]>
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> |
6aa8b732 | 54 | |
e495606d | 55 | #include <asm/processor.h> |
2ea75be3 | 56 | #include <asm/ioctl.h> |
7c0f6ba6 | 57 | #include <linux/uaccess.h> |
6aa8b732 | 58 | |
5f94c174 | 59 | #include "coalesced_mmio.h" |
af585b92 | 60 | #include "async_pf.h" |
531810ca | 61 | #include "mmu_lock.h" |
3c3c29fd | 62 | #include "vfio.h" |
5f94c174 | 63 | |
229456fc MT |
64 | #define CREATE_TRACE_POINTS |
65 | #include <trace/events/kvm.h> | |
66 | ||
fb04a1ed PX |
67 | #include <linux/kvm_dirty_ring.h> |
68 | ||
536a6f88 JF |
69 | /* Worst case buffer size needed for holding an integer. */ |
70 | #define ITOA_MAX_LEN 12 | |
71 | ||
6aa8b732 AK |
72 | MODULE_AUTHOR("Qumranet"); |
73 | MODULE_LICENSE("GPL"); | |
74 | ||
920552b2 | 75 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 76 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
039c5d1b | 77 | module_param(halt_poll_ns, uint, 0644); |
ec76d819 | 78 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 79 | |
aca6ff29 | 80 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 81 | unsigned int halt_poll_ns_grow = 2; |
039c5d1b | 82 | module_param(halt_poll_ns_grow, uint, 0644); |
ec76d819 | 83 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 | 84 | |
49113d36 NW |
85 | /* The start value to grow halt_poll_ns from */ |
86 | unsigned int halt_poll_ns_grow_start = 10000; /* 10us */ | |
87 | module_param(halt_poll_ns_grow_start, uint, 0644); | |
88 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start); | |
89 | ||
aca6ff29 | 90 | /* Default resets per-vcpu halt_poll_ns . */ |
ec76d819 | 91 | unsigned int halt_poll_ns_shrink; |
039c5d1b | 92 | module_param(halt_poll_ns_shrink, uint, 0644); |
ec76d819 | 93 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 94 | |
fa40a821 MT |
95 | /* |
96 | * Ordering of locks: | |
97 | * | |
b7d409de | 98 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
99 | */ |
100 | ||
0d9ce162 | 101 | DEFINE_MUTEX(kvm_lock); |
4a937f96 | 102 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 103 | LIST_HEAD(vm_list); |
133de902 | 104 | |
7f59f492 | 105 | static cpumask_var_t cpus_hardware_enabled; |
f4fee932 | 106 | static int kvm_usage_count; |
10474ae8 | 107 | static atomic_t hardware_enable_failed; |
1b6c0168 | 108 | |
aaba298c | 109 | static struct kmem_cache *kvm_vcpu_cache; |
1165f5fe | 110 | |
15ad7146 | 111 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
7495e22b | 112 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu); |
15ad7146 | 113 | |
76f7c879 | 114 | struct dentry *kvm_debugfs_dir; |
e23a808b | 115 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 116 | |
536a6f88 | 117 | static int kvm_debugfs_num_entries; |
09cbcef6 | 118 | static const struct file_operations stat_fops_per_vm; |
536a6f88 | 119 | |
bccf2150 AK |
120 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
121 | unsigned long arg); | |
de8e5d74 | 122 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
123 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
124 | unsigned long arg); | |
7ddfd3e0 MZ |
125 | #define KVM_COMPAT(c) .compat_ioctl = (c) |
126 | #else | |
9cb09e7c MZ |
127 | /* |
128 | * For architectures that don't implement a compat infrastructure, | |
129 | * adopt a double line of defense: | |
130 | * - Prevent a compat task from opening /dev/kvm | |
131 | * - If the open has been done by a 64bit task, and the KVM fd | |
132 | * passed to a compat task, let the ioctls fail. | |
133 | */ | |
7ddfd3e0 MZ |
134 | static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl, |
135 | unsigned long arg) { return -EINVAL; } | |
b9876e6d MZ |
136 | |
137 | static int kvm_no_compat_open(struct inode *inode, struct file *file) | |
138 | { | |
139 | return is_compat_task() ? -ENODEV : 0; | |
140 | } | |
141 | #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \ | |
142 | .open = kvm_no_compat_open | |
1dda606c | 143 | #endif |
10474ae8 AG |
144 | static int hardware_enable_all(void); |
145 | static void hardware_disable_all(void); | |
bccf2150 | 146 | |
e93f8a0f | 147 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 148 | |
52480137 | 149 | __visible bool kvm_rebooting; |
b7c4145b | 150 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 151 | |
286de8f6 CI |
152 | #define KVM_EVENT_CREATE_VM 0 |
153 | #define KVM_EVENT_DESTROY_VM 1 | |
154 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); | |
155 | static unsigned long long kvm_createvm_count; | |
156 | static unsigned long long kvm_active_vms; | |
157 | ||
e649b3f0 ET |
158 | __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
159 | unsigned long start, unsigned long end) | |
b1394e74 RK |
160 | { |
161 | } | |
162 | ||
a78986aa SC |
163 | bool kvm_is_zone_device_pfn(kvm_pfn_t pfn) |
164 | { | |
165 | /* | |
166 | * The metadata used by is_zone_device_page() to determine whether or | |
167 | * not a page is ZONE_DEVICE is guaranteed to be valid if and only if | |
168 | * the device has been pinned, e.g. by get_user_pages(). WARN if the | |
169 | * page_count() is zero to help detect bad usage of this helper. | |
170 | */ | |
171 | if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn)))) | |
172 | return false; | |
173 | ||
174 | return is_zone_device_page(pfn_to_page(pfn)); | |
175 | } | |
176 | ||
ba049e93 | 177 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
cbff90a7 | 178 | { |
a78986aa SC |
179 | /* |
180 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting | |
181 | * perspective they are "normal" pages, albeit with slightly different | |
182 | * usage rules. | |
183 | */ | |
11feeb49 | 184 | if (pfn_valid(pfn)) |
a78986aa | 185 | return PageReserved(pfn_to_page(pfn)) && |
7df003c8 | 186 | !is_zero_pfn(pfn) && |
a78986aa | 187 | !kvm_is_zone_device_pfn(pfn); |
cbff90a7 BAY |
188 | |
189 | return true; | |
190 | } | |
191 | ||
005ba37c SC |
192 | bool kvm_is_transparent_hugepage(kvm_pfn_t pfn) |
193 | { | |
194 | struct page *page = pfn_to_page(pfn); | |
195 | ||
196 | if (!PageTransCompoundMap(page)) | |
197 | return false; | |
198 | ||
199 | return is_transparent_hugepage(compound_head(page)); | |
200 | } | |
201 | ||
bccf2150 AK |
202 | /* |
203 | * Switches to specified vcpu, until a matching vcpu_put() | |
204 | */ | |
ec7660cc | 205 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 206 | { |
ec7660cc | 207 | int cpu = get_cpu(); |
7495e22b PB |
208 | |
209 | __this_cpu_write(kvm_running_vcpu, vcpu); | |
15ad7146 | 210 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 211 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 212 | put_cpu(); |
6aa8b732 | 213 | } |
2f1fe811 | 214 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 215 | |
313a3dc7 | 216 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 217 | { |
15ad7146 | 218 | preempt_disable(); |
313a3dc7 | 219 | kvm_arch_vcpu_put(vcpu); |
15ad7146 | 220 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
7495e22b | 221 | __this_cpu_write(kvm_running_vcpu, NULL); |
15ad7146 | 222 | preempt_enable(); |
6aa8b732 | 223 | } |
2f1fe811 | 224 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 225 | |
7a97cec2 PB |
226 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
227 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
228 | { | |
229 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
230 | ||
231 | /* | |
232 | * We need to wait for the VCPU to reenable interrupts and get out of | |
233 | * READING_SHADOW_PAGE_TABLES mode. | |
234 | */ | |
235 | if (req & KVM_REQUEST_WAIT) | |
236 | return mode != OUTSIDE_GUEST_MODE; | |
237 | ||
238 | /* | |
239 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
240 | */ | |
241 | return mode == IN_GUEST_MODE; | |
242 | } | |
243 | ||
d9e368d6 AK |
244 | static void ack_flush(void *_completed) |
245 | { | |
d9e368d6 AK |
246 | } |
247 | ||
b49defe8 PB |
248 | static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait) |
249 | { | |
250 | if (unlikely(!cpus)) | |
251 | cpus = cpu_online_mask; | |
252 | ||
253 | if (cpumask_empty(cpus)) | |
254 | return false; | |
255 | ||
256 | smp_call_function_many(cpus, ack_flush, NULL, wait); | |
257 | return true; | |
258 | } | |
259 | ||
7053df4e | 260 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
54163a34 | 261 | struct kvm_vcpu *except, |
7053df4e | 262 | unsigned long *vcpu_bitmap, cpumask_var_t tmp) |
d9e368d6 | 263 | { |
597a5f55 | 264 | int i, cpu, me; |
d9e368d6 | 265 | struct kvm_vcpu *vcpu; |
7053df4e | 266 | bool called; |
6ef7a1bc | 267 | |
3cba4130 | 268 | me = get_cpu(); |
7053df4e | 269 | |
988a2cae | 270 | kvm_for_each_vcpu(i, vcpu, kvm) { |
54163a34 SS |
271 | if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) || |
272 | vcpu == except) | |
7053df4e VK |
273 | continue; |
274 | ||
3cba4130 | 275 | kvm_make_request(req, vcpu); |
d9e368d6 | 276 | cpu = vcpu->cpu; |
6b7e2d09 | 277 | |
178f02ff RK |
278 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) |
279 | continue; | |
6c6e8360 | 280 | |
7053df4e | 281 | if (tmp != NULL && cpu != -1 && cpu != me && |
7a97cec2 | 282 | kvm_request_needs_ipi(vcpu, req)) |
7053df4e | 283 | __cpumask_set_cpu(cpu, tmp); |
49846896 | 284 | } |
7053df4e VK |
285 | |
286 | called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT)); | |
3cba4130 | 287 | put_cpu(); |
7053df4e VK |
288 | |
289 | return called; | |
290 | } | |
291 | ||
54163a34 SS |
292 | bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req, |
293 | struct kvm_vcpu *except) | |
7053df4e VK |
294 | { |
295 | cpumask_var_t cpus; | |
296 | bool called; | |
7053df4e VK |
297 | |
298 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); | |
299 | ||
54163a34 | 300 | called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus); |
7053df4e | 301 | |
6ef7a1bc | 302 | free_cpumask_var(cpus); |
49846896 | 303 | return called; |
d9e368d6 AK |
304 | } |
305 | ||
54163a34 SS |
306 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
307 | { | |
308 | return kvm_make_all_cpus_request_except(kvm, req, NULL); | |
309 | } | |
310 | ||
a6d51016 | 311 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 312 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 313 | { |
4ae3cb3a LT |
314 | /* |
315 | * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in | |
316 | * kvm_make_all_cpus_request. | |
317 | */ | |
318 | long dirty_count = smp_load_acquire(&kvm->tlbs_dirty); | |
319 | ||
320 | /* | |
321 | * We want to publish modifications to the page tables before reading | |
322 | * mode. Pairs with a memory barrier in arch-specific code. | |
323 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
324 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
325 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
326 | * | |
327 | * There is already an smp_mb__after_atomic() before | |
328 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
329 | * barrier here. | |
330 | */ | |
b08660e5 TL |
331 | if (!kvm_arch_flush_remote_tlb(kvm) |
332 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
49846896 | 333 | ++kvm->stat.remote_tlb_flush; |
a086f6a1 | 334 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 335 | } |
2ba9f0d8 | 336 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 337 | #endif |
2e53d63a | 338 | |
49846896 RR |
339 | void kvm_reload_remote_mmus(struct kvm *kvm) |
340 | { | |
445b8236 | 341 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 342 | } |
2e53d63a | 343 | |
6926f95a SC |
344 | #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE |
345 | static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc, | |
346 | gfp_t gfp_flags) | |
347 | { | |
348 | gfp_flags |= mc->gfp_zero; | |
349 | ||
350 | if (mc->kmem_cache) | |
351 | return kmem_cache_alloc(mc->kmem_cache, gfp_flags); | |
352 | else | |
353 | return (void *)__get_free_page(gfp_flags); | |
354 | } | |
355 | ||
356 | int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min) | |
357 | { | |
358 | void *obj; | |
359 | ||
360 | if (mc->nobjs >= min) | |
361 | return 0; | |
362 | while (mc->nobjs < ARRAY_SIZE(mc->objects)) { | |
363 | obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT); | |
364 | if (!obj) | |
365 | return mc->nobjs >= min ? 0 : -ENOMEM; | |
366 | mc->objects[mc->nobjs++] = obj; | |
367 | } | |
368 | return 0; | |
369 | } | |
370 | ||
371 | int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc) | |
372 | { | |
373 | return mc->nobjs; | |
374 | } | |
375 | ||
376 | void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
377 | { | |
378 | while (mc->nobjs) { | |
379 | if (mc->kmem_cache) | |
380 | kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]); | |
381 | else | |
382 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
383 | } | |
384 | } | |
385 | ||
386 | void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
387 | { | |
388 | void *p; | |
389 | ||
390 | if (WARN_ON(!mc->nobjs)) | |
391 | p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT); | |
392 | else | |
393 | p = mc->objects[--mc->nobjs]; | |
394 | BUG_ON(!p); | |
395 | return p; | |
396 | } | |
397 | #endif | |
398 | ||
8bd826d6 | 399 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 400 | { |
fb3f0f51 RR |
401 | mutex_init(&vcpu->mutex); |
402 | vcpu->cpu = -1; | |
fb3f0f51 RR |
403 | vcpu->kvm = kvm; |
404 | vcpu->vcpu_id = id; | |
34bb10b7 | 405 | vcpu->pid = NULL; |
da4ad88c | 406 | rcuwait_init(&vcpu->wait); |
af585b92 | 407 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 408 | |
bf9f6ac8 FW |
409 | vcpu->pre_pcpu = -1; |
410 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
411 | ||
4c088493 R |
412 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
413 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 414 | vcpu->preempted = false; |
d73eb57b | 415 | vcpu->ready = false; |
d5c48deb | 416 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
fb3f0f51 | 417 | } |
fb3f0f51 | 418 | |
4543bdc0 SC |
419 | void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
420 | { | |
fb04a1ed | 421 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
4543bdc0 | 422 | kvm_arch_vcpu_destroy(vcpu); |
e529ef66 | 423 | |
9941d224 SC |
424 | /* |
425 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
426 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
427 | * are already gone. | |
428 | */ | |
429 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
430 | ||
8bd826d6 | 431 | free_page((unsigned long)vcpu->run); |
e529ef66 | 432 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 SC |
433 | } |
434 | EXPORT_SYMBOL_GPL(kvm_vcpu_destroy); | |
435 | ||
e930bffe AA |
436 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
437 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
438 | { | |
439 | return container_of(mn, struct kvm, mmu_notifier); | |
440 | } | |
441 | ||
e649b3f0 ET |
442 | static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn, |
443 | struct mm_struct *mm, | |
444 | unsigned long start, unsigned long end) | |
445 | { | |
446 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
447 | int idx; | |
448 | ||
449 | idx = srcu_read_lock(&kvm->srcu); | |
450 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
451 | srcu_read_unlock(&kvm->srcu, idx); | |
452 | } | |
453 | ||
3da0dd43 IE |
454 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
455 | struct mm_struct *mm, | |
456 | unsigned long address, | |
457 | pte_t pte) | |
458 | { | |
459 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 460 | int idx; |
3da0dd43 | 461 | |
bc6678a3 | 462 | idx = srcu_read_lock(&kvm->srcu); |
531810ca BG |
463 | |
464 | KVM_MMU_LOCK(kvm); | |
465 | ||
3da0dd43 | 466 | kvm->mmu_notifier_seq++; |
0cf853c5 LT |
467 | |
468 | if (kvm_set_spte_hva(kvm, address, pte)) | |
469 | kvm_flush_remote_tlbs(kvm); | |
470 | ||
531810ca | 471 | KVM_MMU_UNLOCK(kvm); |
bc6678a3 | 472 | srcu_read_unlock(&kvm->srcu, idx); |
3da0dd43 IE |
473 | } |
474 | ||
93065ac7 | 475 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
5d6527a7 | 476 | const struct mmu_notifier_range *range) |
e930bffe AA |
477 | { |
478 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 479 | int need_tlb_flush = 0, idx; |
e930bffe | 480 | |
bc6678a3 | 481 | idx = srcu_read_lock(&kvm->srcu); |
531810ca | 482 | KVM_MMU_LOCK(kvm); |
e930bffe AA |
483 | /* |
484 | * The count increase must become visible at unlock time as no | |
485 | * spte can be established without taking the mmu_lock and | |
486 | * count is also read inside the mmu_lock critical section. | |
487 | */ | |
488 | kvm->mmu_notifier_count++; | |
fdfe7cbd WD |
489 | need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end, |
490 | range->flags); | |
e930bffe | 491 | /* we've to flush the tlb before the pages can be freed */ |
88bf56d0 | 492 | if (need_tlb_flush || kvm->tlbs_dirty) |
e930bffe | 493 | kvm_flush_remote_tlbs(kvm); |
565f3be2 | 494 | |
531810ca | 495 | KVM_MMU_UNLOCK(kvm); |
565f3be2 | 496 | srcu_read_unlock(&kvm->srcu, idx); |
93065ac7 | 497 | |
e649b3f0 | 498 | return 0; |
e930bffe AA |
499 | } |
500 | ||
501 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
5d6527a7 | 502 | const struct mmu_notifier_range *range) |
e930bffe AA |
503 | { |
504 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
505 | ||
531810ca | 506 | KVM_MMU_LOCK(kvm); |
e930bffe AA |
507 | /* |
508 | * This sequence increase will notify the kvm page fault that | |
509 | * the page that is going to be mapped in the spte could have | |
510 | * been freed. | |
511 | */ | |
512 | kvm->mmu_notifier_seq++; | |
a355aa54 | 513 | smp_wmb(); |
e930bffe AA |
514 | /* |
515 | * The above sequence increase must be visible before the | |
a355aa54 PM |
516 | * below count decrease, which is ensured by the smp_wmb above |
517 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
518 | */ |
519 | kvm->mmu_notifier_count--; | |
531810ca | 520 | KVM_MMU_UNLOCK(kvm); |
e930bffe AA |
521 | |
522 | BUG_ON(kvm->mmu_notifier_count < 0); | |
523 | } | |
524 | ||
525 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
526 | struct mm_struct *mm, | |
57128468 ALC |
527 | unsigned long start, |
528 | unsigned long end) | |
e930bffe AA |
529 | { |
530 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
bc6678a3 | 531 | int young, idx; |
e930bffe | 532 | |
bc6678a3 | 533 | idx = srcu_read_lock(&kvm->srcu); |
531810ca | 534 | KVM_MMU_LOCK(kvm); |
e930bffe | 535 | |
57128468 | 536 | young = kvm_age_hva(kvm, start, end); |
e930bffe AA |
537 | if (young) |
538 | kvm_flush_remote_tlbs(kvm); | |
539 | ||
531810ca | 540 | KVM_MMU_UNLOCK(kvm); |
565f3be2 TY |
541 | srcu_read_unlock(&kvm->srcu, idx); |
542 | ||
e930bffe AA |
543 | return young; |
544 | } | |
545 | ||
1d7715c6 VD |
546 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
547 | struct mm_struct *mm, | |
548 | unsigned long start, | |
549 | unsigned long end) | |
550 | { | |
551 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
552 | int young, idx; | |
553 | ||
554 | idx = srcu_read_lock(&kvm->srcu); | |
531810ca | 555 | KVM_MMU_LOCK(kvm); |
1d7715c6 VD |
556 | /* |
557 | * Even though we do not flush TLB, this will still adversely | |
558 | * affect performance on pre-Haswell Intel EPT, where there is | |
559 | * no EPT Access Bit to clear so that we have to tear down EPT | |
560 | * tables instead. If we find this unacceptable, we can always | |
561 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
562 | * do anything on clear_young. | |
563 | * | |
564 | * Also note that currently we never issue secondary TLB flushes | |
565 | * from clear_young, leaving this job up to the regular system | |
566 | * cadence. If we find this inaccurate, we might come up with a | |
567 | * more sophisticated heuristic later. | |
568 | */ | |
569 | young = kvm_age_hva(kvm, start, end); | |
531810ca | 570 | KVM_MMU_UNLOCK(kvm); |
1d7715c6 VD |
571 | srcu_read_unlock(&kvm->srcu, idx); |
572 | ||
573 | return young; | |
574 | } | |
575 | ||
8ee53820 AA |
576 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
577 | struct mm_struct *mm, | |
578 | unsigned long address) | |
579 | { | |
580 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
581 | int young, idx; | |
582 | ||
583 | idx = srcu_read_lock(&kvm->srcu); | |
531810ca | 584 | KVM_MMU_LOCK(kvm); |
8ee53820 | 585 | young = kvm_test_age_hva(kvm, address); |
531810ca | 586 | KVM_MMU_UNLOCK(kvm); |
8ee53820 AA |
587 | srcu_read_unlock(&kvm->srcu, idx); |
588 | ||
589 | return young; | |
590 | } | |
591 | ||
85db06e5 MT |
592 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
593 | struct mm_struct *mm) | |
594 | { | |
595 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
596 | int idx; |
597 | ||
598 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 599 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 600 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
601 | } |
602 | ||
e930bffe | 603 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e649b3f0 | 604 | .invalidate_range = kvm_mmu_notifier_invalidate_range, |
e930bffe AA |
605 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
606 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
607 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 608 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 609 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 610 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 611 | .release = kvm_mmu_notifier_release, |
e930bffe | 612 | }; |
4c07b0a4 AK |
613 | |
614 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
615 | { | |
616 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
617 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
618 | } | |
619 | ||
620 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
621 | ||
622 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
623 | { | |
624 | return 0; | |
625 | } | |
626 | ||
e930bffe AA |
627 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
628 | ||
a47d2b07 | 629 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
630 | { |
631 | int i; | |
a47d2b07 | 632 | struct kvm_memslots *slots; |
bf3e05bc | 633 | |
b12ce36a | 634 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
a47d2b07 PB |
635 | if (!slots) |
636 | return NULL; | |
637 | ||
bf3e05bc | 638 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
36947254 | 639 | slots->id_to_index[i] = -1; |
a47d2b07 PB |
640 | |
641 | return slots; | |
642 | } | |
643 | ||
644 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
645 | { | |
646 | if (!memslot->dirty_bitmap) | |
647 | return; | |
648 | ||
649 | kvfree(memslot->dirty_bitmap); | |
650 | memslot->dirty_bitmap = NULL; | |
651 | } | |
652 | ||
e96c81ee | 653 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 654 | { |
e96c81ee | 655 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 656 | |
e96c81ee | 657 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 658 | |
e96c81ee SC |
659 | slot->flags = 0; |
660 | slot->npages = 0; | |
a47d2b07 PB |
661 | } |
662 | ||
663 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
664 | { | |
665 | struct kvm_memory_slot *memslot; | |
666 | ||
667 | if (!slots) | |
668 | return; | |
669 | ||
670 | kvm_for_each_memslot(memslot, slots) | |
e96c81ee | 671 | kvm_free_memslot(kvm, memslot); |
a47d2b07 PB |
672 | |
673 | kvfree(slots); | |
bf3e05bc XG |
674 | } |
675 | ||
536a6f88 JF |
676 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
677 | { | |
678 | int i; | |
679 | ||
680 | if (!kvm->debugfs_dentry) | |
681 | return; | |
682 | ||
683 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
684 | ||
9d5a1dce LC |
685 | if (kvm->debugfs_stat_data) { |
686 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
687 | kfree(kvm->debugfs_stat_data[i]); | |
688 | kfree(kvm->debugfs_stat_data); | |
689 | } | |
536a6f88 JF |
690 | } |
691 | ||
692 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
693 | { | |
694 | char dir_name[ITOA_MAX_LEN * 2]; | |
695 | struct kvm_stat_data *stat_data; | |
696 | struct kvm_stats_debugfs_item *p; | |
697 | ||
698 | if (!debugfs_initialized()) | |
699 | return 0; | |
700 | ||
701 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
929f45e3 | 702 | kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir); |
536a6f88 JF |
703 | |
704 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, | |
705 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 706 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
707 | if (!kvm->debugfs_stat_data) |
708 | return -ENOMEM; | |
709 | ||
710 | for (p = debugfs_entries; p->name; p++) { | |
b12ce36a | 711 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
712 | if (!stat_data) |
713 | return -ENOMEM; | |
714 | ||
715 | stat_data->kvm = kvm; | |
09cbcef6 | 716 | stat_data->dbgfs_item = p; |
536a6f88 | 717 | kvm->debugfs_stat_data[p - debugfs_entries] = stat_data; |
09cbcef6 MP |
718 | debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p), |
719 | kvm->debugfs_dentry, stat_data, | |
720 | &stat_fops_per_vm); | |
536a6f88 JF |
721 | } |
722 | return 0; | |
723 | } | |
724 | ||
1aa9b957 JS |
725 | /* |
726 | * Called after the VM is otherwise initialized, but just before adding it to | |
727 | * the vm_list. | |
728 | */ | |
729 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
730 | { | |
731 | return 0; | |
732 | } | |
733 | ||
734 | /* | |
735 | * Called just after removing the VM from the vm_list, but before doing any | |
736 | * other destruction. | |
737 | */ | |
738 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
739 | { | |
740 | } | |
741 | ||
e08b9637 | 742 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 743 | { |
d89f5eff | 744 | struct kvm *kvm = kvm_arch_alloc_vm(); |
9121923c JM |
745 | int r = -ENOMEM; |
746 | int i; | |
6aa8b732 | 747 | |
d89f5eff JK |
748 | if (!kvm) |
749 | return ERR_PTR(-ENOMEM); | |
750 | ||
531810ca | 751 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 752 | mmgrab(current->mm); |
e9ad4ec8 PB |
753 | kvm->mm = current->mm; |
754 | kvm_eventfd_init(kvm); | |
755 | mutex_init(&kvm->lock); | |
756 | mutex_init(&kvm->irq_lock); | |
757 | mutex_init(&kvm->slots_lock); | |
e9ad4ec8 PB |
758 | INIT_LIST_HEAD(&kvm->devices); |
759 | ||
1e702d9a AW |
760 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
761 | ||
8a44119a PB |
762 | if (init_srcu_struct(&kvm->srcu)) |
763 | goto out_err_no_srcu; | |
764 | if (init_srcu_struct(&kvm->irq_srcu)) | |
765 | goto out_err_no_irq_srcu; | |
766 | ||
e2d3fcaf | 767 | refcount_set(&kvm->users_count, 1); |
f481b069 | 768 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4bd518f1 | 769 | struct kvm_memslots *slots = kvm_alloc_memslots(); |
9121923c | 770 | |
4bd518f1 | 771 | if (!slots) |
a97b0e77 | 772 | goto out_err_no_arch_destroy_vm; |
0e32958e | 773 | /* Generations must be different for each address space. */ |
164bf7e5 | 774 | slots->generation = i; |
4bd518f1 | 775 | rcu_assign_pointer(kvm->memslots[i], slots); |
f481b069 | 776 | } |
00f034a1 | 777 | |
e93f8a0f | 778 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 779 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 780 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 781 | if (!kvm->buses[i]) |
a97b0e77 | 782 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 783 | } |
e930bffe | 784 | |
acd05785 DM |
785 | kvm->max_halt_poll_ns = halt_poll_ns; |
786 | ||
e08b9637 | 787 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 788 | if (r) |
a97b0e77 | 789 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
790 | |
791 | r = hardware_enable_all(); | |
792 | if (r) | |
719d93cd | 793 | goto out_err_no_disable; |
10474ae8 | 794 | |
c77dcacb | 795 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 796 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 797 | #endif |
6aa8b732 | 798 | |
74b5c5bf | 799 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
800 | if (r) |
801 | goto out_err_no_mmu_notifier; | |
802 | ||
803 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
804 | if (r) |
805 | goto out_err; | |
806 | ||
0d9ce162 | 807 | mutex_lock(&kvm_lock); |
5e58cfe4 | 808 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 809 | mutex_unlock(&kvm_lock); |
d89f5eff | 810 | |
2ecd9d29 PZ |
811 | preempt_notifier_inc(); |
812 | ||
f17abe9a | 813 | return kvm; |
10474ae8 AG |
814 | |
815 | out_err: | |
1aa9b957 JS |
816 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
817 | if (kvm->mmu_notifier.ops) | |
818 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
819 | #endif | |
820 | out_err_no_mmu_notifier: | |
10474ae8 | 821 | hardware_disable_all(); |
719d93cd | 822 | out_err_no_disable: |
a97b0e77 | 823 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 824 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 825 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 826 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 827 | kfree(kvm_get_bus(kvm, i)); |
f481b069 | 828 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 829 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
8a44119a PB |
830 | cleanup_srcu_struct(&kvm->irq_srcu); |
831 | out_err_no_irq_srcu: | |
832 | cleanup_srcu_struct(&kvm->srcu); | |
833 | out_err_no_srcu: | |
d89f5eff | 834 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 835 | mmdrop(current->mm); |
10474ae8 | 836 | return ERR_PTR(r); |
f17abe9a AK |
837 | } |
838 | ||
07f0a7bd SW |
839 | static void kvm_destroy_devices(struct kvm *kvm) |
840 | { | |
e6e3b5a6 | 841 | struct kvm_device *dev, *tmp; |
07f0a7bd | 842 | |
a28ebea2 CD |
843 | /* |
844 | * We do not need to take the kvm->lock here, because nobody else | |
845 | * has a reference to the struct kvm at this point and therefore | |
846 | * cannot access the devices list anyhow. | |
847 | */ | |
e6e3b5a6 GT |
848 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
849 | list_del(&dev->vm_node); | |
07f0a7bd SW |
850 | dev->ops->destroy(dev); |
851 | } | |
852 | } | |
853 | ||
f17abe9a AK |
854 | static void kvm_destroy_vm(struct kvm *kvm) |
855 | { | |
e93f8a0f | 856 | int i; |
6d4e4c4f AK |
857 | struct mm_struct *mm = kvm->mm; |
858 | ||
286de8f6 | 859 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 860 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 861 | kvm_arch_sync_events(kvm); |
0d9ce162 | 862 | mutex_lock(&kvm_lock); |
133de902 | 863 | list_del(&kvm->vm_list); |
0d9ce162 | 864 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
865 | kvm_arch_pre_destroy_vm(kvm); |
866 | ||
399ec807 | 867 | kvm_free_irq_routing(kvm); |
df630b8c | 868 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 869 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 870 | |
4a12f951 CB |
871 | if (bus) |
872 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
873 | kvm->buses[i] = NULL; |
874 | } | |
980da6ce | 875 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
876 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
877 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 878 | #else |
2df72e9b | 879 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 880 | #endif |
d19a9cd2 | 881 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 882 | kvm_destroy_devices(kvm); |
f481b069 | 883 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 884 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
820b3fcd | 885 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
886 | cleanup_srcu_struct(&kvm->srcu); |
887 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 888 | preempt_notifier_dec(); |
10474ae8 | 889 | hardware_disable_all(); |
6d4e4c4f | 890 | mmdrop(mm); |
f17abe9a AK |
891 | } |
892 | ||
d39f13b0 IE |
893 | void kvm_get_kvm(struct kvm *kvm) |
894 | { | |
e3736c3e | 895 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
896 | } |
897 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
898 | ||
899 | void kvm_put_kvm(struct kvm *kvm) | |
900 | { | |
e3736c3e | 901 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
902 | kvm_destroy_vm(kvm); |
903 | } | |
904 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
905 | ||
149487bd SC |
906 | /* |
907 | * Used to put a reference that was taken on behalf of an object associated | |
908 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
909 | * of the new file descriptor fails and the reference cannot be transferred to | |
910 | * its final owner. In such cases, the caller is still actively using @kvm and | |
911 | * will fail miserably if the refcount unexpectedly hits zero. | |
912 | */ | |
913 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
914 | { | |
915 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
916 | } | |
917 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 918 | |
f17abe9a AK |
919 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
920 | { | |
921 | struct kvm *kvm = filp->private_data; | |
922 | ||
721eecbf GH |
923 | kvm_irqfd_release(kvm); |
924 | ||
d39f13b0 | 925 | kvm_put_kvm(kvm); |
6aa8b732 AK |
926 | return 0; |
927 | } | |
928 | ||
515a0127 TY |
929 | /* |
930 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 931 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 932 | */ |
3c9bd400 | 933 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 934 | { |
515a0127 | 935 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 936 | |
b12ce36a | 937 | memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
938 | if (!memslot->dirty_bitmap) |
939 | return -ENOMEM; | |
940 | ||
a36a57b1 TY |
941 | return 0; |
942 | } | |
943 | ||
bf3e05bc | 944 | /* |
0577d1ab SC |
945 | * Delete a memslot by decrementing the number of used slots and shifting all |
946 | * other entries in the array forward one spot. | |
bf3e05bc | 947 | */ |
0577d1ab SC |
948 | static inline void kvm_memslot_delete(struct kvm_memslots *slots, |
949 | struct kvm_memory_slot *memslot) | |
bf3e05bc | 950 | { |
063584d4 | 951 | struct kvm_memory_slot *mslots = slots->memslots; |
0577d1ab | 952 | int i; |
f85e2cb5 | 953 | |
0577d1ab SC |
954 | if (WARN_ON(slots->id_to_index[memslot->id] == -1)) |
955 | return; | |
0e60b079 | 956 | |
0577d1ab SC |
957 | slots->used_slots--; |
958 | ||
0774a964 SC |
959 | if (atomic_read(&slots->lru_slot) >= slots->used_slots) |
960 | atomic_set(&slots->lru_slot, 0); | |
961 | ||
0577d1ab | 962 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) { |
7f379cff IM |
963 | mslots[i] = mslots[i + 1]; |
964 | slots->id_to_index[mslots[i].id] = i; | |
7f379cff | 965 | } |
0577d1ab SC |
966 | mslots[i] = *memslot; |
967 | slots->id_to_index[memslot->id] = -1; | |
968 | } | |
969 | ||
970 | /* | |
971 | * "Insert" a new memslot by incrementing the number of used slots. Returns | |
972 | * the new slot's initial index into the memslots array. | |
973 | */ | |
974 | static inline int kvm_memslot_insert_back(struct kvm_memslots *slots) | |
975 | { | |
976 | return slots->used_slots++; | |
977 | } | |
978 | ||
979 | /* | |
980 | * Move a changed memslot backwards in the array by shifting existing slots | |
981 | * with a higher GFN toward the front of the array. Note, the changed memslot | |
982 | * itself is not preserved in the array, i.e. not swapped at this time, only | |
983 | * its new index into the array is tracked. Returns the changed memslot's | |
984 | * current index into the memslots array. | |
985 | */ | |
986 | static inline int kvm_memslot_move_backward(struct kvm_memslots *slots, | |
987 | struct kvm_memory_slot *memslot) | |
988 | { | |
989 | struct kvm_memory_slot *mslots = slots->memslots; | |
990 | int i; | |
991 | ||
992 | if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) || | |
993 | WARN_ON_ONCE(!slots->used_slots)) | |
994 | return -1; | |
efbeec70 PB |
995 | |
996 | /* | |
0577d1ab SC |
997 | * Move the target memslot backward in the array by shifting existing |
998 | * memslots with a higher GFN (than the target memslot) towards the | |
999 | * front of the array. | |
efbeec70 | 1000 | */ |
0577d1ab SC |
1001 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) { |
1002 | if (memslot->base_gfn > mslots[i + 1].base_gfn) | |
1003 | break; | |
1004 | ||
1005 | WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn); | |
f85e2cb5 | 1006 | |
0577d1ab SC |
1007 | /* Shift the next memslot forward one and update its index. */ |
1008 | mslots[i] = mslots[i + 1]; | |
1009 | slots->id_to_index[mslots[i].id] = i; | |
1010 | } | |
1011 | return i; | |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * Move a changed memslot forwards in the array by shifting existing slots with | |
1016 | * a lower GFN toward the back of the array. Note, the changed memslot itself | |
1017 | * is not preserved in the array, i.e. not swapped at this time, only its new | |
1018 | * index into the array is tracked. Returns the changed memslot's final index | |
1019 | * into the memslots array. | |
1020 | */ | |
1021 | static inline int kvm_memslot_move_forward(struct kvm_memslots *slots, | |
1022 | struct kvm_memory_slot *memslot, | |
1023 | int start) | |
1024 | { | |
1025 | struct kvm_memory_slot *mslots = slots->memslots; | |
1026 | int i; | |
1027 | ||
1028 | for (i = start; i > 0; i--) { | |
1029 | if (memslot->base_gfn < mslots[i - 1].base_gfn) | |
1030 | break; | |
1031 | ||
1032 | WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn); | |
1033 | ||
1034 | /* Shift the next memslot back one and update its index. */ | |
1035 | mslots[i] = mslots[i - 1]; | |
1036 | slots->id_to_index[mslots[i].id] = i; | |
1037 | } | |
1038 | return i; | |
1039 | } | |
1040 | ||
1041 | /* | |
1042 | * Re-sort memslots based on their GFN to account for an added, deleted, or | |
1043 | * moved memslot. Sorting memslots by GFN allows using a binary search during | |
1044 | * memslot lookup. | |
1045 | * | |
1046 | * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry | |
1047 | * at memslots[0] has the highest GFN. | |
1048 | * | |
1049 | * The sorting algorithm takes advantage of having initially sorted memslots | |
1050 | * and knowing the position of the changed memslot. Sorting is also optimized | |
1051 | * by not swapping the updated memslot and instead only shifting other memslots | |
1052 | * and tracking the new index for the update memslot. Only once its final | |
1053 | * index is known is the updated memslot copied into its position in the array. | |
1054 | * | |
1055 | * - When deleting a memslot, the deleted memslot simply needs to be moved to | |
1056 | * the end of the array. | |
1057 | * | |
1058 | * - When creating a memslot, the algorithm "inserts" the new memslot at the | |
1059 | * end of the array and then it forward to its correct location. | |
1060 | * | |
1061 | * - When moving a memslot, the algorithm first moves the updated memslot | |
1062 | * backward to handle the scenario where the memslot's GFN was changed to a | |
1063 | * lower value. update_memslots() then falls through and runs the same flow | |
1064 | * as creating a memslot to move the memslot forward to handle the scenario | |
1065 | * where its GFN was changed to a higher value. | |
1066 | * | |
1067 | * Note, slots are sorted from highest->lowest instead of lowest->highest for | |
1068 | * historical reasons. Originally, invalid memslots where denoted by having | |
1069 | * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots | |
1070 | * to the end of the array. The current algorithm uses dedicated logic to | |
1071 | * delete a memslot and thus does not rely on invalid memslots having GFN=0. | |
1072 | * | |
1073 | * The other historical motiviation for highest->lowest was to improve the | |
1074 | * performance of memslot lookup. KVM originally used a linear search starting | |
1075 | * at memslots[0]. On x86, the largest memslot usually has one of the highest, | |
1076 | * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a | |
1077 | * single memslot above the 4gb boundary. As the largest memslot is also the | |
1078 | * most likely to be referenced, sorting it to the front of the array was | |
1079 | * advantageous. The current binary search starts from the middle of the array | |
1080 | * and uses an LRU pointer to improve performance for all memslots and GFNs. | |
1081 | */ | |
1082 | static void update_memslots(struct kvm_memslots *slots, | |
1083 | struct kvm_memory_slot *memslot, | |
1084 | enum kvm_mr_change change) | |
1085 | { | |
1086 | int i; | |
1087 | ||
1088 | if (change == KVM_MR_DELETE) { | |
1089 | kvm_memslot_delete(slots, memslot); | |
1090 | } else { | |
1091 | if (change == KVM_MR_CREATE) | |
1092 | i = kvm_memslot_insert_back(slots); | |
1093 | else | |
1094 | i = kvm_memslot_move_backward(slots, memslot); | |
1095 | i = kvm_memslot_move_forward(slots, memslot, i); | |
1096 | ||
1097 | /* | |
1098 | * Copy the memslot to its new position in memslots and update | |
1099 | * its index accordingly. | |
1100 | */ | |
1101 | slots->memslots[i] = *memslot; | |
1102 | slots->id_to_index[memslot->id] = i; | |
1103 | } | |
bf3e05bc XG |
1104 | } |
1105 | ||
09170a49 | 1106 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 1107 | { |
4d8b81ab XG |
1108 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1109 | ||
0f8a4de3 | 1110 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
1111 | valid_flags |= KVM_MEM_READONLY; |
1112 | #endif | |
1113 | ||
1114 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1115 | return -EINVAL; |
1116 | ||
1117 | return 0; | |
1118 | } | |
1119 | ||
7ec4fb44 | 1120 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 1121 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 1122 | { |
f481b069 | 1123 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
361209e0 | 1124 | u64 gen = old_memslots->generation; |
7ec4fb44 | 1125 | |
361209e0 SC |
1126 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1127 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1128 | |
f481b069 | 1129 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
7ec4fb44 | 1130 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1131 | |
ee3d1570 | 1132 | /* |
361209e0 | 1133 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1134 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1135 | * the number of address spaces. This provides a unique and easily |
1136 | * identifiable generation number while the memslots are in flux. | |
1137 | */ | |
1138 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1139 | ||
1140 | /* | |
4bd518f1 PB |
1141 | * Generations must be unique even across address spaces. We do not need |
1142 | * a global counter for that, instead the generation space is evenly split | |
1143 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1144 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1145 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1146 | */ |
164bf7e5 | 1147 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 1148 | |
15248258 | 1149 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1150 | |
15248258 | 1151 | slots->generation = gen; |
e59dbe09 TY |
1152 | |
1153 | return old_memslots; | |
7ec4fb44 GN |
1154 | } |
1155 | ||
36947254 SC |
1156 | /* |
1157 | * Note, at a minimum, the current number of used slots must be allocated, even | |
1158 | * when deleting a memslot, as we need a complete duplicate of the memslots for | |
1159 | * use when invalidating a memslot prior to deleting/moving the memslot. | |
1160 | */ | |
1161 | static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old, | |
1162 | enum kvm_mr_change change) | |
1163 | { | |
1164 | struct kvm_memslots *slots; | |
1165 | size_t old_size, new_size; | |
1166 | ||
1167 | old_size = sizeof(struct kvm_memslots) + | |
1168 | (sizeof(struct kvm_memory_slot) * old->used_slots); | |
1169 | ||
1170 | if (change == KVM_MR_CREATE) | |
1171 | new_size = old_size + sizeof(struct kvm_memory_slot); | |
1172 | else | |
1173 | new_size = old_size; | |
1174 | ||
1175 | slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT); | |
1176 | if (likely(slots)) | |
1177 | memcpy(slots, old, old_size); | |
1178 | ||
1179 | return slots; | |
1180 | } | |
1181 | ||
cf47f50b SC |
1182 | static int kvm_set_memslot(struct kvm *kvm, |
1183 | const struct kvm_userspace_memory_region *mem, | |
9d4c197c | 1184 | struct kvm_memory_slot *old, |
cf47f50b SC |
1185 | struct kvm_memory_slot *new, int as_id, |
1186 | enum kvm_mr_change change) | |
1187 | { | |
1188 | struct kvm_memory_slot *slot; | |
1189 | struct kvm_memslots *slots; | |
1190 | int r; | |
1191 | ||
36947254 | 1192 | slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change); |
cf47f50b SC |
1193 | if (!slots) |
1194 | return -ENOMEM; | |
cf47f50b SC |
1195 | |
1196 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { | |
1197 | /* | |
1198 | * Note, the INVALID flag needs to be in the appropriate entry | |
1199 | * in the freshly allocated memslots, not in @old or @new. | |
1200 | */ | |
1201 | slot = id_to_memslot(slots, old->id); | |
1202 | slot->flags |= KVM_MEMSLOT_INVALID; | |
1203 | ||
1204 | /* | |
1205 | * We can re-use the old memslots, the only difference from the | |
1206 | * newly installed memslots is the invalid flag, which will get | |
1207 | * dropped by update_memslots anyway. We'll also revert to the | |
1208 | * old memslots if preparing the new memory region fails. | |
1209 | */ | |
1210 | slots = install_new_memslots(kvm, as_id, slots); | |
1211 | ||
1212 | /* From this point no new shadow pages pointing to a deleted, | |
1213 | * or moved, memslot will be created. | |
1214 | * | |
1215 | * validation of sp->gfn happens in: | |
1216 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1217 | * - kvm_is_visible_gfn (mmu_check_root) | |
1218 | */ | |
1219 | kvm_arch_flush_shadow_memslot(kvm, slot); | |
1220 | } | |
1221 | ||
1222 | r = kvm_arch_prepare_memory_region(kvm, new, mem, change); | |
1223 | if (r) | |
1224 | goto out_slots; | |
1225 | ||
1226 | update_memslots(slots, new, change); | |
1227 | slots = install_new_memslots(kvm, as_id, slots); | |
1228 | ||
1229 | kvm_arch_commit_memory_region(kvm, mem, old, new, change); | |
1230 | ||
1231 | kvfree(slots); | |
1232 | return 0; | |
1233 | ||
1234 | out_slots: | |
1235 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) | |
1236 | slots = install_new_memslots(kvm, as_id, slots); | |
1237 | kvfree(slots); | |
1238 | return r; | |
1239 | } | |
1240 | ||
5c0b4f3d SC |
1241 | static int kvm_delete_memslot(struct kvm *kvm, |
1242 | const struct kvm_userspace_memory_region *mem, | |
1243 | struct kvm_memory_slot *old, int as_id) | |
1244 | { | |
1245 | struct kvm_memory_slot new; | |
1246 | int r; | |
1247 | ||
1248 | if (!old->npages) | |
1249 | return -EINVAL; | |
1250 | ||
1251 | memset(&new, 0, sizeof(new)); | |
1252 | new.id = old->id; | |
9e9eb226 PX |
1253 | /* |
1254 | * This is only for debugging purpose; it should never be referenced | |
1255 | * for a removed memslot. | |
1256 | */ | |
1257 | new.as_id = as_id; | |
5c0b4f3d SC |
1258 | |
1259 | r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE); | |
1260 | if (r) | |
1261 | return r; | |
1262 | ||
e96c81ee | 1263 | kvm_free_memslot(kvm, old); |
5c0b4f3d SC |
1264 | return 0; |
1265 | } | |
1266 | ||
6aa8b732 AK |
1267 | /* |
1268 | * Allocate some memory and give it an address in the guest physical address | |
1269 | * space. | |
1270 | * | |
1271 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1272 | * |
02d5d55b | 1273 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1274 | */ |
f78e0e2e | 1275 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 1276 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 1277 | { |
6aa8b732 | 1278 | struct kvm_memory_slot old, new; |
163da372 | 1279 | struct kvm_memory_slot *tmp; |
f64c0398 | 1280 | enum kvm_mr_change change; |
163da372 SC |
1281 | int as_id, id; |
1282 | int r; | |
6aa8b732 | 1283 | |
a50d64d6 XG |
1284 | r = check_memory_region_flags(mem); |
1285 | if (r) | |
71a4c30b | 1286 | return r; |
a50d64d6 | 1287 | |
f481b069 PB |
1288 | as_id = mem->slot >> 16; |
1289 | id = (u16)mem->slot; | |
1290 | ||
6aa8b732 AK |
1291 | /* General sanity checks */ |
1292 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
71a4c30b | 1293 | return -EINVAL; |
6aa8b732 | 1294 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 1295 | return -EINVAL; |
fa3d315a | 1296 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 1297 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 1298 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 1299 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 1300 | mem->memory_size)) |
71a4c30b | 1301 | return -EINVAL; |
f481b069 | 1302 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 1303 | return -EINVAL; |
6aa8b732 | 1304 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 1305 | return -EINVAL; |
6aa8b732 | 1306 | |
5c0b4f3d SC |
1307 | /* |
1308 | * Make a full copy of the old memslot, the pointer will become stale | |
1309 | * when the memslots are re-sorted by update_memslots(), and the old | |
1310 | * memslot needs to be referenced after calling update_memslots(), e.g. | |
0dff0846 | 1311 | * to free its resources and for arch specific behavior. |
5c0b4f3d | 1312 | */ |
0577d1ab SC |
1313 | tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
1314 | if (tmp) { | |
1315 | old = *tmp; | |
1316 | tmp = NULL; | |
1317 | } else { | |
1318 | memset(&old, 0, sizeof(old)); | |
1319 | old.id = id; | |
1320 | } | |
163da372 | 1321 | |
5c0b4f3d SC |
1322 | if (!mem->memory_size) |
1323 | return kvm_delete_memslot(kvm, mem, &old, as_id); | |
1324 | ||
9e9eb226 | 1325 | new.as_id = as_id; |
f481b069 | 1326 | new.id = id; |
163da372 SC |
1327 | new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
1328 | new.npages = mem->memory_size >> PAGE_SHIFT; | |
6aa8b732 | 1329 | new.flags = mem->flags; |
414de7ab | 1330 | new.userspace_addr = mem->userspace_addr; |
6aa8b732 | 1331 | |
163da372 SC |
1332 | if (new.npages > KVM_MEM_MAX_NR_PAGES) |
1333 | return -EINVAL; | |
1334 | ||
5c0b4f3d SC |
1335 | if (!old.npages) { |
1336 | change = KVM_MR_CREATE; | |
163da372 SC |
1337 | new.dirty_bitmap = NULL; |
1338 | memset(&new.arch, 0, sizeof(new.arch)); | |
5c0b4f3d SC |
1339 | } else { /* Modify an existing slot. */ |
1340 | if ((new.userspace_addr != old.userspace_addr) || | |
163da372 | 1341 | (new.npages != old.npages) || |
5c0b4f3d | 1342 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) |
71a4c30b | 1343 | return -EINVAL; |
09170a49 | 1344 | |
163da372 | 1345 | if (new.base_gfn != old.base_gfn) |
5c0b4f3d SC |
1346 | change = KVM_MR_MOVE; |
1347 | else if (new.flags != old.flags) | |
1348 | change = KVM_MR_FLAGS_ONLY; | |
1349 | else /* Nothing to change. */ | |
1350 | return 0; | |
163da372 SC |
1351 | |
1352 | /* Copy dirty_bitmap and arch from the current memslot. */ | |
1353 | new.dirty_bitmap = old.dirty_bitmap; | |
1354 | memcpy(&new.arch, &old.arch, sizeof(new.arch)); | |
09170a49 | 1355 | } |
6aa8b732 | 1356 | |
f64c0398 | 1357 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee | 1358 | /* Check for overlaps */ |
163da372 SC |
1359 | kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) { |
1360 | if (tmp->id == id) | |
0a706bee | 1361 | continue; |
163da372 SC |
1362 | if (!((new.base_gfn + new.npages <= tmp->base_gfn) || |
1363 | (new.base_gfn >= tmp->base_gfn + tmp->npages))) | |
71a4c30b | 1364 | return -EEXIST; |
0a706bee | 1365 | } |
6aa8b732 | 1366 | } |
6aa8b732 | 1367 | |
414de7ab SC |
1368 | /* Allocate/free page dirty bitmap as needed */ |
1369 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1370 | new.dirty_bitmap = NULL; | |
044c59c4 | 1371 | else if (!new.dirty_bitmap && !kvm->dirty_ring_size) { |
3c9bd400 | 1372 | r = kvm_alloc_dirty_bitmap(&new); |
71a4c30b SC |
1373 | if (r) |
1374 | return r; | |
3c9bd400 JZ |
1375 | |
1376 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1377 | bitmap_set(new.dirty_bitmap, 0, new.npages); | |
6aa8b732 AK |
1378 | } |
1379 | ||
cf47f50b SC |
1380 | r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change); |
1381 | if (r) | |
1382 | goto out_bitmap; | |
82ce2c96 | 1383 | |
5c0b4f3d SC |
1384 | if (old.dirty_bitmap && !new.dirty_bitmap) |
1385 | kvm_destroy_dirty_bitmap(&old); | |
6aa8b732 AK |
1386 | return 0; |
1387 | ||
bd0e96fd SC |
1388 | out_bitmap: |
1389 | if (new.dirty_bitmap && !old.dirty_bitmap) | |
1390 | kvm_destroy_dirty_bitmap(&new); | |
6aa8b732 | 1391 | return r; |
210c7c4d | 1392 | } |
f78e0e2e SY |
1393 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1394 | ||
1395 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1396 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1397 | { |
1398 | int r; | |
1399 | ||
79fac95e | 1400 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1401 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1402 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1403 | return r; |
1404 | } | |
210c7c4d IE |
1405 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1406 | ||
7940876e SH |
1407 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1408 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1409 | { |
f481b069 | 1410 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1411 | return -EINVAL; |
09170a49 | 1412 | |
47ae31e2 | 1413 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1414 | } |
1415 | ||
0dff0846 | 1416 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
1417 | /** |
1418 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
1419 | * @kvm: pointer to kvm instance | |
1420 | * @log: slot id and address to which we copy the log | |
1421 | * @is_dirty: set to '1' if any dirty pages were found | |
1422 | * @memslot: set to the associated memslot, always valid on success | |
1423 | */ | |
1424 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
1425 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 1426 | { |
9f6b8029 | 1427 | struct kvm_memslots *slots; |
843574a3 | 1428 | int i, as_id, id; |
87bf6e7d | 1429 | unsigned long n; |
6aa8b732 AK |
1430 | unsigned long any = 0; |
1431 | ||
b2cc64c4 PX |
1432 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1433 | if (kvm->dirty_ring_size) | |
1434 | return -ENXIO; | |
1435 | ||
2a49f61d SC |
1436 | *memslot = NULL; |
1437 | *is_dirty = 0; | |
1438 | ||
f481b069 PB |
1439 | as_id = log->slot >> 16; |
1440 | id = (u16)log->slot; | |
1441 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 1442 | return -EINVAL; |
6aa8b732 | 1443 | |
f481b069 | 1444 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 1445 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 1446 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 1447 | return -ENOENT; |
6aa8b732 | 1448 | |
2a49f61d SC |
1449 | kvm_arch_sync_dirty_log(kvm, *memslot); |
1450 | ||
1451 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 1452 | |
cd1a4a98 | 1453 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 1454 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 1455 | |
2a49f61d | 1456 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 1457 | return -EFAULT; |
6aa8b732 | 1458 | |
5bb064dc ZX |
1459 | if (any) |
1460 | *is_dirty = 1; | |
843574a3 | 1461 | return 0; |
6aa8b732 | 1462 | } |
2ba9f0d8 | 1463 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1464 | |
0dff0846 | 1465 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 1466 | /** |
b8b00220 | 1467 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 1468 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
1469 | * @kvm: pointer to kvm instance |
1470 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
1471 | * |
1472 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1473 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1474 | * following order: | |
1475 | * | |
1476 | * 1. Take a snapshot of the bit and clear it if needed. | |
1477 | * 2. Write protect the corresponding page. | |
1478 | * 3. Copy the snapshot to the userspace. | |
1479 | * 4. Upon return caller flushes TLB's if needed. | |
1480 | * | |
1481 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1482 | * entry. This is not a problem because the page is reported dirty using | |
1483 | * the snapshot taken before and step 4 ensures that writes done after | |
1484 | * exiting to userspace will be logged for the next call. | |
1485 | * | |
1486 | */ | |
0dff0846 | 1487 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 1488 | { |
9f6b8029 | 1489 | struct kvm_memslots *slots; |
ba0513b5 | 1490 | struct kvm_memory_slot *memslot; |
58d6db34 | 1491 | int i, as_id, id; |
ba0513b5 MS |
1492 | unsigned long n; |
1493 | unsigned long *dirty_bitmap; | |
1494 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 1495 | bool flush; |
ba0513b5 | 1496 | |
b2cc64c4 PX |
1497 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1498 | if (kvm->dirty_ring_size) | |
1499 | return -ENXIO; | |
1500 | ||
f481b069 PB |
1501 | as_id = log->slot >> 16; |
1502 | id = (u16)log->slot; | |
1503 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 1504 | return -EINVAL; |
ba0513b5 | 1505 | |
f481b069 PB |
1506 | slots = __kvm_memslots(kvm, as_id); |
1507 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
1508 | if (!memslot || !memslot->dirty_bitmap) |
1509 | return -ENOENT; | |
ba0513b5 MS |
1510 | |
1511 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 1512 | |
0dff0846 SC |
1513 | kvm_arch_sync_dirty_log(kvm, memslot); |
1514 | ||
ba0513b5 | 1515 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 1516 | flush = false; |
2a31b9db PB |
1517 | if (kvm->manual_dirty_log_protect) { |
1518 | /* | |
1519 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
1520 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
1521 | * is some code duplication between this function and | |
1522 | * kvm_get_dirty_log, but hopefully all architecture | |
1523 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
1524 | * can be eliminated. | |
1525 | */ | |
1526 | dirty_bitmap_buffer = dirty_bitmap; | |
1527 | } else { | |
1528 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
1529 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 1530 | |
531810ca | 1531 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
1532 | for (i = 0; i < n / sizeof(long); i++) { |
1533 | unsigned long mask; | |
1534 | gfn_t offset; | |
ba0513b5 | 1535 | |
2a31b9db PB |
1536 | if (!dirty_bitmap[i]) |
1537 | continue; | |
1538 | ||
0dff0846 | 1539 | flush = true; |
2a31b9db PB |
1540 | mask = xchg(&dirty_bitmap[i], 0); |
1541 | dirty_bitmap_buffer[i] = mask; | |
1542 | ||
a67794ca LT |
1543 | offset = i * BITS_PER_LONG; |
1544 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1545 | offset, mask); | |
2a31b9db | 1546 | } |
531810ca | 1547 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
1548 | } |
1549 | ||
0dff0846 SC |
1550 | if (flush) |
1551 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
1552 | ||
2a31b9db PB |
1553 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
1554 | return -EFAULT; | |
1555 | return 0; | |
1556 | } | |
0dff0846 SC |
1557 | |
1558 | ||
1559 | /** | |
1560 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
1561 | * @kvm: kvm instance | |
1562 | * @log: slot id and address to which we copy the log | |
1563 | * | |
1564 | * Steps 1-4 below provide general overview of dirty page logging. See | |
1565 | * kvm_get_dirty_log_protect() function description for additional details. | |
1566 | * | |
1567 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
1568 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
1569 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
1570 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
1571 | * writes will be marked dirty for next log read. | |
1572 | * | |
1573 | * 1. Take a snapshot of the bit and clear it if needed. | |
1574 | * 2. Write protect the corresponding page. | |
1575 | * 3. Copy the snapshot to the userspace. | |
1576 | * 4. Flush TLB's if needed. | |
1577 | */ | |
1578 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
1579 | struct kvm_dirty_log *log) | |
1580 | { | |
1581 | int r; | |
1582 | ||
1583 | mutex_lock(&kvm->slots_lock); | |
1584 | ||
1585 | r = kvm_get_dirty_log_protect(kvm, log); | |
1586 | ||
1587 | mutex_unlock(&kvm->slots_lock); | |
1588 | return r; | |
1589 | } | |
2a31b9db PB |
1590 | |
1591 | /** | |
1592 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
1593 | * and reenable dirty page tracking for the corresponding pages. | |
1594 | * @kvm: pointer to kvm instance | |
1595 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
1596 | */ | |
0dff0846 SC |
1597 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
1598 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
1599 | { |
1600 | struct kvm_memslots *slots; | |
1601 | struct kvm_memory_slot *memslot; | |
98938aa8 | 1602 | int as_id, id; |
2a31b9db | 1603 | gfn_t offset; |
98938aa8 | 1604 | unsigned long i, n; |
2a31b9db PB |
1605 | unsigned long *dirty_bitmap; |
1606 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 1607 | bool flush; |
2a31b9db | 1608 | |
b2cc64c4 PX |
1609 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1610 | if (kvm->dirty_ring_size) | |
1611 | return -ENXIO; | |
1612 | ||
2a31b9db PB |
1613 | as_id = log->slot >> 16; |
1614 | id = (u16)log->slot; | |
1615 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
1616 | return -EINVAL; | |
1617 | ||
76d58e0f | 1618 | if (log->first_page & 63) |
2a31b9db PB |
1619 | return -EINVAL; |
1620 | ||
1621 | slots = __kvm_memslots(kvm, as_id); | |
1622 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
1623 | if (!memslot || !memslot->dirty_bitmap) |
1624 | return -ENOENT; | |
2a31b9db PB |
1625 | |
1626 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 1627 | |
4ddc9204 | 1628 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
1629 | |
1630 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
1631 | log->num_pages > memslot->npages - log->first_page || |
1632 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
1633 | return -EINVAL; | |
98938aa8 | 1634 | |
0dff0846 SC |
1635 | kvm_arch_sync_dirty_log(kvm, memslot); |
1636 | ||
1637 | flush = false; | |
2a31b9db PB |
1638 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
1639 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
1640 | return -EFAULT; | |
ba0513b5 | 1641 | |
531810ca | 1642 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
1643 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
1644 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
1645 | i++, offset += BITS_PER_LONG) { |
1646 | unsigned long mask = *dirty_bitmap_buffer++; | |
1647 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
1648 | if (!mask) | |
ba0513b5 MS |
1649 | continue; |
1650 | ||
2a31b9db | 1651 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 1652 | |
2a31b9db PB |
1653 | /* |
1654 | * mask contains the bits that really have been cleared. This | |
1655 | * never includes any bits beyond the length of the memslot (if | |
1656 | * the length is not aligned to 64 pages), therefore it is not | |
1657 | * a problem if userspace sets them in log->dirty_bitmap. | |
1658 | */ | |
58d2930f | 1659 | if (mask) { |
0dff0846 | 1660 | flush = true; |
58d2930f TY |
1661 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
1662 | offset, mask); | |
1663 | } | |
ba0513b5 | 1664 | } |
531810ca | 1665 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 1666 | |
0dff0846 SC |
1667 | if (flush) |
1668 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
1669 | ||
58d6db34 | 1670 | return 0; |
ba0513b5 | 1671 | } |
0dff0846 SC |
1672 | |
1673 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
1674 | struct kvm_clear_dirty_log *log) | |
1675 | { | |
1676 | int r; | |
1677 | ||
1678 | mutex_lock(&kvm->slots_lock); | |
1679 | ||
1680 | r = kvm_clear_dirty_log_protect(kvm, log); | |
1681 | ||
1682 | mutex_unlock(&kvm->slots_lock); | |
1683 | return r; | |
1684 | } | |
1685 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 1686 | |
49c7754c GN |
1687 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1688 | { | |
1689 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1690 | } | |
a1f4d395 | 1691 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1692 | |
8e73485c PB |
1693 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
1694 | { | |
1695 | return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn); | |
1696 | } | |
e72436bc | 1697 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot); |
8e73485c | 1698 | |
33e94154 | 1699 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 1700 | { |
bf3e05bc | 1701 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1702 | |
c36b7150 | 1703 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
1704 | } |
1705 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1706 | ||
995decb6 VK |
1707 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
1708 | { | |
1709 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1710 | ||
1711 | return kvm_is_visible_memslot(memslot); | |
1712 | } | |
1713 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
1714 | ||
f9b84e19 | 1715 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
1716 | { |
1717 | struct vm_area_struct *vma; | |
1718 | unsigned long addr, size; | |
1719 | ||
1720 | size = PAGE_SIZE; | |
1721 | ||
42cde48b | 1722 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
1723 | if (kvm_is_error_hva(addr)) |
1724 | return PAGE_SIZE; | |
1725 | ||
d8ed45c5 | 1726 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
1727 | vma = find_vma(current->mm, addr); |
1728 | if (!vma) | |
1729 | goto out; | |
1730 | ||
1731 | size = vma_kernel_pagesize(vma); | |
1732 | ||
1733 | out: | |
d8ed45c5 | 1734 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
1735 | |
1736 | return size; | |
1737 | } | |
1738 | ||
4d8b81ab XG |
1739 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
1740 | { | |
1741 | return slot->flags & KVM_MEM_READONLY; | |
1742 | } | |
1743 | ||
4d8b81ab XG |
1744 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1745 | gfn_t *nr_pages, bool write) | |
539cb660 | 1746 | { |
bc6678a3 | 1747 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 1748 | return KVM_HVA_ERR_BAD; |
48987781 | 1749 | |
4d8b81ab XG |
1750 | if (memslot_is_readonly(slot) && write) |
1751 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
1752 | |
1753 | if (nr_pages) | |
1754 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
1755 | ||
4d8b81ab | 1756 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 1757 | } |
48987781 | 1758 | |
4d8b81ab XG |
1759 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
1760 | gfn_t *nr_pages) | |
1761 | { | |
1762 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 1763 | } |
48987781 | 1764 | |
4d8b81ab | 1765 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 1766 | gfn_t gfn) |
4d8b81ab XG |
1767 | { |
1768 | return gfn_to_hva_many(slot, gfn, NULL); | |
1769 | } | |
1770 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
1771 | ||
48987781 XG |
1772 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
1773 | { | |
49c7754c | 1774 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 1775 | } |
0d150298 | 1776 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 1777 | |
8e73485c PB |
1778 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
1779 | { | |
1780 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
1781 | } | |
1782 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
1783 | ||
86ab8cff | 1784 | /* |
970c0d4b WY |
1785 | * Return the hva of a @gfn and the R/W attribute if possible. |
1786 | * | |
1787 | * @slot: the kvm_memory_slot which contains @gfn | |
1788 | * @gfn: the gfn to be translated | |
1789 | * @writable: used to return the read/write attribute of the @slot if the hva | |
1790 | * is valid and @writable is not NULL | |
86ab8cff | 1791 | */ |
64d83126 CD |
1792 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
1793 | gfn_t gfn, bool *writable) | |
86ab8cff | 1794 | { |
a2ac07fe GN |
1795 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
1796 | ||
1797 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
1798 | *writable = !memslot_is_readonly(slot); |
1799 | ||
a2ac07fe | 1800 | return hva; |
86ab8cff XG |
1801 | } |
1802 | ||
64d83126 CD |
1803 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
1804 | { | |
1805 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
1806 | ||
1807 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1808 | } | |
1809 | ||
8e73485c PB |
1810 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
1811 | { | |
1812 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1813 | ||
1814 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
1815 | } | |
1816 | ||
fafc3dba HY |
1817 | static inline int check_user_page_hwpoison(unsigned long addr) |
1818 | { | |
0d731759 | 1819 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 1820 | |
0d731759 | 1821 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
1822 | return rc == -EHWPOISON; |
1823 | } | |
1824 | ||
2fc84311 | 1825 | /* |
b9b33da2 PB |
1826 | * The fast path to get the writable pfn which will be stored in @pfn, |
1827 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 1828 | * only part that runs if we can in atomic context. |
2fc84311 | 1829 | */ |
b9b33da2 PB |
1830 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
1831 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 1832 | { |
8d4e1288 | 1833 | struct page *page[1]; |
954bbbc2 | 1834 | |
12ce13fe XG |
1835 | /* |
1836 | * Fast pin a writable pfn only if it is a write fault request | |
1837 | * or the caller allows to map a writable pfn for a read fault | |
1838 | * request. | |
1839 | */ | |
1840 | if (!(write_fault || writable)) | |
1841 | return false; | |
612819c3 | 1842 | |
dadbb612 | 1843 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 1844 | *pfn = page_to_pfn(page[0]); |
612819c3 | 1845 | |
2fc84311 XG |
1846 | if (writable) |
1847 | *writable = true; | |
1848 | return true; | |
1849 | } | |
af585b92 | 1850 | |
2fc84311 XG |
1851 | return false; |
1852 | } | |
612819c3 | 1853 | |
2fc84311 XG |
1854 | /* |
1855 | * The slow path to get the pfn of the specified host virtual address, | |
1856 | * 1 indicates success, -errno is returned if error is detected. | |
1857 | */ | |
1858 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 1859 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 1860 | { |
ce53053c AV |
1861 | unsigned int flags = FOLL_HWPOISON; |
1862 | struct page *page; | |
2fc84311 | 1863 | int npages = 0; |
612819c3 | 1864 | |
2fc84311 XG |
1865 | might_sleep(); |
1866 | ||
1867 | if (writable) | |
1868 | *writable = write_fault; | |
1869 | ||
ce53053c AV |
1870 | if (write_fault) |
1871 | flags |= FOLL_WRITE; | |
1872 | if (async) | |
1873 | flags |= FOLL_NOWAIT; | |
d4944b0e | 1874 | |
ce53053c | 1875 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
1876 | if (npages != 1) |
1877 | return npages; | |
1878 | ||
1879 | /* map read fault as writable if possible */ | |
12ce13fe | 1880 | if (unlikely(!write_fault) && writable) { |
ce53053c | 1881 | struct page *wpage; |
2fc84311 | 1882 | |
dadbb612 | 1883 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 1884 | *writable = true; |
ce53053c AV |
1885 | put_page(page); |
1886 | page = wpage; | |
612819c3 | 1887 | } |
887c08ac | 1888 | } |
ce53053c | 1889 | *pfn = page_to_pfn(page); |
2fc84311 XG |
1890 | return npages; |
1891 | } | |
539cb660 | 1892 | |
4d8b81ab XG |
1893 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
1894 | { | |
1895 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
1896 | return false; | |
2e2e3738 | 1897 | |
4d8b81ab XG |
1898 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
1899 | return false; | |
887c08ac | 1900 | |
4d8b81ab XG |
1901 | return true; |
1902 | } | |
bf998156 | 1903 | |
92176a8e PB |
1904 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1905 | unsigned long addr, bool *async, | |
a340b3e2 KA |
1906 | bool write_fault, bool *writable, |
1907 | kvm_pfn_t *p_pfn) | |
92176a8e | 1908 | { |
add6a0cd | 1909 | unsigned long pfn; |
bd2fae8d PB |
1910 | pte_t *ptep; |
1911 | spinlock_t *ptl; | |
add6a0cd PB |
1912 | int r; |
1913 | ||
bd2fae8d | 1914 | r = follow_pte(vma->vm_mm, addr, NULL, &ptep, NULL, &ptl); |
add6a0cd PB |
1915 | if (r) { |
1916 | /* | |
1917 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
1918 | * not call the fault handler, so do it here. | |
1919 | */ | |
1920 | bool unlocked = false; | |
64019a2e | 1921 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
1922 | (write_fault ? FAULT_FLAG_WRITE : 0), |
1923 | &unlocked); | |
a8387d0b PB |
1924 | if (unlocked) |
1925 | return -EAGAIN; | |
add6a0cd PB |
1926 | if (r) |
1927 | return r; | |
1928 | ||
bd2fae8d | 1929 | r = follow_pte(vma->vm_mm, addr, NULL, &ptep, NULL, &ptl); |
add6a0cd PB |
1930 | if (r) |
1931 | return r; | |
bd2fae8d | 1932 | } |
add6a0cd | 1933 | |
bd2fae8d PB |
1934 | if (write_fault && !pte_write(*ptep)) { |
1935 | pfn = KVM_PFN_ERR_RO_FAULT; | |
1936 | goto out; | |
add6a0cd PB |
1937 | } |
1938 | ||
a340b3e2 | 1939 | if (writable) |
bd2fae8d PB |
1940 | *writable = pte_write(*ptep); |
1941 | pfn = pte_pfn(*ptep); | |
add6a0cd PB |
1942 | |
1943 | /* | |
1944 | * Get a reference here because callers of *hva_to_pfn* and | |
1945 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
1946 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
1947 | * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will | |
1948 | * simply do nothing for reserved pfns. | |
1949 | * | |
1950 | * Whoever called remap_pfn_range is also going to call e.g. | |
1951 | * unmap_mapping_range before the underlying pages are freed, | |
1952 | * causing a call to our MMU notifier. | |
1953 | */ | |
1954 | kvm_get_pfn(pfn); | |
1955 | ||
bd2fae8d PB |
1956 | out: |
1957 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 1958 | *p_pfn = pfn; |
92176a8e PB |
1959 | return 0; |
1960 | } | |
1961 | ||
12ce13fe XG |
1962 | /* |
1963 | * Pin guest page in memory and return its pfn. | |
1964 | * @addr: host virtual address which maps memory to the guest | |
1965 | * @atomic: whether this function can sleep | |
1966 | * @async: whether this function need to wait IO complete if the | |
1967 | * host page is not in the memory | |
1968 | * @write_fault: whether we should get a writable host page | |
1969 | * @writable: whether it allows to map a writable host page for !@write_fault | |
1970 | * | |
1971 | * The function will map a writable host page for these two cases: | |
1972 | * 1): @write_fault = true | |
1973 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
1974 | * whether the mapping is writable. | |
1975 | */ | |
ba049e93 | 1976 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
1977 | bool write_fault, bool *writable) |
1978 | { | |
1979 | struct vm_area_struct *vma; | |
ba049e93 | 1980 | kvm_pfn_t pfn = 0; |
92176a8e | 1981 | int npages, r; |
2e2e3738 | 1982 | |
2fc84311 XG |
1983 | /* we can do it either atomically or asynchronously, not both */ |
1984 | BUG_ON(atomic && async); | |
8d4e1288 | 1985 | |
b9b33da2 | 1986 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
1987 | return pfn; |
1988 | ||
1989 | if (atomic) | |
1990 | return KVM_PFN_ERR_FAULT; | |
1991 | ||
1992 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
1993 | if (npages == 1) | |
1994 | return pfn; | |
8d4e1288 | 1995 | |
d8ed45c5 | 1996 | mmap_read_lock(current->mm); |
2fc84311 XG |
1997 | if (npages == -EHWPOISON || |
1998 | (!async && check_user_page_hwpoison(addr))) { | |
1999 | pfn = KVM_PFN_ERR_HWPOISON; | |
2000 | goto exit; | |
2001 | } | |
2002 | ||
a8387d0b | 2003 | retry: |
2fc84311 XG |
2004 | vma = find_vma_intersection(current->mm, addr, addr + 1); |
2005 | ||
2006 | if (vma == NULL) | |
2007 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 2008 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
a340b3e2 | 2009 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn); |
a8387d0b PB |
2010 | if (r == -EAGAIN) |
2011 | goto retry; | |
92176a8e PB |
2012 | if (r < 0) |
2013 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 2014 | } else { |
4d8b81ab | 2015 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
2016 | *async = true; |
2017 | pfn = KVM_PFN_ERR_FAULT; | |
2018 | } | |
2019 | exit: | |
d8ed45c5 | 2020 | mmap_read_unlock(current->mm); |
2e2e3738 | 2021 | return pfn; |
35149e21 AL |
2022 | } |
2023 | ||
ba049e93 DW |
2024 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
2025 | bool atomic, bool *async, bool write_fault, | |
2026 | bool *writable) | |
887c08ac | 2027 | { |
4d8b81ab XG |
2028 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
2029 | ||
b2740d35 PB |
2030 | if (addr == KVM_HVA_ERR_RO_BAD) { |
2031 | if (writable) | |
2032 | *writable = false; | |
4d8b81ab | 2033 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 2034 | } |
4d8b81ab | 2035 | |
b2740d35 PB |
2036 | if (kvm_is_error_hva(addr)) { |
2037 | if (writable) | |
2038 | *writable = false; | |
81c52c56 | 2039 | return KVM_PFN_NOSLOT; |
b2740d35 | 2040 | } |
4d8b81ab XG |
2041 | |
2042 | /* Do not map writable pfn in the readonly memslot. */ | |
2043 | if (writable && memslot_is_readonly(slot)) { | |
2044 | *writable = false; | |
2045 | writable = NULL; | |
2046 | } | |
2047 | ||
2048 | return hva_to_pfn(addr, atomic, async, write_fault, | |
2049 | writable); | |
887c08ac | 2050 | } |
3520469d | 2051 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 2052 | |
ba049e93 | 2053 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
2054 | bool *writable) |
2055 | { | |
e37afc6e PB |
2056 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
2057 | write_fault, writable); | |
612819c3 MT |
2058 | } |
2059 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
2060 | ||
ba049e93 | 2061 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2062 | { |
4d8b81ab | 2063 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
506f0d6f | 2064 | } |
e37afc6e | 2065 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 2066 | |
ba049e93 | 2067 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2068 | { |
4d8b81ab | 2069 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
506f0d6f | 2070 | } |
037d92dc | 2071 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 2072 | |
ba049e93 | 2073 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2074 | { |
2075 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2076 | } | |
2077 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
2078 | ||
ba049e93 | 2079 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
2080 | { |
2081 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
2082 | } | |
2083 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
2084 | ||
ba049e93 | 2085 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2086 | { |
2087 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2088 | } | |
2089 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
2090 | ||
d9ef13c2 PB |
2091 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2092 | struct page **pages, int nr_pages) | |
48987781 XG |
2093 | { |
2094 | unsigned long addr; | |
076b925d | 2095 | gfn_t entry = 0; |
48987781 | 2096 | |
d9ef13c2 | 2097 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
2098 | if (kvm_is_error_hva(addr)) |
2099 | return -1; | |
2100 | ||
2101 | if (entry < nr_pages) | |
2102 | return 0; | |
2103 | ||
dadbb612 | 2104 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
2105 | } |
2106 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
2107 | ||
ba049e93 | 2108 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 2109 | { |
81c52c56 | 2110 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 2111 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2112 | |
bf4bea8e | 2113 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 2114 | WARN_ON(1); |
6cede2e6 | 2115 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 2116 | } |
a2766325 XG |
2117 | |
2118 | return pfn_to_page(pfn); | |
2119 | } | |
2120 | ||
35149e21 AL |
2121 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
2122 | { | |
ba049e93 | 2123 | kvm_pfn_t pfn; |
2e2e3738 AL |
2124 | |
2125 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 2126 | |
a2766325 | 2127 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
2128 | } |
2129 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
2130 | ||
91724814 BO |
2131 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache) |
2132 | { | |
2133 | if (pfn == 0) | |
2134 | return; | |
2135 | ||
2136 | if (cache) | |
2137 | cache->pfn = cache->gfn = 0; | |
2138 | ||
2139 | if (dirty) | |
2140 | kvm_release_pfn_dirty(pfn); | |
2141 | else | |
2142 | kvm_release_pfn_clean(pfn); | |
2143 | } | |
2144 | ||
2145 | static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn, | |
2146 | struct gfn_to_pfn_cache *cache, u64 gen) | |
2147 | { | |
2148 | kvm_release_pfn(cache->pfn, cache->dirty, cache); | |
2149 | ||
2150 | cache->pfn = gfn_to_pfn_memslot(slot, gfn); | |
2151 | cache->gfn = gfn; | |
2152 | cache->dirty = false; | |
2153 | cache->generation = gen; | |
2154 | } | |
2155 | ||
1eff70a9 | 2156 | static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn, |
91724814 BO |
2157 | struct kvm_host_map *map, |
2158 | struct gfn_to_pfn_cache *cache, | |
2159 | bool atomic) | |
e45adf66 KA |
2160 | { |
2161 | kvm_pfn_t pfn; | |
2162 | void *hva = NULL; | |
2163 | struct page *page = KVM_UNMAPPED_PAGE; | |
1eff70a9 | 2164 | struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn); |
91724814 | 2165 | u64 gen = slots->generation; |
e45adf66 KA |
2166 | |
2167 | if (!map) | |
2168 | return -EINVAL; | |
2169 | ||
91724814 BO |
2170 | if (cache) { |
2171 | if (!cache->pfn || cache->gfn != gfn || | |
2172 | cache->generation != gen) { | |
2173 | if (atomic) | |
2174 | return -EAGAIN; | |
2175 | kvm_cache_gfn_to_pfn(slot, gfn, cache, gen); | |
2176 | } | |
2177 | pfn = cache->pfn; | |
2178 | } else { | |
2179 | if (atomic) | |
2180 | return -EAGAIN; | |
2181 | pfn = gfn_to_pfn_memslot(slot, gfn); | |
2182 | } | |
e45adf66 KA |
2183 | if (is_error_noslot_pfn(pfn)) |
2184 | return -EINVAL; | |
2185 | ||
2186 | if (pfn_valid(pfn)) { | |
2187 | page = pfn_to_page(pfn); | |
91724814 BO |
2188 | if (atomic) |
2189 | hva = kmap_atomic(page); | |
2190 | else | |
2191 | hva = kmap(page); | |
d30b214d | 2192 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2193 | } else if (!atomic) { |
e45adf66 | 2194 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
91724814 BO |
2195 | } else { |
2196 | return -EINVAL; | |
d30b214d | 2197 | #endif |
e45adf66 KA |
2198 | } |
2199 | ||
2200 | if (!hva) | |
2201 | return -EFAULT; | |
2202 | ||
2203 | map->page = page; | |
2204 | map->hva = hva; | |
2205 | map->pfn = pfn; | |
2206 | map->gfn = gfn; | |
2207 | ||
2208 | return 0; | |
2209 | } | |
2210 | ||
91724814 BO |
2211 | int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map, |
2212 | struct gfn_to_pfn_cache *cache, bool atomic) | |
1eff70a9 | 2213 | { |
91724814 BO |
2214 | return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map, |
2215 | cache, atomic); | |
1eff70a9 BO |
2216 | } |
2217 | EXPORT_SYMBOL_GPL(kvm_map_gfn); | |
2218 | ||
e45adf66 KA |
2219 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
2220 | { | |
91724814 BO |
2221 | return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map, |
2222 | NULL, false); | |
e45adf66 KA |
2223 | } |
2224 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); | |
2225 | ||
28bd726a PX |
2226 | static void __kvm_unmap_gfn(struct kvm *kvm, |
2227 | struct kvm_memory_slot *memslot, | |
91724814 BO |
2228 | struct kvm_host_map *map, |
2229 | struct gfn_to_pfn_cache *cache, | |
2230 | bool dirty, bool atomic) | |
e45adf66 KA |
2231 | { |
2232 | if (!map) | |
2233 | return; | |
2234 | ||
2235 | if (!map->hva) | |
2236 | return; | |
2237 | ||
91724814 BO |
2238 | if (map->page != KVM_UNMAPPED_PAGE) { |
2239 | if (atomic) | |
2240 | kunmap_atomic(map->hva); | |
2241 | else | |
2242 | kunmap(map->page); | |
2243 | } | |
eb1f2f38 | 2244 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2245 | else if (!atomic) |
e45adf66 | 2246 | memunmap(map->hva); |
91724814 BO |
2247 | else |
2248 | WARN_ONCE(1, "Unexpected unmapping in atomic context"); | |
eb1f2f38 | 2249 | #endif |
e45adf66 | 2250 | |
91724814 | 2251 | if (dirty) |
28bd726a | 2252 | mark_page_dirty_in_slot(kvm, memslot, map->gfn); |
91724814 BO |
2253 | |
2254 | if (cache) | |
2255 | cache->dirty |= dirty; | |
2256 | else | |
2257 | kvm_release_pfn(map->pfn, dirty, NULL); | |
e45adf66 KA |
2258 | |
2259 | map->hva = NULL; | |
2260 | map->page = NULL; | |
2261 | } | |
1eff70a9 | 2262 | |
91724814 BO |
2263 | int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map, |
2264 | struct gfn_to_pfn_cache *cache, bool dirty, bool atomic) | |
1eff70a9 | 2265 | { |
28bd726a | 2266 | __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map, |
91724814 | 2267 | cache, dirty, atomic); |
1eff70a9 BO |
2268 | return 0; |
2269 | } | |
2270 | EXPORT_SYMBOL_GPL(kvm_unmap_gfn); | |
2271 | ||
2272 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) | |
2273 | { | |
28bd726a PX |
2274 | __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), |
2275 | map, NULL, dirty, false); | |
1eff70a9 | 2276 | } |
e45adf66 KA |
2277 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); |
2278 | ||
8e73485c PB |
2279 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
2280 | { | |
ba049e93 | 2281 | kvm_pfn_t pfn; |
8e73485c PB |
2282 | |
2283 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
2284 | ||
2285 | return kvm_pfn_to_page(pfn); | |
2286 | } | |
2287 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
2288 | ||
b4231d61 IE |
2289 | void kvm_release_page_clean(struct page *page) |
2290 | { | |
32cad84f XG |
2291 | WARN_ON(is_error_page(page)); |
2292 | ||
35149e21 | 2293 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
2294 | } |
2295 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
2296 | ||
ba049e93 | 2297 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 2298 | { |
bf4bea8e | 2299 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 2300 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
2301 | } |
2302 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
2303 | ||
b4231d61 | 2304 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 2305 | { |
a2766325 XG |
2306 | WARN_ON(is_error_page(page)); |
2307 | ||
35149e21 AL |
2308 | kvm_release_pfn_dirty(page_to_pfn(page)); |
2309 | } | |
2310 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
2311 | ||
f7a6509f | 2312 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
2313 | { |
2314 | kvm_set_pfn_dirty(pfn); | |
2315 | kvm_release_pfn_clean(pfn); | |
2316 | } | |
f7a6509f | 2317 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 2318 | |
ba049e93 | 2319 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2320 | { |
d29c03a5 ML |
2321 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2322 | SetPageDirty(pfn_to_page(pfn)); | |
8a7ae055 | 2323 | } |
35149e21 AL |
2324 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
2325 | ||
ba049e93 | 2326 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 2327 | { |
a78986aa | 2328 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2e2e3738 | 2329 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
2330 | } |
2331 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
2332 | ||
ba049e93 | 2333 | void kvm_get_pfn(kvm_pfn_t pfn) |
35149e21 | 2334 | { |
bf4bea8e | 2335 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 2336 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
2337 | } |
2338 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 2339 | |
195aefde IE |
2340 | static int next_segment(unsigned long len, int offset) |
2341 | { | |
2342 | if (len > PAGE_SIZE - offset) | |
2343 | return PAGE_SIZE - offset; | |
2344 | else | |
2345 | return len; | |
2346 | } | |
2347 | ||
8e73485c PB |
2348 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
2349 | void *data, int offset, int len) | |
195aefde | 2350 | { |
e0506bcb IE |
2351 | int r; |
2352 | unsigned long addr; | |
195aefde | 2353 | |
8e73485c | 2354 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
2355 | if (kvm_is_error_hva(addr)) |
2356 | return -EFAULT; | |
3180a7fc | 2357 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 2358 | if (r) |
195aefde | 2359 | return -EFAULT; |
195aefde IE |
2360 | return 0; |
2361 | } | |
8e73485c PB |
2362 | |
2363 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
2364 | int len) | |
2365 | { | |
2366 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2367 | ||
2368 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2369 | } | |
195aefde IE |
2370 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
2371 | ||
8e73485c PB |
2372 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
2373 | int offset, int len) | |
2374 | { | |
2375 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2376 | ||
2377 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2378 | } | |
2379 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
2380 | ||
195aefde IE |
2381 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
2382 | { | |
2383 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2384 | int seg; | |
2385 | int offset = offset_in_page(gpa); | |
2386 | int ret; | |
2387 | ||
2388 | while ((seg = next_segment(len, offset)) != 0) { | |
2389 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
2390 | if (ret < 0) | |
2391 | return ret; | |
2392 | offset = 0; | |
2393 | len -= seg; | |
2394 | data += seg; | |
2395 | ++gfn; | |
2396 | } | |
2397 | return 0; | |
2398 | } | |
2399 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
2400 | ||
8e73485c | 2401 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 2402 | { |
7ec54588 | 2403 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 2404 | int seg; |
7ec54588 | 2405 | int offset = offset_in_page(gpa); |
8e73485c PB |
2406 | int ret; |
2407 | ||
2408 | while ((seg = next_segment(len, offset)) != 0) { | |
2409 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
2410 | if (ret < 0) | |
2411 | return ret; | |
2412 | offset = 0; | |
2413 | len -= seg; | |
2414 | data += seg; | |
2415 | ++gfn; | |
2416 | } | |
2417 | return 0; | |
2418 | } | |
2419 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 2420 | |
8e73485c PB |
2421 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2422 | void *data, int offset, unsigned long len) | |
2423 | { | |
2424 | int r; | |
2425 | unsigned long addr; | |
2426 | ||
2427 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
2428 | if (kvm_is_error_hva(addr)) |
2429 | return -EFAULT; | |
0aac03f0 | 2430 | pagefault_disable(); |
3180a7fc | 2431 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 2432 | pagefault_enable(); |
7ec54588 MT |
2433 | if (r) |
2434 | return -EFAULT; | |
2435 | return 0; | |
2436 | } | |
7ec54588 | 2437 | |
8e73485c PB |
2438 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
2439 | void *data, unsigned long len) | |
2440 | { | |
2441 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2442 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2443 | int offset = offset_in_page(gpa); | |
2444 | ||
2445 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2446 | } | |
2447 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
2448 | ||
28bd726a PX |
2449 | static int __kvm_write_guest_page(struct kvm *kvm, |
2450 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 2451 | const void *data, int offset, int len) |
195aefde | 2452 | { |
e0506bcb IE |
2453 | int r; |
2454 | unsigned long addr; | |
195aefde | 2455 | |
251eb841 | 2456 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
2457 | if (kvm_is_error_hva(addr)) |
2458 | return -EFAULT; | |
8b0cedff | 2459 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 2460 | if (r) |
195aefde | 2461 | return -EFAULT; |
28bd726a | 2462 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
2463 | return 0; |
2464 | } | |
8e73485c PB |
2465 | |
2466 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
2467 | const void *data, int offset, int len) | |
2468 | { | |
2469 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2470 | ||
28bd726a | 2471 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 2472 | } |
195aefde IE |
2473 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
2474 | ||
8e73485c PB |
2475 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
2476 | const void *data, int offset, int len) | |
2477 | { | |
2478 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2479 | ||
28bd726a | 2480 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
2481 | } |
2482 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
2483 | ||
195aefde IE |
2484 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
2485 | unsigned long len) | |
2486 | { | |
2487 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2488 | int seg; | |
2489 | int offset = offset_in_page(gpa); | |
2490 | int ret; | |
2491 | ||
2492 | while ((seg = next_segment(len, offset)) != 0) { | |
2493 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
2494 | if (ret < 0) | |
2495 | return ret; | |
2496 | offset = 0; | |
2497 | len -= seg; | |
2498 | data += seg; | |
2499 | ++gfn; | |
2500 | } | |
2501 | return 0; | |
2502 | } | |
ff651cb6 | 2503 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 2504 | |
8e73485c PB |
2505 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
2506 | unsigned long len) | |
2507 | { | |
2508 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2509 | int seg; | |
2510 | int offset = offset_in_page(gpa); | |
2511 | int ret; | |
2512 | ||
2513 | while ((seg = next_segment(len, offset)) != 0) { | |
2514 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
2515 | if (ret < 0) | |
2516 | return ret; | |
2517 | offset = 0; | |
2518 | len -= seg; | |
2519 | data += seg; | |
2520 | ++gfn; | |
2521 | } | |
2522 | return 0; | |
2523 | } | |
2524 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
2525 | ||
5a2d4365 PB |
2526 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
2527 | struct gfn_to_hva_cache *ghc, | |
2528 | gpa_t gpa, unsigned long len) | |
49c7754c | 2529 | { |
49c7754c | 2530 | int offset = offset_in_page(gpa); |
8f964525 AH |
2531 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
2532 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
2533 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
2534 | gfn_t nr_pages_avail; | |
49c7754c | 2535 | |
6ad1e29f | 2536 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 2537 | ghc->generation = slots->generation; |
6ad1e29f SC |
2538 | |
2539 | if (start_gfn > end_gfn) { | |
2540 | ghc->hva = KVM_HVA_ERR_BAD; | |
2541 | return -EINVAL; | |
2542 | } | |
f1b9dd5e JM |
2543 | |
2544 | /* | |
2545 | * If the requested region crosses two memslots, we still | |
2546 | * verify that the entire region is valid here. | |
2547 | */ | |
6ad1e29f | 2548 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
2549 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
2550 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
2551 | &nr_pages_avail); | |
2552 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 2553 | return -EFAULT; |
f1b9dd5e JM |
2554 | } |
2555 | ||
2556 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 2557 | if (nr_pages_needed == 1) |
49c7754c | 2558 | ghc->hva += offset; |
f1b9dd5e | 2559 | else |
8f964525 | 2560 | ghc->memslot = NULL; |
f1b9dd5e | 2561 | |
6ad1e29f SC |
2562 | ghc->gpa = gpa; |
2563 | ghc->len = len; | |
2564 | return 0; | |
49c7754c | 2565 | } |
5a2d4365 | 2566 | |
4e335d9e | 2567 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
2568 | gpa_t gpa, unsigned long len) |
2569 | { | |
4e335d9e | 2570 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
2571 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
2572 | } | |
4e335d9e | 2573 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 2574 | |
4e335d9e | 2575 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
2576 | void *data, unsigned int offset, |
2577 | unsigned long len) | |
49c7754c | 2578 | { |
4e335d9e | 2579 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 2580 | int r; |
4ec6e863 | 2581 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 2582 | |
4ec6e863 | 2583 | BUG_ON(len + offset > ghc->len); |
8f964525 | 2584 | |
dc9ce71e SC |
2585 | if (slots->generation != ghc->generation) { |
2586 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
2587 | return -EFAULT; | |
2588 | } | |
8f964525 | 2589 | |
49c7754c GN |
2590 | if (kvm_is_error_hva(ghc->hva)) |
2591 | return -EFAULT; | |
2592 | ||
fcfbc617 SC |
2593 | if (unlikely(!ghc->memslot)) |
2594 | return kvm_write_guest(kvm, gpa, data, len); | |
2595 | ||
4ec6e863 | 2596 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
2597 | if (r) |
2598 | return -EFAULT; | |
28bd726a | 2599 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
2600 | |
2601 | return 0; | |
2602 | } | |
4e335d9e | 2603 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 2604 | |
4e335d9e PB |
2605 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2606 | void *data, unsigned long len) | |
4ec6e863 | 2607 | { |
4e335d9e | 2608 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 2609 | } |
4e335d9e | 2610 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 2611 | |
0958f0ce VK |
2612 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2613 | void *data, unsigned int offset, | |
2614 | unsigned long len) | |
e03b644f | 2615 | { |
4e335d9e | 2616 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 2617 | int r; |
0958f0ce | 2618 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 2619 | |
0958f0ce | 2620 | BUG_ON(len + offset > ghc->len); |
8f964525 | 2621 | |
dc9ce71e SC |
2622 | if (slots->generation != ghc->generation) { |
2623 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
2624 | return -EFAULT; | |
2625 | } | |
8f964525 | 2626 | |
e03b644f GN |
2627 | if (kvm_is_error_hva(ghc->hva)) |
2628 | return -EFAULT; | |
2629 | ||
fcfbc617 | 2630 | if (unlikely(!ghc->memslot)) |
0958f0ce | 2631 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 2632 | |
0958f0ce | 2633 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
2634 | if (r) |
2635 | return -EFAULT; | |
2636 | ||
2637 | return 0; | |
2638 | } | |
0958f0ce VK |
2639 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
2640 | ||
2641 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
2642 | void *data, unsigned long len) | |
2643 | { | |
2644 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
2645 | } | |
4e335d9e | 2646 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 2647 | |
195aefde IE |
2648 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
2649 | { | |
2f541442 | 2650 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
2651 | gfn_t gfn = gpa >> PAGE_SHIFT; |
2652 | int seg; | |
2653 | int offset = offset_in_page(gpa); | |
2654 | int ret; | |
2655 | ||
bfda0e84 | 2656 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 2657 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
2658 | if (ret < 0) |
2659 | return ret; | |
2660 | offset = 0; | |
2661 | len -= seg; | |
2662 | ++gfn; | |
2663 | } | |
2664 | return 0; | |
2665 | } | |
2666 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
2667 | ||
28bd726a PX |
2668 | void mark_page_dirty_in_slot(struct kvm *kvm, |
2669 | struct kvm_memory_slot *memslot, | |
2670 | gfn_t gfn) | |
6aa8b732 | 2671 | { |
044c59c4 | 2672 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 2673 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 2674 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 2675 | |
fb04a1ed PX |
2676 | if (kvm->dirty_ring_size) |
2677 | kvm_dirty_ring_push(kvm_dirty_ring_get(kvm), | |
2678 | slot, rel_gfn); | |
2679 | else | |
2680 | set_bit_le(rel_gfn, memslot->dirty_bitmap); | |
6aa8b732 AK |
2681 | } |
2682 | } | |
a6a0b05d | 2683 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 2684 | |
49c7754c GN |
2685 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
2686 | { | |
2687 | struct kvm_memory_slot *memslot; | |
2688 | ||
2689 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 2690 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 2691 | } |
2ba9f0d8 | 2692 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 2693 | |
8e73485c PB |
2694 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
2695 | { | |
2696 | struct kvm_memory_slot *memslot; | |
2697 | ||
2698 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 2699 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
2700 | } |
2701 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
2702 | ||
20b7035c JS |
2703 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
2704 | { | |
2705 | if (!vcpu->sigset_active) | |
2706 | return; | |
2707 | ||
2708 | /* | |
2709 | * This does a lockless modification of ->real_blocked, which is fine | |
2710 | * because, only current can change ->real_blocked and all readers of | |
2711 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
2712 | * of ->blocked. | |
2713 | */ | |
2714 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
2715 | } | |
2716 | ||
2717 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
2718 | { | |
2719 | if (!vcpu->sigset_active) | |
2720 | return; | |
2721 | ||
2722 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
2723 | sigemptyset(¤t->real_blocked); | |
2724 | } | |
2725 | ||
aca6ff29 WL |
2726 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
2727 | { | |
dee339b5 | 2728 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 2729 | |
2cbd7824 | 2730 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 2731 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 2732 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
2733 | if (!grow) |
2734 | goto out; | |
2735 | ||
dee339b5 NW |
2736 | val *= grow; |
2737 | if (val < grow_start) | |
2738 | val = grow_start; | |
aca6ff29 | 2739 | |
313f636d DM |
2740 | if (val > halt_poll_ns) |
2741 | val = halt_poll_ns; | |
2742 | ||
aca6ff29 | 2743 | vcpu->halt_poll_ns = val; |
7fa08e71 | 2744 | out: |
2cbd7824 | 2745 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2746 | } |
2747 | ||
2748 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
2749 | { | |
6b6de68c | 2750 | unsigned int old, val, shrink; |
aca6ff29 | 2751 | |
2cbd7824 | 2752 | old = val = vcpu->halt_poll_ns; |
6b6de68c CB |
2753 | shrink = READ_ONCE(halt_poll_ns_shrink); |
2754 | if (shrink == 0) | |
aca6ff29 WL |
2755 | val = 0; |
2756 | else | |
6b6de68c | 2757 | val /= shrink; |
aca6ff29 WL |
2758 | |
2759 | vcpu->halt_poll_ns = val; | |
2cbd7824 | 2760 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
2761 | } |
2762 | ||
f7819512 PB |
2763 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
2764 | { | |
50c28f21 JS |
2765 | int ret = -EINTR; |
2766 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
2767 | ||
f7819512 PB |
2768 | if (kvm_arch_vcpu_runnable(vcpu)) { |
2769 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 2770 | goto out; |
f7819512 PB |
2771 | } |
2772 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 2773 | goto out; |
f7819512 | 2774 | if (signal_pending(current)) |
50c28f21 | 2775 | goto out; |
f7819512 | 2776 | |
50c28f21 JS |
2777 | ret = 0; |
2778 | out: | |
2779 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
2780 | return ret; | |
f7819512 PB |
2781 | } |
2782 | ||
cb953129 DM |
2783 | static inline void |
2784 | update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited) | |
2785 | { | |
2786 | if (waited) | |
2787 | vcpu->stat.halt_poll_fail_ns += poll_ns; | |
2788 | else | |
2789 | vcpu->stat.halt_poll_success_ns += poll_ns; | |
2790 | } | |
2791 | ||
b6958ce4 ED |
2792 | /* |
2793 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
2794 | */ | |
8776e519 | 2795 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 2796 | { |
cb953129 | 2797 | ktime_t start, cur, poll_end; |
f7819512 | 2798 | bool waited = false; |
aca6ff29 | 2799 | u64 block_ns; |
f7819512 | 2800 | |
07ab0f8d MZ |
2801 | kvm_arch_vcpu_blocking(vcpu); |
2802 | ||
cb953129 | 2803 | start = cur = poll_end = ktime_get(); |
cdd6ad3a | 2804 | if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) { |
19020f8a | 2805 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); |
f95ef0cd | 2806 | |
62bea5bf | 2807 | ++vcpu->stat.halt_attempted_poll; |
f7819512 PB |
2808 | do { |
2809 | /* | |
2810 | * This sets KVM_REQ_UNHALT if an interrupt | |
2811 | * arrives. | |
2812 | */ | |
2813 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
2814 | ++vcpu->stat.halt_successful_poll; | |
3491caf2 CB |
2815 | if (!vcpu_valid_wakeup(vcpu)) |
2816 | ++vcpu->stat.halt_poll_invalid; | |
f7819512 PB |
2817 | goto out; |
2818 | } | |
cb953129 | 2819 | poll_end = cur = ktime_get(); |
f7819512 PB |
2820 | } while (single_task_running() && ktime_before(cur, stop)); |
2821 | } | |
e5c239cf | 2822 | |
da4ad88c | 2823 | prepare_to_rcuwait(&vcpu->wait); |
e5c239cf | 2824 | for (;;) { |
da4ad88c | 2825 | set_current_state(TASK_INTERRUPTIBLE); |
e5c239cf | 2826 | |
f7819512 | 2827 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
2828 | break; |
2829 | ||
f7819512 | 2830 | waited = true; |
b6958ce4 | 2831 | schedule(); |
b6958ce4 | 2832 | } |
da4ad88c | 2833 | finish_rcuwait(&vcpu->wait); |
f7819512 | 2834 | cur = ktime_get(); |
f7819512 | 2835 | out: |
07ab0f8d | 2836 | kvm_arch_vcpu_unblocking(vcpu); |
aca6ff29 WL |
2837 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
2838 | ||
cb953129 DM |
2839 | update_halt_poll_stats( |
2840 | vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited); | |
2841 | ||
44551b2f WL |
2842 | if (!kvm_arch_no_poll(vcpu)) { |
2843 | if (!vcpu_valid_wakeup(vcpu)) { | |
aca6ff29 | 2844 | shrink_halt_poll_ns(vcpu); |
acd05785 | 2845 | } else if (vcpu->kvm->max_halt_poll_ns) { |
44551b2f WL |
2846 | if (block_ns <= vcpu->halt_poll_ns) |
2847 | ; | |
2848 | /* we had a long block, shrink polling */ | |
acd05785 DM |
2849 | else if (vcpu->halt_poll_ns && |
2850 | block_ns > vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
2851 | shrink_halt_poll_ns(vcpu); |
2852 | /* we had a short halt and our poll time is too small */ | |
acd05785 DM |
2853 | else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && |
2854 | block_ns < vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
2855 | grow_halt_poll_ns(vcpu); |
2856 | } else { | |
2857 | vcpu->halt_poll_ns = 0; | |
2858 | } | |
2859 | } | |
aca6ff29 | 2860 | |
3491caf2 CB |
2861 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
2862 | kvm_arch_vcpu_block_finish(vcpu); | |
b6958ce4 | 2863 | } |
2ba9f0d8 | 2864 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 2865 | |
178f02ff | 2866 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 2867 | { |
da4ad88c | 2868 | struct rcuwait *waitp; |
b6d33834 | 2869 | |
da4ad88c DB |
2870 | waitp = kvm_arch_vcpu_get_wait(vcpu); |
2871 | if (rcuwait_wake_up(waitp)) { | |
d73eb57b | 2872 | WRITE_ONCE(vcpu->ready, true); |
b6d33834 | 2873 | ++vcpu->stat.halt_wakeup; |
178f02ff | 2874 | return true; |
b6d33834 CD |
2875 | } |
2876 | ||
178f02ff | 2877 | return false; |
dd1a4cc1 RK |
2878 | } |
2879 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
2880 | ||
0266c894 | 2881 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
2882 | /* |
2883 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
2884 | */ | |
2885 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
2886 | { | |
2887 | int me; | |
2888 | int cpu = vcpu->cpu; | |
2889 | ||
178f02ff RK |
2890 | if (kvm_vcpu_wake_up(vcpu)) |
2891 | return; | |
2892 | ||
b6d33834 CD |
2893 | me = get_cpu(); |
2894 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
2895 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
2896 | smp_send_reschedule(cpu); | |
2897 | put_cpu(); | |
2898 | } | |
a20ed54d | 2899 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 2900 | #endif /* !CONFIG_S390 */ |
b6d33834 | 2901 | |
fa93384f | 2902 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
2903 | { |
2904 | struct pid *pid; | |
2905 | struct task_struct *task = NULL; | |
fa93384f | 2906 | int ret = 0; |
41628d33 KW |
2907 | |
2908 | rcu_read_lock(); | |
2909 | pid = rcu_dereference(target->pid); | |
2910 | if (pid) | |
27fbe64b | 2911 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
2912 | rcu_read_unlock(); |
2913 | if (!task) | |
c45c528e | 2914 | return ret; |
c45c528e | 2915 | ret = yield_to(task, 1); |
41628d33 | 2916 | put_task_struct(task); |
c45c528e R |
2917 | |
2918 | return ret; | |
41628d33 KW |
2919 | } |
2920 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
2921 | ||
06e48c51 R |
2922 | /* |
2923 | * Helper that checks whether a VCPU is eligible for directed yield. | |
2924 | * Most eligible candidate to yield is decided by following heuristics: | |
2925 | * | |
2926 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
2927 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 2928 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
2929 | * |
2930 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
2931 | * chance last time (mostly it has become eligible now since we have probably | |
2932 | * yielded to lockholder in last iteration. This is done by toggling | |
2933 | * @dy_eligible each time a VCPU checked for eligibility.) | |
2934 | * | |
2935 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
2936 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
2937 | * burning. Giving priority for a potential lock-holder increases lock | |
2938 | * progress. | |
2939 | * | |
2940 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
2941 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
2942 | * and continue with next VCPU and so on. | |
2943 | */ | |
7940876e | 2944 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 2945 | { |
4a55dd72 | 2946 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
2947 | bool eligible; |
2948 | ||
2949 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 2950 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
2951 | |
2952 | if (vcpu->spin_loop.in_spin_loop) | |
2953 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
2954 | ||
2955 | return eligible; | |
4a55dd72 SW |
2956 | #else |
2957 | return true; | |
06e48c51 | 2958 | #endif |
4a55dd72 | 2959 | } |
c45c528e | 2960 | |
17e433b5 WL |
2961 | /* |
2962 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
2963 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
2964 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
2965 | */ | |
2966 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
2967 | { | |
2968 | return kvm_arch_vcpu_runnable(vcpu); | |
2969 | } | |
2970 | ||
2971 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
2972 | { | |
2973 | if (kvm_arch_dy_runnable(vcpu)) | |
2974 | return true; | |
2975 | ||
2976 | #ifdef CONFIG_KVM_ASYNC_PF | |
2977 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
2978 | return true; | |
2979 | #endif | |
2980 | ||
2981 | return false; | |
2982 | } | |
2983 | ||
199b5763 | 2984 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 2985 | { |
217ece61 RR |
2986 | struct kvm *kvm = me->kvm; |
2987 | struct kvm_vcpu *vcpu; | |
2988 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
2989 | int yielded = 0; | |
c45c528e | 2990 | int try = 3; |
217ece61 RR |
2991 | int pass; |
2992 | int i; | |
d255f4f2 | 2993 | |
4c088493 | 2994 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
2995 | /* |
2996 | * We boost the priority of a VCPU that is runnable but not | |
2997 | * currently running, because it got preempted by something | |
2998 | * else and called schedule in __vcpu_run. Hopefully that | |
2999 | * VCPU is holding the lock that we need and will release it. | |
3000 | * We approximate round-robin by starting at the last boosted VCPU. | |
3001 | */ | |
c45c528e | 3002 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 3003 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 3004 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
3005 | i = last_boosted_vcpu; |
3006 | continue; | |
3007 | } else if (pass && i > last_boosted_vcpu) | |
3008 | break; | |
d73eb57b | 3009 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 3010 | continue; |
217ece61 RR |
3011 | if (vcpu == me) |
3012 | continue; | |
da4ad88c DB |
3013 | if (rcuwait_active(&vcpu->wait) && |
3014 | !vcpu_dy_runnable(vcpu)) | |
217ece61 | 3015 | continue; |
046ddeed WL |
3016 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
3017 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 3018 | continue; |
06e48c51 R |
3019 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
3020 | continue; | |
c45c528e R |
3021 | |
3022 | yielded = kvm_vcpu_yield_to(vcpu); | |
3023 | if (yielded > 0) { | |
217ece61 | 3024 | kvm->last_boosted_vcpu = i; |
217ece61 | 3025 | break; |
c45c528e R |
3026 | } else if (yielded < 0) { |
3027 | try--; | |
3028 | if (!try) | |
3029 | break; | |
217ece61 | 3030 | } |
217ece61 RR |
3031 | } |
3032 | } | |
4c088493 | 3033 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
3034 | |
3035 | /* Ensure vcpu is not eligible during next spinloop */ | |
3036 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
3037 | } |
3038 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
3039 | ||
fb04a1ed PX |
3040 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
3041 | { | |
3042 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
3043 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && | |
3044 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
3045 | kvm->dirty_ring_size / PAGE_SIZE); | |
3046 | #else | |
3047 | return false; | |
3048 | #endif | |
3049 | } | |
3050 | ||
1499fa80 | 3051 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 3052 | { |
11bac800 | 3053 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
3054 | struct page *page; |
3055 | ||
e4a533a4 | 3056 | if (vmf->pgoff == 0) |
039576c0 | 3057 | page = virt_to_page(vcpu->run); |
09566765 | 3058 | #ifdef CONFIG_X86 |
e4a533a4 | 3059 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 3060 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 3061 | #endif |
4b4357e0 | 3062 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3063 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
3064 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 3065 | #endif |
fb04a1ed PX |
3066 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
3067 | page = kvm_dirty_ring_get_page( | |
3068 | &vcpu->dirty_ring, | |
3069 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 3070 | else |
5b1c1493 | 3071 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 3072 | get_page(page); |
e4a533a4 NP |
3073 | vmf->page = page; |
3074 | return 0; | |
9a2bb7f4 AK |
3075 | } |
3076 | ||
f0f37e2f | 3077 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 3078 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
3079 | }; |
3080 | ||
3081 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
3082 | { | |
fb04a1ed PX |
3083 | struct kvm_vcpu *vcpu = file->private_data; |
3084 | unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; | |
3085 | ||
3086 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
3087 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
3088 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
3089 | return -EINVAL; | |
3090 | ||
9a2bb7f4 AK |
3091 | vma->vm_ops = &kvm_vcpu_vm_ops; |
3092 | return 0; | |
3093 | } | |
3094 | ||
bccf2150 AK |
3095 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
3096 | { | |
3097 | struct kvm_vcpu *vcpu = filp->private_data; | |
3098 | ||
66c0b394 | 3099 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
3100 | return 0; |
3101 | } | |
3102 | ||
3d3aab1b | 3103 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
3104 | .release = kvm_vcpu_release, |
3105 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 3106 | .mmap = kvm_vcpu_mmap, |
6038f373 | 3107 | .llseek = noop_llseek, |
7ddfd3e0 | 3108 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
3109 | }; |
3110 | ||
3111 | /* | |
3112 | * Allocates an inode for the vcpu. | |
3113 | */ | |
3114 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
3115 | { | |
e46b4692 MY |
3116 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
3117 | ||
3118 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
3119 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
3120 | } |
3121 | ||
3e7093d0 | 3122 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 3123 | { |
741cbbae | 3124 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
d56f5136 | 3125 | struct dentry *debugfs_dentry; |
45b5939e | 3126 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 3127 | |
45b5939e | 3128 | if (!debugfs_initialized()) |
3e7093d0 | 3129 | return; |
45b5939e LC |
3130 | |
3131 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
3132 | debugfs_dentry = debugfs_create_dir(dir_name, |
3133 | vcpu->kvm->debugfs_dentry); | |
45b5939e | 3134 | |
d56f5136 | 3135 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
741cbbae | 3136 | #endif |
45b5939e LC |
3137 | } |
3138 | ||
c5ea7660 AK |
3139 | /* |
3140 | * Creates some virtual cpus. Good luck creating more than one. | |
3141 | */ | |
73880c80 | 3142 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
3143 | { |
3144 | int r; | |
e09fefde | 3145 | struct kvm_vcpu *vcpu; |
8bd826d6 | 3146 | struct page *page; |
c5ea7660 | 3147 | |
0b1b1dfd | 3148 | if (id >= KVM_MAX_VCPU_ID) |
338c7dba AH |
3149 | return -EINVAL; |
3150 | ||
6c7caebc PB |
3151 | mutex_lock(&kvm->lock); |
3152 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
3153 | mutex_unlock(&kvm->lock); | |
3154 | return -EINVAL; | |
3155 | } | |
3156 | ||
3157 | kvm->created_vcpus++; | |
3158 | mutex_unlock(&kvm->lock); | |
3159 | ||
897cc38e SC |
3160 | r = kvm_arch_vcpu_precreate(kvm, id); |
3161 | if (r) | |
3162 | goto vcpu_decrement; | |
3163 | ||
e529ef66 SC |
3164 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
3165 | if (!vcpu) { | |
3166 | r = -ENOMEM; | |
6c7caebc PB |
3167 | goto vcpu_decrement; |
3168 | } | |
c5ea7660 | 3169 | |
fcd97ad5 | 3170 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 3171 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
3172 | if (!page) { |
3173 | r = -ENOMEM; | |
e529ef66 | 3174 | goto vcpu_free; |
8bd826d6 SC |
3175 | } |
3176 | vcpu->run = page_address(page); | |
3177 | ||
3178 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
3179 | |
3180 | r = kvm_arch_vcpu_create(vcpu); | |
3181 | if (r) | |
8bd826d6 | 3182 | goto vcpu_free_run_page; |
e529ef66 | 3183 | |
fb04a1ed PX |
3184 | if (kvm->dirty_ring_size) { |
3185 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
3186 | id, kvm->dirty_ring_size); | |
3187 | if (r) | |
3188 | goto arch_vcpu_destroy; | |
3189 | } | |
3190 | ||
11ec2804 | 3191 | mutex_lock(&kvm->lock); |
e09fefde DH |
3192 | if (kvm_get_vcpu_by_id(kvm, id)) { |
3193 | r = -EEXIST; | |
3194 | goto unlock_vcpu_destroy; | |
3195 | } | |
73880c80 | 3196 | |
8750e72a RK |
3197 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
3198 | BUG_ON(kvm->vcpus[vcpu->vcpu_idx]); | |
c5ea7660 | 3199 | |
fb3f0f51 | 3200 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 3201 | kvm_get_kvm(kvm); |
bccf2150 | 3202 | r = create_vcpu_fd(vcpu); |
73880c80 | 3203 | if (r < 0) { |
149487bd | 3204 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 3205 | goto unlock_vcpu_destroy; |
73880c80 GN |
3206 | } |
3207 | ||
8750e72a | 3208 | kvm->vcpus[vcpu->vcpu_idx] = vcpu; |
dd489240 PB |
3209 | |
3210 | /* | |
3211 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
3212 | * before kvm->online_vcpu's incremented value. | |
3213 | */ | |
73880c80 GN |
3214 | smp_wmb(); |
3215 | atomic_inc(&kvm->online_vcpus); | |
3216 | ||
73880c80 | 3217 | mutex_unlock(&kvm->lock); |
42897d86 | 3218 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 3219 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 3220 | return r; |
39c3b86e | 3221 | |
d780592b | 3222 | unlock_vcpu_destroy: |
7d8fece6 | 3223 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
3224 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
3225 | arch_vcpu_destroy: | |
d40ccc62 | 3226 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
3227 | vcpu_free_run_page: |
3228 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
3229 | vcpu_free: |
3230 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
3231 | vcpu_decrement: |
3232 | mutex_lock(&kvm->lock); | |
3233 | kvm->created_vcpus--; | |
3234 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
3235 | return r; |
3236 | } | |
3237 | ||
1961d276 AK |
3238 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
3239 | { | |
3240 | if (sigset) { | |
3241 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
3242 | vcpu->sigset_active = 1; | |
3243 | vcpu->sigset = *sigset; | |
3244 | } else | |
3245 | vcpu->sigset_active = 0; | |
3246 | return 0; | |
3247 | } | |
3248 | ||
bccf2150 AK |
3249 | static long kvm_vcpu_ioctl(struct file *filp, |
3250 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 3251 | { |
bccf2150 | 3252 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 3253 | void __user *argp = (void __user *)arg; |
313a3dc7 | 3254 | int r; |
fa3795a7 DH |
3255 | struct kvm_fpu *fpu = NULL; |
3256 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 3257 | |
6d4e4c4f AK |
3258 | if (vcpu->kvm->mm != current->mm) |
3259 | return -EIO; | |
2122ff5e | 3260 | |
2ea75be3 DM |
3261 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
3262 | return -EINVAL; | |
3263 | ||
2122ff5e | 3264 | /* |
5cb0944c PB |
3265 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
3266 | * execution; mutex_lock() would break them. | |
2122ff5e | 3267 | */ |
5cb0944c PB |
3268 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
3269 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 3270 | return r; |
2122ff5e | 3271 | |
ec7660cc CD |
3272 | if (mutex_lock_killable(&vcpu->mutex)) |
3273 | return -EINTR; | |
6aa8b732 | 3274 | switch (ioctl) { |
0e4524a5 CB |
3275 | case KVM_RUN: { |
3276 | struct pid *oldpid; | |
f0fe5108 AK |
3277 | r = -EINVAL; |
3278 | if (arg) | |
3279 | goto out; | |
0e4524a5 | 3280 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 3281 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 3282 | /* The thread running this VCPU changed. */ |
bd2a6394 | 3283 | struct pid *newpid; |
f95ef0cd | 3284 | |
bd2a6394 CD |
3285 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
3286 | if (r) | |
3287 | break; | |
3288 | ||
3289 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
3290 | rcu_assign_pointer(vcpu->pid, newpid); |
3291 | if (oldpid) | |
3292 | synchronize_rcu(); | |
3293 | put_pid(oldpid); | |
3294 | } | |
1b94f6f8 | 3295 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 3296 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 3297 | break; |
0e4524a5 | 3298 | } |
6aa8b732 | 3299 | case KVM_GET_REGS: { |
3e4bb3ac | 3300 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3301 | |
3e4bb3ac | 3302 | r = -ENOMEM; |
b12ce36a | 3303 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 3304 | if (!kvm_regs) |
6aa8b732 | 3305 | goto out; |
3e4bb3ac XZ |
3306 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
3307 | if (r) | |
3308 | goto out_free1; | |
6aa8b732 | 3309 | r = -EFAULT; |
3e4bb3ac XZ |
3310 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
3311 | goto out_free1; | |
6aa8b732 | 3312 | r = 0; |
3e4bb3ac XZ |
3313 | out_free1: |
3314 | kfree(kvm_regs); | |
6aa8b732 AK |
3315 | break; |
3316 | } | |
3317 | case KVM_SET_REGS: { | |
3e4bb3ac | 3318 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3319 | |
ff5c2c03 SL |
3320 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
3321 | if (IS_ERR(kvm_regs)) { | |
3322 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 3323 | goto out; |
ff5c2c03 | 3324 | } |
3e4bb3ac | 3325 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 3326 | kfree(kvm_regs); |
6aa8b732 AK |
3327 | break; |
3328 | } | |
3329 | case KVM_GET_SREGS: { | |
b12ce36a BG |
3330 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
3331 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
3332 | r = -ENOMEM; |
3333 | if (!kvm_sregs) | |
3334 | goto out; | |
3335 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
3336 | if (r) |
3337 | goto out; | |
3338 | r = -EFAULT; | |
fa3795a7 | 3339 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
3340 | goto out; |
3341 | r = 0; | |
3342 | break; | |
3343 | } | |
3344 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
3345 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
3346 | if (IS_ERR(kvm_sregs)) { | |
3347 | r = PTR_ERR(kvm_sregs); | |
18595411 | 3348 | kvm_sregs = NULL; |
6aa8b732 | 3349 | goto out; |
ff5c2c03 | 3350 | } |
fa3795a7 | 3351 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
3352 | break; |
3353 | } | |
62d9f0db MT |
3354 | case KVM_GET_MP_STATE: { |
3355 | struct kvm_mp_state mp_state; | |
3356 | ||
3357 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
3358 | if (r) | |
3359 | goto out; | |
3360 | r = -EFAULT; | |
893bdbf1 | 3361 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
3362 | goto out; |
3363 | r = 0; | |
3364 | break; | |
3365 | } | |
3366 | case KVM_SET_MP_STATE: { | |
3367 | struct kvm_mp_state mp_state; | |
3368 | ||
3369 | r = -EFAULT; | |
893bdbf1 | 3370 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
3371 | goto out; |
3372 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
3373 | break; |
3374 | } | |
6aa8b732 AK |
3375 | case KVM_TRANSLATE: { |
3376 | struct kvm_translation tr; | |
3377 | ||
3378 | r = -EFAULT; | |
893bdbf1 | 3379 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 3380 | goto out; |
8b006791 | 3381 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
3382 | if (r) |
3383 | goto out; | |
3384 | r = -EFAULT; | |
893bdbf1 | 3385 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
3386 | goto out; |
3387 | r = 0; | |
3388 | break; | |
3389 | } | |
d0bfb940 JK |
3390 | case KVM_SET_GUEST_DEBUG: { |
3391 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
3392 | |
3393 | r = -EFAULT; | |
893bdbf1 | 3394 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 3395 | goto out; |
d0bfb940 | 3396 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
3397 | break; |
3398 | } | |
1961d276 AK |
3399 | case KVM_SET_SIGNAL_MASK: { |
3400 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3401 | struct kvm_signal_mask kvm_sigmask; | |
3402 | sigset_t sigset, *p; | |
3403 | ||
3404 | p = NULL; | |
3405 | if (argp) { | |
3406 | r = -EFAULT; | |
3407 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3408 | sizeof(kvm_sigmask))) |
1961d276 AK |
3409 | goto out; |
3410 | r = -EINVAL; | |
893bdbf1 | 3411 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
3412 | goto out; |
3413 | r = -EFAULT; | |
3414 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 3415 | sizeof(sigset))) |
1961d276 AK |
3416 | goto out; |
3417 | p = &sigset; | |
3418 | } | |
376d41ff | 3419 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
3420 | break; |
3421 | } | |
b8836737 | 3422 | case KVM_GET_FPU: { |
b12ce36a | 3423 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
3424 | r = -ENOMEM; |
3425 | if (!fpu) | |
3426 | goto out; | |
3427 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
3428 | if (r) |
3429 | goto out; | |
3430 | r = -EFAULT; | |
fa3795a7 | 3431 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
3432 | goto out; |
3433 | r = 0; | |
3434 | break; | |
3435 | } | |
3436 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
3437 | fpu = memdup_user(argp, sizeof(*fpu)); |
3438 | if (IS_ERR(fpu)) { | |
3439 | r = PTR_ERR(fpu); | |
18595411 | 3440 | fpu = NULL; |
b8836737 | 3441 | goto out; |
ff5c2c03 | 3442 | } |
fa3795a7 | 3443 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
3444 | break; |
3445 | } | |
bccf2150 | 3446 | default: |
313a3dc7 | 3447 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
3448 | } |
3449 | out: | |
ec7660cc | 3450 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
3451 | kfree(fpu); |
3452 | kfree(kvm_sregs); | |
bccf2150 AK |
3453 | return r; |
3454 | } | |
3455 | ||
de8e5d74 | 3456 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
3457 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
3458 | unsigned int ioctl, unsigned long arg) | |
3459 | { | |
3460 | struct kvm_vcpu *vcpu = filp->private_data; | |
3461 | void __user *argp = compat_ptr(arg); | |
3462 | int r; | |
3463 | ||
3464 | if (vcpu->kvm->mm != current->mm) | |
3465 | return -EIO; | |
3466 | ||
3467 | switch (ioctl) { | |
3468 | case KVM_SET_SIGNAL_MASK: { | |
3469 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3470 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
3471 | sigset_t sigset; |
3472 | ||
3473 | if (argp) { | |
3474 | r = -EFAULT; | |
3475 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3476 | sizeof(kvm_sigmask))) |
1dda606c AG |
3477 | goto out; |
3478 | r = -EINVAL; | |
3968cf62 | 3479 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
3480 | goto out; |
3481 | r = -EFAULT; | |
1393b4aa PB |
3482 | if (get_compat_sigset(&sigset, |
3483 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 3484 | goto out; |
760a9a30 AC |
3485 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
3486 | } else | |
3487 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
3488 | break; |
3489 | } | |
3490 | default: | |
3491 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
3492 | } | |
3493 | ||
3494 | out: | |
3495 | return r; | |
3496 | } | |
3497 | #endif | |
3498 | ||
a1cd3f08 CLG |
3499 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
3500 | { | |
3501 | struct kvm_device *dev = filp->private_data; | |
3502 | ||
3503 | if (dev->ops->mmap) | |
3504 | return dev->ops->mmap(dev, vma); | |
3505 | ||
3506 | return -ENODEV; | |
3507 | } | |
3508 | ||
852b6d57 SW |
3509 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
3510 | int (*accessor)(struct kvm_device *dev, | |
3511 | struct kvm_device_attr *attr), | |
3512 | unsigned long arg) | |
3513 | { | |
3514 | struct kvm_device_attr attr; | |
3515 | ||
3516 | if (!accessor) | |
3517 | return -EPERM; | |
3518 | ||
3519 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
3520 | return -EFAULT; | |
3521 | ||
3522 | return accessor(dev, &attr); | |
3523 | } | |
3524 | ||
3525 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
3526 | unsigned long arg) | |
3527 | { | |
3528 | struct kvm_device *dev = filp->private_data; | |
3529 | ||
ddba9180 SC |
3530 | if (dev->kvm->mm != current->mm) |
3531 | return -EIO; | |
3532 | ||
852b6d57 SW |
3533 | switch (ioctl) { |
3534 | case KVM_SET_DEVICE_ATTR: | |
3535 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
3536 | case KVM_GET_DEVICE_ATTR: | |
3537 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
3538 | case KVM_HAS_DEVICE_ATTR: | |
3539 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
3540 | default: | |
3541 | if (dev->ops->ioctl) | |
3542 | return dev->ops->ioctl(dev, ioctl, arg); | |
3543 | ||
3544 | return -ENOTTY; | |
3545 | } | |
3546 | } | |
3547 | ||
852b6d57 SW |
3548 | static int kvm_device_release(struct inode *inode, struct file *filp) |
3549 | { | |
3550 | struct kvm_device *dev = filp->private_data; | |
3551 | struct kvm *kvm = dev->kvm; | |
3552 | ||
2bde9b3e CLG |
3553 | if (dev->ops->release) { |
3554 | mutex_lock(&kvm->lock); | |
3555 | list_del(&dev->vm_node); | |
3556 | dev->ops->release(dev); | |
3557 | mutex_unlock(&kvm->lock); | |
3558 | } | |
3559 | ||
852b6d57 SW |
3560 | kvm_put_kvm(kvm); |
3561 | return 0; | |
3562 | } | |
3563 | ||
3564 | static const struct file_operations kvm_device_fops = { | |
3565 | .unlocked_ioctl = kvm_device_ioctl, | |
3566 | .release = kvm_device_release, | |
7ddfd3e0 | 3567 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 3568 | .mmap = kvm_device_mmap, |
852b6d57 SW |
3569 | }; |
3570 | ||
3571 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
3572 | { | |
3573 | if (filp->f_op != &kvm_device_fops) | |
3574 | return NULL; | |
3575 | ||
3576 | return filp->private_data; | |
3577 | } | |
3578 | ||
8538cb22 | 3579 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 3580 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
3581 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
3582 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 3583 | #endif |
d60eacb0 WD |
3584 | }; |
3585 | ||
8538cb22 | 3586 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
3587 | { |
3588 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
3589 | return -ENOSPC; | |
3590 | ||
3591 | if (kvm_device_ops_table[type] != NULL) | |
3592 | return -EEXIST; | |
3593 | ||
3594 | kvm_device_ops_table[type] = ops; | |
3595 | return 0; | |
3596 | } | |
3597 | ||
571ee1b6 WL |
3598 | void kvm_unregister_device_ops(u32 type) |
3599 | { | |
3600 | if (kvm_device_ops_table[type] != NULL) | |
3601 | kvm_device_ops_table[type] = NULL; | |
3602 | } | |
3603 | ||
852b6d57 SW |
3604 | static int kvm_ioctl_create_device(struct kvm *kvm, |
3605 | struct kvm_create_device *cd) | |
3606 | { | |
8538cb22 | 3607 | const struct kvm_device_ops *ops = NULL; |
852b6d57 SW |
3608 | struct kvm_device *dev; |
3609 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 3610 | int type; |
852b6d57 SW |
3611 | int ret; |
3612 | ||
d60eacb0 WD |
3613 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
3614 | return -ENODEV; | |
3615 | ||
1d487e9b PB |
3616 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
3617 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 3618 | if (ops == NULL) |
852b6d57 | 3619 | return -ENODEV; |
852b6d57 SW |
3620 | |
3621 | if (test) | |
3622 | return 0; | |
3623 | ||
b12ce36a | 3624 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
3625 | if (!dev) |
3626 | return -ENOMEM; | |
3627 | ||
3628 | dev->ops = ops; | |
3629 | dev->kvm = kvm; | |
852b6d57 | 3630 | |
a28ebea2 | 3631 | mutex_lock(&kvm->lock); |
1d487e9b | 3632 | ret = ops->create(dev, type); |
852b6d57 | 3633 | if (ret < 0) { |
a28ebea2 | 3634 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
3635 | kfree(dev); |
3636 | return ret; | |
3637 | } | |
a28ebea2 CD |
3638 | list_add(&dev->vm_node, &kvm->devices); |
3639 | mutex_unlock(&kvm->lock); | |
852b6d57 | 3640 | |
023e9fdd CD |
3641 | if (ops->init) |
3642 | ops->init(dev); | |
3643 | ||
cfa39381 | 3644 | kvm_get_kvm(kvm); |
24009b05 | 3645 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 3646 | if (ret < 0) { |
149487bd | 3647 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
3648 | mutex_lock(&kvm->lock); |
3649 | list_del(&dev->vm_node); | |
3650 | mutex_unlock(&kvm->lock); | |
a0f1d21c | 3651 | ops->destroy(dev); |
852b6d57 SW |
3652 | return ret; |
3653 | } | |
3654 | ||
852b6d57 SW |
3655 | cd->fd = ret; |
3656 | return 0; | |
3657 | } | |
3658 | ||
92b591a4 AG |
3659 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
3660 | { | |
3661 | switch (arg) { | |
3662 | case KVM_CAP_USER_MEMORY: | |
3663 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
3664 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
3665 | case KVM_CAP_INTERNAL_ERROR_DATA: |
3666 | #ifdef CONFIG_HAVE_KVM_MSI | |
3667 | case KVM_CAP_SIGNAL_MSI: | |
3668 | #endif | |
297e2105 | 3669 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 3670 | case KVM_CAP_IRQFD: |
92b591a4 AG |
3671 | case KVM_CAP_IRQFD_RESAMPLE: |
3672 | #endif | |
e9ea5069 | 3673 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 3674 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 3675 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 3676 | case KVM_CAP_HALT_POLL: |
92b591a4 | 3677 | return 1; |
4b4357e0 | 3678 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
3679 | case KVM_CAP_COALESCED_MMIO: |
3680 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
3681 | case KVM_CAP_COALESCED_PIO: |
3682 | return 1; | |
30422558 | 3683 | #endif |
3c9bd400 JZ |
3684 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3685 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
3686 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
3687 | #endif | |
92b591a4 AG |
3688 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
3689 | case KVM_CAP_IRQ_ROUTING: | |
3690 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
3691 | #endif |
3692 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
3693 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
3694 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 3695 | #endif |
c110ae57 PB |
3696 | case KVM_CAP_NR_MEMSLOTS: |
3697 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed PX |
3698 | case KVM_CAP_DIRTY_LOG_RING: |
3699 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
3700 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); | |
3701 | #else | |
3702 | return 0; | |
3703 | #endif | |
92b591a4 AG |
3704 | default: |
3705 | break; | |
3706 | } | |
3707 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
3708 | } | |
3709 | ||
fb04a1ed PX |
3710 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
3711 | { | |
3712 | int r; | |
3713 | ||
3714 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
3715 | return -EINVAL; | |
3716 | ||
3717 | /* the size should be power of 2 */ | |
3718 | if (!size || (size & (size - 1))) | |
3719 | return -EINVAL; | |
3720 | ||
3721 | /* Should be bigger to keep the reserved entries, or a page */ | |
3722 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
3723 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
3724 | return -EINVAL; | |
3725 | ||
3726 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
3727 | sizeof(struct kvm_dirty_gfn)) | |
3728 | return -E2BIG; | |
3729 | ||
3730 | /* We only allow it to set once */ | |
3731 | if (kvm->dirty_ring_size) | |
3732 | return -EINVAL; | |
3733 | ||
3734 | mutex_lock(&kvm->lock); | |
3735 | ||
3736 | if (kvm->created_vcpus) { | |
3737 | /* We don't allow to change this value after vcpu created */ | |
3738 | r = -EINVAL; | |
3739 | } else { | |
3740 | kvm->dirty_ring_size = size; | |
3741 | r = 0; | |
3742 | } | |
3743 | ||
3744 | mutex_unlock(&kvm->lock); | |
3745 | return r; | |
3746 | } | |
3747 | ||
3748 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
3749 | { | |
3750 | int i; | |
3751 | struct kvm_vcpu *vcpu; | |
3752 | int cleared = 0; | |
3753 | ||
3754 | if (!kvm->dirty_ring_size) | |
3755 | return -EINVAL; | |
3756 | ||
3757 | mutex_lock(&kvm->slots_lock); | |
3758 | ||
3759 | kvm_for_each_vcpu(i, vcpu, kvm) | |
3760 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
3761 | ||
3762 | mutex_unlock(&kvm->slots_lock); | |
3763 | ||
3764 | if (cleared) | |
3765 | kvm_flush_remote_tlbs(kvm); | |
3766 | ||
3767 | return cleared; | |
3768 | } | |
3769 | ||
e5d83c74 PB |
3770 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
3771 | struct kvm_enable_cap *cap) | |
3772 | { | |
3773 | return -EINVAL; | |
3774 | } | |
3775 | ||
3776 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
3777 | struct kvm_enable_cap *cap) | |
3778 | { | |
3779 | switch (cap->cap) { | |
2a31b9db | 3780 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
3781 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
3782 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
3783 | ||
3784 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
3785 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
3786 | ||
3787 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
3788 | return -EINVAL; |
3789 | kvm->manual_dirty_log_protect = cap->args[0]; | |
3790 | return 0; | |
3c9bd400 | 3791 | } |
2a31b9db | 3792 | #endif |
acd05785 DM |
3793 | case KVM_CAP_HALT_POLL: { |
3794 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
3795 | return -EINVAL; | |
3796 | ||
3797 | kvm->max_halt_poll_ns = cap->args[0]; | |
3798 | return 0; | |
3799 | } | |
fb04a1ed PX |
3800 | case KVM_CAP_DIRTY_LOG_RING: |
3801 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); | |
e5d83c74 PB |
3802 | default: |
3803 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
3804 | } | |
3805 | } | |
3806 | ||
bccf2150 AK |
3807 | static long kvm_vm_ioctl(struct file *filp, |
3808 | unsigned int ioctl, unsigned long arg) | |
3809 | { | |
3810 | struct kvm *kvm = filp->private_data; | |
3811 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 3812 | int r; |
bccf2150 | 3813 | |
6d4e4c4f AK |
3814 | if (kvm->mm != current->mm) |
3815 | return -EIO; | |
bccf2150 AK |
3816 | switch (ioctl) { |
3817 | case KVM_CREATE_VCPU: | |
3818 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 3819 | break; |
e5d83c74 PB |
3820 | case KVM_ENABLE_CAP: { |
3821 | struct kvm_enable_cap cap; | |
3822 | ||
3823 | r = -EFAULT; | |
3824 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
3825 | goto out; | |
3826 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
3827 | break; | |
3828 | } | |
6fc138d2 IE |
3829 | case KVM_SET_USER_MEMORY_REGION: { |
3830 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
3831 | ||
3832 | r = -EFAULT; | |
3833 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 3834 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
3835 | goto out; |
3836 | ||
47ae31e2 | 3837 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
3838 | break; |
3839 | } | |
3840 | case KVM_GET_DIRTY_LOG: { | |
3841 | struct kvm_dirty_log log; | |
3842 | ||
3843 | r = -EFAULT; | |
893bdbf1 | 3844 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 3845 | goto out; |
2c6f5df9 | 3846 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
3847 | break; |
3848 | } | |
2a31b9db PB |
3849 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3850 | case KVM_CLEAR_DIRTY_LOG: { | |
3851 | struct kvm_clear_dirty_log log; | |
3852 | ||
3853 | r = -EFAULT; | |
3854 | if (copy_from_user(&log, argp, sizeof(log))) | |
3855 | goto out; | |
3856 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
3857 | break; | |
3858 | } | |
3859 | #endif | |
4b4357e0 | 3860 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3861 | case KVM_REGISTER_COALESCED_MMIO: { |
3862 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 3863 | |
5f94c174 | 3864 | r = -EFAULT; |
893bdbf1 | 3865 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 3866 | goto out; |
5f94c174 | 3867 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
3868 | break; |
3869 | } | |
3870 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
3871 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 3872 | |
5f94c174 | 3873 | r = -EFAULT; |
893bdbf1 | 3874 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 3875 | goto out; |
5f94c174 | 3876 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
3877 | break; |
3878 | } | |
3879 | #endif | |
721eecbf GH |
3880 | case KVM_IRQFD: { |
3881 | struct kvm_irqfd data; | |
3882 | ||
3883 | r = -EFAULT; | |
893bdbf1 | 3884 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 3885 | goto out; |
d4db2935 | 3886 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
3887 | break; |
3888 | } | |
d34e6b17 GH |
3889 | case KVM_IOEVENTFD: { |
3890 | struct kvm_ioeventfd data; | |
3891 | ||
3892 | r = -EFAULT; | |
893bdbf1 | 3893 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
3894 | goto out; |
3895 | r = kvm_ioeventfd(kvm, &data); | |
3896 | break; | |
3897 | } | |
07975ad3 JK |
3898 | #ifdef CONFIG_HAVE_KVM_MSI |
3899 | case KVM_SIGNAL_MSI: { | |
3900 | struct kvm_msi msi; | |
3901 | ||
3902 | r = -EFAULT; | |
893bdbf1 | 3903 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
3904 | goto out; |
3905 | r = kvm_send_userspace_msi(kvm, &msi); | |
3906 | break; | |
3907 | } | |
23d43cf9 CD |
3908 | #endif |
3909 | #ifdef __KVM_HAVE_IRQ_LINE | |
3910 | case KVM_IRQ_LINE_STATUS: | |
3911 | case KVM_IRQ_LINE: { | |
3912 | struct kvm_irq_level irq_event; | |
3913 | ||
3914 | r = -EFAULT; | |
893bdbf1 | 3915 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
3916 | goto out; |
3917 | ||
aa2fbe6d YZ |
3918 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
3919 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
3920 | if (r) |
3921 | goto out; | |
3922 | ||
3923 | r = -EFAULT; | |
3924 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 3925 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
3926 | goto out; |
3927 | } | |
3928 | ||
3929 | r = 0; | |
3930 | break; | |
3931 | } | |
73880c80 | 3932 | #endif |
aa8d5944 AG |
3933 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
3934 | case KVM_SET_GSI_ROUTING: { | |
3935 | struct kvm_irq_routing routing; | |
3936 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 3937 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
3938 | |
3939 | r = -EFAULT; | |
3940 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
3941 | goto out; | |
3942 | r = -EINVAL; | |
5c0aea0e DH |
3943 | if (!kvm_arch_can_set_irq_routing(kvm)) |
3944 | goto out; | |
caf1ff26 | 3945 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
3946 | goto out; |
3947 | if (routing.flags) | |
3948 | goto out; | |
f8c1b85b | 3949 | if (routing.nr) { |
f8c1b85b | 3950 | urouting = argp; |
7ec28e26 DE |
3951 | entries = vmemdup_user(urouting->entries, |
3952 | array_size(sizeof(*entries), | |
3953 | routing.nr)); | |
3954 | if (IS_ERR(entries)) { | |
3955 | r = PTR_ERR(entries); | |
3956 | goto out; | |
3957 | } | |
f8c1b85b | 3958 | } |
aa8d5944 AG |
3959 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
3960 | routing.flags); | |
7ec28e26 | 3961 | kvfree(entries); |
aa8d5944 AG |
3962 | break; |
3963 | } | |
3964 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
3965 | case KVM_CREATE_DEVICE: { |
3966 | struct kvm_create_device cd; | |
3967 | ||
3968 | r = -EFAULT; | |
3969 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
3970 | goto out; | |
3971 | ||
3972 | r = kvm_ioctl_create_device(kvm, &cd); | |
3973 | if (r) | |
3974 | goto out; | |
3975 | ||
3976 | r = -EFAULT; | |
3977 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
3978 | goto out; | |
3979 | ||
3980 | r = 0; | |
3981 | break; | |
3982 | } | |
92b591a4 AG |
3983 | case KVM_CHECK_EXTENSION: |
3984 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
3985 | break; | |
fb04a1ed PX |
3986 | case KVM_RESET_DIRTY_RINGS: |
3987 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
3988 | break; | |
f17abe9a | 3989 | default: |
1fe779f8 | 3990 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
3991 | } |
3992 | out: | |
3993 | return r; | |
3994 | } | |
3995 | ||
de8e5d74 | 3996 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
3997 | struct compat_kvm_dirty_log { |
3998 | __u32 slot; | |
3999 | __u32 padding1; | |
4000 | union { | |
4001 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4002 | __u64 padding2; | |
4003 | }; | |
4004 | }; | |
4005 | ||
4006 | static long kvm_vm_compat_ioctl(struct file *filp, | |
4007 | unsigned int ioctl, unsigned long arg) | |
4008 | { | |
4009 | struct kvm *kvm = filp->private_data; | |
4010 | int r; | |
4011 | ||
4012 | if (kvm->mm != current->mm) | |
4013 | return -EIO; | |
4014 | switch (ioctl) { | |
4015 | case KVM_GET_DIRTY_LOG: { | |
4016 | struct compat_kvm_dirty_log compat_log; | |
4017 | struct kvm_dirty_log log; | |
4018 | ||
6ff5894c AB |
4019 | if (copy_from_user(&compat_log, (void __user *)arg, |
4020 | sizeof(compat_log))) | |
f6a3b168 | 4021 | return -EFAULT; |
6ff5894c AB |
4022 | log.slot = compat_log.slot; |
4023 | log.padding1 = compat_log.padding1; | |
4024 | log.padding2 = compat_log.padding2; | |
4025 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4026 | ||
4027 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
4028 | break; |
4029 | } | |
4030 | default: | |
4031 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
4032 | } | |
6ff5894c AB |
4033 | return r; |
4034 | } | |
4035 | #endif | |
4036 | ||
3d3aab1b | 4037 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
4038 | .release = kvm_vm_release, |
4039 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 4040 | .llseek = noop_llseek, |
7ddfd3e0 | 4041 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
4042 | }; |
4043 | ||
e08b9637 | 4044 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 4045 | { |
aac87636 | 4046 | int r; |
f17abe9a | 4047 | struct kvm *kvm; |
506cfba9 | 4048 | struct file *file; |
f17abe9a | 4049 | |
e08b9637 | 4050 | kvm = kvm_create_vm(type); |
d6d28168 AK |
4051 | if (IS_ERR(kvm)) |
4052 | return PTR_ERR(kvm); | |
4b4357e0 | 4053 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 4054 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
4055 | if (r < 0) |
4056 | goto put_kvm; | |
6ce5a090 | 4057 | #endif |
506cfba9 | 4058 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
4059 | if (r < 0) |
4060 | goto put_kvm; | |
4061 | ||
506cfba9 AV |
4062 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
4063 | if (IS_ERR(file)) { | |
4064 | put_unused_fd(r); | |
78588335 ME |
4065 | r = PTR_ERR(file); |
4066 | goto put_kvm; | |
506cfba9 | 4067 | } |
536a6f88 | 4068 | |
525df861 PB |
4069 | /* |
4070 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
4071 | * already set, with ->release() being kvm_vm_release(). In error | |
4072 | * cases it will be called by the final fput(file) and will take | |
4073 | * care of doing kvm_put_kvm(kvm). | |
4074 | */ | |
536a6f88 | 4075 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
4076 | put_unused_fd(r); |
4077 | fput(file); | |
536a6f88 JF |
4078 | return -ENOMEM; |
4079 | } | |
286de8f6 | 4080 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 4081 | |
506cfba9 | 4082 | fd_install(r, file); |
aac87636 | 4083 | return r; |
78588335 ME |
4084 | |
4085 | put_kvm: | |
4086 | kvm_put_kvm(kvm); | |
4087 | return r; | |
f17abe9a AK |
4088 | } |
4089 | ||
4090 | static long kvm_dev_ioctl(struct file *filp, | |
4091 | unsigned int ioctl, unsigned long arg) | |
4092 | { | |
07c45a36 | 4093 | long r = -EINVAL; |
f17abe9a AK |
4094 | |
4095 | switch (ioctl) { | |
4096 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
4097 | if (arg) |
4098 | goto out; | |
f17abe9a AK |
4099 | r = KVM_API_VERSION; |
4100 | break; | |
4101 | case KVM_CREATE_VM: | |
e08b9637 | 4102 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 4103 | break; |
018d00d2 | 4104 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 4105 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 4106 | break; |
07c45a36 | 4107 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
4108 | if (arg) |
4109 | goto out; | |
adb1ff46 AK |
4110 | r = PAGE_SIZE; /* struct kvm_run */ |
4111 | #ifdef CONFIG_X86 | |
4112 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 4113 | #endif |
4b4357e0 | 4114 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 4115 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 4116 | #endif |
07c45a36 | 4117 | break; |
d4c9ff2d FEL |
4118 | case KVM_TRACE_ENABLE: |
4119 | case KVM_TRACE_PAUSE: | |
4120 | case KVM_TRACE_DISABLE: | |
2023a29c | 4121 | r = -EOPNOTSUPP; |
d4c9ff2d | 4122 | break; |
6aa8b732 | 4123 | default: |
043405e1 | 4124 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
4125 | } |
4126 | out: | |
4127 | return r; | |
4128 | } | |
4129 | ||
6aa8b732 | 4130 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 4131 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 4132 | .llseek = noop_llseek, |
7ddfd3e0 | 4133 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
4134 | }; |
4135 | ||
4136 | static struct miscdevice kvm_dev = { | |
bbe4432e | 4137 | KVM_MINOR, |
6aa8b732 AK |
4138 | "kvm", |
4139 | &kvm_chardev_ops, | |
4140 | }; | |
4141 | ||
75b7127c | 4142 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
4143 | { |
4144 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 4145 | int r; |
1b6c0168 | 4146 | |
7f59f492 | 4147 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4148 | return; |
10474ae8 | 4149 | |
7f59f492 | 4150 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 4151 | |
13a34e06 | 4152 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
4153 | |
4154 | if (r) { | |
4155 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
4156 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 4157 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 4158 | } |
1b6c0168 AK |
4159 | } |
4160 | ||
8c18b2d2 | 4161 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 4162 | { |
4a937f96 | 4163 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4164 | if (kvm_usage_count) |
4165 | hardware_enable_nolock(NULL); | |
4a937f96 | 4166 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4167 | return 0; |
75b7127c TY |
4168 | } |
4169 | ||
4170 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
4171 | { |
4172 | int cpu = raw_smp_processor_id(); | |
4173 | ||
7f59f492 | 4174 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4175 | return; |
7f59f492 | 4176 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 4177 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
4178 | } |
4179 | ||
8c18b2d2 | 4180 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 4181 | { |
4a937f96 | 4182 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4183 | if (kvm_usage_count) |
4184 | hardware_disable_nolock(NULL); | |
4a937f96 | 4185 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4186 | return 0; |
75b7127c TY |
4187 | } |
4188 | ||
10474ae8 AG |
4189 | static void hardware_disable_all_nolock(void) |
4190 | { | |
4191 | BUG_ON(!kvm_usage_count); | |
4192 | ||
4193 | kvm_usage_count--; | |
4194 | if (!kvm_usage_count) | |
75b7127c | 4195 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
4196 | } |
4197 | ||
4198 | static void hardware_disable_all(void) | |
4199 | { | |
4a937f96 | 4200 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 4201 | hardware_disable_all_nolock(); |
4a937f96 | 4202 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4203 | } |
4204 | ||
4205 | static int hardware_enable_all(void) | |
4206 | { | |
4207 | int r = 0; | |
4208 | ||
4a937f96 | 4209 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
4210 | |
4211 | kvm_usage_count++; | |
4212 | if (kvm_usage_count == 1) { | |
4213 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 4214 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
4215 | |
4216 | if (atomic_read(&hardware_enable_failed)) { | |
4217 | hardware_disable_all_nolock(); | |
4218 | r = -EBUSY; | |
4219 | } | |
4220 | } | |
4221 | ||
4a937f96 | 4222 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4223 | |
4224 | return r; | |
4225 | } | |
4226 | ||
9a2b85c6 | 4227 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 4228 | void *v) |
9a2b85c6 | 4229 | { |
8e1c1815 SY |
4230 | /* |
4231 | * Some (well, at least mine) BIOSes hang on reboot if | |
4232 | * in vmx root mode. | |
4233 | * | |
4234 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
4235 | */ | |
1170adc6 | 4236 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 4237 | kvm_rebooting = true; |
75b7127c | 4238 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
4239 | return NOTIFY_OK; |
4240 | } | |
4241 | ||
4242 | static struct notifier_block kvm_reboot_notifier = { | |
4243 | .notifier_call = kvm_reboot, | |
4244 | .priority = 0, | |
4245 | }; | |
4246 | ||
e93f8a0f | 4247 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
4248 | { |
4249 | int i; | |
4250 | ||
4251 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 4252 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
4253 | |
4254 | kvm_iodevice_destructor(pos); | |
4255 | } | |
e93f8a0f | 4256 | kfree(bus); |
2eeb2e94 GH |
4257 | } |
4258 | ||
c21fbff1 | 4259 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 4260 | const struct kvm_io_range *r2) |
743eeb0b | 4261 | { |
8f4216c7 JW |
4262 | gpa_t addr1 = r1->addr; |
4263 | gpa_t addr2 = r2->addr; | |
4264 | ||
4265 | if (addr1 < addr2) | |
743eeb0b | 4266 | return -1; |
8f4216c7 JW |
4267 | |
4268 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
4269 | * accept any overlapping write. Any order is acceptable for | |
4270 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
4271 | * we process all of them. | |
4272 | */ | |
4273 | if (r2->len) { | |
4274 | addr1 += r1->len; | |
4275 | addr2 += r2->len; | |
4276 | } | |
4277 | ||
4278 | if (addr1 > addr2) | |
743eeb0b | 4279 | return 1; |
8f4216c7 | 4280 | |
743eeb0b SL |
4281 | return 0; |
4282 | } | |
4283 | ||
a343c9b7 PB |
4284 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
4285 | { | |
c21fbff1 | 4286 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
4287 | } |
4288 | ||
39369f7a | 4289 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
4290 | gpa_t addr, int len) |
4291 | { | |
4292 | struct kvm_io_range *range, key; | |
4293 | int off; | |
4294 | ||
4295 | key = (struct kvm_io_range) { | |
4296 | .addr = addr, | |
4297 | .len = len, | |
4298 | }; | |
4299 | ||
4300 | range = bsearch(&key, bus->range, bus->dev_count, | |
4301 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
4302 | if (range == NULL) | |
4303 | return -ENOENT; | |
4304 | ||
4305 | off = range - bus->range; | |
4306 | ||
c21fbff1 | 4307 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
4308 | off--; |
4309 | ||
4310 | return off; | |
4311 | } | |
4312 | ||
e32edf4f | 4313 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
4314 | struct kvm_io_range *range, const void *val) |
4315 | { | |
4316 | int idx; | |
4317 | ||
4318 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
4319 | if (idx < 0) | |
4320 | return -EOPNOTSUPP; | |
4321 | ||
4322 | while (idx < bus->dev_count && | |
c21fbff1 | 4323 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4324 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4325 | range->len, val)) |
4326 | return idx; | |
4327 | idx++; | |
4328 | } | |
4329 | ||
4330 | return -EOPNOTSUPP; | |
4331 | } | |
4332 | ||
bda9020e | 4333 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 4334 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 4335 | int len, const void *val) |
2eeb2e94 | 4336 | { |
90d83dc3 | 4337 | struct kvm_io_bus *bus; |
743eeb0b | 4338 | struct kvm_io_range range; |
126a5af5 | 4339 | int r; |
743eeb0b SL |
4340 | |
4341 | range = (struct kvm_io_range) { | |
4342 | .addr = addr, | |
4343 | .len = len, | |
4344 | }; | |
90d83dc3 | 4345 | |
e32edf4f | 4346 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4347 | if (!bus) |
4348 | return -ENOMEM; | |
e32edf4f | 4349 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4350 | return r < 0 ? r : 0; |
4351 | } | |
a2420107 | 4352 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
4353 | |
4354 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
4355 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
4356 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
4357 | { |
4358 | struct kvm_io_bus *bus; | |
4359 | struct kvm_io_range range; | |
4360 | ||
4361 | range = (struct kvm_io_range) { | |
4362 | .addr = addr, | |
4363 | .len = len, | |
4364 | }; | |
4365 | ||
e32edf4f | 4366 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4367 | if (!bus) |
4368 | return -ENOMEM; | |
126a5af5 CH |
4369 | |
4370 | /* First try the device referenced by cookie. */ | |
4371 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 4372 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 4373 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
4374 | val)) |
4375 | return cookie; | |
4376 | ||
4377 | /* | |
4378 | * cookie contained garbage; fall back to search and return the | |
4379 | * correct cookie value. | |
4380 | */ | |
e32edf4f | 4381 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4382 | } |
4383 | ||
e32edf4f NN |
4384 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
4385 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
4386 | { |
4387 | int idx; | |
4388 | ||
4389 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
4390 | if (idx < 0) |
4391 | return -EOPNOTSUPP; | |
4392 | ||
4393 | while (idx < bus->dev_count && | |
c21fbff1 | 4394 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4395 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4396 | range->len, val)) |
4397 | return idx; | |
743eeb0b SL |
4398 | idx++; |
4399 | } | |
4400 | ||
bda9020e MT |
4401 | return -EOPNOTSUPP; |
4402 | } | |
2eeb2e94 | 4403 | |
bda9020e | 4404 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 4405 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 4406 | int len, void *val) |
bda9020e | 4407 | { |
90d83dc3 | 4408 | struct kvm_io_bus *bus; |
743eeb0b | 4409 | struct kvm_io_range range; |
126a5af5 | 4410 | int r; |
743eeb0b SL |
4411 | |
4412 | range = (struct kvm_io_range) { | |
4413 | .addr = addr, | |
4414 | .len = len, | |
4415 | }; | |
e93f8a0f | 4416 | |
e32edf4f | 4417 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4418 | if (!bus) |
4419 | return -ENOMEM; | |
e32edf4f | 4420 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
4421 | return r < 0 ? r : 0; |
4422 | } | |
743eeb0b | 4423 | |
79fac95e | 4424 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
4425 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
4426 | int len, struct kvm_io_device *dev) | |
6c474694 | 4427 | { |
d4c67a7a | 4428 | int i; |
e93f8a0f | 4429 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 4430 | struct kvm_io_range range; |
090b7aff | 4431 | |
4a12f951 | 4432 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
4433 | if (!bus) |
4434 | return -ENOMEM; | |
4435 | ||
6ea34c9b AK |
4436 | /* exclude ioeventfd which is limited by maximum fd */ |
4437 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 4438 | return -ENOSPC; |
2eeb2e94 | 4439 | |
90952cd3 | 4440 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 4441 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
4442 | if (!new_bus) |
4443 | return -ENOMEM; | |
d4c67a7a GH |
4444 | |
4445 | range = (struct kvm_io_range) { | |
4446 | .addr = addr, | |
4447 | .len = len, | |
4448 | .dev = dev, | |
4449 | }; | |
4450 | ||
4451 | for (i = 0; i < bus->dev_count; i++) | |
4452 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
4453 | break; | |
4454 | ||
4455 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
4456 | new_bus->dev_count++; | |
4457 | new_bus->range[i] = range; | |
4458 | memcpy(new_bus->range + i + 1, bus->range + i, | |
4459 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
4460 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
4461 | synchronize_srcu_expedited(&kvm->srcu); | |
4462 | kfree(bus); | |
090b7aff GH |
4463 | |
4464 | return 0; | |
4465 | } | |
4466 | ||
79fac95e | 4467 | /* Caller must hold slots_lock. */ |
90db1043 DH |
4468 | void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
4469 | struct kvm_io_device *dev) | |
090b7aff | 4470 | { |
f6588660 | 4471 | int i, j; |
e93f8a0f | 4472 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 4473 | |
4a12f951 | 4474 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 4475 | if (!bus) |
90db1043 | 4476 | return; |
df630b8c | 4477 | |
a1300716 AK |
4478 | for (i = 0; i < bus->dev_count; i++) |
4479 | if (bus->range[i].dev == dev) { | |
090b7aff GH |
4480 | break; |
4481 | } | |
e93f8a0f | 4482 | |
90db1043 DH |
4483 | if (i == bus->dev_count) |
4484 | return; | |
a1300716 | 4485 | |
90952cd3 | 4486 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 4487 | GFP_KERNEL_ACCOUNT); |
f6588660 | 4488 | if (new_bus) { |
871c433b | 4489 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
4490 | new_bus->dev_count--; |
4491 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 4492 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
f6588660 | 4493 | } else { |
90db1043 | 4494 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
f6588660 RK |
4495 | for (j = 0; j < bus->dev_count; j++) { |
4496 | if (j == i) | |
4497 | continue; | |
4498 | kvm_iodevice_destructor(bus->range[j].dev); | |
4499 | } | |
90db1043 | 4500 | } |
a1300716 | 4501 | |
e93f8a0f MT |
4502 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
4503 | synchronize_srcu_expedited(&kvm->srcu); | |
4504 | kfree(bus); | |
90db1043 | 4505 | return; |
2eeb2e94 GH |
4506 | } |
4507 | ||
8a39d006 AP |
4508 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
4509 | gpa_t addr) | |
4510 | { | |
4511 | struct kvm_io_bus *bus; | |
4512 | int dev_idx, srcu_idx; | |
4513 | struct kvm_io_device *iodev = NULL; | |
4514 | ||
4515 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
4516 | ||
4517 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
4518 | if (!bus) |
4519 | goto out_unlock; | |
8a39d006 AP |
4520 | |
4521 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
4522 | if (dev_idx < 0) | |
4523 | goto out_unlock; | |
4524 | ||
4525 | iodev = bus->range[dev_idx].dev; | |
4526 | ||
4527 | out_unlock: | |
4528 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
4529 | ||
4530 | return iodev; | |
4531 | } | |
4532 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
4533 | ||
536a6f88 JF |
4534 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
4535 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
4536 | const char *fmt) | |
4537 | { | |
4538 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
4539 | inode->i_private; | |
4540 | ||
4541 | /* The debugfs files are a reference to the kvm struct which | |
4542 | * is still valid when kvm_destroy_vm is called. | |
4543 | * To avoid the race between open and the removal of the debugfs | |
4544 | * directory we test against the users count. | |
4545 | */ | |
e3736c3e | 4546 | if (!refcount_inc_not_zero(&stat_data->kvm->users_count)) |
536a6f88 JF |
4547 | return -ENOENT; |
4548 | ||
833b45de | 4549 | if (simple_attr_open(inode, file, get, |
09cbcef6 MP |
4550 | KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222 |
4551 | ? set : NULL, | |
4552 | fmt)) { | |
536a6f88 JF |
4553 | kvm_put_kvm(stat_data->kvm); |
4554 | return -ENOMEM; | |
4555 | } | |
4556 | ||
4557 | return 0; | |
4558 | } | |
4559 | ||
4560 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
4561 | { | |
4562 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
4563 | inode->i_private; | |
4564 | ||
4565 | simple_attr_release(inode, file); | |
4566 | kvm_put_kvm(stat_data->kvm); | |
4567 | ||
4568 | return 0; | |
4569 | } | |
4570 | ||
09cbcef6 | 4571 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 4572 | { |
09cbcef6 | 4573 | *val = *(ulong *)((void *)kvm + offset); |
536a6f88 | 4574 | |
09cbcef6 MP |
4575 | return 0; |
4576 | } | |
4577 | ||
4578 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
4579 | { | |
4580 | *(ulong *)((void *)kvm + offset) = 0; | |
536a6f88 JF |
4581 | |
4582 | return 0; | |
4583 | } | |
4584 | ||
09cbcef6 | 4585 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 4586 | { |
09cbcef6 MP |
4587 | int i; |
4588 | struct kvm_vcpu *vcpu; | |
ce35ef27 | 4589 | |
09cbcef6 | 4590 | *val = 0; |
ce35ef27 | 4591 | |
09cbcef6 MP |
4592 | kvm_for_each_vcpu(i, vcpu, kvm) |
4593 | *val += *(u64 *)((void *)vcpu + offset); | |
ce35ef27 SJS |
4594 | |
4595 | return 0; | |
4596 | } | |
4597 | ||
09cbcef6 | 4598 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 4599 | { |
09cbcef6 MP |
4600 | int i; |
4601 | struct kvm_vcpu *vcpu; | |
536a6f88 | 4602 | |
09cbcef6 MP |
4603 | kvm_for_each_vcpu(i, vcpu, kvm) |
4604 | *(u64 *)((void *)vcpu + offset) = 0; | |
4605 | ||
4606 | return 0; | |
4607 | } | |
536a6f88 | 4608 | |
09cbcef6 | 4609 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 4610 | { |
09cbcef6 | 4611 | int r = -EFAULT; |
536a6f88 | 4612 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
536a6f88 | 4613 | |
09cbcef6 MP |
4614 | switch (stat_data->dbgfs_item->kind) { |
4615 | case KVM_STAT_VM: | |
4616 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
4617 | stat_data->dbgfs_item->offset, val); | |
4618 | break; | |
4619 | case KVM_STAT_VCPU: | |
4620 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
4621 | stat_data->dbgfs_item->offset, val); | |
4622 | break; | |
4623 | } | |
536a6f88 | 4624 | |
09cbcef6 | 4625 | return r; |
536a6f88 JF |
4626 | } |
4627 | ||
09cbcef6 | 4628 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 4629 | { |
09cbcef6 | 4630 | int r = -EFAULT; |
ce35ef27 | 4631 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
ce35ef27 SJS |
4632 | |
4633 | if (val) | |
4634 | return -EINVAL; | |
4635 | ||
09cbcef6 MP |
4636 | switch (stat_data->dbgfs_item->kind) { |
4637 | case KVM_STAT_VM: | |
4638 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
4639 | stat_data->dbgfs_item->offset); | |
4640 | break; | |
4641 | case KVM_STAT_VCPU: | |
4642 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
4643 | stat_data->dbgfs_item->offset); | |
4644 | break; | |
4645 | } | |
ce35ef27 | 4646 | |
09cbcef6 | 4647 | return r; |
ce35ef27 SJS |
4648 | } |
4649 | ||
09cbcef6 | 4650 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
4651 | { |
4652 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
4653 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
4654 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
4655 | } |
4656 | ||
09cbcef6 MP |
4657 | static const struct file_operations stat_fops_per_vm = { |
4658 | .owner = THIS_MODULE, | |
4659 | .open = kvm_stat_data_open, | |
536a6f88 | 4660 | .release = kvm_debugfs_release, |
09cbcef6 MP |
4661 | .read = simple_attr_read, |
4662 | .write = simple_attr_write, | |
4663 | .llseek = no_llseek, | |
536a6f88 JF |
4664 | }; |
4665 | ||
8b88b099 | 4666 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
4667 | { |
4668 | unsigned offset = (long)_offset; | |
ba1389b7 | 4669 | struct kvm *kvm; |
536a6f88 | 4670 | u64 tmp_val; |
ba1389b7 | 4671 | |
8b88b099 | 4672 | *val = 0; |
0d9ce162 | 4673 | mutex_lock(&kvm_lock); |
536a6f88 | 4674 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 4675 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
4676 | *val += tmp_val; |
4677 | } | |
0d9ce162 | 4678 | mutex_unlock(&kvm_lock); |
8b88b099 | 4679 | return 0; |
ba1389b7 AK |
4680 | } |
4681 | ||
ce35ef27 SJS |
4682 | static int vm_stat_clear(void *_offset, u64 val) |
4683 | { | |
4684 | unsigned offset = (long)_offset; | |
4685 | struct kvm *kvm; | |
ce35ef27 SJS |
4686 | |
4687 | if (val) | |
4688 | return -EINVAL; | |
4689 | ||
0d9ce162 | 4690 | mutex_lock(&kvm_lock); |
ce35ef27 | 4691 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 4692 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 4693 | } |
0d9ce162 | 4694 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
4695 | |
4696 | return 0; | |
4697 | } | |
4698 | ||
4699 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
ba1389b7 | 4700 | |
8b88b099 | 4701 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
4702 | { |
4703 | unsigned offset = (long)_offset; | |
1165f5fe | 4704 | struct kvm *kvm; |
536a6f88 | 4705 | u64 tmp_val; |
1165f5fe | 4706 | |
8b88b099 | 4707 | *val = 0; |
0d9ce162 | 4708 | mutex_lock(&kvm_lock); |
536a6f88 | 4709 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 4710 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
4711 | *val += tmp_val; |
4712 | } | |
0d9ce162 | 4713 | mutex_unlock(&kvm_lock); |
8b88b099 | 4714 | return 0; |
1165f5fe AK |
4715 | } |
4716 | ||
ce35ef27 SJS |
4717 | static int vcpu_stat_clear(void *_offset, u64 val) |
4718 | { | |
4719 | unsigned offset = (long)_offset; | |
4720 | struct kvm *kvm; | |
ce35ef27 SJS |
4721 | |
4722 | if (val) | |
4723 | return -EINVAL; | |
4724 | ||
0d9ce162 | 4725 | mutex_lock(&kvm_lock); |
ce35ef27 | 4726 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 4727 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 4728 | } |
0d9ce162 | 4729 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
4730 | |
4731 | return 0; | |
4732 | } | |
4733 | ||
4734 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
4735 | "%llu\n"); | |
ba1389b7 | 4736 | |
828c0950 | 4737 | static const struct file_operations *stat_fops[] = { |
ba1389b7 AK |
4738 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
4739 | [KVM_STAT_VM] = &vm_stat_fops, | |
4740 | }; | |
1165f5fe | 4741 | |
286de8f6 CI |
4742 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
4743 | { | |
4744 | struct kobj_uevent_env *env; | |
286de8f6 CI |
4745 | unsigned long long created, active; |
4746 | ||
4747 | if (!kvm_dev.this_device || !kvm) | |
4748 | return; | |
4749 | ||
0d9ce162 | 4750 | mutex_lock(&kvm_lock); |
286de8f6 CI |
4751 | if (type == KVM_EVENT_CREATE_VM) { |
4752 | kvm_createvm_count++; | |
4753 | kvm_active_vms++; | |
4754 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
4755 | kvm_active_vms--; | |
4756 | } | |
4757 | created = kvm_createvm_count; | |
4758 | active = kvm_active_vms; | |
0d9ce162 | 4759 | mutex_unlock(&kvm_lock); |
286de8f6 | 4760 | |
b12ce36a | 4761 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
4762 | if (!env) |
4763 | return; | |
4764 | ||
4765 | add_uevent_var(env, "CREATED=%llu", created); | |
4766 | add_uevent_var(env, "COUNT=%llu", active); | |
4767 | ||
fdeaf7e3 | 4768 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 4769 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
4770 | kvm->userspace_pid = task_pid_nr(current); |
4771 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 4772 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
4773 | } |
4774 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 4775 | |
8ed0579c | 4776 | if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) { |
b12ce36a | 4777 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
4778 | |
4779 | if (p) { | |
4780 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
4781 | if (!IS_ERR(tmp)) | |
4782 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
4783 | kfree(p); | |
286de8f6 CI |
4784 | } |
4785 | } | |
4786 | /* no need for checks, since we are adding at most only 5 keys */ | |
4787 | env->envp[env->envp_idx++] = NULL; | |
4788 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
4789 | kfree(env); | |
286de8f6 CI |
4790 | } |
4791 | ||
929f45e3 | 4792 | static void kvm_init_debug(void) |
6aa8b732 AK |
4793 | { |
4794 | struct kvm_stats_debugfs_item *p; | |
4795 | ||
76f7c879 | 4796 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 4797 | |
536a6f88 JF |
4798 | kvm_debugfs_num_entries = 0; |
4799 | for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) { | |
09cbcef6 MP |
4800 | debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p), |
4801 | kvm_debugfs_dir, (void *)(long)p->offset, | |
929f45e3 | 4802 | stat_fops[p->kind]); |
4f69b680 | 4803 | } |
6aa8b732 AK |
4804 | } |
4805 | ||
fb3600cc | 4806 | static int kvm_suspend(void) |
59ae6c6b | 4807 | { |
10474ae8 | 4808 | if (kvm_usage_count) |
75b7127c | 4809 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
4810 | return 0; |
4811 | } | |
4812 | ||
fb3600cc | 4813 | static void kvm_resume(void) |
59ae6c6b | 4814 | { |
ca84d1a2 | 4815 | if (kvm_usage_count) { |
2eb06c30 WL |
4816 | #ifdef CONFIG_LOCKDEP |
4817 | WARN_ON(lockdep_is_held(&kvm_count_lock)); | |
4818 | #endif | |
75b7127c | 4819 | hardware_enable_nolock(NULL); |
ca84d1a2 | 4820 | } |
59ae6c6b AK |
4821 | } |
4822 | ||
fb3600cc | 4823 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
4824 | .suspend = kvm_suspend, |
4825 | .resume = kvm_resume, | |
4826 | }; | |
4827 | ||
15ad7146 AK |
4828 | static inline |
4829 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
4830 | { | |
4831 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
4832 | } | |
4833 | ||
4834 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
4835 | { | |
4836 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 4837 | |
046ddeed | 4838 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 4839 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 4840 | |
7495e22b | 4841 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 4842 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 4843 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
4844 | } |
4845 | ||
4846 | static void kvm_sched_out(struct preempt_notifier *pn, | |
4847 | struct task_struct *next) | |
4848 | { | |
4849 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
4850 | ||
d73eb57b | 4851 | if (current->state == TASK_RUNNING) { |
046ddeed | 4852 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
4853 | WRITE_ONCE(vcpu->ready, true); |
4854 | } | |
e9b11c17 | 4855 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
4856 | __this_cpu_write(kvm_running_vcpu, NULL); |
4857 | } | |
4858 | ||
4859 | /** | |
4860 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
4861 | * |
4862 | * We can disable preemption locally around accessing the per-CPU variable, | |
4863 | * and use the resolved vcpu pointer after enabling preemption again, | |
4864 | * because even if the current thread is migrated to another CPU, reading | |
4865 | * the per-CPU value later will give us the same value as we update the | |
4866 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
4867 | */ |
4868 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
4869 | { | |
1f03b2bc MZ |
4870 | struct kvm_vcpu *vcpu; |
4871 | ||
4872 | preempt_disable(); | |
4873 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
4874 | preempt_enable(); | |
4875 | ||
4876 | return vcpu; | |
7495e22b | 4877 | } |
379a3c8e | 4878 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
4879 | |
4880 | /** | |
4881 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
4882 | */ | |
4883 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
4884 | { | |
4885 | return &kvm_running_vcpu; | |
15ad7146 AK |
4886 | } |
4887 | ||
b9904085 SC |
4888 | struct kvm_cpu_compat_check { |
4889 | void *opaque; | |
4890 | int *ret; | |
4891 | }; | |
4892 | ||
4893 | static void check_processor_compat(void *data) | |
f257d6dc | 4894 | { |
b9904085 SC |
4895 | struct kvm_cpu_compat_check *c = data; |
4896 | ||
4897 | *c->ret = kvm_arch_check_processor_compat(c->opaque); | |
f257d6dc SC |
4898 | } |
4899 | ||
0ee75bea | 4900 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 4901 | struct module *module) |
6aa8b732 | 4902 | { |
b9904085 | 4903 | struct kvm_cpu_compat_check c; |
6aa8b732 | 4904 | int r; |
002c7f7c | 4905 | int cpu; |
6aa8b732 | 4906 | |
f8c16bba ZX |
4907 | r = kvm_arch_init(opaque); |
4908 | if (r) | |
d2308784 | 4909 | goto out_fail; |
cb498ea2 | 4910 | |
7dac16c3 AH |
4911 | /* |
4912 | * kvm_arch_init makes sure there's at most one caller | |
4913 | * for architectures that support multiple implementations, | |
4914 | * like intel and amd on x86. | |
36343f6e PB |
4915 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
4916 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 4917 | */ |
36343f6e PB |
4918 | r = kvm_irqfd_init(); |
4919 | if (r) | |
4920 | goto out_irqfd; | |
7dac16c3 | 4921 | |
8437a617 | 4922 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
4923 | r = -ENOMEM; |
4924 | goto out_free_0; | |
4925 | } | |
4926 | ||
b9904085 | 4927 | r = kvm_arch_hardware_setup(opaque); |
6aa8b732 | 4928 | if (r < 0) |
faf0be22 | 4929 | goto out_free_1; |
6aa8b732 | 4930 | |
b9904085 SC |
4931 | c.ret = &r; |
4932 | c.opaque = opaque; | |
002c7f7c | 4933 | for_each_online_cpu(cpu) { |
b9904085 | 4934 | smp_call_function_single(cpu, check_processor_compat, &c, 1); |
002c7f7c | 4935 | if (r < 0) |
faf0be22 | 4936 | goto out_free_2; |
002c7f7c YS |
4937 | } |
4938 | ||
73c1b41e | 4939 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 4940 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 4941 | if (r) |
d2308784 | 4942 | goto out_free_2; |
6aa8b732 AK |
4943 | register_reboot_notifier(&kvm_reboot_notifier); |
4944 | ||
c16f862d | 4945 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
4946 | if (!vcpu_align) |
4947 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
4948 | kvm_vcpu_cache = |
4949 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
4950 | SLAB_ACCOUNT, | |
4951 | offsetof(struct kvm_vcpu, arch), | |
4952 | sizeof_field(struct kvm_vcpu, arch), | |
4953 | NULL); | |
c16f862d RR |
4954 | if (!kvm_vcpu_cache) { |
4955 | r = -ENOMEM; | |
fb3600cc | 4956 | goto out_free_3; |
c16f862d RR |
4957 | } |
4958 | ||
af585b92 GN |
4959 | r = kvm_async_pf_init(); |
4960 | if (r) | |
4961 | goto out_free; | |
4962 | ||
6aa8b732 | 4963 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
4964 | kvm_vm_fops.owner = module; |
4965 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
4966 | |
4967 | r = misc_register(&kvm_dev); | |
4968 | if (r) { | |
1170adc6 | 4969 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 4970 | goto out_unreg; |
6aa8b732 AK |
4971 | } |
4972 | ||
fb3600cc RW |
4973 | register_syscore_ops(&kvm_syscore_ops); |
4974 | ||
15ad7146 AK |
4975 | kvm_preempt_ops.sched_in = kvm_sched_in; |
4976 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
4977 | ||
929f45e3 | 4978 | kvm_init_debug(); |
0ea4ed8e | 4979 | |
3c3c29fd PB |
4980 | r = kvm_vfio_ops_init(); |
4981 | WARN_ON(r); | |
4982 | ||
c7addb90 | 4983 | return 0; |
6aa8b732 | 4984 | |
af585b92 GN |
4985 | out_unreg: |
4986 | kvm_async_pf_deinit(); | |
6aa8b732 | 4987 | out_free: |
c16f862d | 4988 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 4989 | out_free_3: |
6aa8b732 | 4990 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 4991 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 4992 | out_free_2: |
e9b11c17 | 4993 | kvm_arch_hardware_unsetup(); |
faf0be22 | 4994 | out_free_1: |
7f59f492 | 4995 | free_cpumask_var(cpus_hardware_enabled); |
d2308784 | 4996 | out_free_0: |
a0f155e9 | 4997 | kvm_irqfd_exit(); |
36343f6e | 4998 | out_irqfd: |
7dac16c3 AH |
4999 | kvm_arch_exit(); |
5000 | out_fail: | |
6aa8b732 AK |
5001 | return r; |
5002 | } | |
cb498ea2 | 5003 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 5004 | |
cb498ea2 | 5005 | void kvm_exit(void) |
6aa8b732 | 5006 | { |
4bd33b56 | 5007 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 5008 | misc_deregister(&kvm_dev); |
c16f862d | 5009 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 5010 | kvm_async_pf_deinit(); |
fb3600cc | 5011 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 5012 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5013 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 5014 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 5015 | kvm_arch_hardware_unsetup(); |
f8c16bba | 5016 | kvm_arch_exit(); |
a0f155e9 | 5017 | kvm_irqfd_exit(); |
7f59f492 | 5018 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 5019 | kvm_vfio_ops_exit(); |
6aa8b732 | 5020 | } |
cb498ea2 | 5021 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
5022 | |
5023 | struct kvm_vm_worker_thread_context { | |
5024 | struct kvm *kvm; | |
5025 | struct task_struct *parent; | |
5026 | struct completion init_done; | |
5027 | kvm_vm_thread_fn_t thread_fn; | |
5028 | uintptr_t data; | |
5029 | int err; | |
5030 | }; | |
5031 | ||
5032 | static int kvm_vm_worker_thread(void *context) | |
5033 | { | |
5034 | /* | |
5035 | * The init_context is allocated on the stack of the parent thread, so | |
5036 | * we have to locally copy anything that is needed beyond initialization | |
5037 | */ | |
5038 | struct kvm_vm_worker_thread_context *init_context = context; | |
5039 | struct kvm *kvm = init_context->kvm; | |
5040 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
5041 | uintptr_t data = init_context->data; | |
5042 | int err; | |
5043 | ||
5044 | err = kthread_park(current); | |
5045 | /* kthread_park(current) is never supposed to return an error */ | |
5046 | WARN_ON(err != 0); | |
5047 | if (err) | |
5048 | goto init_complete; | |
5049 | ||
5050 | err = cgroup_attach_task_all(init_context->parent, current); | |
5051 | if (err) { | |
5052 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
5053 | __func__, err); | |
5054 | goto init_complete; | |
5055 | } | |
5056 | ||
5057 | set_user_nice(current, task_nice(init_context->parent)); | |
5058 | ||
5059 | init_complete: | |
5060 | init_context->err = err; | |
5061 | complete(&init_context->init_done); | |
5062 | init_context = NULL; | |
5063 | ||
5064 | if (err) | |
5065 | return err; | |
5066 | ||
5067 | /* Wait to be woken up by the spawner before proceeding. */ | |
5068 | kthread_parkme(); | |
5069 | ||
5070 | if (!kthread_should_stop()) | |
5071 | err = thread_fn(kvm, data); | |
5072 | ||
5073 | return err; | |
5074 | } | |
5075 | ||
5076 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
5077 | uintptr_t data, const char *name, | |
5078 | struct task_struct **thread_ptr) | |
5079 | { | |
5080 | struct kvm_vm_worker_thread_context init_context = {}; | |
5081 | struct task_struct *thread; | |
5082 | ||
5083 | *thread_ptr = NULL; | |
5084 | init_context.kvm = kvm; | |
5085 | init_context.parent = current; | |
5086 | init_context.thread_fn = thread_fn; | |
5087 | init_context.data = data; | |
5088 | init_completion(&init_context.init_done); | |
5089 | ||
5090 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
5091 | "%s-%d", name, task_pid_nr(current)); | |
5092 | if (IS_ERR(thread)) | |
5093 | return PTR_ERR(thread); | |
5094 | ||
5095 | /* kthread_run is never supposed to return NULL */ | |
5096 | WARN_ON(thread == NULL); | |
5097 | ||
5098 | wait_for_completion(&init_context.init_done); | |
5099 | ||
5100 | if (!init_context.err) | |
5101 | *thread_ptr = thread; | |
5102 | ||
5103 | return init_context.err; | |
5104 | } |