the current memslots are copied until after |
1878 | * the new memslots are installed with rcu_assign_pointer, then | |
1879 | * released before the synchronize srcu in kvm_swap_active_memslots(). | |
b10a038e BG |
1880 | * |
1881 | * When modifying memslots outside of the slots_lock, must be held | |
1882 | * before reading the pointer to the current memslots until after all | |
1883 | * changes to those memslots are complete. | |
1884 | * | |
1885 | * These rules ensure that installing new memslots does not lose | |
1886 | * changes made to the previous memslots. | |
1887 | */ | |
1888 | mutex_lock(&kvm->slots_arch_lock); | |
1889 | ||
a54d8066 MS |
1890 | /* |
1891 | * Invalidate the old slot if it's being deleted or moved. This is | |
1892 | * done prior to actually deleting/moving the memslot to allow vCPUs to | |
1893 | * continue running by ensuring there are no mappings or shadow pages | |
1894 | * for the memslot when it is deleted/moved. Without pre-invalidation | |
1895 | * (and without a lock), a window would exist between effecting the | |
1896 | * delete/move and committing the changes in arch code where KVM or a | |
1897 | * guest could access a non-existent memslot. | |
244893fa SC |
1898 | * |
1899 | * Modifications are done on a temporary, unreachable slot. The old | |
1900 | * slot needs to be preserved in case a later step fails and the | |
1901 | * invalidation needs to be reverted. | |
a54d8066 | 1902 | */ |
cf47f50b | 1903 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
244893fa SC |
1904 | invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT); |
1905 | if (!invalid_slot) { | |
1906 | mutex_unlock(&kvm->slots_arch_lock); | |
1907 | return -ENOMEM; | |
1908 | } | |
1909 | kvm_invalidate_memslot(kvm, old, invalid_slot); | |
1910 | } | |
b10a038e | 1911 | |
a54d8066 MS |
1912 | r = kvm_prepare_memory_region(kvm, old, new, change); |
1913 | if (r) { | |
b10a038e | 1914 | /* |
a54d8066 MS |
1915 | * For DELETE/MOVE, revert the above INVALID change. No |
1916 | * modifications required since the original slot was preserved | |
1917 | * in the inactive slots. Changing the active memslots also | |
1918 | * release slots_arch_lock. | |
b10a038e | 1919 | */ |
244893fa SC |
1920 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
1921 | kvm_activate_memslot(kvm, invalid_slot, old); | |
1922 | kfree(invalid_slot); | |
1923 | } else { | |
a54d8066 | 1924 | mutex_unlock(&kvm->slots_arch_lock); |
244893fa | 1925 | } |
a54d8066 | 1926 | return r; |
cf47f50b SC |
1927 | } |
1928 | ||
bda44d84 | 1929 | /* |
a54d8066 MS |
1930 | * For DELETE and MOVE, the working slot is now active as the INVALID |
1931 | * version of the old slot. MOVE is particularly special as it reuses | |
1932 | * the old slot and returns a copy of the old slot (in working_slot). | |
1933 | * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the | |
1934 | * old slot is detached but otherwise preserved. | |
bda44d84 | 1935 | */ |
a54d8066 | 1936 | if (change == KVM_MR_CREATE) |
244893fa | 1937 | kvm_create_memslot(kvm, new); |
a54d8066 | 1938 | else if (change == KVM_MR_DELETE) |
244893fa | 1939 | kvm_delete_memslot(kvm, old, invalid_slot); |
a54d8066 | 1940 | else if (change == KVM_MR_MOVE) |
244893fa | 1941 | kvm_move_memslot(kvm, old, new, invalid_slot); |
a54d8066 | 1942 | else if (change == KVM_MR_FLAGS_ONLY) |
244893fa | 1943 | kvm_update_flags_memslot(kvm, old, new); |
a54d8066 MS |
1944 | else |
1945 | BUG(); | |
cf47f50b | 1946 | |
244893fa SC |
1947 | /* Free the temporary INVALID slot used for DELETE and MOVE. */ |
1948 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) | |
1949 | kfree(invalid_slot); | |
bda44d84 | 1950 | |
a54d8066 MS |
1951 | /* |
1952 | * No need to refresh new->arch, changes after dropping slots_arch_lock | |
a413a625 | 1953 | * will directly hit the final, active memslot. Architectures are |
a54d8066 MS |
1954 | * responsible for knowing that new->arch may be stale. |
1955 | */ | |
1956 | kvm_commit_memory_region(kvm, old, new, change); | |
cf47f50b | 1957 | |
cf47f50b | 1958 | return 0; |
cf47f50b SC |
1959 | } |
1960 | ||
44401a20 MS |
1961 | static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id, |
1962 | gfn_t start, gfn_t end) | |
5c0b4f3d | 1963 | { |
44401a20 | 1964 | struct kvm_memslot_iter iter; |
5c0b4f3d | 1965 | |
44401a20 MS |
1966 | kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) { |
1967 | if (iter.slot->id != id) | |
1968 | return true; | |
1969 | } | |
5c0b4f3d | 1970 | |
44401a20 | 1971 | return false; |
5c0b4f3d SC |
1972 | } |
1973 | ||
6aa8b732 AK |
1974 | /* |
1975 | * Allocate some memory and give it an address in the guest physical address | |
1976 | * space. | |
1977 | * | |
1978 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1979 | * |
02d5d55b | 1980 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1981 | */ |
f78e0e2e | 1982 | int __kvm_set_memory_region(struct kvm *kvm, |
bb58b90b | 1983 | const struct kvm_userspace_memory_region2 *mem) |
6aa8b732 | 1984 | { |
244893fa | 1985 | struct kvm_memory_slot *old, *new; |
44401a20 | 1986 | struct kvm_memslots *slots; |
f64c0398 | 1987 | enum kvm_mr_change change; |
0f9bdef3 SC |
1988 | unsigned long npages; |
1989 | gfn_t base_gfn; | |
163da372 SC |
1990 | int as_id, id; |
1991 | int r; | |
6aa8b732 | 1992 | |
a7800aa8 | 1993 | r = check_memory_region_flags(kvm, mem); |
a50d64d6 | 1994 | if (r) |
71a4c30b | 1995 | return r; |
a50d64d6 | 1996 | |
f481b069 PB |
1997 | as_id = mem->slot >> 16; |
1998 | id = (u16)mem->slot; | |
1999 | ||
6aa8b732 | 2000 | /* General sanity checks */ |
6b285a55 SC |
2001 | if ((mem->memory_size & (PAGE_SIZE - 1)) || |
2002 | (mem->memory_size != (unsigned long)mem->memory_size)) | |
71a4c30b | 2003 | return -EINVAL; |
6aa8b732 | 2004 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 2005 | return -EINVAL; |
fa3d315a | 2006 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 2007 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 2008 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 2009 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 2010 | mem->memory_size)) |
71a4c30b | 2011 | return -EINVAL; |
a7800aa8 SC |
2012 | if (mem->flags & KVM_MEM_GUEST_MEMFD && |
2013 | (mem->guest_memfd_offset & (PAGE_SIZE - 1) || | |
2014 | mem->guest_memfd_offset + mem->memory_size < mem->guest_memfd_offset)) | |
2015 | return -EINVAL; | |
eed52e43 | 2016 | if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 2017 | return -EINVAL; |
6aa8b732 | 2018 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 2019 | return -EINVAL; |
0f9bdef3 SC |
2020 | if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES) |
2021 | return -EINVAL; | |
6aa8b732 | 2022 | |
44401a20 | 2023 | slots = __kvm_memslots(kvm, as_id); |
6aa8b732 | 2024 | |
5c0b4f3d | 2025 | /* |
7cd08553 SC |
2026 | * Note, the old memslot (and the pointer itself!) may be invalidated |
2027 | * and/or destroyed by kvm_set_memslot(). | |
5c0b4f3d | 2028 | */ |
44401a20 | 2029 | old = id_to_memslot(slots, id); |
163da372 | 2030 | |
47ea7d90 | 2031 | if (!mem->memory_size) { |
7cd08553 | 2032 | if (!old || !old->npages) |
47ea7d90 | 2033 | return -EINVAL; |
5c0b4f3d | 2034 | |
7cd08553 | 2035 | if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages)) |
47ea7d90 | 2036 | return -EIO; |
6aa8b732 | 2037 | |
244893fa | 2038 | return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE); |
47ea7d90 | 2039 | } |
5c0b4f3d | 2040 | |
0f9bdef3 SC |
2041 | base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT); |
2042 | npages = (mem->memory_size >> PAGE_SHIFT); | |
163da372 | 2043 | |
7cd08553 | 2044 | if (!old || !old->npages) { |
5c0b4f3d | 2045 | change = KVM_MR_CREATE; |
afa319a5 SC |
2046 | |
2047 | /* | |
2048 | * To simplify KVM internals, the total number of pages across | |
2049 | * all memslots must fit in an unsigned long. | |
2050 | */ | |
0f9bdef3 | 2051 | if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages) |
afa319a5 | 2052 | return -EINVAL; |
5c0b4f3d | 2053 | } else { /* Modify an existing slot. */ |
a7800aa8 SC |
2054 | /* Private memslots are immutable, they can only be deleted. */ |
2055 | if (mem->flags & KVM_MEM_GUEST_MEMFD) | |
2056 | return -EINVAL; | |
0f9bdef3 SC |
2057 | if ((mem->userspace_addr != old->userspace_addr) || |
2058 | (npages != old->npages) || | |
2059 | ((mem->flags ^ old->flags) & KVM_MEM_READONLY)) | |
71a4c30b | 2060 | return -EINVAL; |
09170a49 | 2061 | |
0f9bdef3 | 2062 | if (base_gfn != old->base_gfn) |
5c0b4f3d | 2063 | change = KVM_MR_MOVE; |
0f9bdef3 | 2064 | else if (mem->flags != old->flags) |
5c0b4f3d SC |
2065 | change = KVM_MR_FLAGS_ONLY; |
2066 | else /* Nothing to change. */ | |
2067 | return 0; | |
09170a49 | 2068 | } |
6aa8b732 | 2069 | |
44401a20 | 2070 | if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) && |
0f9bdef3 | 2071 | kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages)) |
44401a20 | 2072 | return -EEXIST; |
6aa8b732 | 2073 | |
244893fa SC |
2074 | /* Allocate a slot that will persist in the memslot. */ |
2075 | new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT); | |
2076 | if (!new) | |
2077 | return -ENOMEM; | |
3c9bd400 | 2078 | |
244893fa SC |
2079 | new->as_id = as_id; |
2080 | new->id = id; | |
2081 | new->base_gfn = base_gfn; | |
2082 | new->npages = npages; | |
2083 | new->flags = mem->flags; | |
2084 | new->userspace_addr = mem->userspace_addr; | |
a7800aa8 SC |
2085 | if (mem->flags & KVM_MEM_GUEST_MEMFD) { |
2086 | r = kvm_gmem_bind(kvm, new, mem->guest_memfd, mem->guest_memfd_offset); | |
2087 | if (r) | |
2088 | goto out; | |
2089 | } | |
6aa8b732 | 2090 | |
244893fa | 2091 | r = kvm_set_memslot(kvm, old, new, change); |
cf47f50b | 2092 | if (r) |
a7800aa8 SC |
2093 | goto out_unbind; |
2094 | ||
2095 | return 0; | |
2096 | ||
2097 | out_unbind: | |
2098 | if (mem->flags & KVM_MEM_GUEST_MEMFD) | |
2099 | kvm_gmem_unbind(new); | |
2100 | out: | |
2101 | kfree(new); | |
6aa8b732 | 2102 | return r; |
210c7c4d | 2103 | } |
f78e0e2e SY |
2104 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
2105 | ||
2106 | int kvm_set_memory_region(struct kvm *kvm, | |
bb58b90b | 2107 | const struct kvm_userspace_memory_region2 *mem) |
f78e0e2e SY |
2108 | { |
2109 | int r; | |
2110 | ||
79fac95e | 2111 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 2112 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 2113 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
2114 | return r; |
2115 | } | |
210c7c4d IE |
2116 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
2117 | ||
7940876e | 2118 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
bb58b90b | 2119 | struct kvm_userspace_memory_region2 *mem) |
210c7c4d | 2120 | { |
f481b069 | 2121 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 2122 | return -EINVAL; |
09170a49 | 2123 | |
47ae31e2 | 2124 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
2125 | } |
2126 | ||
0dff0846 | 2127 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
2128 | /** |
2129 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
2130 | * @kvm: pointer to kvm instance | |
2131 | * @log: slot id and address to which we copy the log | |
2132 | * @is_dirty: set to '1' if any dirty pages were found | |
2133 | * @memslot: set to the associated memslot, always valid on success | |
2134 | */ | |
2135 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
2136 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 2137 | { |
9f6b8029 | 2138 | struct kvm_memslots *slots; |
843574a3 | 2139 | int i, as_id, id; |
87bf6e7d | 2140 | unsigned long n; |
6aa8b732 AK |
2141 | unsigned long any = 0; |
2142 | ||
86bdf3eb GS |
2143 | /* Dirty ring tracking may be exclusive to dirty log tracking */ |
2144 | if (!kvm_use_dirty_bitmap(kvm)) | |
b2cc64c4 PX |
2145 | return -ENXIO; |
2146 | ||
2a49f61d SC |
2147 | *memslot = NULL; |
2148 | *is_dirty = 0; | |
2149 | ||
f481b069 PB |
2150 | as_id = log->slot >> 16; |
2151 | id = (u16)log->slot; | |
eed52e43 | 2152 | if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS) |
843574a3 | 2153 | return -EINVAL; |
6aa8b732 | 2154 | |
f481b069 | 2155 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 2156 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 2157 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 2158 | return -ENOENT; |
6aa8b732 | 2159 | |
2a49f61d SC |
2160 | kvm_arch_sync_dirty_log(kvm, *memslot); |
2161 | ||
2162 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 2163 | |
cd1a4a98 | 2164 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 2165 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 2166 | |
2a49f61d | 2167 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 2168 | return -EFAULT; |
6aa8b732 | 2169 | |
5bb064dc ZX |
2170 | if (any) |
2171 | *is_dirty = 1; | |
843574a3 | 2172 | return 0; |
6aa8b732 | 2173 | } |
2ba9f0d8 | 2174 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 2175 | |
0dff0846 | 2176 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 2177 | /** |
b8b00220 | 2178 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 2179 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
2180 | * @kvm: pointer to kvm instance |
2181 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
2182 | * |
2183 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
2184 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
2185 | * following order: | |
2186 | * | |
2187 | * 1. Take a snapshot of the bit and clear it if needed. | |
2188 | * 2. Write protect the corresponding page. | |
2189 | * 3. Copy the snapshot to the userspace. | |
2190 | * 4. Upon return caller flushes TLB's if needed. | |
2191 | * | |
2192 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
2193 | * entry. This is not a problem because the page is reported dirty using | |
2194 | * the snapshot taken before and step 4 ensures that writes done after | |
2195 | * exiting to userspace will be logged for the next call. | |
2196 | * | |
2197 | */ | |
0dff0846 | 2198 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 2199 | { |
9f6b8029 | 2200 | struct kvm_memslots *slots; |
ba0513b5 | 2201 | struct kvm_memory_slot *memslot; |
58d6db34 | 2202 | int i, as_id, id; |
ba0513b5 MS |
2203 | unsigned long n; |
2204 | unsigned long *dirty_bitmap; | |
2205 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2206 | bool flush; |
ba0513b5 | 2207 | |
86bdf3eb GS |
2208 | /* Dirty ring tracking may be exclusive to dirty log tracking */ |
2209 | if (!kvm_use_dirty_bitmap(kvm)) | |
b2cc64c4 PX |
2210 | return -ENXIO; |
2211 | ||
f481b069 PB |
2212 | as_id = log->slot >> 16; |
2213 | id = (u16)log->slot; | |
eed52e43 | 2214 | if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS) |
58d6db34 | 2215 | return -EINVAL; |
ba0513b5 | 2216 | |
f481b069 PB |
2217 | slots = __kvm_memslots(kvm, as_id); |
2218 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2219 | if (!memslot || !memslot->dirty_bitmap) |
2220 | return -ENOENT; | |
ba0513b5 MS |
2221 | |
2222 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 2223 | |
0dff0846 SC |
2224 | kvm_arch_sync_dirty_log(kvm, memslot); |
2225 | ||
ba0513b5 | 2226 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 2227 | flush = false; |
2a31b9db PB |
2228 | if (kvm->manual_dirty_log_protect) { |
2229 | /* | |
2230 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
2231 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
2232 | * is some code duplication between this function and | |
2233 | * kvm_get_dirty_log, but hopefully all architecture | |
2234 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
2235 | * can be eliminated. | |
2236 | */ | |
2237 | dirty_bitmap_buffer = dirty_bitmap; | |
2238 | } else { | |
2239 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
2240 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 2241 | |
531810ca | 2242 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
2243 | for (i = 0; i < n / sizeof(long); i++) { |
2244 | unsigned long mask; | |
2245 | gfn_t offset; | |
ba0513b5 | 2246 | |
2a31b9db PB |
2247 | if (!dirty_bitmap[i]) |
2248 | continue; | |
2249 | ||
0dff0846 | 2250 | flush = true; |
2a31b9db PB |
2251 | mask = xchg(&dirty_bitmap[i], 0); |
2252 | dirty_bitmap_buffer[i] = mask; | |
2253 | ||
a67794ca LT |
2254 | offset = i * BITS_PER_LONG; |
2255 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
2256 | offset, mask); | |
2a31b9db | 2257 | } |
531810ca | 2258 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
2259 | } |
2260 | ||
0dff0846 | 2261 | if (flush) |
619b5072 | 2262 | kvm_flush_remote_tlbs_memslot(kvm, memslot); |
0dff0846 | 2263 | |
2a31b9db PB |
2264 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
2265 | return -EFAULT; | |
2266 | return 0; | |
2267 | } | |
0dff0846 SC |
2268 | |
2269 | ||
2270 | /** | |
2271 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
2272 | * @kvm: kvm instance | |
2273 | * @log: slot id and address to which we copy the log | |
2274 | * | |
2275 | * Steps 1-4 below provide general overview of dirty page logging. See | |
2276 | * kvm_get_dirty_log_protect() function description for additional details. | |
2277 | * | |
2278 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
2279 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
2280 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
2281 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
2282 | * writes will be marked dirty for next log read. | |
2283 | * | |
2284 | * 1. Take a snapshot of the bit and clear it if needed. | |
2285 | * 2. Write protect the corresponding page. | |
2286 | * 3. Copy the snapshot to the userspace. | |
2287 | * 4. Flush TLB's if needed. | |
2288 | */ | |
2289 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
2290 | struct kvm_dirty_log *log) | |
2291 | { | |
2292 | int r; | |
2293 | ||
2294 | mutex_lock(&kvm->slots_lock); | |
2295 | ||
2296 | r = kvm_get_dirty_log_protect(kvm, log); | |
2297 | ||
2298 | mutex_unlock(&kvm->slots_lock); | |
2299 | return r; | |
2300 | } | |
2a31b9db PB |
2301 | |
2302 | /** | |
2303 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
2304 | * and reenable dirty page tracking for the corresponding pages. | |
2305 | * @kvm: pointer to kvm instance | |
2306 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
2307 | */ | |
0dff0846 SC |
2308 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
2309 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
2310 | { |
2311 | struct kvm_memslots *slots; | |
2312 | struct kvm_memory_slot *memslot; | |
98938aa8 | 2313 | int as_id, id; |
2a31b9db | 2314 | gfn_t offset; |
98938aa8 | 2315 | unsigned long i, n; |
2a31b9db PB |
2316 | unsigned long *dirty_bitmap; |
2317 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2318 | bool flush; |
2a31b9db | 2319 | |
86bdf3eb GS |
2320 | /* Dirty ring tracking may be exclusive to dirty log tracking */ |
2321 | if (!kvm_use_dirty_bitmap(kvm)) | |
b2cc64c4 PX |
2322 | return -ENXIO; |
2323 | ||
2a31b9db PB |
2324 | as_id = log->slot >> 16; |
2325 | id = (u16)log->slot; | |
eed52e43 | 2326 | if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS) |
2a31b9db PB |
2327 | return -EINVAL; |
2328 | ||
76d58e0f | 2329 | if (log->first_page & 63) |
2a31b9db PB |
2330 | return -EINVAL; |
2331 | ||
2332 | slots = __kvm_memslots(kvm, as_id); | |
2333 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2334 | if (!memslot || !memslot->dirty_bitmap) |
2335 | return -ENOENT; | |
2a31b9db PB |
2336 | |
2337 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 2338 | |
4ddc9204 | 2339 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
2340 | |
2341 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
2342 | log->num_pages > memslot->npages - log->first_page || |
2343 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
2344 | return -EINVAL; | |
98938aa8 | 2345 | |
0dff0846 SC |
2346 | kvm_arch_sync_dirty_log(kvm, memslot); |
2347 | ||
2348 | flush = false; | |
2a31b9db PB |
2349 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
2350 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
2351 | return -EFAULT; | |
ba0513b5 | 2352 | |
531810ca | 2353 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
2354 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
2355 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
2356 | i++, offset += BITS_PER_LONG) { |
2357 | unsigned long mask = *dirty_bitmap_buffer++; | |
2358 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
2359 | if (!mask) | |
ba0513b5 MS |
2360 | continue; |
2361 | ||
2a31b9db | 2362 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 2363 | |
2a31b9db PB |
2364 | /* |
2365 | * mask contains the bits that really have been cleared. This | |
2366 | * never includes any bits beyond the length of the memslot (if | |
2367 | * the length is not aligned to 64 pages), therefore it is not | |
2368 | * a problem if userspace sets them in log->dirty_bitmap. | |
2369 | */ | |
58d2930f | 2370 | if (mask) { |
0dff0846 | 2371 | flush = true; |
58d2930f TY |
2372 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
2373 | offset, mask); | |
2374 | } | |
ba0513b5 | 2375 | } |
531810ca | 2376 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 2377 | |
0dff0846 | 2378 | if (flush) |
619b5072 | 2379 | kvm_flush_remote_tlbs_memslot(kvm, memslot); |
0dff0846 | 2380 | |
58d6db34 | 2381 | return 0; |
ba0513b5 | 2382 | } |
0dff0846 SC |
2383 | |
2384 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
2385 | struct kvm_clear_dirty_log *log) | |
2386 | { | |
2387 | int r; | |
2388 | ||
2389 | mutex_lock(&kvm->slots_lock); | |
2390 | ||
2391 | r = kvm_clear_dirty_log_protect(kvm, log); | |
2392 | ||
2393 | mutex_unlock(&kvm->slots_lock); | |
2394 | return r; | |
2395 | } | |
2396 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 2397 | |
5a475554 CP |
2398 | #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES |
2399 | /* | |
2400 | * Returns true if _all_ gfns in the range [@start, @end) have attributes | |
2401 | * matching @attrs. | |
2402 | */ | |
2403 | bool kvm_range_has_memory_attributes(struct kvm *kvm, gfn_t start, gfn_t end, | |
2404 | unsigned long attrs) | |
2405 | { | |
2406 | XA_STATE(xas, &kvm->mem_attr_array, start); | |
2407 | unsigned long index; | |
2408 | bool has_attrs; | |
2409 | void *entry; | |
2410 | ||
2411 | rcu_read_lock(); | |
2412 | ||
2413 | if (!attrs) { | |
2414 | has_attrs = !xas_find(&xas, end - 1); | |
2415 | goto out; | |
2416 | } | |
2417 | ||
2418 | has_attrs = true; | |
2419 | for (index = start; index < end; index++) { | |
2420 | do { | |
2421 | entry = xas_next(&xas); | |
2422 | } while (xas_retry(&xas, entry)); | |
2423 | ||
2424 | if (xas.xa_index != index || xa_to_value(entry) != attrs) { | |
2425 | has_attrs = false; | |
2426 | break; | |
2427 | } | |
2428 | } | |
2429 | ||
2430 | out: | |
2431 | rcu_read_unlock(); | |
2432 | return has_attrs; | |
2433 | } | |
2434 | ||
2435 | static u64 kvm_supported_mem_attributes(struct kvm *kvm) | |
2436 | { | |
a7800aa8 | 2437 | if (!kvm || kvm_arch_has_private_mem(kvm)) |
5a475554 CP |
2438 | return KVM_MEMORY_ATTRIBUTE_PRIVATE; |
2439 | ||
2440 | return 0; | |
2441 | } | |
2442 | ||
2443 | static __always_inline void kvm_handle_gfn_range(struct kvm *kvm, | |
2444 | struct kvm_mmu_notifier_range *range) | |
2445 | { | |
2446 | struct kvm_gfn_range gfn_range; | |
2447 | struct kvm_memory_slot *slot; | |
2448 | struct kvm_memslots *slots; | |
2449 | struct kvm_memslot_iter iter; | |
2450 | bool found_memslot = false; | |
2451 | bool ret = false; | |
2452 | int i; | |
2453 | ||
2454 | gfn_range.arg = range->arg; | |
2455 | gfn_range.may_block = range->may_block; | |
2456 | ||
eed52e43 | 2457 | for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { |
5a475554 CP |
2458 | slots = __kvm_memslots(kvm, i); |
2459 | ||
2460 | kvm_for_each_memslot_in_gfn_range(&iter, slots, range->start, range->end) { | |
2461 | slot = iter.slot; | |
2462 | gfn_range.slot = slot; | |
2463 | ||
2464 | gfn_range.start = max(range->start, slot->base_gfn); | |
2465 | gfn_range.end = min(range->end, slot->base_gfn + slot->npages); | |
2466 | if (gfn_range.start >= gfn_range.end) | |
2467 | continue; | |
2468 | ||
2469 | if (!found_memslot) { | |
2470 | found_memslot = true; | |
2471 | KVM_MMU_LOCK(kvm); | |
2472 | if (!IS_KVM_NULL_FN(range->on_lock)) | |
2473 | range->on_lock(kvm); | |
2474 | } | |
2475 | ||
2476 | ret |= range->handler(kvm, &gfn_range); | |
2477 | } | |
2478 | } | |
2479 | ||
2480 | if (range->flush_on_ret && ret) | |
2481 | kvm_flush_remote_tlbs(kvm); | |
2482 | ||
2483 | if (found_memslot) | |
2484 | KVM_MMU_UNLOCK(kvm); | |
2485 | } | |
2486 | ||
2487 | static bool kvm_pre_set_memory_attributes(struct kvm *kvm, | |
2488 | struct kvm_gfn_range *range) | |
2489 | { | |
2490 | /* | |
2491 | * Unconditionally add the range to the invalidation set, regardless of | |
2492 | * whether or not the arch callback actually needs to zap SPTEs. E.g. | |
2493 | * if KVM supports RWX attributes in the future and the attributes are | |
2494 | * going from R=>RW, zapping isn't strictly necessary. Unconditionally | |
2495 | * adding the range allows KVM to require that MMU invalidations add at | |
2496 | * least one range between begin() and end(), e.g. allows KVM to detect | |
2497 | * bugs where the add() is missed. Relaxing the rule *might* be safe, | |
2498 | * but it's not obvious that allowing new mappings while the attributes | |
2499 | * are in flux is desirable or worth the complexity. | |
2500 | */ | |
2501 | kvm_mmu_invalidate_range_add(kvm, range->start, range->end); | |
2502 | ||
2503 | return kvm_arch_pre_set_memory_attributes(kvm, range); | |
2504 | } | |
2505 | ||
2506 | /* Set @attributes for the gfn range [@start, @end). */ | |
2507 | static int kvm_vm_set_mem_attributes(struct kvm *kvm, gfn_t start, gfn_t end, | |
2508 | unsigned long attributes) | |
2509 | { | |
2510 | struct kvm_mmu_notifier_range pre_set_range = { | |
2511 | .start = start, | |
2512 | .end = end, | |
2513 | .handler = kvm_pre_set_memory_attributes, | |
2514 | .on_lock = kvm_mmu_invalidate_begin, | |
2515 | .flush_on_ret = true, | |
2516 | .may_block = true, | |
2517 | }; | |
2518 | struct kvm_mmu_notifier_range post_set_range = { | |
2519 | .start = start, | |
2520 | .end = end, | |
2521 | .arg.attributes = attributes, | |
2522 | .handler = kvm_arch_post_set_memory_attributes, | |
2523 | .on_lock = kvm_mmu_invalidate_end, | |
2524 | .may_block = true, | |
2525 | }; | |
2526 | unsigned long i; | |
2527 | void *entry; | |
2528 | int r = 0; | |
2529 | ||
2530 | entry = attributes ? xa_mk_value(attributes) : NULL; | |
2531 | ||
2532 | mutex_lock(&kvm->slots_lock); | |
2533 | ||
2534 | /* Nothing to do if the entire range as the desired attributes. */ | |
2535 | if (kvm_range_has_memory_attributes(kvm, start, end, attributes)) | |
2536 | goto out_unlock; | |
2537 | ||
2538 | /* | |
2539 | * Reserve memory ahead of time to avoid having to deal with failures | |
2540 | * partway through setting the new attributes. | |
2541 | */ | |
2542 | for (i = start; i < end; i++) { | |
2543 | r = xa_reserve(&kvm->mem_attr_array, i, GFP_KERNEL_ACCOUNT); | |
2544 | if (r) | |
2545 | goto out_unlock; | |
2546 | } | |
2547 | ||
2548 | kvm_handle_gfn_range(kvm, &pre_set_range); | |
2549 | ||
2550 | for (i = start; i < end; i++) { | |
2551 | r = xa_err(xa_store(&kvm->mem_attr_array, i, entry, | |
2552 | GFP_KERNEL_ACCOUNT)); | |
2553 | KVM_BUG_ON(r, kvm); | |
2554 | } | |
2555 | ||
2556 | kvm_handle_gfn_range(kvm, &post_set_range); | |
2557 | ||
2558 | out_unlock: | |
2559 | mutex_unlock(&kvm->slots_lock); | |
2560 | ||
2561 | return r; | |
2562 | } | |
2563 | static int kvm_vm_ioctl_set_mem_attributes(struct kvm *kvm, | |
2564 | struct kvm_memory_attributes *attrs) | |
2565 | { | |
2566 | gfn_t start, end; | |
2567 | ||
2568 | /* flags is currently not used. */ | |
2569 | if (attrs->flags) | |
2570 | return -EINVAL; | |
2571 | if (attrs->attributes & ~kvm_supported_mem_attributes(kvm)) | |
2572 | return -EINVAL; | |
2573 | if (attrs->size == 0 || attrs->address + attrs->size < attrs->address) | |
2574 | return -EINVAL; | |
2575 | if (!PAGE_ALIGNED(attrs->address) || !PAGE_ALIGNED(attrs->size)) | |
2576 | return -EINVAL; | |
2577 | ||
2578 | start = attrs->address >> PAGE_SHIFT; | |
2579 | end = (attrs->address + attrs->size) >> PAGE_SHIFT; | |
2580 | ||
2581 | /* | |
2582 | * xarray tracks data using "unsigned long", and as a result so does | |
2583 | * KVM. For simplicity, supports generic attributes only on 64-bit | |
2584 | * architectures. | |
2585 | */ | |
2586 | BUILD_BUG_ON(sizeof(attrs->attributes) != sizeof(unsigned long)); | |
2587 | ||
2588 | return kvm_vm_set_mem_attributes(kvm, start, end, attrs->attributes); | |
2589 | } | |
2590 | #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */ | |
2591 | ||
49c7754c GN |
2592 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
2593 | { | |
2594 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
2595 | } | |
a1f4d395 | 2596 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 2597 | |
8e73485c PB |
2598 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
2599 | { | |
fe22ed82 | 2600 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
a54d8066 | 2601 | u64 gen = slots->generation; |
fe22ed82 | 2602 | struct kvm_memory_slot *slot; |
fe22ed82 | 2603 | |
a54d8066 MS |
2604 | /* |
2605 | * This also protects against using a memslot from a different address space, | |
2606 | * since different address spaces have different generation numbers. | |
2607 | */ | |
2608 | if (unlikely(gen != vcpu->last_used_slot_gen)) { | |
2609 | vcpu->last_used_slot = NULL; | |
2610 | vcpu->last_used_slot_gen = gen; | |
2611 | } | |
2612 | ||
2613 | slot = try_get_memslot(vcpu->last_used_slot, gfn); | |
fe22ed82 DM |
2614 | if (slot) |
2615 | return slot; | |
2616 | ||
2617 | /* | |
2618 | * Fall back to searching all memslots. We purposely use | |
2619 | * search_memslots() instead of __gfn_to_memslot() to avoid | |
a54d8066 | 2620 | * thrashing the VM-wide last_used_slot in kvm_memslots. |
fe22ed82 | 2621 | */ |
a54d8066 | 2622 | slot = search_memslots(slots, gfn, false); |
fe22ed82 | 2623 | if (slot) { |
a54d8066 | 2624 | vcpu->last_used_slot = slot; |
fe22ed82 DM |
2625 | return slot; |
2626 | } | |
2627 | ||
2628 | return NULL; | |
8e73485c PB |
2629 | } |
2630 | ||
33e94154 | 2631 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 2632 | { |
bf3e05bc | 2633 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 2634 | |
c36b7150 | 2635 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
2636 | } |
2637 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
2638 | ||
995decb6 VK |
2639 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
2640 | { | |
2641 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2642 | ||
2643 | return kvm_is_visible_memslot(memslot); | |
2644 | } | |
2645 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
2646 | ||
f9b84e19 | 2647 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
2648 | { |
2649 | struct vm_area_struct *vma; | |
2650 | unsigned long addr, size; | |
2651 | ||
2652 | size = PAGE_SIZE; | |
2653 | ||
42cde48b | 2654 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
2655 | if (kvm_is_error_hva(addr)) |
2656 | return PAGE_SIZE; | |
2657 | ||
d8ed45c5 | 2658 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
2659 | vma = find_vma(current->mm, addr); |
2660 | if (!vma) | |
2661 | goto out; | |
2662 | ||
2663 | size = vma_kernel_pagesize(vma); | |
2664 | ||
2665 | out: | |
d8ed45c5 | 2666 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
2667 | |
2668 | return size; | |
2669 | } | |
2670 | ||
8283e36a | 2671 | static bool memslot_is_readonly(const struct kvm_memory_slot *slot) |
4d8b81ab XG |
2672 | { |
2673 | return slot->flags & KVM_MEM_READONLY; | |
2674 | } | |
2675 | ||
8283e36a | 2676 | static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn, |
4d8b81ab | 2677 | gfn_t *nr_pages, bool write) |
539cb660 | 2678 | { |
bc6678a3 | 2679 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 2680 | return KVM_HVA_ERR_BAD; |
48987781 | 2681 | |
4d8b81ab XG |
2682 | if (memslot_is_readonly(slot) && write) |
2683 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
2684 | |
2685 | if (nr_pages) | |
2686 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
2687 | ||
4d8b81ab | 2688 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 2689 | } |
48987781 | 2690 | |
4d8b81ab XG |
2691 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2692 | gfn_t *nr_pages) | |
2693 | { | |
2694 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 2695 | } |
48987781 | 2696 | |
4d8b81ab | 2697 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 2698 | gfn_t gfn) |
4d8b81ab XG |
2699 | { |
2700 | return gfn_to_hva_many(slot, gfn, NULL); | |
2701 | } | |
2702 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
2703 | ||
48987781 XG |
2704 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
2705 | { | |
49c7754c | 2706 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 2707 | } |
0d150298 | 2708 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 2709 | |
8e73485c PB |
2710 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
2711 | { | |
2712 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
2713 | } | |
2714 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
2715 | ||
86ab8cff | 2716 | /* |
970c0d4b WY |
2717 | * Return the hva of a @gfn and the R/W attribute if possible. |
2718 | * | |
2719 | * @slot: the kvm_memory_slot which contains @gfn | |
2720 | * @gfn: the gfn to be translated | |
2721 | * @writable: used to return the read/write attribute of the @slot if the hva | |
2722 | * is valid and @writable is not NULL | |
86ab8cff | 2723 | */ |
64d83126 CD |
2724 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
2725 | gfn_t gfn, bool *writable) | |
86ab8cff | 2726 | { |
a2ac07fe GN |
2727 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
2728 | ||
2729 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
2730 | *writable = !memslot_is_readonly(slot); |
2731 | ||
a2ac07fe | 2732 | return hva; |
86ab8cff XG |
2733 | } |
2734 | ||
64d83126 CD |
2735 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
2736 | { | |
2737 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2738 | ||
2739 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2740 | } | |
2741 | ||
8e73485c PB |
2742 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
2743 | { | |
2744 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2745 | ||
2746 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2747 | } | |
2748 | ||
fafc3dba HY |
2749 | static inline int check_user_page_hwpoison(unsigned long addr) |
2750 | { | |
0d731759 | 2751 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 2752 | |
54d02069 | 2753 | rc = get_user_pages(addr, 1, flags, NULL); |
fafc3dba HY |
2754 | return rc == -EHWPOISON; |
2755 | } | |
2756 | ||
2fc84311 | 2757 | /* |
b9b33da2 PB |
2758 | * The fast path to get the writable pfn which will be stored in @pfn, |
2759 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 2760 | * only part that runs if we can in atomic context. |
2fc84311 | 2761 | */ |
b9b33da2 PB |
2762 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
2763 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 2764 | { |
8d4e1288 | 2765 | struct page *page[1]; |
954bbbc2 | 2766 | |
12ce13fe XG |
2767 | /* |
2768 | * Fast pin a writable pfn only if it is a write fault request | |
2769 | * or the caller allows to map a writable pfn for a read fault | |
2770 | * request. | |
2771 | */ | |
2772 | if (!(write_fault || writable)) | |
2773 | return false; | |
612819c3 | 2774 | |
dadbb612 | 2775 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 2776 | *pfn = page_to_pfn(page[0]); |
612819c3 | 2777 | |
2fc84311 XG |
2778 | if (writable) |
2779 | *writable = true; | |
2780 | return true; | |
2781 | } | |
af585b92 | 2782 | |
2fc84311 XG |
2783 | return false; |
2784 | } | |
612819c3 | 2785 | |
2fc84311 XG |
2786 | /* |
2787 | * The slow path to get the pfn of the specified host virtual address, | |
2788 | * 1 indicates success, -errno is returned if error is detected. | |
2789 | */ | |
2790 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
c8b88b33 | 2791 | bool interruptible, bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 2792 | { |
b1e1296d DH |
2793 | /* |
2794 | * When a VCPU accesses a page that is not mapped into the secondary | |
2795 | * MMU, we lookup the page using GUP to map it, so the guest VCPU can | |
2796 | * make progress. We always want to honor NUMA hinting faults in that | |
2797 | * case, because GUP usage corresponds to memory accesses from the VCPU. | |
2798 | * Otherwise, we'd not trigger NUMA hinting faults once a page is | |
2799 | * mapped into the secondary MMU and gets accessed by a VCPU. | |
2800 | * | |
2801 | * Note that get_user_page_fast_only() and FOLL_WRITE for now | |
2802 | * implicitly honor NUMA hinting faults and don't need this flag. | |
2803 | */ | |
2804 | unsigned int flags = FOLL_HWPOISON | FOLL_HONOR_NUMA_FAULT; | |
ce53053c | 2805 | struct page *page; |
28249139 | 2806 | int npages; |
612819c3 | 2807 | |
2fc84311 XG |
2808 | might_sleep(); |
2809 | ||
2810 | if (writable) | |
2811 | *writable = write_fault; | |
2812 | ||
ce53053c AV |
2813 | if (write_fault) |
2814 | flags |= FOLL_WRITE; | |
2815 | if (async) | |
2816 | flags |= FOLL_NOWAIT; | |
c8b88b33 PX |
2817 | if (interruptible) |
2818 | flags |= FOLL_INTERRUPTIBLE; | |
d4944b0e | 2819 | |
ce53053c | 2820 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
2821 | if (npages != 1) |
2822 | return npages; | |
2823 | ||
2824 | /* map read fault as writable if possible */ | |
12ce13fe | 2825 | if (unlikely(!write_fault) && writable) { |
ce53053c | 2826 | struct page *wpage; |
2fc84311 | 2827 | |
dadbb612 | 2828 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 2829 | *writable = true; |
ce53053c AV |
2830 | put_page(page); |
2831 | page = wpage; | |
612819c3 | 2832 | } |
887c08ac | 2833 | } |
ce53053c | 2834 | *pfn = page_to_pfn(page); |
2fc84311 XG |
2835 | return npages; |
2836 | } | |
539cb660 | 2837 | |
4d8b81ab XG |
2838 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
2839 | { | |
2840 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
2841 | return false; | |
2e2e3738 | 2842 | |
4d8b81ab XG |
2843 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
2844 | return false; | |
887c08ac | 2845 | |
4d8b81ab XG |
2846 | return true; |
2847 | } | |
bf998156 | 2848 | |
f8be156b NP |
2849 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
2850 | { | |
b14b2690 SC |
2851 | struct page *page = kvm_pfn_to_refcounted_page(pfn); |
2852 | ||
2853 | if (!page) | |
f8be156b | 2854 | return 1; |
b14b2690 SC |
2855 | |
2856 | return get_page_unless_zero(page); | |
f8be156b NP |
2857 | } |
2858 | ||
92176a8e | 2859 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1625566e XT |
2860 | unsigned long addr, bool write_fault, |
2861 | bool *writable, kvm_pfn_t *p_pfn) | |
92176a8e | 2862 | { |
a9545779 | 2863 | kvm_pfn_t pfn; |
bd2fae8d | 2864 | pte_t *ptep; |
c33c7948 | 2865 | pte_t pte; |
bd2fae8d | 2866 | spinlock_t *ptl; |
add6a0cd PB |
2867 | int r; |
2868 | ||
9fd6dad1 | 2869 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2870 | if (r) { |
2871 | /* | |
2872 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
2873 | * not call the fault handler, so do it here. | |
2874 | */ | |
2875 | bool unlocked = false; | |
64019a2e | 2876 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
2877 | (write_fault ? FAULT_FLAG_WRITE : 0), |
2878 | &unlocked); | |
a8387d0b PB |
2879 | if (unlocked) |
2880 | return -EAGAIN; | |
add6a0cd PB |
2881 | if (r) |
2882 | return r; | |
2883 | ||
9fd6dad1 | 2884 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2885 | if (r) |
2886 | return r; | |
bd2fae8d | 2887 | } |
add6a0cd | 2888 | |
c33c7948 RR |
2889 | pte = ptep_get(ptep); |
2890 | ||
2891 | if (write_fault && !pte_write(pte)) { | |
bd2fae8d PB |
2892 | pfn = KVM_PFN_ERR_RO_FAULT; |
2893 | goto out; | |
add6a0cd PB |
2894 | } |
2895 | ||
a340b3e2 | 2896 | if (writable) |
c33c7948 RR |
2897 | *writable = pte_write(pte); |
2898 | pfn = pte_pfn(pte); | |
add6a0cd PB |
2899 | |
2900 | /* | |
2901 | * Get a reference here because callers of *hva_to_pfn* and | |
2902 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
2903 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
36c3ce6c | 2904 | * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will |
add6a0cd PB |
2905 | * simply do nothing for reserved pfns. |
2906 | * | |
2907 | * Whoever called remap_pfn_range is also going to call e.g. | |
2908 | * unmap_mapping_range before the underlying pages are freed, | |
2909 | * causing a call to our MMU notifier. | |
f8be156b NP |
2910 | * |
2911 | * Certain IO or PFNMAP mappings can be backed with valid | |
2912 | * struct pages, but be allocated without refcounting e.g., | |
2913 | * tail pages of non-compound higher order allocations, which | |
2914 | * would then underflow the refcount when the caller does the | |
2915 | * required put_page. Don't allow those pages here. | |
c33c7948 | 2916 | */ |
f8be156b NP |
2917 | if (!kvm_try_get_pfn(pfn)) |
2918 | r = -EFAULT; | |
add6a0cd | 2919 | |
bd2fae8d PB |
2920 | out: |
2921 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 2922 | *p_pfn = pfn; |
f8be156b NP |
2923 | |
2924 | return r; | |
92176a8e PB |
2925 | } |
2926 | ||
12ce13fe XG |
2927 | /* |
2928 | * Pin guest page in memory and return its pfn. | |
2929 | * @addr: host virtual address which maps memory to the guest | |
2930 | * @atomic: whether this function can sleep | |
c8b88b33 | 2931 | * @interruptible: whether the process can be interrupted by non-fatal signals |
12ce13fe XG |
2932 | * @async: whether this function need to wait IO complete if the |
2933 | * host page is not in the memory | |
2934 | * @write_fault: whether we should get a writable host page | |
2935 | * @writable: whether it allows to map a writable host page for !@write_fault | |
2936 | * | |
2937 | * The function will map a writable host page for these two cases: | |
2938 | * 1): @write_fault = true | |
2939 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
2940 | * whether the mapping is writable. | |
2941 | */ | |
c8b88b33 PX |
2942 | kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool interruptible, |
2943 | bool *async, bool write_fault, bool *writable) | |
2fc84311 XG |
2944 | { |
2945 | struct vm_area_struct *vma; | |
943dfea8 | 2946 | kvm_pfn_t pfn; |
92176a8e | 2947 | int npages, r; |
2e2e3738 | 2948 | |
2fc84311 XG |
2949 | /* we can do it either atomically or asynchronously, not both */ |
2950 | BUG_ON(atomic && async); | |
8d4e1288 | 2951 | |
b9b33da2 | 2952 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
2953 | return pfn; |
2954 | ||
2955 | if (atomic) | |
2956 | return KVM_PFN_ERR_FAULT; | |
2957 | ||
c8b88b33 PX |
2958 | npages = hva_to_pfn_slow(addr, async, write_fault, interruptible, |
2959 | writable, &pfn); | |
2fc84311 XG |
2960 | if (npages == 1) |
2961 | return pfn; | |
fe5ed56c PX |
2962 | if (npages == -EINTR) |
2963 | return KVM_PFN_ERR_SIGPENDING; | |
8d4e1288 | 2964 | |
d8ed45c5 | 2965 | mmap_read_lock(current->mm); |
2fc84311 XG |
2966 | if (npages == -EHWPOISON || |
2967 | (!async && check_user_page_hwpoison(addr))) { | |
2968 | pfn = KVM_PFN_ERR_HWPOISON; | |
2969 | goto exit; | |
2970 | } | |
2971 | ||
a8387d0b | 2972 | retry: |
fc98c03b | 2973 | vma = vma_lookup(current->mm, addr); |
2fc84311 XG |
2974 | |
2975 | if (vma == NULL) | |
2976 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 2977 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
1625566e | 2978 | r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn); |
a8387d0b PB |
2979 | if (r == -EAGAIN) |
2980 | goto retry; | |
92176a8e PB |
2981 | if (r < 0) |
2982 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 2983 | } else { |
4d8b81ab | 2984 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
2985 | *async = true; |
2986 | pfn = KVM_PFN_ERR_FAULT; | |
2987 | } | |
2988 | exit: | |
d8ed45c5 | 2989 | mmap_read_unlock(current->mm); |
2e2e3738 | 2990 | return pfn; |
35149e21 AL |
2991 | } |
2992 | ||
8283e36a | 2993 | kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn, |
c8b88b33 PX |
2994 | bool atomic, bool interruptible, bool *async, |
2995 | bool write_fault, bool *writable, hva_t *hva) | |
887c08ac | 2996 | { |
4d8b81ab XG |
2997 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
2998 | ||
4a42d848 DS |
2999 | if (hva) |
3000 | *hva = addr; | |
3001 | ||
b2740d35 PB |
3002 | if (addr == KVM_HVA_ERR_RO_BAD) { |
3003 | if (writable) | |
3004 | *writable = false; | |
4d8b81ab | 3005 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 3006 | } |
4d8b81ab | 3007 | |
b2740d35 PB |
3008 | if (kvm_is_error_hva(addr)) { |
3009 | if (writable) | |
3010 | *writable = false; | |
81c52c56 | 3011 | return KVM_PFN_NOSLOT; |
b2740d35 | 3012 | } |
4d8b81ab XG |
3013 | |
3014 | /* Do not map writable pfn in the readonly memslot. */ | |
3015 | if (writable && memslot_is_readonly(slot)) { | |
3016 | *writable = false; | |
3017 | writable = NULL; | |
3018 | } | |
3019 | ||
c8b88b33 | 3020 | return hva_to_pfn(addr, atomic, interruptible, async, write_fault, |
4d8b81ab | 3021 | writable); |
887c08ac | 3022 | } |
3520469d | 3023 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 3024 | |
ba049e93 | 3025 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
3026 | bool *writable) |
3027 | { | |
c8b88b33 PX |
3028 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, false, |
3029 | NULL, write_fault, writable, NULL); | |
612819c3 MT |
3030 | } |
3031 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
3032 | ||
8283e36a | 3033 | kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 3034 | { |
c8b88b33 PX |
3035 | return __gfn_to_pfn_memslot(slot, gfn, false, false, NULL, true, |
3036 | NULL, NULL); | |
506f0d6f | 3037 | } |
e37afc6e | 3038 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 3039 | |
8283e36a | 3040 | kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 3041 | { |
c8b88b33 PX |
3042 | return __gfn_to_pfn_memslot(slot, gfn, true, false, NULL, true, |
3043 | NULL, NULL); | |
506f0d6f | 3044 | } |
037d92dc | 3045 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 3046 | |
ba049e93 | 3047 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
3048 | { |
3049 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
3050 | } | |
3051 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
3052 | ||
ba049e93 | 3053 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
3054 | { |
3055 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
3056 | } | |
3057 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
3058 | ||
ba049e93 | 3059 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
3060 | { |
3061 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
3062 | } | |
3063 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
3064 | ||
d9ef13c2 PB |
3065 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
3066 | struct page **pages, int nr_pages) | |
48987781 XG |
3067 | { |
3068 | unsigned long addr; | |
076b925d | 3069 | gfn_t entry = 0; |
48987781 | 3070 | |
d9ef13c2 | 3071 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
3072 | if (kvm_is_error_hva(addr)) |
3073 | return -1; | |
3074 | ||
3075 | if (entry < nr_pages) | |
3076 | return 0; | |
3077 | ||
dadbb612 | 3078 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
3079 | } |
3080 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
3081 | ||
b1624f99 SC |
3082 | /* |
3083 | * Do not use this helper unless you are absolutely certain the gfn _must_ be | |
3084 | * backed by 'struct page'. A valid example is if the backing memslot is | |
3085 | * controlled by KVM. Note, if the returned page is valid, it's refcount has | |
3086 | * been elevated by gfn_to_pfn(). | |
3087 | */ | |
35149e21 AL |
3088 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
3089 | { | |
b14b2690 | 3090 | struct page *page; |
ba049e93 | 3091 | kvm_pfn_t pfn; |
2e2e3738 AL |
3092 | |
3093 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 3094 | |
81c52c56 | 3095 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 3096 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 3097 | |
b14b2690 SC |
3098 | page = kvm_pfn_to_refcounted_page(pfn); |
3099 | if (!page) | |
6cede2e6 | 3100 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 3101 | |
b14b2690 | 3102 | return page; |
954bbbc2 AK |
3103 | } |
3104 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
3105 | ||
357a18ad | 3106 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty) |
91724814 | 3107 | { |
91724814 BO |
3108 | if (dirty) |
3109 | kvm_release_pfn_dirty(pfn); | |
3110 | else | |
3111 | kvm_release_pfn_clean(pfn); | |
3112 | } | |
3113 | ||
357a18ad | 3114 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
e45adf66 KA |
3115 | { |
3116 | kvm_pfn_t pfn; | |
3117 | void *hva = NULL; | |
3118 | struct page *page = KVM_UNMAPPED_PAGE; | |
3119 | ||
3120 | if (!map) | |
3121 | return -EINVAL; | |
3122 | ||
357a18ad | 3123 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
e45adf66 KA |
3124 | if (is_error_noslot_pfn(pfn)) |
3125 | return -EINVAL; | |
3126 | ||
3127 | if (pfn_valid(pfn)) { | |
3128 | page = pfn_to_page(pfn); | |
357a18ad | 3129 | hva = kmap(page); |
d30b214d | 3130 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 3131 | } else { |
357a18ad | 3132 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
d30b214d | 3133 | #endif |
e45adf66 KA |
3134 | } |
3135 | ||
3136 | if (!hva) | |
3137 | return -EFAULT; | |
3138 | ||
3139 | map->page = page; | |
3140 | map->hva = hva; | |
3141 | map->pfn = pfn; | |
3142 | map->gfn = gfn; | |
3143 | ||
3144 | return 0; | |
3145 | } | |
e45adf66 KA |
3146 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); |
3147 | ||
357a18ad | 3148 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) |
e45adf66 KA |
3149 | { |
3150 | if (!map) | |
3151 | return; | |
3152 | ||
3153 | if (!map->hva) | |
3154 | return; | |
3155 | ||
357a18ad DW |
3156 | if (map->page != KVM_UNMAPPED_PAGE) |
3157 | kunmap(map->page); | |
eb1f2f38 | 3158 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 3159 | else |
357a18ad | 3160 | memunmap(map->hva); |
eb1f2f38 | 3161 | #endif |
e45adf66 | 3162 | |
91724814 | 3163 | if (dirty) |
357a18ad | 3164 | kvm_vcpu_mark_page_dirty(vcpu, map->gfn); |
91724814 | 3165 | |
357a18ad | 3166 | kvm_release_pfn(map->pfn, dirty); |
e45adf66 KA |
3167 | |
3168 | map->hva = NULL; | |
3169 | map->page = NULL; | |
3170 | } | |
3171 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); | |
3172 | ||
8e1c6914 | 3173 | static bool kvm_is_ad_tracked_page(struct page *page) |
8e73485c | 3174 | { |
8e1c6914 SC |
3175 | /* |
3176 | * Per page-flags.h, pages tagged PG_reserved "should in general not be | |
3177 | * touched (e.g. set dirty) except by its owner". | |
3178 | */ | |
3179 | return !PageReserved(page); | |
3180 | } | |
8e73485c | 3181 | |
8e1c6914 SC |
3182 | static void kvm_set_page_dirty(struct page *page) |
3183 | { | |
3184 | if (kvm_is_ad_tracked_page(page)) | |
3185 | SetPageDirty(page); | |
3186 | } | |
8e73485c | 3187 | |
8e1c6914 SC |
3188 | static void kvm_set_page_accessed(struct page *page) |
3189 | { | |
3190 | if (kvm_is_ad_tracked_page(page)) | |
3191 | mark_page_accessed(page); | |
8e73485c | 3192 | } |
8e73485c | 3193 | |
b4231d61 IE |
3194 | void kvm_release_page_clean(struct page *page) |
3195 | { | |
32cad84f XG |
3196 | WARN_ON(is_error_page(page)); |
3197 | ||
8e1c6914 SC |
3198 | kvm_set_page_accessed(page); |
3199 | put_page(page); | |
b4231d61 IE |
3200 | } |
3201 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
3202 | ||
ba049e93 | 3203 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 3204 | { |
b14b2690 SC |
3205 | struct page *page; |
3206 | ||
3207 | if (is_error_noslot_pfn(pfn)) | |
3208 | return; | |
3209 | ||
3210 | page = kvm_pfn_to_refcounted_page(pfn); | |
3211 | if (!page) | |
3212 | return; | |
3213 | ||
3214 | kvm_release_page_clean(page); | |
35149e21 AL |
3215 | } |
3216 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
3217 | ||
b4231d61 | 3218 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 3219 | { |
a2766325 XG |
3220 | WARN_ON(is_error_page(page)); |
3221 | ||
8e1c6914 SC |
3222 | kvm_set_page_dirty(page); |
3223 | kvm_release_page_clean(page); | |
35149e21 AL |
3224 | } |
3225 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
3226 | ||
f7a6509f | 3227 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 3228 | { |
b14b2690 SC |
3229 | struct page *page; |
3230 | ||
3231 | if (is_error_noslot_pfn(pfn)) | |
3232 | return; | |
3233 | ||
3234 | page = kvm_pfn_to_refcounted_page(pfn); | |
3235 | if (!page) | |
3236 | return; | |
3237 | ||
3238 | kvm_release_page_dirty(page); | |
35149e21 | 3239 | } |
f7a6509f | 3240 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 3241 | |
8e1c6914 SC |
3242 | /* |
3243 | * Note, checking for an error/noslot pfn is the caller's responsibility when | |
3244 | * directly marking a page dirty/accessed. Unlike the "release" helpers, the | |
3245 | * "set" helpers are not to be used when the pfn might point at garbage. | |
3246 | */ | |
ba049e93 | 3247 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 3248 | { |
8e1c6914 SC |
3249 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
3250 | return; | |
3251 | ||
3252 | if (pfn_valid(pfn)) | |
3253 | kvm_set_page_dirty(pfn_to_page(pfn)); | |
8a7ae055 | 3254 | } |
35149e21 AL |
3255 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
3256 | ||
ba049e93 | 3257 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 3258 | { |
8e1c6914 SC |
3259 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
3260 | return; | |
3261 | ||
3262 | if (pfn_valid(pfn)) | |
3263 | kvm_set_page_accessed(pfn_to_page(pfn)); | |
35149e21 AL |
3264 | } |
3265 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
3266 | ||
195aefde IE |
3267 | static int next_segment(unsigned long len, int offset) |
3268 | { | |
3269 | if (len > PAGE_SIZE - offset) | |
3270 | return PAGE_SIZE - offset; | |
3271 | else | |
3272 | return len; | |
3273 | } | |
3274 | ||
8e73485c PB |
3275 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
3276 | void *data, int offset, int len) | |
195aefde | 3277 | { |
e0506bcb IE |
3278 | int r; |
3279 | unsigned long addr; | |
195aefde | 3280 | |
8e73485c | 3281 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
3282 | if (kvm_is_error_hva(addr)) |
3283 | return -EFAULT; | |
3180a7fc | 3284 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 3285 | if (r) |
195aefde | 3286 | return -EFAULT; |
195aefde IE |
3287 | return 0; |
3288 | } | |
8e73485c PB |
3289 | |
3290 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
3291 | int len) | |
3292 | { | |
3293 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
3294 | ||
3295 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
3296 | } | |
195aefde IE |
3297 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
3298 | ||
8e73485c PB |
3299 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
3300 | int offset, int len) | |
3301 | { | |
3302 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3303 | ||
3304 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
3305 | } | |
3306 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
3307 | ||
195aefde IE |
3308 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
3309 | { | |
3310 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3311 | int seg; | |
3312 | int offset = offset_in_page(gpa); | |
3313 | int ret; | |
3314 | ||
3315 | while ((seg = next_segment(len, offset)) != 0) { | |
3316 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
3317 | if (ret < 0) | |
3318 | return ret; | |
3319 | offset = 0; | |
3320 | len -= seg; | |
3321 | data += seg; | |
3322 | ++gfn; | |
3323 | } | |
3324 | return 0; | |
3325 | } | |
3326 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
3327 | ||
8e73485c | 3328 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 3329 | { |
7ec54588 | 3330 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 3331 | int seg; |
7ec54588 | 3332 | int offset = offset_in_page(gpa); |
8e73485c PB |
3333 | int ret; |
3334 | ||
3335 | while ((seg = next_segment(len, offset)) != 0) { | |
3336 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
3337 | if (ret < 0) | |
3338 | return ret; | |
3339 | offset = 0; | |
3340 | len -= seg; | |
3341 | data += seg; | |
3342 | ++gfn; | |
3343 | } | |
3344 | return 0; | |
3345 | } | |
3346 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 3347 | |
8e73485c PB |
3348 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
3349 | void *data, int offset, unsigned long len) | |
3350 | { | |
3351 | int r; | |
3352 | unsigned long addr; | |
3353 | ||
3354 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
3355 | if (kvm_is_error_hva(addr)) |
3356 | return -EFAULT; | |
0aac03f0 | 3357 | pagefault_disable(); |
3180a7fc | 3358 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 3359 | pagefault_enable(); |
7ec54588 MT |
3360 | if (r) |
3361 | return -EFAULT; | |
3362 | return 0; | |
3363 | } | |
7ec54588 | 3364 | |
8e73485c PB |
3365 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
3366 | void *data, unsigned long len) | |
3367 | { | |
3368 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3369 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3370 | int offset = offset_in_page(gpa); | |
3371 | ||
3372 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
3373 | } | |
3374 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
3375 | ||
28bd726a PX |
3376 | static int __kvm_write_guest_page(struct kvm *kvm, |
3377 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 3378 | const void *data, int offset, int len) |
195aefde | 3379 | { |
e0506bcb IE |
3380 | int r; |
3381 | unsigned long addr; | |
195aefde | 3382 | |
251eb841 | 3383 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
3384 | if (kvm_is_error_hva(addr)) |
3385 | return -EFAULT; | |
8b0cedff | 3386 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 3387 | if (r) |
195aefde | 3388 | return -EFAULT; |
28bd726a | 3389 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
3390 | return 0; |
3391 | } | |
8e73485c PB |
3392 | |
3393 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
3394 | const void *data, int offset, int len) | |
3395 | { | |
3396 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
3397 | ||
28bd726a | 3398 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 3399 | } |
195aefde IE |
3400 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
3401 | ||
8e73485c PB |
3402 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3403 | const void *data, int offset, int len) | |
3404 | { | |
3405 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3406 | ||
28bd726a | 3407 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
3408 | } |
3409 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
3410 | ||
195aefde IE |
3411 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
3412 | unsigned long len) | |
3413 | { | |
3414 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3415 | int seg; | |
3416 | int offset = offset_in_page(gpa); | |
3417 | int ret; | |
3418 | ||
3419 | while ((seg = next_segment(len, offset)) != 0) { | |
3420 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
3421 | if (ret < 0) | |
3422 | return ret; | |
3423 | offset = 0; | |
3424 | len -= seg; | |
3425 | data += seg; | |
3426 | ++gfn; | |
3427 | } | |
3428 | return 0; | |
3429 | } | |
ff651cb6 | 3430 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 3431 | |
8e73485c PB |
3432 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
3433 | unsigned long len) | |
3434 | { | |
3435 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3436 | int seg; | |
3437 | int offset = offset_in_page(gpa); | |
3438 | int ret; | |
3439 | ||
3440 | while ((seg = next_segment(len, offset)) != 0) { | |
3441 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
3442 | if (ret < 0) | |
3443 | return ret; | |
3444 | offset = 0; | |
3445 | len -= seg; | |
3446 | data += seg; | |
3447 | ++gfn; | |
3448 | } | |
3449 | return 0; | |
3450 | } | |
3451 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
3452 | ||
5a2d4365 PB |
3453 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
3454 | struct gfn_to_hva_cache *ghc, | |
3455 | gpa_t gpa, unsigned long len) | |
49c7754c | 3456 | { |
49c7754c | 3457 | int offset = offset_in_page(gpa); |
8f964525 AH |
3458 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
3459 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
3460 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
3461 | gfn_t nr_pages_avail; | |
49c7754c | 3462 | |
6ad1e29f | 3463 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 3464 | ghc->generation = slots->generation; |
6ad1e29f SC |
3465 | |
3466 | if (start_gfn > end_gfn) { | |
3467 | ghc->hva = KVM_HVA_ERR_BAD; | |
3468 | return -EINVAL; | |
3469 | } | |
f1b9dd5e JM |
3470 | |
3471 | /* | |
3472 | * If the requested region crosses two memslots, we still | |
3473 | * verify that the entire region is valid here. | |
3474 | */ | |
6ad1e29f | 3475 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
3476 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
3477 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
3478 | &nr_pages_avail); | |
3479 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 3480 | return -EFAULT; |
f1b9dd5e JM |
3481 | } |
3482 | ||
3483 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 3484 | if (nr_pages_needed == 1) |
49c7754c | 3485 | ghc->hva += offset; |
f1b9dd5e | 3486 | else |
8f964525 | 3487 | ghc->memslot = NULL; |
f1b9dd5e | 3488 | |
6ad1e29f SC |
3489 | ghc->gpa = gpa; |
3490 | ghc->len = len; | |
3491 | return 0; | |
49c7754c | 3492 | } |
5a2d4365 | 3493 | |
4e335d9e | 3494 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
3495 | gpa_t gpa, unsigned long len) |
3496 | { | |
4e335d9e | 3497 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
3498 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
3499 | } | |
4e335d9e | 3500 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 3501 | |
4e335d9e | 3502 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
3503 | void *data, unsigned int offset, |
3504 | unsigned long len) | |
49c7754c | 3505 | { |
4e335d9e | 3506 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 3507 | int r; |
4ec6e863 | 3508 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 3509 | |
5f25e71e PB |
3510 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3511 | return -EINVAL; | |
8f964525 | 3512 | |
dc9ce71e SC |
3513 | if (slots->generation != ghc->generation) { |
3514 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3515 | return -EFAULT; | |
3516 | } | |
8f964525 | 3517 | |
49c7754c GN |
3518 | if (kvm_is_error_hva(ghc->hva)) |
3519 | return -EFAULT; | |
3520 | ||
fcfbc617 SC |
3521 | if (unlikely(!ghc->memslot)) |
3522 | return kvm_write_guest(kvm, gpa, data, len); | |
3523 | ||
4ec6e863 | 3524 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
3525 | if (r) |
3526 | return -EFAULT; | |
28bd726a | 3527 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
3528 | |
3529 | return 0; | |
3530 | } | |
4e335d9e | 3531 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 3532 | |
4e335d9e PB |
3533 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3534 | void *data, unsigned long len) | |
4ec6e863 | 3535 | { |
4e335d9e | 3536 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 3537 | } |
4e335d9e | 3538 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 3539 | |
0958f0ce VK |
3540 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3541 | void *data, unsigned int offset, | |
3542 | unsigned long len) | |
e03b644f | 3543 | { |
4e335d9e | 3544 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 3545 | int r; |
0958f0ce | 3546 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 3547 | |
5f25e71e PB |
3548 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3549 | return -EINVAL; | |
8f964525 | 3550 | |
dc9ce71e SC |
3551 | if (slots->generation != ghc->generation) { |
3552 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3553 | return -EFAULT; | |
3554 | } | |
8f964525 | 3555 | |
e03b644f GN |
3556 | if (kvm_is_error_hva(ghc->hva)) |
3557 | return -EFAULT; | |
3558 | ||
fcfbc617 | 3559 | if (unlikely(!ghc->memslot)) |
0958f0ce | 3560 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 3561 | |
0958f0ce | 3562 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
3563 | if (r) |
3564 | return -EFAULT; | |
3565 | ||
3566 | return 0; | |
3567 | } | |
0958f0ce VK |
3568 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
3569 | ||
3570 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
3571 | void *data, unsigned long len) | |
3572 | { | |
3573 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
3574 | } | |
4e335d9e | 3575 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 3576 | |
195aefde IE |
3577 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
3578 | { | |
2f541442 | 3579 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
3580 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3581 | int seg; | |
3582 | int offset = offset_in_page(gpa); | |
3583 | int ret; | |
3584 | ||
bfda0e84 | 3585 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 3586 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
3587 | if (ret < 0) |
3588 | return ret; | |
3589 | offset = 0; | |
3590 | len -= seg; | |
3591 | ++gfn; | |
3592 | } | |
3593 | return 0; | |
3594 | } | |
3595 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
3596 | ||
28bd726a | 3597 | void mark_page_dirty_in_slot(struct kvm *kvm, |
8283e36a | 3598 | const struct kvm_memory_slot *memslot, |
28bd726a | 3599 | gfn_t gfn) |
6aa8b732 | 3600 | { |
2efd61a6 DW |
3601 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); |
3602 | ||
e09fccb5 | 3603 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
86bdf3eb | 3604 | if (WARN_ON_ONCE(vcpu && vcpu->kvm != kvm)) |
2efd61a6 | 3605 | return; |
86bdf3eb | 3606 | |
c57351a7 | 3607 | WARN_ON_ONCE(!vcpu && !kvm_arch_allow_write_without_running_vcpu(kvm)); |
e09fccb5 | 3608 | #endif |
2efd61a6 | 3609 | |
044c59c4 | 3610 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 3611 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 3612 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 3613 | |
86bdf3eb | 3614 | if (kvm->dirty_ring_size && vcpu) |
cf87ac73 | 3615 | kvm_dirty_ring_push(vcpu, slot, rel_gfn); |
c57351a7 | 3616 | else if (memslot->dirty_bitmap) |
fb04a1ed | 3617 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
6aa8b732 AK |
3618 | } |
3619 | } | |
a6a0b05d | 3620 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 3621 | |
49c7754c GN |
3622 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
3623 | { | |
3624 | struct kvm_memory_slot *memslot; | |
3625 | ||
3626 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 3627 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 3628 | } |
2ba9f0d8 | 3629 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 3630 | |
8e73485c PB |
3631 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
3632 | { | |
3633 | struct kvm_memory_slot *memslot; | |
3634 | ||
3635 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 3636 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
3637 | } |
3638 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
3639 | ||
20b7035c JS |
3640 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
3641 | { | |
3642 | if (!vcpu->sigset_active) | |
3643 | return; | |
3644 | ||
3645 | /* | |
3646 | * This does a lockless modification of ->real_blocked, which is fine | |
3647 | * because, only current can change ->real_blocked and all readers of | |
3648 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
3649 | * of ->blocked. | |
3650 | */ | |
3651 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
3652 | } | |
3653 | ||
3654 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
3655 | { | |
3656 | if (!vcpu->sigset_active) | |
3657 | return; | |
3658 | ||
3659 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
3660 | sigemptyset(¤t->real_blocked); | |
3661 | } | |
3662 | ||
aca6ff29 WL |
3663 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
3664 | { | |
dee339b5 | 3665 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 3666 | |
2cbd7824 | 3667 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 3668 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3669 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
3670 | if (!grow) |
3671 | goto out; | |
3672 | ||
dee339b5 NW |
3673 | val *= grow; |
3674 | if (val < grow_start) | |
3675 | val = grow_start; | |
aca6ff29 WL |
3676 | |
3677 | vcpu->halt_poll_ns = val; | |
7fa08e71 | 3678 | out: |
2cbd7824 | 3679 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3680 | } |
3681 | ||
3682 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
3683 | { | |
ae232ea4 | 3684 | unsigned int old, val, shrink, grow_start; |
aca6ff29 | 3685 | |
2cbd7824 | 3686 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 3687 | shrink = READ_ONCE(halt_poll_ns_shrink); |
ae232ea4 | 3688 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3689 | if (shrink == 0) |
aca6ff29 WL |
3690 | val = 0; |
3691 | else | |
6b6de68c | 3692 | val /= shrink; |
aca6ff29 | 3693 | |
ae232ea4 SS |
3694 | if (val < grow_start) |
3695 | val = 0; | |
3696 | ||
aca6ff29 | 3697 | vcpu->halt_poll_ns = val; |
2cbd7824 | 3698 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3699 | } |
3700 | ||
f7819512 PB |
3701 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
3702 | { | |
50c28f21 JS |
3703 | int ret = -EINTR; |
3704 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
3705 | ||
c59fb127 | 3706 | if (kvm_arch_vcpu_runnable(vcpu)) |
50c28f21 | 3707 | goto out; |
f7819512 | 3708 | if (kvm_cpu_has_pending_timer(vcpu)) |
50c28f21 | 3709 | goto out; |
f7819512 | 3710 | if (signal_pending(current)) |
50c28f21 | 3711 | goto out; |
084071d5 MT |
3712 | if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu)) |
3713 | goto out; | |
f7819512 | 3714 | |
50c28f21 JS |
3715 | ret = 0; |
3716 | out: | |
3717 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
3718 | return ret; | |
f7819512 PB |
3719 | } |
3720 | ||
fac42688 SC |
3721 | /* |
3722 | * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is | |
3723 | * pending. This is mostly used when halting a vCPU, but may also be used | |
3724 | * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI. | |
3725 | */ | |
3726 | bool kvm_vcpu_block(struct kvm_vcpu *vcpu) | |
cb953129 | 3727 | { |
fac42688 SC |
3728 | struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); |
3729 | bool waited = false; | |
3730 | ||
c3858335 JZ |
3731 | vcpu->stat.generic.blocking = 1; |
3732 | ||
18869f26 | 3733 | preempt_disable(); |
fac42688 | 3734 | kvm_arch_vcpu_blocking(vcpu); |
fac42688 | 3735 | prepare_to_rcuwait(wait); |
18869f26 ML |
3736 | preempt_enable(); |
3737 | ||
fac42688 SC |
3738 | for (;;) { |
3739 | set_current_state(TASK_INTERRUPTIBLE); | |
3740 | ||
3741 | if (kvm_vcpu_check_block(vcpu) < 0) | |
3742 | break; | |
3743 | ||
3744 | waited = true; | |
3745 | schedule(); | |
3746 | } | |
fac42688 | 3747 | |
18869f26 ML |
3748 | preempt_disable(); |
3749 | finish_rcuwait(wait); | |
fac42688 | 3750 | kvm_arch_vcpu_unblocking(vcpu); |
18869f26 | 3751 | preempt_enable(); |
fac42688 | 3752 | |
c3858335 JZ |
3753 | vcpu->stat.generic.blocking = 0; |
3754 | ||
fac42688 SC |
3755 | return waited; |
3756 | } | |
3757 | ||
29e72893 SC |
3758 | static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start, |
3759 | ktime_t end, bool success) | |
cb953129 | 3760 | { |
30c94347 | 3761 | struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic; |
29e72893 SC |
3762 | u64 poll_ns = ktime_to_ns(ktime_sub(end, start)); |
3763 | ||
30c94347 SC |
3764 | ++vcpu->stat.generic.halt_attempted_poll; |
3765 | ||
3766 | if (success) { | |
3767 | ++vcpu->stat.generic.halt_successful_poll; | |
3768 | ||
3769 | if (!vcpu_valid_wakeup(vcpu)) | |
3770 | ++vcpu->stat.generic.halt_poll_invalid; | |
3771 | ||
3772 | stats->halt_poll_success_ns += poll_ns; | |
3773 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns); | |
3774 | } else { | |
3775 | stats->halt_poll_fail_ns += poll_ns; | |
3776 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns); | |
3777 | } | |
cb953129 DM |
3778 | } |
3779 | ||
175d5dc7 DM |
3780 | static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu *vcpu) |
3781 | { | |
9eb8ca04 DM |
3782 | struct kvm *kvm = vcpu->kvm; |
3783 | ||
3784 | if (kvm->override_halt_poll_ns) { | |
3785 | /* | |
3786 | * Ensure kvm->max_halt_poll_ns is not read before | |
3787 | * kvm->override_halt_poll_ns. | |
3788 | * | |
3789 | * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL. | |
3790 | */ | |
3791 | smp_rmb(); | |
3792 | return READ_ONCE(kvm->max_halt_poll_ns); | |
3793 | } | |
3794 | ||
3795 | return READ_ONCE(halt_poll_ns); | |
175d5dc7 DM |
3796 | } |
3797 | ||
b6958ce4 | 3798 | /* |
fac42688 SC |
3799 | * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt |
3800 | * polling is enabled, busy wait for a short time before blocking to avoid the | |
3801 | * expensive block+unblock sequence if a wake event arrives soon after the vCPU | |
3802 | * is halted. | |
b6958ce4 | 3803 | */ |
91b99ea7 | 3804 | void kvm_vcpu_halt(struct kvm_vcpu *vcpu) |
d3bef15f | 3805 | { |
175d5dc7 | 3806 | unsigned int max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu); |
6f390916 | 3807 | bool halt_poll_allowed = !kvm_arch_no_poll(vcpu); |
cb953129 | 3808 | ktime_t start, cur, poll_end; |
f7819512 | 3809 | bool waited = false; |
97b6847a | 3810 | bool do_halt_poll; |
91b99ea7 | 3811 | u64 halt_ns; |
07ab0f8d | 3812 | |
175d5dc7 DM |
3813 | if (vcpu->halt_poll_ns > max_halt_poll_ns) |
3814 | vcpu->halt_poll_ns = max_halt_poll_ns; | |
97b6847a DM |
3815 | |
3816 | do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns; | |
3817 | ||
cb953129 | 3818 | start = cur = poll_end = ktime_get(); |
8df6a61c | 3819 | if (do_halt_poll) { |
109a9826 | 3820 | ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns); |
f95ef0cd | 3821 | |
f7819512 | 3822 | do { |
30c94347 | 3823 | if (kvm_vcpu_check_block(vcpu) < 0) |
f7819512 | 3824 | goto out; |
74775654 | 3825 | cpu_relax(); |
cb953129 | 3826 | poll_end = cur = ktime_get(); |
6bd5b743 | 3827 | } while (kvm_vcpu_can_poll(cur, stop)); |
f7819512 | 3828 | } |
e5c239cf | 3829 | |
fac42688 | 3830 | waited = kvm_vcpu_block(vcpu); |
8ccba534 | 3831 | |
f7819512 | 3832 | cur = ktime_get(); |
87bcc5fa JZ |
3833 | if (waited) { |
3834 | vcpu->stat.generic.halt_wait_ns += | |
3835 | ktime_to_ns(cur) - ktime_to_ns(poll_end); | |
8ccba534 JZ |
3836 | KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist, |
3837 | ktime_to_ns(cur) - ktime_to_ns(poll_end)); | |
87bcc5fa | 3838 | } |
f7819512 | 3839 | out: |
91b99ea7 SC |
3840 | /* The total time the vCPU was "halted", including polling time. */ |
3841 | halt_ns = ktime_to_ns(cur) - ktime_to_ns(start); | |
aca6ff29 | 3842 | |
29e72893 SC |
3843 | /* |
3844 | * Note, halt-polling is considered successful so long as the vCPU was | |
3845 | * never actually scheduled out, i.e. even if the wake event arrived | |
3846 | * after of the halt-polling loop itself, but before the full wait. | |
3847 | */ | |
8df6a61c | 3848 | if (do_halt_poll) |
29e72893 | 3849 | update_halt_poll_stats(vcpu, start, poll_end, !waited); |
cb953129 | 3850 | |
6f390916 | 3851 | if (halt_poll_allowed) { |
175d5dc7 DM |
3852 | /* Recompute the max halt poll time in case it changed. */ |
3853 | max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu); | |
3854 | ||
44551b2f | 3855 | if (!vcpu_valid_wakeup(vcpu)) { |
aca6ff29 | 3856 | shrink_halt_poll_ns(vcpu); |
175d5dc7 | 3857 | } else if (max_halt_poll_ns) { |
91b99ea7 | 3858 | if (halt_ns <= vcpu->halt_poll_ns) |
44551b2f WL |
3859 | ; |
3860 | /* we had a long block, shrink polling */ | |
acd05785 | 3861 | else if (vcpu->halt_poll_ns && |
175d5dc7 | 3862 | halt_ns > max_halt_poll_ns) |
44551b2f WL |
3863 | shrink_halt_poll_ns(vcpu); |
3864 | /* we had a short halt and our poll time is too small */ | |
175d5dc7 DM |
3865 | else if (vcpu->halt_poll_ns < max_halt_poll_ns && |
3866 | halt_ns < max_halt_poll_ns) | |
44551b2f WL |
3867 | grow_halt_poll_ns(vcpu); |
3868 | } else { | |
3869 | vcpu->halt_poll_ns = 0; | |
3870 | } | |
3871 | } | |
aca6ff29 | 3872 | |
91b99ea7 | 3873 | trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu)); |
b6958ce4 | 3874 | } |
91b99ea7 | 3875 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
b6958ce4 | 3876 | |
178f02ff | 3877 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 3878 | { |
d92a5d1c | 3879 | if (__kvm_vcpu_wake_up(vcpu)) { |
d73eb57b | 3880 | WRITE_ONCE(vcpu->ready, true); |
0193cc90 | 3881 | ++vcpu->stat.generic.halt_wakeup; |
178f02ff | 3882 | return true; |
b6d33834 CD |
3883 | } |
3884 | ||
178f02ff | 3885 | return false; |
dd1a4cc1 RK |
3886 | } |
3887 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
3888 | ||
0266c894 | 3889 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
3890 | /* |
3891 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
3892 | */ | |
3893 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
3894 | { | |
85b64045 | 3895 | int me, cpu; |
dd1a4cc1 | 3896 | |
178f02ff RK |
3897 | if (kvm_vcpu_wake_up(vcpu)) |
3898 | return; | |
3899 | ||
aefdc2ed PB |
3900 | me = get_cpu(); |
3901 | /* | |
3902 | * The only state change done outside the vcpu mutex is IN_GUEST_MODE | |
3903 | * to EXITING_GUEST_MODE. Therefore the moderately expensive "should | |
3904 | * kick" check does not need atomic operations if kvm_vcpu_kick is used | |
3905 | * within the vCPU thread itself. | |
3906 | */ | |
3907 | if (vcpu == __this_cpu_read(kvm_running_vcpu)) { | |
3908 | if (vcpu->mode == IN_GUEST_MODE) | |
3909 | WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE); | |
3910 | goto out; | |
3911 | } | |
3912 | ||
85b64045 SC |
3913 | /* |
3914 | * Note, the vCPU could get migrated to a different pCPU at any point | |
3915 | * after kvm_arch_vcpu_should_kick(), which could result in sending an | |
3916 | * IPI to the previous pCPU. But, that's ok because the purpose of the | |
3917 | * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the | |
3918 | * vCPU also requires it to leave IN_GUEST_MODE. | |
3919 | */ | |
85b64045 SC |
3920 | if (kvm_arch_vcpu_should_kick(vcpu)) { |
3921 | cpu = READ_ONCE(vcpu->cpu); | |
3922 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
b6d33834 | 3923 | smp_send_reschedule(cpu); |
85b64045 | 3924 | } |
aefdc2ed | 3925 | out: |
b6d33834 CD |
3926 | put_cpu(); |
3927 | } | |
a20ed54d | 3928 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 3929 | #endif /* !CONFIG_S390 */ |
b6d33834 | 3930 | |
fa93384f | 3931 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
3932 | { |
3933 | struct pid *pid; | |
3934 | struct task_struct *task = NULL; | |
fa93384f | 3935 | int ret = 0; |
41628d33 KW |
3936 | |
3937 | rcu_read_lock(); | |
3938 | pid = rcu_dereference(target->pid); | |
3939 | if (pid) | |
27fbe64b | 3940 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
3941 | rcu_read_unlock(); |
3942 | if (!task) | |
c45c528e | 3943 | return ret; |
c45c528e | 3944 | ret = yield_to(task, 1); |
41628d33 | 3945 | put_task_struct(task); |
c45c528e R |
3946 | |
3947 | return ret; | |
41628d33 KW |
3948 | } |
3949 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
3950 | ||
06e48c51 R |
3951 | /* |
3952 | * Helper that checks whether a VCPU is eligible for directed yield. | |
3953 | * Most eligible candidate to yield is decided by following heuristics: | |
3954 | * | |
3955 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
3956 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 3957 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
3958 | * |
3959 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
3960 | * chance last time (mostly it has become eligible now since we have probably | |
3961 | * yielded to lockholder in last iteration. This is done by toggling | |
3962 | * @dy_eligible each time a VCPU checked for eligibility.) | |
3963 | * | |
3964 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
3965 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
3966 | * burning. Giving priority for a potential lock-holder increases lock | |
3967 | * progress. | |
3968 | * | |
3969 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
3970 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
3971 | * and continue with next VCPU and so on. | |
3972 | */ | |
7940876e | 3973 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 3974 | { |
4a55dd72 | 3975 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
3976 | bool eligible; |
3977 | ||
3978 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 3979 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
3980 | |
3981 | if (vcpu->spin_loop.in_spin_loop) | |
3982 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
3983 | ||
3984 | return eligible; | |
4a55dd72 SW |
3985 | #else |
3986 | return true; | |
06e48c51 | 3987 | #endif |
4a55dd72 | 3988 | } |
c45c528e | 3989 | |
17e433b5 WL |
3990 | /* |
3991 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
3992 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
3993 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
3994 | */ | |
3995 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
3996 | { | |
3997 | return kvm_arch_vcpu_runnable(vcpu); | |
3998 | } | |
3999 | ||
4000 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
4001 | { | |
4002 | if (kvm_arch_dy_runnable(vcpu)) | |
4003 | return true; | |
4004 | ||
4005 | #ifdef CONFIG_KVM_ASYNC_PF | |
4006 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
4007 | return true; | |
4008 | #endif | |
4009 | ||
4010 | return false; | |
4011 | } | |
4012 | ||
77bcd9e6 SC |
4013 | /* |
4014 | * By default, simply query the target vCPU's current mode when checking if a | |
4015 | * vCPU was preempted in kernel mode. All architectures except x86 (or more | |
4016 | * specifical, except VMX) allow querying whether or not a vCPU is in kernel | |
4017 | * mode even if the vCPU is NOT loaded, i.e. using kvm_arch_vcpu_in_kernel() | |
4018 | * directly for cross-vCPU checks is functionally correct and accurate. | |
4019 | */ | |
4020 | bool __weak kvm_arch_vcpu_preempted_in_kernel(struct kvm_vcpu *vcpu) | |
4021 | { | |
4022 | return kvm_arch_vcpu_in_kernel(vcpu); | |
4023 | } | |
4024 | ||
52acd22f WL |
4025 | bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
4026 | { | |
4027 | return false; | |
4028 | } | |
4029 | ||
199b5763 | 4030 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 4031 | { |
217ece61 RR |
4032 | struct kvm *kvm = me->kvm; |
4033 | struct kvm_vcpu *vcpu; | |
4034 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
46808a4c | 4035 | unsigned long i; |
217ece61 | 4036 | int yielded = 0; |
c45c528e | 4037 | int try = 3; |
217ece61 | 4038 | int pass; |
d255f4f2 | 4039 | |
4c088493 | 4040 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
4041 | /* |
4042 | * We boost the priority of a VCPU that is runnable but not | |
4043 | * currently running, because it got preempted by something | |
4044 | * else and called schedule in __vcpu_run. Hopefully that | |
4045 | * VCPU is holding the lock that we need and will release it. | |
4046 | * We approximate round-robin by starting at the last boosted VCPU. | |
4047 | */ | |
c45c528e | 4048 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 4049 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 4050 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
4051 | i = last_boosted_vcpu; |
4052 | continue; | |
4053 | } else if (pass && i > last_boosted_vcpu) | |
4054 | break; | |
d73eb57b | 4055 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 4056 | continue; |
217ece61 RR |
4057 | if (vcpu == me) |
4058 | continue; | |
d92a5d1c | 4059 | if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu)) |
217ece61 | 4060 | continue; |
dafc17dd SC |
4061 | |
4062 | /* | |
4063 | * Treat the target vCPU as being in-kernel if it has a | |
4064 | * pending interrupt, as the vCPU trying to yield may | |
4065 | * be spinning waiting on IPI delivery, i.e. the target | |
4066 | * vCPU is in-kernel for the purposes of directed yield. | |
4067 | */ | |
046ddeed | 4068 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
52acd22f | 4069 | !kvm_arch_dy_has_pending_interrupt(vcpu) && |
77bcd9e6 | 4070 | !kvm_arch_vcpu_preempted_in_kernel(vcpu)) |
199b5763 | 4071 | continue; |
06e48c51 R |
4072 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
4073 | continue; | |
c45c528e R |
4074 | |
4075 | yielded = kvm_vcpu_yield_to(vcpu); | |
4076 | if (yielded > 0) { | |
217ece61 | 4077 | kvm->last_boosted_vcpu = i; |
217ece61 | 4078 | break; |
c45c528e R |
4079 | } else if (yielded < 0) { |
4080 | try--; | |
4081 | if (!try) | |
4082 | break; | |
217ece61 | 4083 | } |
217ece61 RR |
4084 | } |
4085 | } | |
4c088493 | 4086 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
4087 | |
4088 | /* Ensure vcpu is not eligible during next spinloop */ | |
4089 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
4090 | } |
4091 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
4092 | ||
fb04a1ed PX |
4093 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
4094 | { | |
dc70ec21 | 4095 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
fb04a1ed PX |
4096 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && |
4097 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
4098 | kvm->dirty_ring_size / PAGE_SIZE); | |
4099 | #else | |
4100 | return false; | |
4101 | #endif | |
4102 | } | |
4103 | ||
1499fa80 | 4104 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 4105 | { |
11bac800 | 4106 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
4107 | struct page *page; |
4108 | ||
e4a533a4 | 4109 | if (vmf->pgoff == 0) |
039576c0 | 4110 | page = virt_to_page(vcpu->run); |
09566765 | 4111 | #ifdef CONFIG_X86 |
e4a533a4 | 4112 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 4113 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 4114 | #endif |
4b4357e0 | 4115 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
4116 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
4117 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 4118 | #endif |
fb04a1ed PX |
4119 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
4120 | page = kvm_dirty_ring_get_page( | |
4121 | &vcpu->dirty_ring, | |
4122 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 4123 | else |
5b1c1493 | 4124 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 4125 | get_page(page); |
e4a533a4 | 4126 | vmf->page = page; |
4127 | return 0; | |
9a2bb7f4 AK |
4128 | } |
4129 | ||
f0f37e2f | 4130 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 4131 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
4132 | }; |
4133 | ||
4134 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
4135 | { | |
fb04a1ed | 4136 | struct kvm_vcpu *vcpu = file->private_data; |
11476d27 | 4137 | unsigned long pages = vma_pages(vma); |
fb04a1ed PX |
4138 | |
4139 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
4140 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
4141 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
4142 | return -EINVAL; | |
4143 | ||
9a2bb7f4 AK |
4144 | vma->vm_ops = &kvm_vcpu_vm_ops; |
4145 | return 0; | |
4146 | } | |
4147 | ||
bccf2150 AK |
4148 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
4149 | { | |
4150 | struct kvm_vcpu *vcpu = filp->private_data; | |
4151 | ||
66c0b394 | 4152 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
4153 | return 0; |
4154 | } | |
4155 | ||
087e1520 | 4156 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
4157 | .release = kvm_vcpu_release, |
4158 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 4159 | .mmap = kvm_vcpu_mmap, |
6038f373 | 4160 | .llseek = noop_llseek, |
7ddfd3e0 | 4161 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
4162 | }; |
4163 | ||
4164 | /* | |
4165 | * Allocates an inode for the vcpu. | |
4166 | */ | |
4167 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
4168 | { | |
e46b4692 MY |
4169 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
4170 | ||
4171 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
4172 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
4173 | } |
4174 | ||
e36de87d VP |
4175 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
4176 | static int vcpu_get_pid(void *data, u64 *val) | |
4177 | { | |
14aa40a1 | 4178 | struct kvm_vcpu *vcpu = data; |
76021e96 SC |
4179 | |
4180 | rcu_read_lock(); | |
4181 | *val = pid_nr(rcu_dereference(vcpu->pid)); | |
4182 | rcu_read_unlock(); | |
e36de87d VP |
4183 | return 0; |
4184 | } | |
4185 | ||
4186 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n"); | |
4187 | ||
3e7093d0 | 4188 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 4189 | { |
d56f5136 | 4190 | struct dentry *debugfs_dentry; |
45b5939e | 4191 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 4192 | |
45b5939e | 4193 | if (!debugfs_initialized()) |
3e7093d0 | 4194 | return; |
45b5939e LC |
4195 | |
4196 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
4197 | debugfs_dentry = debugfs_create_dir(dir_name, |
4198 | vcpu->kvm->debugfs_dentry); | |
e36de87d VP |
4199 | debugfs_create_file("pid", 0444, debugfs_dentry, vcpu, |
4200 | &vcpu_get_pid_fops); | |
45b5939e | 4201 | |
d56f5136 | 4202 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
45b5939e | 4203 | } |
e36de87d | 4204 | #endif |
45b5939e | 4205 | |
c5ea7660 AK |
4206 | /* |
4207 | * Creates some virtual cpus. Good luck creating more than one. | |
4208 | */ | |
73880c80 | 4209 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
4210 | { |
4211 | int r; | |
e09fefde | 4212 | struct kvm_vcpu *vcpu; |
8bd826d6 | 4213 | struct page *page; |
c5ea7660 | 4214 | |
a1c42dde | 4215 | if (id >= KVM_MAX_VCPU_IDS) |
338c7dba AH |
4216 | return -EINVAL; |
4217 | ||
6c7caebc | 4218 | mutex_lock(&kvm->lock); |
f502cc56 | 4219 | if (kvm->created_vcpus >= kvm->max_vcpus) { |
6c7caebc PB |
4220 | mutex_unlock(&kvm->lock); |
4221 | return -EINVAL; | |
4222 | } | |
4223 | ||
1d5e740d ZG |
4224 | r = kvm_arch_vcpu_precreate(kvm, id); |
4225 | if (r) { | |
4226 | mutex_unlock(&kvm->lock); | |
4227 | return r; | |
4228 | } | |
4229 | ||
6c7caebc PB |
4230 | kvm->created_vcpus++; |
4231 | mutex_unlock(&kvm->lock); | |
4232 | ||
85f47930 | 4233 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
e529ef66 SC |
4234 | if (!vcpu) { |
4235 | r = -ENOMEM; | |
6c7caebc PB |
4236 | goto vcpu_decrement; |
4237 | } | |
c5ea7660 | 4238 | |
fcd97ad5 | 4239 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 4240 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
4241 | if (!page) { |
4242 | r = -ENOMEM; | |
e529ef66 | 4243 | goto vcpu_free; |
8bd826d6 SC |
4244 | } |
4245 | vcpu->run = page_address(page); | |
4246 | ||
4247 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
4248 | |
4249 | r = kvm_arch_vcpu_create(vcpu); | |
4250 | if (r) | |
8bd826d6 | 4251 | goto vcpu_free_run_page; |
e529ef66 | 4252 | |
fb04a1ed PX |
4253 | if (kvm->dirty_ring_size) { |
4254 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
4255 | id, kvm->dirty_ring_size); | |
4256 | if (r) | |
4257 | goto arch_vcpu_destroy; | |
4258 | } | |
4259 | ||
11ec2804 | 4260 | mutex_lock(&kvm->lock); |
42a90008 DW |
4261 | |
4262 | #ifdef CONFIG_LOCKDEP | |
4263 | /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */ | |
4264 | mutex_lock(&vcpu->mutex); | |
4265 | mutex_unlock(&vcpu->mutex); | |
4266 | #endif | |
4267 | ||
e09fefde DH |
4268 | if (kvm_get_vcpu_by_id(kvm, id)) { |
4269 | r = -EEXIST; | |
4270 | goto unlock_vcpu_destroy; | |
4271 | } | |
73880c80 | 4272 | |
8750e72a | 4273 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
afb2acb2 | 4274 | r = xa_reserve(&kvm->vcpu_array, vcpu->vcpu_idx, GFP_KERNEL_ACCOUNT); |
c5b07754 MZ |
4275 | if (r) |
4276 | goto unlock_vcpu_destroy; | |
c5ea7660 | 4277 | |
fb3f0f51 | 4278 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 4279 | kvm_get_kvm(kvm); |
bccf2150 | 4280 | r = create_vcpu_fd(vcpu); |
afb2acb2 ML |
4281 | if (r < 0) |
4282 | goto kvm_put_xa_release; | |
4283 | ||
5f643e46 | 4284 | if (KVM_BUG_ON(xa_store(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, 0), kvm)) { |
afb2acb2 ML |
4285 | r = -EINVAL; |
4286 | goto kvm_put_xa_release; | |
73880c80 GN |
4287 | } |
4288 | ||
dd489240 | 4289 | /* |
c5b07754 MZ |
4290 | * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu |
4291 | * pointer before kvm->online_vcpu's incremented value. | |
dd489240 | 4292 | */ |
73880c80 GN |
4293 | smp_wmb(); |
4294 | atomic_inc(&kvm->online_vcpus); | |
4295 | ||
73880c80 | 4296 | mutex_unlock(&kvm->lock); |
42897d86 | 4297 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 4298 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 4299 | return r; |
39c3b86e | 4300 | |
afb2acb2 ML |
4301 | kvm_put_xa_release: |
4302 | kvm_put_kvm_no_destroy(kvm); | |
4303 | xa_release(&kvm->vcpu_array, vcpu->vcpu_idx); | |
d780592b | 4304 | unlock_vcpu_destroy: |
7d8fece6 | 4305 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
4306 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
4307 | arch_vcpu_destroy: | |
d40ccc62 | 4308 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
4309 | vcpu_free_run_page: |
4310 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
4311 | vcpu_free: |
4312 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
4313 | vcpu_decrement: |
4314 | mutex_lock(&kvm->lock); | |
4315 | kvm->created_vcpus--; | |
4316 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
4317 | return r; |
4318 | } | |
4319 | ||
1961d276 AK |
4320 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
4321 | { | |
4322 | if (sigset) { | |
4323 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
4324 | vcpu->sigset_active = 1; | |
4325 | vcpu->sigset = *sigset; | |
4326 | } else | |
4327 | vcpu->sigset_active = 0; | |
4328 | return 0; | |
4329 | } | |
4330 | ||
ce55c049 JZ |
4331 | static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer, |
4332 | size_t size, loff_t *offset) | |
4333 | { | |
4334 | struct kvm_vcpu *vcpu = file->private_data; | |
4335 | ||
4336 | return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header, | |
4337 | &kvm_vcpu_stats_desc[0], &vcpu->stat, | |
4338 | sizeof(vcpu->stat), user_buffer, size, offset); | |
4339 | } | |
4340 | ||
eed3013f SC |
4341 | static int kvm_vcpu_stats_release(struct inode *inode, struct file *file) |
4342 | { | |
4343 | struct kvm_vcpu *vcpu = file->private_data; | |
4344 | ||
4345 | kvm_put_kvm(vcpu->kvm); | |
4346 | return 0; | |
4347 | } | |
4348 | ||
ce55c049 | 4349 | static const struct file_operations kvm_vcpu_stats_fops = { |
087e1520 | 4350 | .owner = THIS_MODULE, |
ce55c049 | 4351 | .read = kvm_vcpu_stats_read, |
eed3013f | 4352 | .release = kvm_vcpu_stats_release, |
ce55c049 JZ |
4353 | .llseek = noop_llseek, |
4354 | }; | |
4355 | ||
4356 | static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu) | |
4357 | { | |
4358 | int fd; | |
4359 | struct file *file; | |
4360 | char name[15 + ITOA_MAX_LEN + 1]; | |
4361 | ||
4362 | snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id); | |
4363 | ||
4364 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4365 | if (fd < 0) | |
4366 | return fd; | |
4367 | ||
4368 | file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY); | |
4369 | if (IS_ERR(file)) { | |
4370 | put_unused_fd(fd); | |
4371 | return PTR_ERR(file); | |
4372 | } | |
eed3013f SC |
4373 | |
4374 | kvm_get_kvm(vcpu->kvm); | |
4375 | ||
ce55c049 JZ |
4376 | file->f_mode |= FMODE_PREAD; |
4377 | fd_install(fd, file); | |
4378 | ||
4379 | return fd; | |
4380 | } | |
4381 | ||
bccf2150 AK |
4382 | static long kvm_vcpu_ioctl(struct file *filp, |
4383 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 4384 | { |
bccf2150 | 4385 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 4386 | void __user *argp = (void __user *)arg; |
313a3dc7 | 4387 | int r; |
fa3795a7 DH |
4388 | struct kvm_fpu *fpu = NULL; |
4389 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 4390 | |
f4d31653 | 4391 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
6d4e4c4f | 4392 | return -EIO; |
2122ff5e | 4393 | |
2ea75be3 DM |
4394 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
4395 | return -EINVAL; | |
4396 | ||
2122ff5e | 4397 | /* |
5cb0944c PB |
4398 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
4399 | * execution; mutex_lock() would break them. | |
2122ff5e | 4400 | */ |
5cb0944c PB |
4401 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
4402 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 4403 | return r; |
2122ff5e | 4404 | |
ec7660cc CD |
4405 | if (mutex_lock_killable(&vcpu->mutex)) |
4406 | return -EINTR; | |
6aa8b732 | 4407 | switch (ioctl) { |
0e4524a5 CB |
4408 | case KVM_RUN: { |
4409 | struct pid *oldpid; | |
f0fe5108 AK |
4410 | r = -EINVAL; |
4411 | if (arg) | |
4412 | goto out; | |
0e4524a5 | 4413 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 4414 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 4415 | /* The thread running this VCPU changed. */ |
bd2a6394 | 4416 | struct pid *newpid; |
f95ef0cd | 4417 | |
bd2a6394 CD |
4418 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
4419 | if (r) | |
4420 | break; | |
4421 | ||
4422 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
4423 | rcu_assign_pointer(vcpu->pid, newpid); |
4424 | if (oldpid) | |
4425 | synchronize_rcu(); | |
4426 | put_pid(oldpid); | |
4427 | } | |
1b94f6f8 | 4428 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 4429 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 4430 | break; |
0e4524a5 | 4431 | } |
6aa8b732 | 4432 | case KVM_GET_REGS: { |
3e4bb3ac | 4433 | struct kvm_regs *kvm_regs; |
6aa8b732 | 4434 | |
3e4bb3ac | 4435 | r = -ENOMEM; |
b12ce36a | 4436 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 4437 | if (!kvm_regs) |
6aa8b732 | 4438 | goto out; |
3e4bb3ac XZ |
4439 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
4440 | if (r) | |
4441 | goto out_free1; | |
6aa8b732 | 4442 | r = -EFAULT; |
3e4bb3ac XZ |
4443 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
4444 | goto out_free1; | |
6aa8b732 | 4445 | r = 0; |
3e4bb3ac XZ |
4446 | out_free1: |
4447 | kfree(kvm_regs); | |
6aa8b732 AK |
4448 | break; |
4449 | } | |
4450 | case KVM_SET_REGS: { | |
3e4bb3ac | 4451 | struct kvm_regs *kvm_regs; |
6aa8b732 | 4452 | |
ff5c2c03 SL |
4453 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
4454 | if (IS_ERR(kvm_regs)) { | |
4455 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 4456 | goto out; |
ff5c2c03 | 4457 | } |
3e4bb3ac | 4458 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 4459 | kfree(kvm_regs); |
6aa8b732 AK |
4460 | break; |
4461 | } | |
4462 | case KVM_GET_SREGS: { | |
b12ce36a BG |
4463 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
4464 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
4465 | r = -ENOMEM; |
4466 | if (!kvm_sregs) | |
4467 | goto out; | |
4468 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
4469 | if (r) |
4470 | goto out; | |
4471 | r = -EFAULT; | |
fa3795a7 | 4472 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
4473 | goto out; |
4474 | r = 0; | |
4475 | break; | |
4476 | } | |
4477 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
4478 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
4479 | if (IS_ERR(kvm_sregs)) { | |
4480 | r = PTR_ERR(kvm_sregs); | |
18595411 | 4481 | kvm_sregs = NULL; |
6aa8b732 | 4482 | goto out; |
ff5c2c03 | 4483 | } |
fa3795a7 | 4484 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
4485 | break; |
4486 | } | |
62d9f0db MT |
4487 | case KVM_GET_MP_STATE: { |
4488 | struct kvm_mp_state mp_state; | |
4489 | ||
4490 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
4491 | if (r) | |
4492 | goto out; | |
4493 | r = -EFAULT; | |
893bdbf1 | 4494 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
4495 | goto out; |
4496 | r = 0; | |
4497 | break; | |
4498 | } | |
4499 | case KVM_SET_MP_STATE: { | |
4500 | struct kvm_mp_state mp_state; | |
4501 | ||
4502 | r = -EFAULT; | |
893bdbf1 | 4503 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
4504 | goto out; |
4505 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
4506 | break; |
4507 | } | |
6aa8b732 AK |
4508 | case KVM_TRANSLATE: { |
4509 | struct kvm_translation tr; | |
4510 | ||
4511 | r = -EFAULT; | |
893bdbf1 | 4512 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 4513 | goto out; |
8b006791 | 4514 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
4515 | if (r) |
4516 | goto out; | |
4517 | r = -EFAULT; | |
893bdbf1 | 4518 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
4519 | goto out; |
4520 | r = 0; | |
4521 | break; | |
4522 | } | |
d0bfb940 JK |
4523 | case KVM_SET_GUEST_DEBUG: { |
4524 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
4525 | |
4526 | r = -EFAULT; | |
893bdbf1 | 4527 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 4528 | goto out; |
d0bfb940 | 4529 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
4530 | break; |
4531 | } | |
1961d276 AK |
4532 | case KVM_SET_SIGNAL_MASK: { |
4533 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4534 | struct kvm_signal_mask kvm_sigmask; | |
4535 | sigset_t sigset, *p; | |
4536 | ||
4537 | p = NULL; | |
4538 | if (argp) { | |
4539 | r = -EFAULT; | |
4540 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4541 | sizeof(kvm_sigmask))) |
1961d276 AK |
4542 | goto out; |
4543 | r = -EINVAL; | |
893bdbf1 | 4544 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
4545 | goto out; |
4546 | r = -EFAULT; | |
4547 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 4548 | sizeof(sigset))) |
1961d276 AK |
4549 | goto out; |
4550 | p = &sigset; | |
4551 | } | |
376d41ff | 4552 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
4553 | break; |
4554 | } | |
b8836737 | 4555 | case KVM_GET_FPU: { |
b12ce36a | 4556 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
4557 | r = -ENOMEM; |
4558 | if (!fpu) | |
4559 | goto out; | |
4560 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
4561 | if (r) |
4562 | goto out; | |
4563 | r = -EFAULT; | |
fa3795a7 | 4564 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
4565 | goto out; |
4566 | r = 0; | |
4567 | break; | |
4568 | } | |
4569 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
4570 | fpu = memdup_user(argp, sizeof(*fpu)); |
4571 | if (IS_ERR(fpu)) { | |
4572 | r = PTR_ERR(fpu); | |
18595411 | 4573 | fpu = NULL; |
b8836737 | 4574 | goto out; |
ff5c2c03 | 4575 | } |
fa3795a7 | 4576 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
4577 | break; |
4578 | } | |
ce55c049 JZ |
4579 | case KVM_GET_STATS_FD: { |
4580 | r = kvm_vcpu_ioctl_get_stats_fd(vcpu); | |
4581 | break; | |
4582 | } | |
bccf2150 | 4583 | default: |
313a3dc7 | 4584 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
4585 | } |
4586 | out: | |
ec7660cc | 4587 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
4588 | kfree(fpu); |
4589 | kfree(kvm_sregs); | |
bccf2150 AK |
4590 | return r; |
4591 | } | |
4592 | ||
de8e5d74 | 4593 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
4594 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
4595 | unsigned int ioctl, unsigned long arg) | |
4596 | { | |
4597 | struct kvm_vcpu *vcpu = filp->private_data; | |
4598 | void __user *argp = compat_ptr(arg); | |
4599 | int r; | |
4600 | ||
f4d31653 | 4601 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
1dda606c AG |
4602 | return -EIO; |
4603 | ||
4604 | switch (ioctl) { | |
4605 | case KVM_SET_SIGNAL_MASK: { | |
4606 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4607 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
4608 | sigset_t sigset; |
4609 | ||
4610 | if (argp) { | |
4611 | r = -EFAULT; | |
4612 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4613 | sizeof(kvm_sigmask))) |
1dda606c AG |
4614 | goto out; |
4615 | r = -EINVAL; | |
3968cf62 | 4616 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
4617 | goto out; |
4618 | r = -EFAULT; | |
1393b4aa PB |
4619 | if (get_compat_sigset(&sigset, |
4620 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 4621 | goto out; |
760a9a30 AC |
4622 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
4623 | } else | |
4624 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
4625 | break; |
4626 | } | |
4627 | default: | |
4628 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
4629 | } | |
4630 | ||
4631 | out: | |
4632 | return r; | |
4633 | } | |
4634 | #endif | |
4635 | ||
a1cd3f08 CLG |
4636 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
4637 | { | |
4638 | struct kvm_device *dev = filp->private_data; | |
4639 | ||
4640 | if (dev->ops->mmap) | |
4641 | return dev->ops->mmap(dev, vma); | |
4642 | ||
4643 | return -ENODEV; | |
4644 | } | |
4645 | ||
852b6d57 SW |
4646 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
4647 | int (*accessor)(struct kvm_device *dev, | |
4648 | struct kvm_device_attr *attr), | |
4649 | unsigned long arg) | |
4650 | { | |
4651 | struct kvm_device_attr attr; | |
4652 | ||
4653 | if (!accessor) | |
4654 | return -EPERM; | |
4655 | ||
4656 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
4657 | return -EFAULT; | |
4658 | ||
4659 | return accessor(dev, &attr); | |
4660 | } | |
4661 | ||
4662 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
4663 | unsigned long arg) | |
4664 | { | |
4665 | struct kvm_device *dev = filp->private_data; | |
4666 | ||
f4d31653 | 4667 | if (dev->kvm->mm != current->mm || dev->kvm->vm_dead) |
ddba9180 SC |
4668 | return -EIO; |
4669 | ||
852b6d57 SW |
4670 | switch (ioctl) { |
4671 | case KVM_SET_DEVICE_ATTR: | |
4672 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
4673 | case KVM_GET_DEVICE_ATTR: | |
4674 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
4675 | case KVM_HAS_DEVICE_ATTR: | |
4676 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
4677 | default: | |
4678 | if (dev->ops->ioctl) | |
4679 | return dev->ops->ioctl(dev, ioctl, arg); | |
4680 | ||
4681 | return -ENOTTY; | |
4682 | } | |
4683 | } | |
4684 | ||
852b6d57 SW |
4685 | static int kvm_device_release(struct inode *inode, struct file *filp) |
4686 | { | |
4687 | struct kvm_device *dev = filp->private_data; | |
4688 | struct kvm *kvm = dev->kvm; | |
4689 | ||
2bde9b3e CLG |
4690 | if (dev->ops->release) { |
4691 | mutex_lock(&kvm->lock); | |
4692 | list_del(&dev->vm_node); | |
4693 | dev->ops->release(dev); | |
4694 | mutex_unlock(&kvm->lock); | |
4695 | } | |
4696 | ||
852b6d57 SW |
4697 | kvm_put_kvm(kvm); |
4698 | return 0; | |
4699 | } | |
4700 | ||
087e1520 | 4701 | static struct file_operations kvm_device_fops = { |
852b6d57 SW |
4702 | .unlocked_ioctl = kvm_device_ioctl, |
4703 | .release = kvm_device_release, | |
7ddfd3e0 | 4704 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 4705 | .mmap = kvm_device_mmap, |
852b6d57 SW |
4706 | }; |
4707 | ||
4708 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
4709 | { | |
4710 | if (filp->f_op != &kvm_device_fops) | |
4711 | return NULL; | |
4712 | ||
4713 | return filp->private_data; | |
4714 | } | |
4715 | ||
8538cb22 | 4716 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 4717 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
4718 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
4719 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 4720 | #endif |
d60eacb0 WD |
4721 | }; |
4722 | ||
8538cb22 | 4723 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
4724 | { |
4725 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
4726 | return -ENOSPC; | |
4727 | ||
4728 | if (kvm_device_ops_table[type] != NULL) | |
4729 | return -EEXIST; | |
4730 | ||
4731 | kvm_device_ops_table[type] = ops; | |
4732 | return 0; | |
4733 | } | |
4734 | ||
571ee1b6 WL |
4735 | void kvm_unregister_device_ops(u32 type) |
4736 | { | |
4737 | if (kvm_device_ops_table[type] != NULL) | |
4738 | kvm_device_ops_table[type] = NULL; | |
4739 | } | |
4740 | ||
852b6d57 SW |
4741 | static int kvm_ioctl_create_device(struct kvm *kvm, |
4742 | struct kvm_create_device *cd) | |
4743 | { | |
eceb6e1d | 4744 | const struct kvm_device_ops *ops; |
852b6d57 SW |
4745 | struct kvm_device *dev; |
4746 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 4747 | int type; |
852b6d57 SW |
4748 | int ret; |
4749 | ||
d60eacb0 WD |
4750 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
4751 | return -ENODEV; | |
4752 | ||
1d487e9b PB |
4753 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
4754 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 4755 | if (ops == NULL) |
852b6d57 | 4756 | return -ENODEV; |
852b6d57 SW |
4757 | |
4758 | if (test) | |
4759 | return 0; | |
4760 | ||
b12ce36a | 4761 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
4762 | if (!dev) |
4763 | return -ENOMEM; | |
4764 | ||
4765 | dev->ops = ops; | |
4766 | dev->kvm = kvm; | |
852b6d57 | 4767 | |
a28ebea2 | 4768 | mutex_lock(&kvm->lock); |
1d487e9b | 4769 | ret = ops->create(dev, type); |
852b6d57 | 4770 | if (ret < 0) { |
a28ebea2 | 4771 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
4772 | kfree(dev); |
4773 | return ret; | |
4774 | } | |
a28ebea2 CD |
4775 | list_add(&dev->vm_node, &kvm->devices); |
4776 | mutex_unlock(&kvm->lock); | |
852b6d57 | 4777 | |
023e9fdd CD |
4778 | if (ops->init) |
4779 | ops->init(dev); | |
4780 | ||
cfa39381 | 4781 | kvm_get_kvm(kvm); |
24009b05 | 4782 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 4783 | if (ret < 0) { |
149487bd | 4784 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
4785 | mutex_lock(&kvm->lock); |
4786 | list_del(&dev->vm_node); | |
e8bc2427 AK |
4787 | if (ops->release) |
4788 | ops->release(dev); | |
a28ebea2 | 4789 | mutex_unlock(&kvm->lock); |
e8bc2427 AK |
4790 | if (ops->destroy) |
4791 | ops->destroy(dev); | |
852b6d57 SW |
4792 | return ret; |
4793 | } | |
4794 | ||
852b6d57 SW |
4795 | cd->fd = ret; |
4796 | return 0; | |
4797 | } | |
4798 | ||
f15ba52b | 4799 | static int kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
92b591a4 AG |
4800 | { |
4801 | switch (arg) { | |
4802 | case KVM_CAP_USER_MEMORY: | |
bb58b90b | 4803 | case KVM_CAP_USER_MEMORY2: |
92b591a4 AG |
4804 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: |
4805 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
4806 | case KVM_CAP_INTERNAL_ERROR_DATA: |
4807 | #ifdef CONFIG_HAVE_KVM_MSI | |
4808 | case KVM_CAP_SIGNAL_MSI: | |
4809 | #endif | |
c5b31cc2 | 4810 | #ifdef CONFIG_HAVE_KVM_IRQCHIP |
dc9be0fa | 4811 | case KVM_CAP_IRQFD: |
92b591a4 | 4812 | #endif |
e9ea5069 | 4813 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 4814 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 4815 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 4816 | case KVM_CAP_HALT_POLL: |
92b591a4 | 4817 | return 1; |
4b4357e0 | 4818 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
4819 | case KVM_CAP_COALESCED_MMIO: |
4820 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
4821 | case KVM_CAP_COALESCED_PIO: |
4822 | return 1; | |
30422558 | 4823 | #endif |
3c9bd400 JZ |
4824 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4825 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
4826 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
4827 | #endif | |
92b591a4 AG |
4828 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4829 | case KVM_CAP_IRQ_ROUTING: | |
4830 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 | 4831 | #endif |
eed52e43 | 4832 | #if KVM_MAX_NR_ADDRESS_SPACES > 1 |
f481b069 | 4833 | case KVM_CAP_MULTI_ADDRESS_SPACE: |
eed52e43 SC |
4834 | if (kvm) |
4835 | return kvm_arch_nr_memslot_as_ids(kvm); | |
4836 | return KVM_MAX_NR_ADDRESS_SPACES; | |
92b591a4 | 4837 | #endif |
c110ae57 PB |
4838 | case KVM_CAP_NR_MEMSLOTS: |
4839 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed | 4840 | case KVM_CAP_DIRTY_LOG_RING: |
17601bfe MZ |
4841 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO |
4842 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); | |
4843 | #else | |
4844 | return 0; | |
4845 | #endif | |
4846 | case KVM_CAP_DIRTY_LOG_RING_ACQ_REL: | |
4847 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL | |
fb04a1ed PX |
4848 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); |
4849 | #else | |
4850 | return 0; | |
86bdf3eb GS |
4851 | #endif |
4852 | #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP | |
4853 | case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: | |
fb04a1ed | 4854 | #endif |
ce55c049 | 4855 | case KVM_CAP_BINARY_STATS_FD: |
d495f942 | 4856 | case KVM_CAP_SYSTEM_EVENT_DATA: |
63912245 | 4857 | case KVM_CAP_DEVICE_CTRL: |
ce55c049 | 4858 | return 1; |
5a475554 CP |
4859 | #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES |
4860 | case KVM_CAP_MEMORY_ATTRIBUTES: | |
4861 | return kvm_supported_mem_attributes(kvm); | |
a7800aa8 SC |
4862 | #endif |
4863 | #ifdef CONFIG_KVM_PRIVATE_MEM | |
4864 | case KVM_CAP_GUEST_MEMFD: | |
4865 | return !kvm || kvm_arch_has_private_mem(kvm); | |
5a475554 | 4866 | #endif |
92b591a4 AG |
4867 | default: |
4868 | break; | |
4869 | } | |
4870 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
4871 | } | |
4872 | ||
fb04a1ed PX |
4873 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
4874 | { | |
4875 | int r; | |
4876 | ||
4877 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
4878 | return -EINVAL; | |
4879 | ||
4880 | /* the size should be power of 2 */ | |
4881 | if (!size || (size & (size - 1))) | |
4882 | return -EINVAL; | |
4883 | ||
4884 | /* Should be bigger to keep the reserved entries, or a page */ | |
4885 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
4886 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
4887 | return -EINVAL; | |
4888 | ||
4889 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
4890 | sizeof(struct kvm_dirty_gfn)) | |
4891 | return -E2BIG; | |
4892 | ||
4893 | /* We only allow it to set once */ | |
4894 | if (kvm->dirty_ring_size) | |
4895 | return -EINVAL; | |
4896 | ||
4897 | mutex_lock(&kvm->lock); | |
4898 | ||
4899 | if (kvm->created_vcpus) { | |
4900 | /* We don't allow to change this value after vcpu created */ | |
4901 | r = -EINVAL; | |
4902 | } else { | |
4903 | kvm->dirty_ring_size = size; | |
4904 | r = 0; | |
4905 | } | |
4906 | ||
4907 | mutex_unlock(&kvm->lock); | |
4908 | return r; | |
4909 | } | |
4910 | ||
4911 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
4912 | { | |
46808a4c | 4913 | unsigned long i; |
fb04a1ed PX |
4914 | struct kvm_vcpu *vcpu; |
4915 | int cleared = 0; | |
4916 | ||
4917 | if (!kvm->dirty_ring_size) | |
4918 | return -EINVAL; | |
4919 | ||
4920 | mutex_lock(&kvm->slots_lock); | |
4921 | ||
4922 | kvm_for_each_vcpu(i, vcpu, kvm) | |
4923 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
4924 | ||
4925 | mutex_unlock(&kvm->slots_lock); | |
4926 | ||
4927 | if (cleared) | |
4928 | kvm_flush_remote_tlbs(kvm); | |
4929 | ||
4930 | return cleared; | |
4931 | } | |
4932 | ||
e5d83c74 PB |
4933 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4934 | struct kvm_enable_cap *cap) | |
4935 | { | |
4936 | return -EINVAL; | |
4937 | } | |
4938 | ||
26f45714 | 4939 | bool kvm_are_all_memslots_empty(struct kvm *kvm) |
86bdf3eb GS |
4940 | { |
4941 | int i; | |
4942 | ||
4943 | lockdep_assert_held(&kvm->slots_lock); | |
4944 | ||
eed52e43 | 4945 | for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { |
86bdf3eb GS |
4946 | if (!kvm_memslots_empty(__kvm_memslots(kvm, i))) |
4947 | return false; | |
4948 | } | |
4949 | ||
4950 | return true; | |
4951 | } | |
26f45714 | 4952 | EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty); |
86bdf3eb | 4953 | |
e5d83c74 PB |
4954 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, |
4955 | struct kvm_enable_cap *cap) | |
4956 | { | |
4957 | switch (cap->cap) { | |
2a31b9db | 4958 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
4959 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
4960 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
4961 | ||
4962 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
4963 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
4964 | ||
4965 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
4966 | return -EINVAL; |
4967 | kvm->manual_dirty_log_protect = cap->args[0]; | |
4968 | return 0; | |
3c9bd400 | 4969 | } |
2a31b9db | 4970 | #endif |
acd05785 DM |
4971 | case KVM_CAP_HALT_POLL: { |
4972 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
4973 | return -EINVAL; | |
4974 | ||
4975 | kvm->max_halt_poll_ns = cap->args[0]; | |
9eb8ca04 DM |
4976 | |
4977 | /* | |
4978 | * Ensure kvm->override_halt_poll_ns does not become visible | |
4979 | * before kvm->max_halt_poll_ns. | |
4980 | * | |
4981 | * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns(). | |
4982 | */ | |
4983 | smp_wmb(); | |
4984 | kvm->override_halt_poll_ns = true; | |
4985 | ||
acd05785 DM |
4986 | return 0; |
4987 | } | |
fb04a1ed | 4988 | case KVM_CAP_DIRTY_LOG_RING: |
17601bfe | 4989 | case KVM_CAP_DIRTY_LOG_RING_ACQ_REL: |
7a2726ec GS |
4990 | if (!kvm_vm_ioctl_check_extension_generic(kvm, cap->cap)) |
4991 | return -EINVAL; | |
4992 | ||
fb04a1ed | 4993 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); |
86bdf3eb GS |
4994 | case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: { |
4995 | int r = -EINVAL; | |
4996 | ||
4997 | if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP) || | |
4998 | !kvm->dirty_ring_size || cap->flags) | |
4999 | return r; | |
5000 | ||
5001 | mutex_lock(&kvm->slots_lock); | |
5002 | ||
5003 | /* | |
5004 | * For simplicity, allow enabling ring+bitmap if and only if | |
5005 | * there are no memslots, e.g. to ensure all memslots allocate | |
5006 | * a bitmap after the capability is enabled. | |
5007 | */ | |
5008 | if (kvm_are_all_memslots_empty(kvm)) { | |
5009 | kvm->dirty_ring_with_bitmap = true; | |
5010 | r = 0; | |
5011 | } | |
5012 | ||
5013 | mutex_unlock(&kvm->slots_lock); | |
5014 | ||
5015 | return r; | |
5016 | } | |
e5d83c74 PB |
5017 | default: |
5018 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
5019 | } | |
5020 | } | |
5021 | ||
fcfe1bae JZ |
5022 | static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer, |
5023 | size_t size, loff_t *offset) | |
5024 | { | |
5025 | struct kvm *kvm = file->private_data; | |
5026 | ||
5027 | return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header, | |
5028 | &kvm_vm_stats_desc[0], &kvm->stat, | |
5029 | sizeof(kvm->stat), user_buffer, size, offset); | |
5030 | } | |
5031 | ||
eed3013f SC |
5032 | static int kvm_vm_stats_release(struct inode *inode, struct file *file) |
5033 | { | |
5034 | struct kvm *kvm = file->private_data; | |
5035 | ||
5036 | kvm_put_kvm(kvm); | |
5037 | return 0; | |
5038 | } | |
5039 | ||
fcfe1bae | 5040 | static const struct file_operations kvm_vm_stats_fops = { |
087e1520 | 5041 | .owner = THIS_MODULE, |
fcfe1bae | 5042 | .read = kvm_vm_stats_read, |
eed3013f | 5043 | .release = kvm_vm_stats_release, |
fcfe1bae JZ |
5044 | .llseek = noop_llseek, |
5045 | }; | |
5046 | ||
5047 | static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm) | |
5048 | { | |
5049 | int fd; | |
5050 | struct file *file; | |
5051 | ||
5052 | fd = get_unused_fd_flags(O_CLOEXEC); | |
5053 | if (fd < 0) | |
5054 | return fd; | |
5055 | ||
5056 | file = anon_inode_getfile("kvm-vm-stats", | |
5057 | &kvm_vm_stats_fops, kvm, O_RDONLY); | |
5058 | if (IS_ERR(file)) { | |
5059 | put_unused_fd(fd); | |
5060 | return PTR_ERR(file); | |
5061 | } | |
eed3013f SC |
5062 | |
5063 | kvm_get_kvm(kvm); | |
5064 | ||
fcfe1bae JZ |
5065 | file->f_mode |= FMODE_PREAD; |
5066 | fd_install(fd, file); | |
5067 | ||
5068 | return fd; | |
5069 | } | |
5070 | ||
bb58b90b SC |
5071 | #define SANITY_CHECK_MEM_REGION_FIELD(field) \ |
5072 | do { \ | |
5073 | BUILD_BUG_ON(offsetof(struct kvm_userspace_memory_region, field) != \ | |
5074 | offsetof(struct kvm_userspace_memory_region2, field)); \ | |
5075 | BUILD_BUG_ON(sizeof_field(struct kvm_userspace_memory_region, field) != \ | |
5076 | sizeof_field(struct kvm_userspace_memory_region2, field)); \ | |
5077 | } while (0) | |
5078 | ||
bccf2150 AK |
5079 | static long kvm_vm_ioctl(struct file *filp, |
5080 | unsigned int ioctl, unsigned long arg) | |
5081 | { | |
5082 | struct kvm *kvm = filp->private_data; | |
5083 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 5084 | int r; |
bccf2150 | 5085 | |
f4d31653 | 5086 | if (kvm->mm != current->mm || kvm->vm_dead) |
6d4e4c4f | 5087 | return -EIO; |
bccf2150 AK |
5088 | switch (ioctl) { |
5089 | case KVM_CREATE_VCPU: | |
5090 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 5091 | break; |
e5d83c74 PB |
5092 | case KVM_ENABLE_CAP: { |
5093 | struct kvm_enable_cap cap; | |
5094 | ||
5095 | r = -EFAULT; | |
5096 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
5097 | goto out; | |
5098 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
5099 | break; | |
5100 | } | |
bb58b90b | 5101 | case KVM_SET_USER_MEMORY_REGION2: |
6fc138d2 | 5102 | case KVM_SET_USER_MEMORY_REGION: { |
bb58b90b SC |
5103 | struct kvm_userspace_memory_region2 mem; |
5104 | unsigned long size; | |
5105 | ||
5106 | if (ioctl == KVM_SET_USER_MEMORY_REGION) { | |
5107 | /* | |
5108 | * Fields beyond struct kvm_userspace_memory_region shouldn't be | |
5109 | * accessed, but avoid leaking kernel memory in case of a bug. | |
5110 | */ | |
5111 | memset(&mem, 0, sizeof(mem)); | |
5112 | size = sizeof(struct kvm_userspace_memory_region); | |
5113 | } else { | |
5114 | size = sizeof(struct kvm_userspace_memory_region2); | |
5115 | } | |
5116 | ||
5117 | /* Ensure the common parts of the two structs are identical. */ | |
5118 | SANITY_CHECK_MEM_REGION_FIELD(slot); | |
5119 | SANITY_CHECK_MEM_REGION_FIELD(flags); | |
5120 | SANITY_CHECK_MEM_REGION_FIELD(guest_phys_addr); | |
5121 | SANITY_CHECK_MEM_REGION_FIELD(memory_size); | |
5122 | SANITY_CHECK_MEM_REGION_FIELD(userspace_addr); | |
6fc138d2 IE |
5123 | |
5124 | r = -EFAULT; | |
bb58b90b SC |
5125 | if (copy_from_user(&mem, argp, size)) |
5126 | goto out; | |
5127 | ||
5128 | r = -EINVAL; | |
5129 | if (ioctl == KVM_SET_USER_MEMORY_REGION && | |
5130 | (mem.flags & ~KVM_SET_USER_MEMORY_REGION_V1_FLAGS)) | |
6fc138d2 IE |
5131 | goto out; |
5132 | ||
bb58b90b | 5133 | r = kvm_vm_ioctl_set_memory_region(kvm, &mem); |
6aa8b732 AK |
5134 | break; |
5135 | } | |
5136 | case KVM_GET_DIRTY_LOG: { | |
5137 | struct kvm_dirty_log log; | |
5138 | ||
5139 | r = -EFAULT; | |
893bdbf1 | 5140 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 5141 | goto out; |
2c6f5df9 | 5142 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
5143 | break; |
5144 | } | |
2a31b9db PB |
5145 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
5146 | case KVM_CLEAR_DIRTY_LOG: { | |
5147 | struct kvm_clear_dirty_log log; | |
5148 | ||
5149 | r = -EFAULT; | |
5150 | if (copy_from_user(&log, argp, sizeof(log))) | |
5151 | goto out; | |
5152 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
5153 | break; | |
5154 | } | |
5155 | #endif | |
4b4357e0 | 5156 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
5157 | case KVM_REGISTER_COALESCED_MMIO: { |
5158 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 5159 | |
5f94c174 | 5160 | r = -EFAULT; |
893bdbf1 | 5161 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 5162 | goto out; |
5f94c174 | 5163 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
5164 | break; |
5165 | } | |
5166 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
5167 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 5168 | |
5f94c174 | 5169 | r = -EFAULT; |
893bdbf1 | 5170 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 5171 | goto out; |
5f94c174 | 5172 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
5173 | break; |
5174 | } | |
5175 | #endif | |
721eecbf GH |
5176 | case KVM_IRQFD: { |
5177 | struct kvm_irqfd data; | |
5178 | ||
5179 | r = -EFAULT; | |
893bdbf1 | 5180 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 5181 | goto out; |
d4db2935 | 5182 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
5183 | break; |
5184 | } | |
d34e6b17 GH |
5185 | case KVM_IOEVENTFD: { |
5186 | struct kvm_ioeventfd data; | |
5187 | ||
5188 | r = -EFAULT; | |
893bdbf1 | 5189 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
5190 | goto out; |
5191 | r = kvm_ioeventfd(kvm, &data); | |
5192 | break; | |
5193 | } | |
07975ad3 JK |
5194 | #ifdef CONFIG_HAVE_KVM_MSI |
5195 | case KVM_SIGNAL_MSI: { | |
5196 | struct kvm_msi msi; | |
5197 | ||
5198 | r = -EFAULT; | |
893bdbf1 | 5199 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
5200 | goto out; |
5201 | r = kvm_send_userspace_msi(kvm, &msi); | |
5202 | break; | |
5203 | } | |
23d43cf9 CD |
5204 | #endif |
5205 | #ifdef __KVM_HAVE_IRQ_LINE | |
5206 | case KVM_IRQ_LINE_STATUS: | |
5207 | case KVM_IRQ_LINE: { | |
5208 | struct kvm_irq_level irq_event; | |
5209 | ||
5210 | r = -EFAULT; | |
893bdbf1 | 5211 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
5212 | goto out; |
5213 | ||
aa2fbe6d YZ |
5214 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
5215 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
5216 | if (r) |
5217 | goto out; | |
5218 | ||
5219 | r = -EFAULT; | |
5220 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 5221 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
5222 | goto out; |
5223 | } | |
5224 | ||
5225 | r = 0; | |
5226 | break; | |
5227 | } | |
73880c80 | 5228 | #endif |
aa8d5944 AG |
5229 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
5230 | case KVM_SET_GSI_ROUTING: { | |
5231 | struct kvm_irq_routing routing; | |
5232 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 5233 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
5234 | |
5235 | r = -EFAULT; | |
5236 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
5237 | goto out; | |
5238 | r = -EINVAL; | |
5c0aea0e DH |
5239 | if (!kvm_arch_can_set_irq_routing(kvm)) |
5240 | goto out; | |
caf1ff26 | 5241 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
5242 | goto out; |
5243 | if (routing.flags) | |
5244 | goto out; | |
f8c1b85b | 5245 | if (routing.nr) { |
f8c1b85b | 5246 | urouting = argp; |
1f829359 PS |
5247 | entries = vmemdup_array_user(urouting->entries, |
5248 | routing.nr, sizeof(*entries)); | |
7ec28e26 DE |
5249 | if (IS_ERR(entries)) { |
5250 | r = PTR_ERR(entries); | |
5251 | goto out; | |
5252 | } | |
f8c1b85b | 5253 | } |
aa8d5944 AG |
5254 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
5255 | routing.flags); | |
7ec28e26 | 5256 | kvfree(entries); |
aa8d5944 AG |
5257 | break; |
5258 | } | |
5259 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
5a475554 CP |
5260 | #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES |
5261 | case KVM_SET_MEMORY_ATTRIBUTES: { | |
5262 | struct kvm_memory_attributes attrs; | |
5263 | ||
5264 | r = -EFAULT; | |
5265 | if (copy_from_user(&attrs, argp, sizeof(attrs))) | |
5266 | goto out; | |
5267 | ||
5268 | r = kvm_vm_ioctl_set_mem_attributes(kvm, &attrs); | |
5269 | break; | |
5270 | } | |
5271 | #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */ | |
852b6d57 SW |
5272 | case KVM_CREATE_DEVICE: { |
5273 | struct kvm_create_device cd; | |
5274 | ||
5275 | r = -EFAULT; | |
5276 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
5277 | goto out; | |
5278 | ||
5279 | r = kvm_ioctl_create_device(kvm, &cd); | |
5280 | if (r) | |
5281 | goto out; | |
5282 | ||
5283 | r = -EFAULT; | |
5284 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
5285 | goto out; | |
5286 | ||
5287 | r = 0; | |
5288 | break; | |
5289 | } | |
92b591a4 AG |
5290 | case KVM_CHECK_EXTENSION: |
5291 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
5292 | break; | |
fb04a1ed PX |
5293 | case KVM_RESET_DIRTY_RINGS: |
5294 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
5295 | break; | |
fcfe1bae JZ |
5296 | case KVM_GET_STATS_FD: |
5297 | r = kvm_vm_ioctl_get_stats_fd(kvm); | |
5298 | break; | |
a7800aa8 SC |
5299 | #ifdef CONFIG_KVM_PRIVATE_MEM |
5300 | case KVM_CREATE_GUEST_MEMFD: { | |
5301 | struct kvm_create_guest_memfd guest_memfd; | |
5302 | ||
5303 | r = -EFAULT; | |
5304 | if (copy_from_user(&guest_memfd, argp, sizeof(guest_memfd))) | |
5305 | goto out; | |
5306 | ||
5307 | r = kvm_gmem_create(kvm, &guest_memfd); | |
5308 | break; | |
5309 | } | |
5310 | #endif | |
f17abe9a | 5311 | default: |
1fe779f8 | 5312 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
5313 | } |
5314 | out: | |
5315 | return r; | |
5316 | } | |
5317 | ||
de8e5d74 | 5318 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
5319 | struct compat_kvm_dirty_log { |
5320 | __u32 slot; | |
5321 | __u32 padding1; | |
5322 | union { | |
5323 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
5324 | __u64 padding2; | |
5325 | }; | |
5326 | }; | |
5327 | ||
8750f9bb PB |
5328 | struct compat_kvm_clear_dirty_log { |
5329 | __u32 slot; | |
5330 | __u32 num_pages; | |
5331 | __u64 first_page; | |
5332 | union { | |
5333 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
5334 | __u64 padding2; | |
5335 | }; | |
5336 | }; | |
5337 | ||
ed51862f AG |
5338 | long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl, |
5339 | unsigned long arg) | |
5340 | { | |
5341 | return -ENOTTY; | |
5342 | } | |
5343 | ||
6ff5894c AB |
5344 | static long kvm_vm_compat_ioctl(struct file *filp, |
5345 | unsigned int ioctl, unsigned long arg) | |
5346 | { | |
5347 | struct kvm *kvm = filp->private_data; | |
5348 | int r; | |
5349 | ||
f4d31653 | 5350 | if (kvm->mm != current->mm || kvm->vm_dead) |
6ff5894c | 5351 | return -EIO; |
ed51862f AG |
5352 | |
5353 | r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg); | |
5354 | if (r != -ENOTTY) | |
5355 | return r; | |
5356 | ||
6ff5894c | 5357 | switch (ioctl) { |
8750f9bb PB |
5358 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
5359 | case KVM_CLEAR_DIRTY_LOG: { | |
5360 | struct compat_kvm_clear_dirty_log compat_log; | |
5361 | struct kvm_clear_dirty_log log; | |
5362 | ||
5363 | if (copy_from_user(&compat_log, (void __user *)arg, | |
5364 | sizeof(compat_log))) | |
5365 | return -EFAULT; | |
5366 | log.slot = compat_log.slot; | |
5367 | log.num_pages = compat_log.num_pages; | |
5368 | log.first_page = compat_log.first_page; | |
5369 | log.padding2 = compat_log.padding2; | |
5370 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
5371 | ||
5372 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
5373 | break; | |
5374 | } | |
5375 | #endif | |
6ff5894c AB |
5376 | case KVM_GET_DIRTY_LOG: { |
5377 | struct compat_kvm_dirty_log compat_log; | |
5378 | struct kvm_dirty_log log; | |
5379 | ||
6ff5894c AB |
5380 | if (copy_from_user(&compat_log, (void __user *)arg, |
5381 | sizeof(compat_log))) | |
f6a3b168 | 5382 | return -EFAULT; |
6ff5894c AB |
5383 | log.slot = compat_log.slot; |
5384 | log.padding1 = compat_log.padding1; | |
5385 | log.padding2 = compat_log.padding2; | |
5386 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
5387 | ||
5388 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
5389 | break; |
5390 | } | |
5391 | default: | |
5392 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
5393 | } | |
6ff5894c AB |
5394 | return r; |
5395 | } | |
5396 | #endif | |
5397 | ||
087e1520 | 5398 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
5399 | .release = kvm_vm_release, |
5400 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 5401 | .llseek = noop_llseek, |
7ddfd3e0 | 5402 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
5403 | }; |
5404 | ||
54526d1f NT |
5405 | bool file_is_kvm(struct file *file) |
5406 | { | |
5407 | return file && file->f_op == &kvm_vm_fops; | |
5408 | } | |
5409 | EXPORT_SYMBOL_GPL(file_is_kvm); | |
5410 | ||
e08b9637 | 5411 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 5412 | { |
59f82aad | 5413 | char fdname[ITOA_MAX_LEN + 1]; |
20020f4c | 5414 | int r, fd; |
f17abe9a | 5415 | struct kvm *kvm; |
506cfba9 | 5416 | struct file *file; |
f17abe9a | 5417 | |
20020f4c OU |
5418 | fd = get_unused_fd_flags(O_CLOEXEC); |
5419 | if (fd < 0) | |
5420 | return fd; | |
5421 | ||
59f82aad OU |
5422 | snprintf(fdname, sizeof(fdname), "%d", fd); |
5423 | ||
b74ed7a6 | 5424 | kvm = kvm_create_vm(type, fdname); |
20020f4c OU |
5425 | if (IS_ERR(kvm)) { |
5426 | r = PTR_ERR(kvm); | |
5427 | goto put_fd; | |
5428 | } | |
5429 | ||
506cfba9 AV |
5430 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
5431 | if (IS_ERR(file)) { | |
78588335 ME |
5432 | r = PTR_ERR(file); |
5433 | goto put_kvm; | |
506cfba9 | 5434 | } |
536a6f88 | 5435 | |
525df861 PB |
5436 | /* |
5437 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
5438 | * already set, with ->release() being kvm_vm_release(). In error | |
5439 | * cases it will be called by the final fput(file) and will take | |
5440 | * care of doing kvm_put_kvm(kvm). | |
5441 | */ | |
286de8f6 | 5442 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 5443 | |
20020f4c OU |
5444 | fd_install(fd, file); |
5445 | return fd; | |
78588335 ME |
5446 | |
5447 | put_kvm: | |
5448 | kvm_put_kvm(kvm); | |
20020f4c OU |
5449 | put_fd: |
5450 | put_unused_fd(fd); | |
78588335 | 5451 | return r; |
f17abe9a AK |
5452 | } |
5453 | ||
5454 | static long kvm_dev_ioctl(struct file *filp, | |
5455 | unsigned int ioctl, unsigned long arg) | |
5456 | { | |
f15ba52b | 5457 | int r = -EINVAL; |
f17abe9a AK |
5458 | |
5459 | switch (ioctl) { | |
5460 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
5461 | if (arg) |
5462 | goto out; | |
f17abe9a AK |
5463 | r = KVM_API_VERSION; |
5464 | break; | |
5465 | case KVM_CREATE_VM: | |
e08b9637 | 5466 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 5467 | break; |
018d00d2 | 5468 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 5469 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 5470 | break; |
07c45a36 | 5471 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
5472 | if (arg) |
5473 | goto out; | |
adb1ff46 AK |
5474 | r = PAGE_SIZE; /* struct kvm_run */ |
5475 | #ifdef CONFIG_X86 | |
5476 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 5477 | #endif |
4b4357e0 | 5478 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 5479 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 5480 | #endif |
07c45a36 | 5481 | break; |
6aa8b732 | 5482 | default: |
043405e1 | 5483 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
5484 | } |
5485 | out: | |
5486 | return r; | |
5487 | } | |
5488 | ||
6aa8b732 | 5489 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 5490 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 5491 | .llseek = noop_llseek, |
7ddfd3e0 | 5492 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
5493 | }; |
5494 | ||
5495 | static struct miscdevice kvm_dev = { | |
bbe4432e | 5496 | KVM_MINOR, |
6aa8b732 AK |
5497 | "kvm", |
5498 | &kvm_chardev_ops, | |
5499 | }; | |
5500 | ||
441f7bfa SC |
5501 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
5502 | __visible bool kvm_rebooting; | |
5503 | EXPORT_SYMBOL_GPL(kvm_rebooting); | |
5504 | ||
5505 | static DEFINE_PER_CPU(bool, hardware_enabled); | |
5506 | static int kvm_usage_count; | |
5507 | ||
e6fb7d6e | 5508 | static int __hardware_enable_nolock(void) |
1b6c0168 | 5509 | { |
37d25881 | 5510 | if (__this_cpu_read(hardware_enabled)) |
e6fb7d6e | 5511 | return 0; |
10474ae8 | 5512 | |
37d25881 | 5513 | if (kvm_arch_hardware_enable()) { |
37d25881 SC |
5514 | pr_info("kvm: enabling virtualization on CPU%d failed\n", |
5515 | raw_smp_processor_id()); | |
e6fb7d6e | 5516 | return -EIO; |
10474ae8 | 5517 | } |
37d25881 SC |
5518 | |
5519 | __this_cpu_write(hardware_enabled, true); | |
e6fb7d6e IY |
5520 | return 0; |
5521 | } | |
5522 | ||
5523 | static void hardware_enable_nolock(void *failed) | |
5524 | { | |
5525 | if (__hardware_enable_nolock()) | |
5526 | atomic_inc(failed); | |
1b6c0168 AK |
5527 | } |
5528 | ||
aaf12a7b | 5529 | static int kvm_online_cpu(unsigned int cpu) |
75b7127c | 5530 | { |
aaf12a7b CG |
5531 | int ret = 0; |
5532 | ||
5533 | /* | |
5534 | * Abort the CPU online process if hardware virtualization cannot | |
5535 | * be enabled. Otherwise running VMs would encounter unrecoverable | |
5536 | * errors when scheduled to this CPU. | |
5537 | */ | |
0bf50497 | 5538 | mutex_lock(&kvm_lock); |
e6fb7d6e IY |
5539 | if (kvm_usage_count) |
5540 | ret = __hardware_enable_nolock(); | |
0bf50497 | 5541 | mutex_unlock(&kvm_lock); |
aaf12a7b | 5542 | return ret; |
75b7127c TY |
5543 | } |
5544 | ||
5545 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 | 5546 | { |
37d25881 SC |
5547 | /* |
5548 | * Note, hardware_disable_all_nolock() tells all online CPUs to disable | |
5549 | * hardware, not just CPUs that successfully enabled hardware! | |
5550 | */ | |
5551 | if (!__this_cpu_read(hardware_enabled)) | |
1b6c0168 | 5552 | return; |
37d25881 | 5553 | |
13a34e06 | 5554 | kvm_arch_hardware_disable(); |
37d25881 SC |
5555 | |
5556 | __this_cpu_write(hardware_enabled, false); | |
1b6c0168 AK |
5557 | } |
5558 | ||
aaf12a7b | 5559 | static int kvm_offline_cpu(unsigned int cpu) |
75b7127c | 5560 | { |
0bf50497 | 5561 | mutex_lock(&kvm_lock); |
4fa92fb2 PB |
5562 | if (kvm_usage_count) |
5563 | hardware_disable_nolock(NULL); | |
0bf50497 | 5564 | mutex_unlock(&kvm_lock); |
8c18b2d2 | 5565 | return 0; |
75b7127c TY |
5566 | } |
5567 | ||
10474ae8 AG |
5568 | static void hardware_disable_all_nolock(void) |
5569 | { | |
5570 | BUG_ON(!kvm_usage_count); | |
5571 | ||
5572 | kvm_usage_count--; | |
5573 | if (!kvm_usage_count) | |
75b7127c | 5574 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
5575 | } |
5576 | ||
5577 | static void hardware_disable_all(void) | |
5578 | { | |
e4aa7f88 | 5579 | cpus_read_lock(); |
0bf50497 | 5580 | mutex_lock(&kvm_lock); |
10474ae8 | 5581 | hardware_disable_all_nolock(); |
0bf50497 | 5582 | mutex_unlock(&kvm_lock); |
e4aa7f88 | 5583 | cpus_read_unlock(); |
10474ae8 AG |
5584 | } |
5585 | ||
5586 | static int hardware_enable_all(void) | |
5587 | { | |
e6fb7d6e | 5588 | atomic_t failed = ATOMIC_INIT(0); |
e0ceec22 SC |
5589 | int r; |
5590 | ||
5591 | /* | |
5592 | * Do not enable hardware virtualization if the system is going down. | |
5593 | * If userspace initiated a forced reboot, e.g. reboot -f, then it's | |
5594 | * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling | |
5595 | * after kvm_reboot() is called. Note, this relies on system_state | |
5596 | * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops | |
5597 | * hook instead of registering a dedicated reboot notifier (the latter | |
5598 | * runs before system_state is updated). | |
5599 | */ | |
5600 | if (system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF || | |
5601 | system_state == SYSTEM_RESTART) | |
5602 | return -EBUSY; | |
10474ae8 | 5603 | |
e4aa7f88 CG |
5604 | /* |
5605 | * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu() | |
5606 | * is called, and so on_each_cpu() between them includes the CPU that | |
5607 | * is being onlined. As a result, hardware_enable_nolock() may get | |
5608 | * invoked before kvm_online_cpu(), which also enables hardware if the | |
5609 | * usage count is non-zero. Disable CPU hotplug to avoid attempting to | |
5610 | * enable hardware multiple times. | |
5611 | */ | |
5612 | cpus_read_lock(); | |
0bf50497 | 5613 | mutex_lock(&kvm_lock); |
10474ae8 | 5614 | |
e0ceec22 SC |
5615 | r = 0; |
5616 | ||
10474ae8 AG |
5617 | kvm_usage_count++; |
5618 | if (kvm_usage_count == 1) { | |
e6fb7d6e | 5619 | on_each_cpu(hardware_enable_nolock, &failed, 1); |
10474ae8 | 5620 | |
e6fb7d6e | 5621 | if (atomic_read(&failed)) { |
10474ae8 AG |
5622 | hardware_disable_all_nolock(); |
5623 | r = -EBUSY; | |
5624 | } | |
5625 | } | |
5626 | ||
0bf50497 | 5627 | mutex_unlock(&kvm_lock); |
e4aa7f88 | 5628 | cpus_read_unlock(); |
10474ae8 AG |
5629 | |
5630 | return r; | |
5631 | } | |
5632 | ||
6735150b | 5633 | static void kvm_shutdown(void) |
9a2b85c6 | 5634 | { |
8e1c1815 | 5635 | /* |
6735150b SC |
5636 | * Disable hardware virtualization and set kvm_rebooting to indicate |
5637 | * that KVM has asynchronously disabled hardware virtualization, i.e. | |
5638 | * that relevant errors and exceptions aren't entirely unexpected. | |
5639 | * Some flavors of hardware virtualization need to be disabled before | |
5640 | * transferring control to firmware (to perform shutdown/reboot), e.g. | |
5641 | * on x86, virtualization can block INIT interrupts, which are used by | |
5642 | * firmware to pull APs back under firmware control. Note, this path | |
5643 | * is used for both shutdown and reboot scenarios, i.e. neither name is | |
5644 | * 100% comprehensive. | |
8e1c1815 | 5645 | */ |
1170adc6 | 5646 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 5647 | kvm_rebooting = true; |
75b7127c | 5648 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
5649 | } |
5650 | ||
35774a9f SC |
5651 | static int kvm_suspend(void) |
5652 | { | |
5653 | /* | |
5654 | * Secondary CPUs and CPU hotplug are disabled across the suspend/resume | |
5655 | * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count | |
5656 | * is stable. Assert that kvm_lock is not held to ensure the system | |
5657 | * isn't suspended while KVM is enabling hardware. Hardware enabling | |
5658 | * can be preempted, but the task cannot be frozen until it has dropped | |
5659 | * all locks (userspace tasks are frozen via a fake signal). | |
5660 | */ | |
5661 | lockdep_assert_not_held(&kvm_lock); | |
5662 | lockdep_assert_irqs_disabled(); | |
5663 | ||
5664 | if (kvm_usage_count) | |
5665 | hardware_disable_nolock(NULL); | |
5666 | return 0; | |
5667 | } | |
5668 | ||
5669 | static void kvm_resume(void) | |
5670 | { | |
5671 | lockdep_assert_not_held(&kvm_lock); | |
5672 | lockdep_assert_irqs_disabled(); | |
5673 | ||
5674 | if (kvm_usage_count) | |
5675 | WARN_ON_ONCE(__hardware_enable_nolock()); | |
5676 | } | |
5677 | ||
5678 | static struct syscore_ops kvm_syscore_ops = { | |
5679 | .suspend = kvm_suspend, | |
5680 | .resume = kvm_resume, | |
6735150b | 5681 | .shutdown = kvm_shutdown, |
35774a9f | 5682 | }; |
441f7bfa SC |
5683 | #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */ |
5684 | static int hardware_enable_all(void) | |
5685 | { | |
5686 | return 0; | |
5687 | } | |
5688 | ||
5689 | static void hardware_disable_all(void) | |
5690 | { | |
5691 | ||
5692 | } | |
5693 | #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */ | |
35774a9f | 5694 | |
5ea5ca3c WW |
5695 | static void kvm_iodevice_destructor(struct kvm_io_device *dev) |
5696 | { | |
5697 | if (dev->ops->destructor) | |
5698 | dev->ops->destructor(dev); | |
5699 | } | |
5700 | ||
e93f8a0f | 5701 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
5702 | { |
5703 | int i; | |
5704 | ||
5705 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 5706 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
5707 | |
5708 | kvm_iodevice_destructor(pos); | |
5709 | } | |
e93f8a0f | 5710 | kfree(bus); |
2eeb2e94 GH |
5711 | } |
5712 | ||
c21fbff1 | 5713 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 5714 | const struct kvm_io_range *r2) |
743eeb0b | 5715 | { |
8f4216c7 JW |
5716 | gpa_t addr1 = r1->addr; |
5717 | gpa_t addr2 = r2->addr; | |
5718 | ||
5719 | if (addr1 < addr2) | |
743eeb0b | 5720 | return -1; |
8f4216c7 JW |
5721 | |
5722 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
5723 | * accept any overlapping write. Any order is acceptable for | |
5724 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
5725 | * we process all of them. | |
5726 | */ | |
5727 | if (r2->len) { | |
5728 | addr1 += r1->len; | |
5729 | addr2 += r2->len; | |
5730 | } | |
5731 | ||
5732 | if (addr1 > addr2) | |
743eeb0b | 5733 | return 1; |
8f4216c7 | 5734 | |
743eeb0b SL |
5735 | return 0; |
5736 | } | |
5737 | ||
a343c9b7 PB |
5738 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
5739 | { | |
c21fbff1 | 5740 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
5741 | } |
5742 | ||
39369f7a | 5743 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
5744 | gpa_t addr, int len) |
5745 | { | |
5746 | struct kvm_io_range *range, key; | |
5747 | int off; | |
5748 | ||
5749 | key = (struct kvm_io_range) { | |
5750 | .addr = addr, | |
5751 | .len = len, | |
5752 | }; | |
5753 | ||
5754 | range = bsearch(&key, bus->range, bus->dev_count, | |
5755 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
5756 | if (range == NULL) | |
5757 | return -ENOENT; | |
5758 | ||
5759 | off = range - bus->range; | |
5760 | ||
c21fbff1 | 5761 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
5762 | off--; |
5763 | ||
5764 | return off; | |
5765 | } | |
5766 | ||
e32edf4f | 5767 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
5768 | struct kvm_io_range *range, const void *val) |
5769 | { | |
5770 | int idx; | |
5771 | ||
5772 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
5773 | if (idx < 0) | |
5774 | return -EOPNOTSUPP; | |
5775 | ||
5776 | while (idx < bus->dev_count && | |
c21fbff1 | 5777 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5778 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5779 | range->len, val)) |
5780 | return idx; | |
5781 | idx++; | |
5782 | } | |
5783 | ||
5784 | return -EOPNOTSUPP; | |
5785 | } | |
5786 | ||
bda9020e | 5787 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 5788 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 5789 | int len, const void *val) |
2eeb2e94 | 5790 | { |
90d83dc3 | 5791 | struct kvm_io_bus *bus; |
743eeb0b | 5792 | struct kvm_io_range range; |
126a5af5 | 5793 | int r; |
743eeb0b SL |
5794 | |
5795 | range = (struct kvm_io_range) { | |
5796 | .addr = addr, | |
5797 | .len = len, | |
5798 | }; | |
90d83dc3 | 5799 | |
e32edf4f | 5800 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5801 | if (!bus) |
5802 | return -ENOMEM; | |
e32edf4f | 5803 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5804 | return r < 0 ? r : 0; |
5805 | } | |
a2420107 | 5806 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
5807 | |
5808 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
5809 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
5810 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
5811 | { |
5812 | struct kvm_io_bus *bus; | |
5813 | struct kvm_io_range range; | |
5814 | ||
5815 | range = (struct kvm_io_range) { | |
5816 | .addr = addr, | |
5817 | .len = len, | |
5818 | }; | |
5819 | ||
e32edf4f | 5820 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5821 | if (!bus) |
5822 | return -ENOMEM; | |
126a5af5 CH |
5823 | |
5824 | /* First try the device referenced by cookie. */ | |
5825 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 5826 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 5827 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
5828 | val)) |
5829 | return cookie; | |
5830 | ||
5831 | /* | |
5832 | * cookie contained garbage; fall back to search and return the | |
5833 | * correct cookie value. | |
5834 | */ | |
e32edf4f | 5835 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5836 | } |
5837 | ||
e32edf4f NN |
5838 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
5839 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
5840 | { |
5841 | int idx; | |
5842 | ||
5843 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
5844 | if (idx < 0) |
5845 | return -EOPNOTSUPP; | |
5846 | ||
5847 | while (idx < bus->dev_count && | |
c21fbff1 | 5848 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5849 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5850 | range->len, val)) |
5851 | return idx; | |
743eeb0b SL |
5852 | idx++; |
5853 | } | |
5854 | ||
bda9020e MT |
5855 | return -EOPNOTSUPP; |
5856 | } | |
2eeb2e94 | 5857 | |
bda9020e | 5858 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 5859 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 5860 | int len, void *val) |
bda9020e | 5861 | { |
90d83dc3 | 5862 | struct kvm_io_bus *bus; |
743eeb0b | 5863 | struct kvm_io_range range; |
126a5af5 | 5864 | int r; |
743eeb0b SL |
5865 | |
5866 | range = (struct kvm_io_range) { | |
5867 | .addr = addr, | |
5868 | .len = len, | |
5869 | }; | |
e93f8a0f | 5870 | |
e32edf4f | 5871 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5872 | if (!bus) |
5873 | return -ENOMEM; | |
e32edf4f | 5874 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
5875 | return r < 0 ? r : 0; |
5876 | } | |
743eeb0b | 5877 | |
743eeb0b SL |
5878 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
5879 | int len, struct kvm_io_device *dev) | |
6c474694 | 5880 | { |
d4c67a7a | 5881 | int i; |
e93f8a0f | 5882 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 5883 | struct kvm_io_range range; |
090b7aff | 5884 | |
b1a39a71 MZ |
5885 | lockdep_assert_held(&kvm->slots_lock); |
5886 | ||
4a12f951 | 5887 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
5888 | if (!bus) |
5889 | return -ENOMEM; | |
5890 | ||
6ea34c9b AK |
5891 | /* exclude ioeventfd which is limited by maximum fd */ |
5892 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 5893 | return -ENOSPC; |
2eeb2e94 | 5894 | |
90952cd3 | 5895 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 5896 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
5897 | if (!new_bus) |
5898 | return -ENOMEM; | |
d4c67a7a GH |
5899 | |
5900 | range = (struct kvm_io_range) { | |
5901 | .addr = addr, | |
5902 | .len = len, | |
5903 | .dev = dev, | |
5904 | }; | |
5905 | ||
5906 | for (i = 0; i < bus->dev_count; i++) | |
5907 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
5908 | break; | |
5909 | ||
5910 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
5911 | new_bus->dev_count++; | |
5912 | new_bus->range[i] = range; | |
5913 | memcpy(new_bus->range + i + 1, bus->range + i, | |
5914 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
5915 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
5916 | synchronize_srcu_expedited(&kvm->srcu); | |
5917 | kfree(bus); | |
090b7aff GH |
5918 | |
5919 | return 0; | |
5920 | } | |
5921 | ||
5d3c4c79 SC |
5922 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5923 | struct kvm_io_device *dev) | |
090b7aff | 5924 | { |
5ea5ca3c | 5925 | int i; |
e93f8a0f | 5926 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 5927 | |
7c896d37 SC |
5928 | lockdep_assert_held(&kvm->slots_lock); |
5929 | ||
4a12f951 | 5930 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 5931 | if (!bus) |
5d3c4c79 | 5932 | return 0; |
df630b8c | 5933 | |
7c896d37 | 5934 | for (i = 0; i < bus->dev_count; i++) { |
a1300716 | 5935 | if (bus->range[i].dev == dev) { |
090b7aff GH |
5936 | break; |
5937 | } | |
7c896d37 | 5938 | } |
e93f8a0f | 5939 | |
90db1043 | 5940 | if (i == bus->dev_count) |
5d3c4c79 | 5941 | return 0; |
a1300716 | 5942 | |
90952cd3 | 5943 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 5944 | GFP_KERNEL_ACCOUNT); |
f6588660 | 5945 | if (new_bus) { |
871c433b | 5946 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
5947 | new_bus->dev_count--; |
5948 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 5949 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
2ee37574 SC |
5950 | } |
5951 | ||
5952 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
5953 | synchronize_srcu_expedited(&kvm->srcu); | |
5954 | ||
5ea5ca3c WW |
5955 | /* |
5956 | * If NULL bus is installed, destroy the old bus, including all the | |
5957 | * attached devices. Otherwise, destroy the caller's device only. | |
5958 | */ | |
2ee37574 | 5959 | if (!new_bus) { |
90db1043 | 5960 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
5ea5ca3c WW |
5961 | kvm_io_bus_destroy(bus); |
5962 | return -ENOMEM; | |
90db1043 | 5963 | } |
a1300716 | 5964 | |
5ea5ca3c | 5965 | kvm_iodevice_destructor(dev); |
e93f8a0f | 5966 | kfree(bus); |
5ea5ca3c | 5967 | return 0; |
2eeb2e94 GH |
5968 | } |
5969 | ||
8a39d006 AP |
5970 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5971 | gpa_t addr) | |
5972 | { | |
5973 | struct kvm_io_bus *bus; | |
5974 | int dev_idx, srcu_idx; | |
5975 | struct kvm_io_device *iodev = NULL; | |
5976 | ||
5977 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
5978 | ||
5979 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
5980 | if (!bus) |
5981 | goto out_unlock; | |
8a39d006 AP |
5982 | |
5983 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
5984 | if (dev_idx < 0) | |
5985 | goto out_unlock; | |
5986 | ||
5987 | iodev = bus->range[dev_idx].dev; | |
5988 | ||
5989 | out_unlock: | |
5990 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
5991 | ||
5992 | return iodev; | |
5993 | } | |
5994 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
5995 | ||
536a6f88 JF |
5996 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
5997 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
5998 | const char *fmt) | |
5999 | { | |
180418e2 | 6000 | int ret; |
14aa40a1 | 6001 | struct kvm_stat_data *stat_data = inode->i_private; |
536a6f88 | 6002 | |
605c7130 PX |
6003 | /* |
6004 | * The debugfs files are a reference to the kvm struct which | |
6005 | * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe | |
6006 | * avoids the race between open and the removal of the debugfs directory. | |
536a6f88 | 6007 | */ |
605c7130 | 6008 | if (!kvm_get_kvm_safe(stat_data->kvm)) |
536a6f88 JF |
6009 | return -ENOENT; |
6010 | ||
180418e2 HW |
6011 | ret = simple_attr_open(inode, file, get, |
6012 | kvm_stats_debugfs_mode(stat_data->desc) & 0222 | |
6013 | ? set : NULL, fmt); | |
6014 | if (ret) | |
536a6f88 | 6015 | kvm_put_kvm(stat_data->kvm); |
536a6f88 | 6016 | |
180418e2 | 6017 | return ret; |
536a6f88 JF |
6018 | } |
6019 | ||
6020 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
6021 | { | |
14aa40a1 | 6022 | struct kvm_stat_data *stat_data = inode->i_private; |
536a6f88 JF |
6023 | |
6024 | simple_attr_release(inode, file); | |
6025 | kvm_put_kvm(stat_data->kvm); | |
6026 | ||
6027 | return 0; | |
6028 | } | |
6029 | ||
09cbcef6 | 6030 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 6031 | { |
bc9e9e67 | 6032 | *val = *(u64 *)((void *)(&kvm->stat) + offset); |
536a6f88 | 6033 | |
09cbcef6 MP |
6034 | return 0; |
6035 | } | |
6036 | ||
6037 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
6038 | { | |
bc9e9e67 | 6039 | *(u64 *)((void *)(&kvm->stat) + offset) = 0; |
536a6f88 JF |
6040 | |
6041 | return 0; | |
6042 | } | |
6043 | ||
09cbcef6 | 6044 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 6045 | { |
46808a4c | 6046 | unsigned long i; |
09cbcef6 | 6047 | struct kvm_vcpu *vcpu; |
ce35ef27 | 6048 | |
09cbcef6 | 6049 | *val = 0; |
ce35ef27 | 6050 | |
09cbcef6 | 6051 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 6052 | *val += *(u64 *)((void *)(&vcpu->stat) + offset); |
ce35ef27 SJS |
6053 | |
6054 | return 0; | |
6055 | } | |
6056 | ||
09cbcef6 | 6057 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 6058 | { |
46808a4c | 6059 | unsigned long i; |
09cbcef6 | 6060 | struct kvm_vcpu *vcpu; |
536a6f88 | 6061 | |
09cbcef6 | 6062 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 6063 | *(u64 *)((void *)(&vcpu->stat) + offset) = 0; |
09cbcef6 MP |
6064 | |
6065 | return 0; | |
6066 | } | |
536a6f88 | 6067 | |
09cbcef6 | 6068 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 6069 | { |
09cbcef6 | 6070 | int r = -EFAULT; |
14aa40a1 | 6071 | struct kvm_stat_data *stat_data = data; |
536a6f88 | 6072 | |
bc9e9e67 | 6073 | switch (stat_data->kind) { |
09cbcef6 MP |
6074 | case KVM_STAT_VM: |
6075 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 6076 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
6077 | break; |
6078 | case KVM_STAT_VCPU: | |
6079 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 6080 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
6081 | break; |
6082 | } | |
536a6f88 | 6083 | |
09cbcef6 | 6084 | return r; |
536a6f88 JF |
6085 | } |
6086 | ||
09cbcef6 | 6087 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 6088 | { |
09cbcef6 | 6089 | int r = -EFAULT; |
14aa40a1 | 6090 | struct kvm_stat_data *stat_data = data; |
ce35ef27 SJS |
6091 | |
6092 | if (val) | |
6093 | return -EINVAL; | |
6094 | ||
bc9e9e67 | 6095 | switch (stat_data->kind) { |
09cbcef6 MP |
6096 | case KVM_STAT_VM: |
6097 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 6098 | stat_data->desc->desc.offset); |
09cbcef6 MP |
6099 | break; |
6100 | case KVM_STAT_VCPU: | |
6101 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 6102 | stat_data->desc->desc.offset); |
09cbcef6 MP |
6103 | break; |
6104 | } | |
ce35ef27 | 6105 | |
09cbcef6 | 6106 | return r; |
ce35ef27 SJS |
6107 | } |
6108 | ||
09cbcef6 | 6109 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
6110 | { |
6111 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
6112 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
6113 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
6114 | } |
6115 | ||
09cbcef6 MP |
6116 | static const struct file_operations stat_fops_per_vm = { |
6117 | .owner = THIS_MODULE, | |
6118 | .open = kvm_stat_data_open, | |
536a6f88 | 6119 | .release = kvm_debugfs_release, |
09cbcef6 MP |
6120 | .read = simple_attr_read, |
6121 | .write = simple_attr_write, | |
6122 | .llseek = no_llseek, | |
536a6f88 JF |
6123 | }; |
6124 | ||
8b88b099 | 6125 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
6126 | { |
6127 | unsigned offset = (long)_offset; | |
ba1389b7 | 6128 | struct kvm *kvm; |
536a6f88 | 6129 | u64 tmp_val; |
ba1389b7 | 6130 | |
8b88b099 | 6131 | *val = 0; |
0d9ce162 | 6132 | mutex_lock(&kvm_lock); |
536a6f88 | 6133 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 6134 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
6135 | *val += tmp_val; |
6136 | } | |
0d9ce162 | 6137 | mutex_unlock(&kvm_lock); |
8b88b099 | 6138 | return 0; |
ba1389b7 AK |
6139 | } |
6140 | ||
ce35ef27 SJS |
6141 | static int vm_stat_clear(void *_offset, u64 val) |
6142 | { | |
6143 | unsigned offset = (long)_offset; | |
6144 | struct kvm *kvm; | |
ce35ef27 SJS |
6145 | |
6146 | if (val) | |
6147 | return -EINVAL; | |
6148 | ||
0d9ce162 | 6149 | mutex_lock(&kvm_lock); |
ce35ef27 | 6150 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 6151 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 6152 | } |
0d9ce162 | 6153 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
6154 | |
6155 | return 0; | |
6156 | } | |
6157 | ||
6158 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
bc9e9e67 | 6159 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n"); |
ba1389b7 | 6160 | |
8b88b099 | 6161 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
6162 | { |
6163 | unsigned offset = (long)_offset; | |
1165f5fe | 6164 | struct kvm *kvm; |
536a6f88 | 6165 | u64 tmp_val; |
1165f5fe | 6166 | |
8b88b099 | 6167 | *val = 0; |
0d9ce162 | 6168 | mutex_lock(&kvm_lock); |
536a6f88 | 6169 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 6170 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
6171 | *val += tmp_val; |
6172 | } | |
0d9ce162 | 6173 | mutex_unlock(&kvm_lock); |
8b88b099 | 6174 | return 0; |
1165f5fe AK |
6175 | } |
6176 | ||
ce35ef27 SJS |
6177 | static int vcpu_stat_clear(void *_offset, u64 val) |
6178 | { | |
6179 | unsigned offset = (long)_offset; | |
6180 | struct kvm *kvm; | |
ce35ef27 SJS |
6181 | |
6182 | if (val) | |
6183 | return -EINVAL; | |
6184 | ||
0d9ce162 | 6185 | mutex_lock(&kvm_lock); |
ce35ef27 | 6186 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 6187 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 6188 | } |
0d9ce162 | 6189 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
6190 | |
6191 | return 0; | |
6192 | } | |
6193 | ||
6194 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
6195 | "%llu\n"); | |
bc9e9e67 | 6196 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n"); |
1165f5fe | 6197 | |
286de8f6 CI |
6198 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
6199 | { | |
6200 | struct kobj_uevent_env *env; | |
286de8f6 CI |
6201 | unsigned long long created, active; |
6202 | ||
6203 | if (!kvm_dev.this_device || !kvm) | |
6204 | return; | |
6205 | ||
0d9ce162 | 6206 | mutex_lock(&kvm_lock); |
286de8f6 CI |
6207 | if (type == KVM_EVENT_CREATE_VM) { |
6208 | kvm_createvm_count++; | |
6209 | kvm_active_vms++; | |
6210 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
6211 | kvm_active_vms--; | |
6212 | } | |
6213 | created = kvm_createvm_count; | |
6214 | active = kvm_active_vms; | |
0d9ce162 | 6215 | mutex_unlock(&kvm_lock); |
286de8f6 | 6216 | |
b12ce36a | 6217 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
6218 | if (!env) |
6219 | return; | |
6220 | ||
6221 | add_uevent_var(env, "CREATED=%llu", created); | |
6222 | add_uevent_var(env, "COUNT=%llu", active); | |
6223 | ||
fdeaf7e3 | 6224 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 6225 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
6226 | kvm->userspace_pid = task_pid_nr(current); |
6227 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 6228 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
6229 | } |
6230 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 6231 | |
a44a4cc1 | 6232 | if (!IS_ERR(kvm->debugfs_dentry)) { |
b12ce36a | 6233 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
6234 | |
6235 | if (p) { | |
6236 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
6237 | if (!IS_ERR(tmp)) | |
6238 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
6239 | kfree(p); | |
286de8f6 CI |
6240 | } |
6241 | } | |
6242 | /* no need for checks, since we are adding at most only 5 keys */ | |
6243 | env->envp[env->envp_idx++] = NULL; | |
6244 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
6245 | kfree(env); | |
286de8f6 CI |
6246 | } |
6247 | ||
929f45e3 | 6248 | static void kvm_init_debug(void) |
6aa8b732 | 6249 | { |
bc9e9e67 JZ |
6250 | const struct file_operations *fops; |
6251 | const struct _kvm_stats_desc *pdesc; | |
6252 | int i; | |
6aa8b732 | 6253 | |
76f7c879 | 6254 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 6255 | |
bc9e9e67 JZ |
6256 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
6257 | pdesc = &kvm_vm_stats_desc[i]; | |
6258 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
6259 | fops = &vm_stat_fops; | |
6260 | else | |
6261 | fops = &vm_stat_readonly_fops; | |
6262 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
6263 | kvm_debugfs_dir, | |
6264 | (void *)(long)pdesc->desc.offset, fops); | |
6265 | } | |
6266 | ||
6267 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
6268 | pdesc = &kvm_vcpu_stats_desc[i]; | |
6269 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
6270 | fops = &vcpu_stat_fops; | |
6271 | else | |
6272 | fops = &vcpu_stat_readonly_fops; | |
6273 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
6274 | kvm_debugfs_dir, | |
6275 | (void *)(long)pdesc->desc.offset, fops); | |
4f69b680 | 6276 | } |
6aa8b732 AK |
6277 | } |
6278 | ||
15ad7146 AK |
6279 | static inline |
6280 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
6281 | { | |
6282 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
6283 | } | |
6284 | ||
6285 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
6286 | { | |
6287 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 6288 | |
046ddeed | 6289 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 6290 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 6291 | |
7495e22b | 6292 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 6293 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 6294 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
6295 | } |
6296 | ||
6297 | static void kvm_sched_out(struct preempt_notifier *pn, | |
6298 | struct task_struct *next) | |
6299 | { | |
6300 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
6301 | ||
3ba9f93b | 6302 | if (current->on_rq) { |
046ddeed | 6303 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
6304 | WRITE_ONCE(vcpu->ready, true); |
6305 | } | |
e9b11c17 | 6306 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
6307 | __this_cpu_write(kvm_running_vcpu, NULL); |
6308 | } | |
6309 | ||
6310 | /** | |
6311 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
6312 | * |
6313 | * We can disable preemption locally around accessing the per-CPU variable, | |
6314 | * and use the resolved vcpu pointer after enabling preemption again, | |
6315 | * because even if the current thread is migrated to another CPU, reading | |
6316 | * the per-CPU value later will give us the same value as we update the | |
6317 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
6318 | */ |
6319 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
6320 | { | |
1f03b2bc MZ |
6321 | struct kvm_vcpu *vcpu; |
6322 | ||
6323 | preempt_disable(); | |
6324 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
6325 | preempt_enable(); | |
6326 | ||
6327 | return vcpu; | |
7495e22b | 6328 | } |
379a3c8e | 6329 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
6330 | |
6331 | /** | |
6332 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
6333 | */ | |
6334 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
6335 | { | |
6336 | return &kvm_running_vcpu; | |
15ad7146 AK |
6337 | } |
6338 | ||
e1bfc245 SC |
6339 | #ifdef CONFIG_GUEST_PERF_EVENTS |
6340 | static unsigned int kvm_guest_state(void) | |
6341 | { | |
6342 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
6343 | unsigned int state; | |
6344 | ||
6345 | if (!kvm_arch_pmi_in_guest(vcpu)) | |
6346 | return 0; | |
6347 | ||
6348 | state = PERF_GUEST_ACTIVE; | |
6349 | if (!kvm_arch_vcpu_in_kernel(vcpu)) | |
6350 | state |= PERF_GUEST_USER; | |
6351 | ||
6352 | return state; | |
6353 | } | |
6354 | ||
6355 | static unsigned long kvm_guest_get_ip(void) | |
6356 | { | |
6357 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
6358 | ||
6359 | /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */ | |
6360 | if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu))) | |
6361 | return 0; | |
6362 | ||
6363 | return kvm_arch_vcpu_get_ip(vcpu); | |
6364 | } | |
6365 | ||
6366 | static struct perf_guest_info_callbacks kvm_guest_cbs = { | |
6367 | .state = kvm_guest_state, | |
6368 | .get_ip = kvm_guest_get_ip, | |
6369 | .handle_intel_pt_intr = NULL, | |
6370 | }; | |
6371 | ||
6372 | void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void)) | |
6373 | { | |
6374 | kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler; | |
6375 | perf_register_guest_info_callbacks(&kvm_guest_cbs); | |
6376 | } | |
6377 | void kvm_unregister_perf_callbacks(void) | |
6378 | { | |
6379 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); | |
6380 | } | |
6381 | #endif | |
6382 | ||
81a1cf9f | 6383 | int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module) |
f257d6dc | 6384 | { |
6aa8b732 | 6385 | int r; |
002c7f7c | 6386 | int cpu; |
6aa8b732 | 6387 | |
441f7bfa | 6388 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
aaf12a7b CG |
6389 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE, "kvm/cpu:online", |
6390 | kvm_online_cpu, kvm_offline_cpu); | |
774c47f1 | 6391 | if (r) |
37d25881 SC |
6392 | return r; |
6393 | ||
35774a9f | 6394 | register_syscore_ops(&kvm_syscore_ops); |
441f7bfa | 6395 | #endif |
6aa8b732 | 6396 | |
c16f862d | 6397 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
6398 | if (!vcpu_align) |
6399 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
6400 | kvm_vcpu_cache = |
6401 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
6402 | SLAB_ACCOUNT, | |
6403 | offsetof(struct kvm_vcpu, arch), | |
ce55c049 JZ |
6404 | offsetofend(struct kvm_vcpu, stats_id) |
6405 | - offsetof(struct kvm_vcpu, arch), | |
46515736 | 6406 | NULL); |
c16f862d RR |
6407 | if (!kvm_vcpu_cache) { |
6408 | r = -ENOMEM; | |
9f1a4c00 | 6409 | goto err_vcpu_cache; |
c16f862d RR |
6410 | } |
6411 | ||
baff59cc VK |
6412 | for_each_possible_cpu(cpu) { |
6413 | if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu), | |
6414 | GFP_KERNEL, cpu_to_node(cpu))) { | |
6415 | r = -ENOMEM; | |
9f1a4c00 | 6416 | goto err_cpu_kick_mask; |
baff59cc VK |
6417 | } |
6418 | } | |
6419 | ||
5910ccf0 SC |
6420 | r = kvm_irqfd_init(); |
6421 | if (r) | |
6422 | goto err_irqfd; | |
6423 | ||
af585b92 GN |
6424 | r = kvm_async_pf_init(); |
6425 | if (r) | |
5910ccf0 | 6426 | goto err_async_pf; |
af585b92 | 6427 | |
6aa8b732 | 6428 | kvm_chardev_ops.owner = module; |
087e1520 SC |
6429 | kvm_vm_fops.owner = module; |
6430 | kvm_vcpu_fops.owner = module; | |
6431 | kvm_device_fops.owner = module; | |
6aa8b732 | 6432 | |
15ad7146 AK |
6433 | kvm_preempt_ops.sched_in = kvm_sched_in; |
6434 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
6435 | ||
929f45e3 | 6436 | kvm_init_debug(); |
0ea4ed8e | 6437 | |
3c3c29fd | 6438 | r = kvm_vfio_ops_init(); |
2b012812 SC |
6439 | if (WARN_ON_ONCE(r)) |
6440 | goto err_vfio; | |
6441 | ||
a7800aa8 SC |
6442 | kvm_gmem_init(module); |
6443 | ||
2b012812 SC |
6444 | /* |
6445 | * Registration _must_ be the very last thing done, as this exposes | |
6446 | * /dev/kvm to userspace, i.e. all infrastructure must be setup! | |
6447 | */ | |
6448 | r = misc_register(&kvm_dev); | |
6449 | if (r) { | |
6450 | pr_err("kvm: misc device register failed\n"); | |
6451 | goto err_register; | |
6452 | } | |
3c3c29fd | 6453 | |
c7addb90 | 6454 | return 0; |
6aa8b732 | 6455 | |
2b012812 SC |
6456 | err_register: |
6457 | kvm_vfio_ops_exit(); | |
6458 | err_vfio: | |
af585b92 | 6459 | kvm_async_pf_deinit(); |
5910ccf0 SC |
6460 | err_async_pf: |
6461 | kvm_irqfd_exit(); | |
6462 | err_irqfd: | |
9f1a4c00 | 6463 | err_cpu_kick_mask: |
baff59cc VK |
6464 | for_each_possible_cpu(cpu) |
6465 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
c16f862d | 6466 | kmem_cache_destroy(kvm_vcpu_cache); |
9f1a4c00 | 6467 | err_vcpu_cache: |
441f7bfa | 6468 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
35774a9f | 6469 | unregister_syscore_ops(&kvm_syscore_ops); |
aaf12a7b | 6470 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE); |
441f7bfa | 6471 | #endif |
6aa8b732 AK |
6472 | return r; |
6473 | } | |
cb498ea2 | 6474 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 6475 | |
cb498ea2 | 6476 | void kvm_exit(void) |
6aa8b732 | 6477 | { |
baff59cc VK |
6478 | int cpu; |
6479 | ||
2b012812 SC |
6480 | /* |
6481 | * Note, unregistering /dev/kvm doesn't strictly need to come first, | |
6482 | * fops_get(), a.k.a. try_module_get(), prevents acquiring references | |
6483 | * to KVM while the module is being stopped. | |
6484 | */ | |
6aa8b732 | 6485 | misc_deregister(&kvm_dev); |
2b012812 SC |
6486 | |
6487 | debugfs_remove_recursive(kvm_debugfs_dir); | |
baff59cc VK |
6488 | for_each_possible_cpu(cpu) |
6489 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
c16f862d | 6490 | kmem_cache_destroy(kvm_vcpu_cache); |
73b8dc04 | 6491 | kvm_vfio_ops_exit(); |
af585b92 | 6492 | kvm_async_pf_deinit(); |
441f7bfa | 6493 | #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING |
fb3600cc | 6494 | unregister_syscore_ops(&kvm_syscore_ops); |
aaf12a7b | 6495 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE); |
441f7bfa | 6496 | #endif |
5910ccf0 | 6497 | kvm_irqfd_exit(); |
6aa8b732 | 6498 | } |
cb498ea2 | 6499 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
6500 | |
6501 | struct kvm_vm_worker_thread_context { | |
6502 | struct kvm *kvm; | |
6503 | struct task_struct *parent; | |
6504 | struct completion init_done; | |
6505 | kvm_vm_thread_fn_t thread_fn; | |
6506 | uintptr_t data; | |
6507 | int err; | |
6508 | }; | |
6509 | ||
6510 | static int kvm_vm_worker_thread(void *context) | |
6511 | { | |
6512 | /* | |
6513 | * The init_context is allocated on the stack of the parent thread, so | |
6514 | * we have to locally copy anything that is needed beyond initialization | |
6515 | */ | |
6516 | struct kvm_vm_worker_thread_context *init_context = context; | |
e45cce30 | 6517 | struct task_struct *parent; |
c57c8046 JS |
6518 | struct kvm *kvm = init_context->kvm; |
6519 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
6520 | uintptr_t data = init_context->data; | |
6521 | int err; | |
6522 | ||
6523 | err = kthread_park(current); | |
6524 | /* kthread_park(current) is never supposed to return an error */ | |
6525 | WARN_ON(err != 0); | |
6526 | if (err) | |
6527 | goto init_complete; | |
6528 | ||
6529 | err = cgroup_attach_task_all(init_context->parent, current); | |
6530 | if (err) { | |
6531 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
6532 | __func__, err); | |
6533 | goto init_complete; | |
6534 | } | |
6535 | ||
6536 | set_user_nice(current, task_nice(init_context->parent)); | |
6537 | ||
6538 | init_complete: | |
6539 | init_context->err = err; | |
6540 | complete(&init_context->init_done); | |
6541 | init_context = NULL; | |
6542 | ||
6543 | if (err) | |
e45cce30 | 6544 | goto out; |
c57c8046 JS |
6545 | |
6546 | /* Wait to be woken up by the spawner before proceeding. */ | |
6547 | kthread_parkme(); | |
6548 | ||
6549 | if (!kthread_should_stop()) | |
6550 | err = thread_fn(kvm, data); | |
6551 | ||
e45cce30 VS |
6552 | out: |
6553 | /* | |
6554 | * Move kthread back to its original cgroup to prevent it lingering in | |
6555 | * the cgroup of the VM process, after the latter finishes its | |
6556 | * execution. | |
6557 | * | |
6558 | * kthread_stop() waits on the 'exited' completion condition which is | |
6559 | * set in exit_mm(), via mm_release(), in do_exit(). However, the | |
6560 | * kthread is removed from the cgroup in the cgroup_exit() which is | |
6561 | * called after the exit_mm(). This causes the kthread_stop() to return | |
6562 | * before the kthread actually quits the cgroup. | |
6563 | */ | |
6564 | rcu_read_lock(); | |
6565 | parent = rcu_dereference(current->real_parent); | |
6566 | get_task_struct(parent); | |
6567 | rcu_read_unlock(); | |
6568 | cgroup_attach_task_all(parent, current); | |
6569 | put_task_struct(parent); | |
6570 | ||
c57c8046 JS |
6571 | return err; |
6572 | } | |
6573 | ||
6574 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
6575 | uintptr_t data, const char *name, | |
6576 | struct task_struct **thread_ptr) | |
6577 | { | |
6578 | struct kvm_vm_worker_thread_context init_context = {}; | |
6579 | struct task_struct *thread; | |
6580 | ||
6581 | *thread_ptr = NULL; | |
6582 | init_context.kvm = kvm; | |
6583 | init_context.parent = current; | |
6584 | init_context.thread_fn = thread_fn; | |
6585 | init_context.data = data; | |
6586 | init_completion(&init_context.init_done); | |
6587 | ||
6588 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
6589 | "%s-%d", name, task_pid_nr(current)); | |
6590 | if (IS_ERR(thread)) | |
6591 | return PTR_ERR(thread); | |
6592 | ||
6593 | /* kthread_run is never supposed to return NULL */ | |
6594 | WARN_ON(thread == NULL); | |
6595 | ||
6596 | wait_for_completion(&init_context.init_done); | |
6597 | ||
6598 | if (!init_context.err) | |
6599 | *thread_ptr = thread; | |
6600 | ||
6601 | return init_context.err; | |
6602 | } |