Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
6aa8b732 AK |
2 | /* |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * This module enables machines with Intel VT-x extensions to run virtual | |
6 | * machines without emulation or binary translation. | |
7 | * | |
8 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 9 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
10 | * |
11 | * Authors: | |
12 | * Avi Kivity <avi@qumranet.com> | |
13 | * Yaniv Kamay <yaniv@qumranet.com> | |
6aa8b732 AK |
14 | */ |
15 | ||
af669ac6 | 16 | #include <kvm/iodev.h> |
6aa8b732 | 17 | |
edf88417 | 18 | #include <linux/kvm_host.h> |
6aa8b732 AK |
19 | #include <linux/kvm.h> |
20 | #include <linux/module.h> | |
21 | #include <linux/errno.h> | |
6aa8b732 | 22 | #include <linux/percpu.h> |
6aa8b732 AK |
23 | #include <linux/mm.h> |
24 | #include <linux/miscdevice.h> | |
25 | #include <linux/vmalloc.h> | |
6aa8b732 | 26 | #include <linux/reboot.h> |
6aa8b732 AK |
27 | #include <linux/debugfs.h> |
28 | #include <linux/highmem.h> | |
29 | #include <linux/file.h> | |
fb3600cc | 30 | #include <linux/syscore_ops.h> |
774c47f1 | 31 | #include <linux/cpu.h> |
174cd4b1 | 32 | #include <linux/sched/signal.h> |
6e84f315 | 33 | #include <linux/sched/mm.h> |
03441a34 | 34 | #include <linux/sched/stat.h> |
d9e368d6 AK |
35 | #include <linux/cpumask.h> |
36 | #include <linux/smp.h> | |
d6d28168 | 37 | #include <linux/anon_inodes.h> |
04d2cc77 | 38 | #include <linux/profile.h> |
7aa81cc0 | 39 | #include <linux/kvm_para.h> |
6fc138d2 | 40 | #include <linux/pagemap.h> |
8d4e1288 | 41 | #include <linux/mman.h> |
35149e21 | 42 | #include <linux/swap.h> |
e56d532f | 43 | #include <linux/bitops.h> |
547de29e | 44 | #include <linux/spinlock.h> |
6ff5894c | 45 | #include <linux/compat.h> |
bc6678a3 | 46 | #include <linux/srcu.h> |
8f0b1ab6 | 47 | #include <linux/hugetlb.h> |
5a0e3ad6 | 48 | #include <linux/slab.h> |
743eeb0b SL |
49 | #include <linux/sort.h> |
50 | #include <linux/bsearch.h> | |
c011d23b | 51 | #include <linux/io.h> |
2eb06c30 | 52 | #include <linux/lockdep.h> |
c57c8046 | 53 | #include <linux/kthread.h> |
2fdef3a2 | 54 | #include <linux/suspend.h> |
6aa8b732 | 55 | |
e495606d | 56 | #include <asm/processor.h> |
2ea75be3 | 57 | #include <asm/ioctl.h> |
7c0f6ba6 | 58 | #include <linux/uaccess.h> |
6aa8b732 | 59 | |
5f94c174 | 60 | #include "coalesced_mmio.h" |
af585b92 | 61 | #include "async_pf.h" |
531810ca | 62 | #include "mmu_lock.h" |
3c3c29fd | 63 | #include "vfio.h" |
5f94c174 | 64 | |
229456fc MT |
65 | #define CREATE_TRACE_POINTS |
66 | #include <trace/events/kvm.h> | |
67 | ||
fb04a1ed PX |
68 | #include <linux/kvm_dirty_ring.h> |
69 | ||
536a6f88 JF |
70 | /* Worst case buffer size needed for holding an integer. */ |
71 | #define ITOA_MAX_LEN 12 | |
72 | ||
6aa8b732 AK |
73 | MODULE_AUTHOR("Qumranet"); |
74 | MODULE_LICENSE("GPL"); | |
75 | ||
920552b2 | 76 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 77 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
039c5d1b | 78 | module_param(halt_poll_ns, uint, 0644); |
ec76d819 | 79 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 80 | |
aca6ff29 | 81 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 82 | unsigned int halt_poll_ns_grow = 2; |
039c5d1b | 83 | module_param(halt_poll_ns_grow, uint, 0644); |
ec76d819 | 84 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 | 85 | |
49113d36 NW |
86 | /* The start value to grow halt_poll_ns from */ |
87 | unsigned int halt_poll_ns_grow_start = 10000; /* 10us */ | |
88 | module_param(halt_poll_ns_grow_start, uint, 0644); | |
89 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start); | |
90 | ||
aca6ff29 | 91 | /* Default resets per-vcpu halt_poll_ns . */ |
ec76d819 | 92 | unsigned int halt_poll_ns_shrink; |
039c5d1b | 93 | module_param(halt_poll_ns_shrink, uint, 0644); |
ec76d819 | 94 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 95 | |
fa40a821 MT |
96 | /* |
97 | * Ordering of locks: | |
98 | * | |
b7d409de | 99 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
100 | */ |
101 | ||
0d9ce162 | 102 | DEFINE_MUTEX(kvm_lock); |
4a937f96 | 103 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 104 | LIST_HEAD(vm_list); |
133de902 | 105 | |
7f59f492 | 106 | static cpumask_var_t cpus_hardware_enabled; |
f4fee932 | 107 | static int kvm_usage_count; |
10474ae8 | 108 | static atomic_t hardware_enable_failed; |
1b6c0168 | 109 | |
aaba298c | 110 | static struct kmem_cache *kvm_vcpu_cache; |
1165f5fe | 111 | |
15ad7146 | 112 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
7495e22b | 113 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu); |
15ad7146 | 114 | |
76f7c879 | 115 | struct dentry *kvm_debugfs_dir; |
e23a808b | 116 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 117 | |
09cbcef6 | 118 | static const struct file_operations stat_fops_per_vm; |
536a6f88 | 119 | |
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 | } | |
a2486020 | 310 | EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request); |
54163a34 | 311 | |
a6d51016 | 312 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 313 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 314 | { |
4ae3cb3a LT |
315 | /* |
316 | * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in | |
317 | * kvm_make_all_cpus_request. | |
318 | */ | |
319 | long dirty_count = smp_load_acquire(&kvm->tlbs_dirty); | |
320 | ||
321 | /* | |
322 | * We want to publish modifications to the page tables before reading | |
323 | * mode. Pairs with a memory barrier in arch-specific code. | |
324 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
325 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
326 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
327 | * | |
328 | * There is already an smp_mb__after_atomic() before | |
329 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
330 | * barrier here. | |
331 | */ | |
b08660e5 TL |
332 | if (!kvm_arch_flush_remote_tlb(kvm) |
333 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
0193cc90 | 334 | ++kvm->stat.generic.remote_tlb_flush; |
a086f6a1 | 335 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 336 | } |
2ba9f0d8 | 337 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 338 | #endif |
2e53d63a | 339 | |
49846896 RR |
340 | void kvm_reload_remote_mmus(struct kvm *kvm) |
341 | { | |
445b8236 | 342 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 343 | } |
2e53d63a | 344 | |
6926f95a SC |
345 | #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE |
346 | static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc, | |
347 | gfp_t gfp_flags) | |
348 | { | |
349 | gfp_flags |= mc->gfp_zero; | |
350 | ||
351 | if (mc->kmem_cache) | |
352 | return kmem_cache_alloc(mc->kmem_cache, gfp_flags); | |
353 | else | |
354 | return (void *)__get_free_page(gfp_flags); | |
355 | } | |
356 | ||
357 | int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min) | |
358 | { | |
359 | void *obj; | |
360 | ||
361 | if (mc->nobjs >= min) | |
362 | return 0; | |
363 | while (mc->nobjs < ARRAY_SIZE(mc->objects)) { | |
364 | obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT); | |
365 | if (!obj) | |
366 | return mc->nobjs >= min ? 0 : -ENOMEM; | |
367 | mc->objects[mc->nobjs++] = obj; | |
368 | } | |
369 | return 0; | |
370 | } | |
371 | ||
372 | int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc) | |
373 | { | |
374 | return mc->nobjs; | |
375 | } | |
376 | ||
377 | void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
378 | { | |
379 | while (mc->nobjs) { | |
380 | if (mc->kmem_cache) | |
381 | kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]); | |
382 | else | |
383 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
384 | } | |
385 | } | |
386 | ||
387 | void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
388 | { | |
389 | void *p; | |
390 | ||
391 | if (WARN_ON(!mc->nobjs)) | |
392 | p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT); | |
393 | else | |
394 | p = mc->objects[--mc->nobjs]; | |
395 | BUG_ON(!p); | |
396 | return p; | |
397 | } | |
398 | #endif | |
399 | ||
8bd826d6 | 400 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 401 | { |
fb3f0f51 RR |
402 | mutex_init(&vcpu->mutex); |
403 | vcpu->cpu = -1; | |
fb3f0f51 RR |
404 | vcpu->kvm = kvm; |
405 | vcpu->vcpu_id = id; | |
34bb10b7 | 406 | vcpu->pid = NULL; |
da4ad88c | 407 | rcuwait_init(&vcpu->wait); |
af585b92 | 408 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 409 | |
bf9f6ac8 FW |
410 | vcpu->pre_pcpu = -1; |
411 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
412 | ||
4c088493 R |
413 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
414 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 415 | vcpu->preempted = false; |
d73eb57b | 416 | vcpu->ready = false; |
d5c48deb | 417 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
fb3f0f51 | 418 | } |
fb3f0f51 | 419 | |
4543bdc0 SC |
420 | void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
421 | { | |
fb04a1ed | 422 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
4543bdc0 | 423 | kvm_arch_vcpu_destroy(vcpu); |
e529ef66 | 424 | |
9941d224 SC |
425 | /* |
426 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
427 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
428 | * are already gone. | |
429 | */ | |
430 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
431 | ||
8bd826d6 | 432 | free_page((unsigned long)vcpu->run); |
e529ef66 | 433 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 SC |
434 | } |
435 | EXPORT_SYMBOL_GPL(kvm_vcpu_destroy); | |
436 | ||
e930bffe AA |
437 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
438 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
439 | { | |
440 | return container_of(mn, struct kvm, mmu_notifier); | |
441 | } | |
442 | ||
e649b3f0 ET |
443 | static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn, |
444 | struct mm_struct *mm, | |
445 | unsigned long start, unsigned long end) | |
446 | { | |
447 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
448 | int idx; | |
449 | ||
450 | idx = srcu_read_lock(&kvm->srcu); | |
451 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
452 | srcu_read_unlock(&kvm->srcu, idx); | |
453 | } | |
454 | ||
3039bcc7 SC |
455 | typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range); |
456 | ||
f922bd9b SC |
457 | typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start, |
458 | unsigned long end); | |
459 | ||
3039bcc7 SC |
460 | struct kvm_hva_range { |
461 | unsigned long start; | |
462 | unsigned long end; | |
463 | pte_t pte; | |
464 | hva_handler_t handler; | |
f922bd9b | 465 | on_lock_fn_t on_lock; |
3039bcc7 SC |
466 | bool flush_on_ret; |
467 | bool may_block; | |
468 | }; | |
469 | ||
f922bd9b SC |
470 | /* |
471 | * Use a dedicated stub instead of NULL to indicate that there is no callback | |
472 | * function/handler. The compiler technically can't guarantee that a real | |
473 | * function will have a non-zero address, and so it will generate code to | |
474 | * check for !NULL, whereas comparing against a stub will be elided at compile | |
475 | * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9). | |
476 | */ | |
477 | static void kvm_null_fn(void) | |
478 | { | |
479 | ||
480 | } | |
481 | #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn) | |
482 | ||
3039bcc7 SC |
483 | static __always_inline int __kvm_handle_hva_range(struct kvm *kvm, |
484 | const struct kvm_hva_range *range) | |
485 | { | |
8931a454 | 486 | bool ret = false, locked = false; |
f922bd9b | 487 | struct kvm_gfn_range gfn_range; |
3039bcc7 SC |
488 | struct kvm_memory_slot *slot; |
489 | struct kvm_memslots *slots; | |
3039bcc7 SC |
490 | int i, idx; |
491 | ||
f922bd9b SC |
492 | /* A null handler is allowed if and only if on_lock() is provided. */ |
493 | if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) && | |
494 | IS_KVM_NULL_FN(range->handler))) | |
495 | return 0; | |
496 | ||
3039bcc7 SC |
497 | idx = srcu_read_lock(&kvm->srcu); |
498 | ||
8931a454 | 499 | /* The on_lock() path does not yet support lock elision. */ |
f922bd9b | 500 | if (!IS_KVM_NULL_FN(range->on_lock)) { |
8931a454 SC |
501 | locked = true; |
502 | KVM_MMU_LOCK(kvm); | |
503 | ||
f922bd9b SC |
504 | range->on_lock(kvm, range->start, range->end); |
505 | ||
506 | if (IS_KVM_NULL_FN(range->handler)) | |
507 | goto out_unlock; | |
508 | } | |
509 | ||
3039bcc7 SC |
510 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
511 | slots = __kvm_memslots(kvm, i); | |
512 | kvm_for_each_memslot(slot, slots) { | |
513 | unsigned long hva_start, hva_end; | |
514 | ||
515 | hva_start = max(range->start, slot->userspace_addr); | |
516 | hva_end = min(range->end, slot->userspace_addr + | |
517 | (slot->npages << PAGE_SHIFT)); | |
518 | if (hva_start >= hva_end) | |
519 | continue; | |
520 | ||
521 | /* | |
522 | * To optimize for the likely case where the address | |
523 | * range is covered by zero or one memslots, don't | |
524 | * bother making these conditional (to avoid writes on | |
525 | * the second or later invocation of the handler). | |
526 | */ | |
527 | gfn_range.pte = range->pte; | |
528 | gfn_range.may_block = range->may_block; | |
529 | ||
530 | /* | |
531 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
532 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
533 | */ | |
534 | gfn_range.start = hva_to_gfn_memslot(hva_start, slot); | |
535 | gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot); | |
536 | gfn_range.slot = slot; | |
537 | ||
8931a454 SC |
538 | if (!locked) { |
539 | locked = true; | |
540 | KVM_MMU_LOCK(kvm); | |
541 | } | |
3039bcc7 SC |
542 | ret |= range->handler(kvm, &gfn_range); |
543 | } | |
544 | } | |
545 | ||
546 | if (range->flush_on_ret && (ret || kvm->tlbs_dirty)) | |
547 | kvm_flush_remote_tlbs(kvm); | |
548 | ||
f922bd9b | 549 | out_unlock: |
8931a454 SC |
550 | if (locked) |
551 | KVM_MMU_UNLOCK(kvm); | |
f922bd9b | 552 | |
3039bcc7 SC |
553 | srcu_read_unlock(&kvm->srcu, idx); |
554 | ||
555 | /* The notifiers are averse to booleans. :-( */ | |
556 | return (int)ret; | |
557 | } | |
558 | ||
559 | static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn, | |
560 | unsigned long start, | |
561 | unsigned long end, | |
562 | pte_t pte, | |
563 | hva_handler_t handler) | |
564 | { | |
565 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
566 | const struct kvm_hva_range range = { | |
567 | .start = start, | |
568 | .end = end, | |
569 | .pte = pte, | |
570 | .handler = handler, | |
f922bd9b | 571 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
572 | .flush_on_ret = true, |
573 | .may_block = false, | |
574 | }; | |
3039bcc7 | 575 | |
f922bd9b | 576 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 SC |
577 | } |
578 | ||
579 | static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn, | |
580 | unsigned long start, | |
581 | unsigned long end, | |
582 | hva_handler_t handler) | |
583 | { | |
584 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
585 | const struct kvm_hva_range range = { | |
586 | .start = start, | |
587 | .end = end, | |
588 | .pte = __pte(0), | |
589 | .handler = handler, | |
f922bd9b | 590 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
591 | .flush_on_ret = false, |
592 | .may_block = false, | |
593 | }; | |
3039bcc7 | 594 | |
f922bd9b | 595 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 | 596 | } |
3da0dd43 IE |
597 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
598 | struct mm_struct *mm, | |
599 | unsigned long address, | |
600 | pte_t pte) | |
601 | { | |
602 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
603 | ||
501b9185 SC |
604 | trace_kvm_set_spte_hva(address); |
605 | ||
c13fda23 SC |
606 | /* |
607 | * .change_pte() must be surrounded by .invalidate_range_{start,end}(), | |
608 | * and so always runs with an elevated notifier count. This obviates | |
609 | * the need to bump the sequence count. | |
610 | */ | |
611 | WARN_ON_ONCE(!kvm->mmu_notifier_count); | |
612 | ||
3039bcc7 | 613 | kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn); |
3da0dd43 IE |
614 | } |
615 | ||
f922bd9b SC |
616 | static void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start, |
617 | unsigned long end) | |
e930bffe | 618 | { |
e930bffe AA |
619 | /* |
620 | * The count increase must become visible at unlock time as no | |
621 | * spte can be established without taking the mmu_lock and | |
622 | * count is also read inside the mmu_lock critical section. | |
623 | */ | |
624 | kvm->mmu_notifier_count++; | |
4a42d848 | 625 | if (likely(kvm->mmu_notifier_count == 1)) { |
f922bd9b SC |
626 | kvm->mmu_notifier_range_start = start; |
627 | kvm->mmu_notifier_range_end = end; | |
4a42d848 DS |
628 | } else { |
629 | /* | |
630 | * Fully tracking multiple concurrent ranges has dimishing | |
631 | * returns. Keep things simple and just find the minimal range | |
632 | * which includes the current and new ranges. As there won't be | |
633 | * enough information to subtract a range after its invalidate | |
634 | * completes, any ranges invalidated concurrently will | |
635 | * accumulate and persist until all outstanding invalidates | |
636 | * complete. | |
637 | */ | |
638 | kvm->mmu_notifier_range_start = | |
f922bd9b | 639 | min(kvm->mmu_notifier_range_start, start); |
4a42d848 | 640 | kvm->mmu_notifier_range_end = |
f922bd9b | 641 | max(kvm->mmu_notifier_range_end, end); |
4a42d848 | 642 | } |
f922bd9b | 643 | } |
3039bcc7 | 644 | |
f922bd9b SC |
645 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
646 | const struct mmu_notifier_range *range) | |
647 | { | |
648 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
649 | const struct kvm_hva_range hva_range = { | |
650 | .start = range->start, | |
651 | .end = range->end, | |
652 | .pte = __pte(0), | |
653 | .handler = kvm_unmap_gfn_range, | |
654 | .on_lock = kvm_inc_notifier_count, | |
655 | .flush_on_ret = true, | |
656 | .may_block = mmu_notifier_range_blockable(range), | |
657 | }; | |
565f3be2 | 658 | |
f922bd9b SC |
659 | trace_kvm_unmap_hva_range(range->start, range->end); |
660 | ||
661 | __kvm_handle_hva_range(kvm, &hva_range); | |
93065ac7 | 662 | |
e649b3f0 | 663 | return 0; |
e930bffe AA |
664 | } |
665 | ||
f922bd9b SC |
666 | static void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start, |
667 | unsigned long end) | |
e930bffe | 668 | { |
e930bffe AA |
669 | /* |
670 | * This sequence increase will notify the kvm page fault that | |
671 | * the page that is going to be mapped in the spte could have | |
672 | * been freed. | |
673 | */ | |
674 | kvm->mmu_notifier_seq++; | |
a355aa54 | 675 | smp_wmb(); |
e930bffe AA |
676 | /* |
677 | * The above sequence increase must be visible before the | |
a355aa54 PM |
678 | * below count decrease, which is ensured by the smp_wmb above |
679 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
680 | */ |
681 | kvm->mmu_notifier_count--; | |
f922bd9b SC |
682 | } |
683 | ||
684 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
685 | const struct mmu_notifier_range *range) | |
686 | { | |
687 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
688 | const struct kvm_hva_range hva_range = { | |
689 | .start = range->start, | |
690 | .end = range->end, | |
691 | .pte = __pte(0), | |
692 | .handler = (void *)kvm_null_fn, | |
693 | .on_lock = kvm_dec_notifier_count, | |
694 | .flush_on_ret = false, | |
695 | .may_block = mmu_notifier_range_blockable(range), | |
696 | }; | |
697 | ||
698 | __kvm_handle_hva_range(kvm, &hva_range); | |
e930bffe AA |
699 | |
700 | BUG_ON(kvm->mmu_notifier_count < 0); | |
701 | } | |
702 | ||
703 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
704 | struct mm_struct *mm, | |
57128468 ALC |
705 | unsigned long start, |
706 | unsigned long end) | |
e930bffe | 707 | { |
501b9185 SC |
708 | trace_kvm_age_hva(start, end); |
709 | ||
3039bcc7 | 710 | return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn); |
e930bffe AA |
711 | } |
712 | ||
1d7715c6 VD |
713 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
714 | struct mm_struct *mm, | |
715 | unsigned long start, | |
716 | unsigned long end) | |
717 | { | |
501b9185 SC |
718 | trace_kvm_age_hva(start, end); |
719 | ||
1d7715c6 VD |
720 | /* |
721 | * Even though we do not flush TLB, this will still adversely | |
722 | * affect performance on pre-Haswell Intel EPT, where there is | |
723 | * no EPT Access Bit to clear so that we have to tear down EPT | |
724 | * tables instead. If we find this unacceptable, we can always | |
725 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
726 | * do anything on clear_young. | |
727 | * | |
728 | * Also note that currently we never issue secondary TLB flushes | |
729 | * from clear_young, leaving this job up to the regular system | |
730 | * cadence. If we find this inaccurate, we might come up with a | |
731 | * more sophisticated heuristic later. | |
732 | */ | |
3039bcc7 | 733 | return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn); |
1d7715c6 VD |
734 | } |
735 | ||
8ee53820 AA |
736 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
737 | struct mm_struct *mm, | |
738 | unsigned long address) | |
739 | { | |
501b9185 SC |
740 | trace_kvm_test_age_hva(address); |
741 | ||
3039bcc7 SC |
742 | return kvm_handle_hva_range_no_flush(mn, address, address + 1, |
743 | kvm_test_age_gfn); | |
8ee53820 AA |
744 | } |
745 | ||
85db06e5 MT |
746 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
747 | struct mm_struct *mm) | |
748 | { | |
749 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
750 | int idx; |
751 | ||
752 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 753 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 754 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
755 | } |
756 | ||
e930bffe | 757 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e649b3f0 | 758 | .invalidate_range = kvm_mmu_notifier_invalidate_range, |
e930bffe AA |
759 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
760 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
761 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 762 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 763 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 764 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 765 | .release = kvm_mmu_notifier_release, |
e930bffe | 766 | }; |
4c07b0a4 AK |
767 | |
768 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
769 | { | |
770 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
771 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
772 | } | |
773 | ||
774 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
775 | ||
776 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
777 | { | |
778 | return 0; | |
779 | } | |
780 | ||
e930bffe AA |
781 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
782 | ||
2fdef3a2 SS |
783 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
784 | static int kvm_pm_notifier_call(struct notifier_block *bl, | |
785 | unsigned long state, | |
786 | void *unused) | |
787 | { | |
788 | struct kvm *kvm = container_of(bl, struct kvm, pm_notifier); | |
789 | ||
790 | return kvm_arch_pm_notifier(kvm, state); | |
791 | } | |
792 | ||
793 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
794 | { | |
795 | kvm->pm_notifier.notifier_call = kvm_pm_notifier_call; | |
796 | /* Suspend KVM before we suspend ftrace, RCU, etc. */ | |
797 | kvm->pm_notifier.priority = INT_MAX; | |
798 | register_pm_notifier(&kvm->pm_notifier); | |
799 | } | |
800 | ||
801 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
802 | { | |
803 | unregister_pm_notifier(&kvm->pm_notifier); | |
804 | } | |
805 | #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
806 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
807 | { | |
808 | } | |
809 | ||
810 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
811 | { | |
812 | } | |
813 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
814 | ||
a47d2b07 | 815 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
816 | { |
817 | int i; | |
a47d2b07 | 818 | struct kvm_memslots *slots; |
bf3e05bc | 819 | |
b12ce36a | 820 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
a47d2b07 PB |
821 | if (!slots) |
822 | return NULL; | |
823 | ||
bf3e05bc | 824 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
36947254 | 825 | slots->id_to_index[i] = -1; |
a47d2b07 PB |
826 | |
827 | return slots; | |
828 | } | |
829 | ||
830 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
831 | { | |
832 | if (!memslot->dirty_bitmap) | |
833 | return; | |
834 | ||
835 | kvfree(memslot->dirty_bitmap); | |
836 | memslot->dirty_bitmap = NULL; | |
837 | } | |
838 | ||
e96c81ee | 839 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 840 | { |
e96c81ee | 841 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 842 | |
e96c81ee | 843 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 844 | |
e96c81ee SC |
845 | slot->flags = 0; |
846 | slot->npages = 0; | |
a47d2b07 PB |
847 | } |
848 | ||
849 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
850 | { | |
851 | struct kvm_memory_slot *memslot; | |
852 | ||
853 | if (!slots) | |
854 | return; | |
855 | ||
856 | kvm_for_each_memslot(memslot, slots) | |
e96c81ee | 857 | kvm_free_memslot(kvm, memslot); |
a47d2b07 PB |
858 | |
859 | kvfree(slots); | |
bf3e05bc XG |
860 | } |
861 | ||
bc9e9e67 JZ |
862 | static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc) |
863 | { | |
864 | switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) { | |
865 | case KVM_STATS_TYPE_INSTANT: | |
866 | return 0444; | |
867 | case KVM_STATS_TYPE_CUMULATIVE: | |
868 | case KVM_STATS_TYPE_PEAK: | |
869 | default: | |
870 | return 0644; | |
871 | } | |
872 | } | |
873 | ||
874 | ||
536a6f88 JF |
875 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
876 | { | |
877 | int i; | |
bc9e9e67 JZ |
878 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
879 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
880 | |
881 | if (!kvm->debugfs_dentry) | |
882 | return; | |
883 | ||
884 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
885 | ||
9d5a1dce LC |
886 | if (kvm->debugfs_stat_data) { |
887 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
888 | kfree(kvm->debugfs_stat_data[i]); | |
889 | kfree(kvm->debugfs_stat_data); | |
890 | } | |
536a6f88 JF |
891 | } |
892 | ||
893 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
894 | { | |
895 | char dir_name[ITOA_MAX_LEN * 2]; | |
896 | struct kvm_stat_data *stat_data; | |
bc9e9e67 JZ |
897 | const struct _kvm_stats_desc *pdesc; |
898 | int i; | |
899 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + | |
900 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
901 | |
902 | if (!debugfs_initialized()) | |
903 | return 0; | |
904 | ||
905 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
929f45e3 | 906 | kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir); |
536a6f88 JF |
907 | |
908 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, | |
909 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 910 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
911 | if (!kvm->debugfs_stat_data) |
912 | return -ENOMEM; | |
913 | ||
bc9e9e67 JZ |
914 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
915 | pdesc = &kvm_vm_stats_desc[i]; | |
b12ce36a | 916 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
917 | if (!stat_data) |
918 | return -ENOMEM; | |
919 | ||
920 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
921 | stat_data->desc = pdesc; |
922 | stat_data->kind = KVM_STAT_VM; | |
923 | kvm->debugfs_stat_data[i] = stat_data; | |
924 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
925 | kvm->debugfs_dentry, stat_data, | |
926 | &stat_fops_per_vm); | |
927 | } | |
928 | ||
929 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
930 | pdesc = &kvm_vcpu_stats_desc[i]; | |
b12ce36a | 931 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
932 | if (!stat_data) |
933 | return -ENOMEM; | |
934 | ||
935 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
936 | stat_data->desc = pdesc; |
937 | stat_data->kind = KVM_STAT_VCPU; | |
938 | kvm->debugfs_stat_data[i] = stat_data; | |
939 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
09cbcef6 MP |
940 | kvm->debugfs_dentry, stat_data, |
941 | &stat_fops_per_vm); | |
536a6f88 JF |
942 | } |
943 | return 0; | |
944 | } | |
945 | ||
1aa9b957 JS |
946 | /* |
947 | * Called after the VM is otherwise initialized, but just before adding it to | |
948 | * the vm_list. | |
949 | */ | |
950 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
951 | { | |
952 | return 0; | |
953 | } | |
954 | ||
955 | /* | |
956 | * Called just after removing the VM from the vm_list, but before doing any | |
957 | * other destruction. | |
958 | */ | |
959 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
960 | { | |
961 | } | |
962 | ||
e08b9637 | 963 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 964 | { |
d89f5eff | 965 | struct kvm *kvm = kvm_arch_alloc_vm(); |
9121923c JM |
966 | int r = -ENOMEM; |
967 | int i; | |
6aa8b732 | 968 | |
d89f5eff JK |
969 | if (!kvm) |
970 | return ERR_PTR(-ENOMEM); | |
971 | ||
531810ca | 972 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 973 | mmgrab(current->mm); |
e9ad4ec8 PB |
974 | kvm->mm = current->mm; |
975 | kvm_eventfd_init(kvm); | |
976 | mutex_init(&kvm->lock); | |
977 | mutex_init(&kvm->irq_lock); | |
978 | mutex_init(&kvm->slots_lock); | |
b10a038e | 979 | mutex_init(&kvm->slots_arch_lock); |
e9ad4ec8 PB |
980 | INIT_LIST_HEAD(&kvm->devices); |
981 | ||
1e702d9a AW |
982 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
983 | ||
8a44119a PB |
984 | if (init_srcu_struct(&kvm->srcu)) |
985 | goto out_err_no_srcu; | |
986 | if (init_srcu_struct(&kvm->irq_srcu)) | |
987 | goto out_err_no_irq_srcu; | |
988 | ||
e2d3fcaf | 989 | refcount_set(&kvm->users_count, 1); |
f481b069 | 990 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4bd518f1 | 991 | struct kvm_memslots *slots = kvm_alloc_memslots(); |
9121923c | 992 | |
4bd518f1 | 993 | if (!slots) |
a97b0e77 | 994 | goto out_err_no_arch_destroy_vm; |
0e32958e | 995 | /* Generations must be different for each address space. */ |
164bf7e5 | 996 | slots->generation = i; |
4bd518f1 | 997 | rcu_assign_pointer(kvm->memslots[i], slots); |
f481b069 | 998 | } |
00f034a1 | 999 | |
e93f8a0f | 1000 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 1001 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 1002 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 1003 | if (!kvm->buses[i]) |
a97b0e77 | 1004 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 1005 | } |
e930bffe | 1006 | |
acd05785 DM |
1007 | kvm->max_halt_poll_ns = halt_poll_ns; |
1008 | ||
e08b9637 | 1009 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 1010 | if (r) |
a97b0e77 | 1011 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
1012 | |
1013 | r = hardware_enable_all(); | |
1014 | if (r) | |
719d93cd | 1015 | goto out_err_no_disable; |
10474ae8 | 1016 | |
c77dcacb | 1017 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 1018 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 1019 | #endif |
6aa8b732 | 1020 | |
74b5c5bf | 1021 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
1022 | if (r) |
1023 | goto out_err_no_mmu_notifier; | |
1024 | ||
1025 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
1026 | if (r) |
1027 | goto out_err; | |
1028 | ||
0d9ce162 | 1029 | mutex_lock(&kvm_lock); |
5e58cfe4 | 1030 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 1031 | mutex_unlock(&kvm_lock); |
d89f5eff | 1032 | |
2ecd9d29 | 1033 | preempt_notifier_inc(); |
2fdef3a2 | 1034 | kvm_init_pm_notifier(kvm); |
2ecd9d29 | 1035 | |
f17abe9a | 1036 | return kvm; |
10474ae8 AG |
1037 | |
1038 | out_err: | |
1aa9b957 JS |
1039 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1040 | if (kvm->mmu_notifier.ops) | |
1041 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
1042 | #endif | |
1043 | out_err_no_mmu_notifier: | |
10474ae8 | 1044 | hardware_disable_all(); |
719d93cd | 1045 | out_err_no_disable: |
a97b0e77 | 1046 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 1047 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 1048 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 1049 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 1050 | kfree(kvm_get_bus(kvm, i)); |
f481b069 | 1051 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 1052 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
8a44119a PB |
1053 | cleanup_srcu_struct(&kvm->irq_srcu); |
1054 | out_err_no_irq_srcu: | |
1055 | cleanup_srcu_struct(&kvm->srcu); | |
1056 | out_err_no_srcu: | |
d89f5eff | 1057 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 1058 | mmdrop(current->mm); |
10474ae8 | 1059 | return ERR_PTR(r); |
f17abe9a AK |
1060 | } |
1061 | ||
07f0a7bd SW |
1062 | static void kvm_destroy_devices(struct kvm *kvm) |
1063 | { | |
e6e3b5a6 | 1064 | struct kvm_device *dev, *tmp; |
07f0a7bd | 1065 | |
a28ebea2 CD |
1066 | /* |
1067 | * We do not need to take the kvm->lock here, because nobody else | |
1068 | * has a reference to the struct kvm at this point and therefore | |
1069 | * cannot access the devices list anyhow. | |
1070 | */ | |
e6e3b5a6 GT |
1071 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
1072 | list_del(&dev->vm_node); | |
07f0a7bd SW |
1073 | dev->ops->destroy(dev); |
1074 | } | |
1075 | } | |
1076 | ||
f17abe9a AK |
1077 | static void kvm_destroy_vm(struct kvm *kvm) |
1078 | { | |
e93f8a0f | 1079 | int i; |
6d4e4c4f AK |
1080 | struct mm_struct *mm = kvm->mm; |
1081 | ||
2fdef3a2 | 1082 | kvm_destroy_pm_notifier(kvm); |
286de8f6 | 1083 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 1084 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 1085 | kvm_arch_sync_events(kvm); |
0d9ce162 | 1086 | mutex_lock(&kvm_lock); |
133de902 | 1087 | list_del(&kvm->vm_list); |
0d9ce162 | 1088 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
1089 | kvm_arch_pre_destroy_vm(kvm); |
1090 | ||
399ec807 | 1091 | kvm_free_irq_routing(kvm); |
df630b8c | 1092 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 1093 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 1094 | |
4a12f951 CB |
1095 | if (bus) |
1096 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
1097 | kvm->buses[i] = NULL; |
1098 | } | |
980da6ce | 1099 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
1100 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1101 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
f00be0ca | 1102 | #else |
2df72e9b | 1103 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 1104 | #endif |
d19a9cd2 | 1105 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 1106 | kvm_destroy_devices(kvm); |
f481b069 | 1107 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 1108 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
820b3fcd | 1109 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
1110 | cleanup_srcu_struct(&kvm->srcu); |
1111 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 1112 | preempt_notifier_dec(); |
10474ae8 | 1113 | hardware_disable_all(); |
6d4e4c4f | 1114 | mmdrop(mm); |
f17abe9a AK |
1115 | } |
1116 | ||
d39f13b0 IE |
1117 | void kvm_get_kvm(struct kvm *kvm) |
1118 | { | |
e3736c3e | 1119 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
1120 | } |
1121 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
1122 | ||
1123 | void kvm_put_kvm(struct kvm *kvm) | |
1124 | { | |
e3736c3e | 1125 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
1126 | kvm_destroy_vm(kvm); |
1127 | } | |
1128 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
1129 | ||
149487bd SC |
1130 | /* |
1131 | * Used to put a reference that was taken on behalf of an object associated | |
1132 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
1133 | * of the new file descriptor fails and the reference cannot be transferred to | |
1134 | * its final owner. In such cases, the caller is still actively using @kvm and | |
1135 | * will fail miserably if the refcount unexpectedly hits zero. | |
1136 | */ | |
1137 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
1138 | { | |
1139 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
1140 | } | |
1141 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 1142 | |
f17abe9a AK |
1143 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
1144 | { | |
1145 | struct kvm *kvm = filp->private_data; | |
1146 | ||
721eecbf GH |
1147 | kvm_irqfd_release(kvm); |
1148 | ||
d39f13b0 | 1149 | kvm_put_kvm(kvm); |
6aa8b732 AK |
1150 | return 0; |
1151 | } | |
1152 | ||
515a0127 TY |
1153 | /* |
1154 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 1155 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 1156 | */ |
3c9bd400 | 1157 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 1158 | { |
515a0127 | 1159 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 1160 | |
b12ce36a | 1161 | memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
1162 | if (!memslot->dirty_bitmap) |
1163 | return -ENOMEM; | |
1164 | ||
a36a57b1 TY |
1165 | return 0; |
1166 | } | |
1167 | ||
bf3e05bc | 1168 | /* |
0577d1ab SC |
1169 | * Delete a memslot by decrementing the number of used slots and shifting all |
1170 | * other entries in the array forward one spot. | |
bf3e05bc | 1171 | */ |
0577d1ab SC |
1172 | static inline void kvm_memslot_delete(struct kvm_memslots *slots, |
1173 | struct kvm_memory_slot *memslot) | |
bf3e05bc | 1174 | { |
063584d4 | 1175 | struct kvm_memory_slot *mslots = slots->memslots; |
0577d1ab | 1176 | int i; |
f85e2cb5 | 1177 | |
0577d1ab SC |
1178 | if (WARN_ON(slots->id_to_index[memslot->id] == -1)) |
1179 | return; | |
0e60b079 | 1180 | |
0577d1ab SC |
1181 | slots->used_slots--; |
1182 | ||
0774a964 SC |
1183 | if (atomic_read(&slots->lru_slot) >= slots->used_slots) |
1184 | atomic_set(&slots->lru_slot, 0); | |
1185 | ||
0577d1ab | 1186 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) { |
7f379cff IM |
1187 | mslots[i] = mslots[i + 1]; |
1188 | slots->id_to_index[mslots[i].id] = i; | |
7f379cff | 1189 | } |
0577d1ab SC |
1190 | mslots[i] = *memslot; |
1191 | slots->id_to_index[memslot->id] = -1; | |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * "Insert" a new memslot by incrementing the number of used slots. Returns | |
1196 | * the new slot's initial index into the memslots array. | |
1197 | */ | |
1198 | static inline int kvm_memslot_insert_back(struct kvm_memslots *slots) | |
1199 | { | |
1200 | return slots->used_slots++; | |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * Move a changed memslot backwards in the array by shifting existing slots | |
1205 | * with a higher GFN toward the front of the array. Note, the changed memslot | |
1206 | * itself is not preserved in the array, i.e. not swapped at this time, only | |
1207 | * its new index into the array is tracked. Returns the changed memslot's | |
1208 | * current index into the memslots array. | |
1209 | */ | |
1210 | static inline int kvm_memslot_move_backward(struct kvm_memslots *slots, | |
1211 | struct kvm_memory_slot *memslot) | |
1212 | { | |
1213 | struct kvm_memory_slot *mslots = slots->memslots; | |
1214 | int i; | |
1215 | ||
1216 | if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) || | |
1217 | WARN_ON_ONCE(!slots->used_slots)) | |
1218 | return -1; | |
efbeec70 PB |
1219 | |
1220 | /* | |
0577d1ab SC |
1221 | * Move the target memslot backward in the array by shifting existing |
1222 | * memslots with a higher GFN (than the target memslot) towards the | |
1223 | * front of the array. | |
efbeec70 | 1224 | */ |
0577d1ab SC |
1225 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) { |
1226 | if (memslot->base_gfn > mslots[i + 1].base_gfn) | |
1227 | break; | |
1228 | ||
1229 | WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn); | |
f85e2cb5 | 1230 | |
0577d1ab SC |
1231 | /* Shift the next memslot forward one and update its index. */ |
1232 | mslots[i] = mslots[i + 1]; | |
1233 | slots->id_to_index[mslots[i].id] = i; | |
1234 | } | |
1235 | return i; | |
1236 | } | |
1237 | ||
1238 | /* | |
1239 | * Move a changed memslot forwards in the array by shifting existing slots with | |
1240 | * a lower GFN toward the back of the array. Note, the changed memslot itself | |
1241 | * is not preserved in the array, i.e. not swapped at this time, only its new | |
1242 | * index into the array is tracked. Returns the changed memslot's final index | |
1243 | * into the memslots array. | |
1244 | */ | |
1245 | static inline int kvm_memslot_move_forward(struct kvm_memslots *slots, | |
1246 | struct kvm_memory_slot *memslot, | |
1247 | int start) | |
1248 | { | |
1249 | struct kvm_memory_slot *mslots = slots->memslots; | |
1250 | int i; | |
1251 | ||
1252 | for (i = start; i > 0; i--) { | |
1253 | if (memslot->base_gfn < mslots[i - 1].base_gfn) | |
1254 | break; | |
1255 | ||
1256 | WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn); | |
1257 | ||
1258 | /* Shift the next memslot back one and update its index. */ | |
1259 | mslots[i] = mslots[i - 1]; | |
1260 | slots->id_to_index[mslots[i].id] = i; | |
1261 | } | |
1262 | return i; | |
1263 | } | |
1264 | ||
1265 | /* | |
1266 | * Re-sort memslots based on their GFN to account for an added, deleted, or | |
1267 | * moved memslot. Sorting memslots by GFN allows using a binary search during | |
1268 | * memslot lookup. | |
1269 | * | |
1270 | * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry | |
1271 | * at memslots[0] has the highest GFN. | |
1272 | * | |
1273 | * The sorting algorithm takes advantage of having initially sorted memslots | |
1274 | * and knowing the position of the changed memslot. Sorting is also optimized | |
1275 | * by not swapping the updated memslot and instead only shifting other memslots | |
1276 | * and tracking the new index for the update memslot. Only once its final | |
1277 | * index is known is the updated memslot copied into its position in the array. | |
1278 | * | |
1279 | * - When deleting a memslot, the deleted memslot simply needs to be moved to | |
1280 | * the end of the array. | |
1281 | * | |
1282 | * - When creating a memslot, the algorithm "inserts" the new memslot at the | |
1283 | * end of the array and then it forward to its correct location. | |
1284 | * | |
1285 | * - When moving a memslot, the algorithm first moves the updated memslot | |
1286 | * backward to handle the scenario where the memslot's GFN was changed to a | |
1287 | * lower value. update_memslots() then falls through and runs the same flow | |
1288 | * as creating a memslot to move the memslot forward to handle the scenario | |
1289 | * where its GFN was changed to a higher value. | |
1290 | * | |
1291 | * Note, slots are sorted from highest->lowest instead of lowest->highest for | |
1292 | * historical reasons. Originally, invalid memslots where denoted by having | |
1293 | * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots | |
1294 | * to the end of the array. The current algorithm uses dedicated logic to | |
1295 | * delete a memslot and thus does not rely on invalid memslots having GFN=0. | |
1296 | * | |
1297 | * The other historical motiviation for highest->lowest was to improve the | |
1298 | * performance of memslot lookup. KVM originally used a linear search starting | |
1299 | * at memslots[0]. On x86, the largest memslot usually has one of the highest, | |
1300 | * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a | |
1301 | * single memslot above the 4gb boundary. As the largest memslot is also the | |
1302 | * most likely to be referenced, sorting it to the front of the array was | |
1303 | * advantageous. The current binary search starts from the middle of the array | |
1304 | * and uses an LRU pointer to improve performance for all memslots and GFNs. | |
1305 | */ | |
1306 | static void update_memslots(struct kvm_memslots *slots, | |
1307 | struct kvm_memory_slot *memslot, | |
1308 | enum kvm_mr_change change) | |
1309 | { | |
1310 | int i; | |
1311 | ||
1312 | if (change == KVM_MR_DELETE) { | |
1313 | kvm_memslot_delete(slots, memslot); | |
1314 | } else { | |
1315 | if (change == KVM_MR_CREATE) | |
1316 | i = kvm_memslot_insert_back(slots); | |
1317 | else | |
1318 | i = kvm_memslot_move_backward(slots, memslot); | |
1319 | i = kvm_memslot_move_forward(slots, memslot, i); | |
1320 | ||
1321 | /* | |
1322 | * Copy the memslot to its new position in memslots and update | |
1323 | * its index accordingly. | |
1324 | */ | |
1325 | slots->memslots[i] = *memslot; | |
1326 | slots->id_to_index[memslot->id] = i; | |
1327 | } | |
bf3e05bc XG |
1328 | } |
1329 | ||
09170a49 | 1330 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 1331 | { |
4d8b81ab XG |
1332 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1333 | ||
0f8a4de3 | 1334 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
1335 | valid_flags |= KVM_MEM_READONLY; |
1336 | #endif | |
1337 | ||
1338 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1339 | return -EINVAL; |
1340 | ||
1341 | return 0; | |
1342 | } | |
1343 | ||
7ec4fb44 | 1344 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 1345 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 1346 | { |
f481b069 | 1347 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
361209e0 | 1348 | u64 gen = old_memslots->generation; |
7ec4fb44 | 1349 | |
361209e0 SC |
1350 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1351 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1352 | |
f481b069 | 1353 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
b10a038e BG |
1354 | |
1355 | /* | |
1356 | * Acquired in kvm_set_memslot. Must be released before synchronize | |
1357 | * SRCU below in order to avoid deadlock with another thread | |
1358 | * acquiring the slots_arch_lock in an srcu critical section. | |
1359 | */ | |
1360 | mutex_unlock(&kvm->slots_arch_lock); | |
1361 | ||
7ec4fb44 | 1362 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1363 | |
ee3d1570 | 1364 | /* |
361209e0 | 1365 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1366 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1367 | * the number of address spaces. This provides a unique and easily |
1368 | * identifiable generation number while the memslots are in flux. | |
1369 | */ | |
1370 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1371 | ||
1372 | /* | |
4bd518f1 PB |
1373 | * Generations must be unique even across address spaces. We do not need |
1374 | * a global counter for that, instead the generation space is evenly split | |
1375 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1376 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1377 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1378 | */ |
164bf7e5 | 1379 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 1380 | |
15248258 | 1381 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1382 | |
15248258 | 1383 | slots->generation = gen; |
e59dbe09 TY |
1384 | |
1385 | return old_memslots; | |
7ec4fb44 GN |
1386 | } |
1387 | ||
ddc12f2a BG |
1388 | static size_t kvm_memslots_size(int slots) |
1389 | { | |
1390 | return sizeof(struct kvm_memslots) + | |
1391 | (sizeof(struct kvm_memory_slot) * slots); | |
1392 | } | |
1393 | ||
1394 | static void kvm_copy_memslots(struct kvm_memslots *to, | |
1395 | struct kvm_memslots *from) | |
1396 | { | |
1397 | memcpy(to, from, kvm_memslots_size(from->used_slots)); | |
1398 | } | |
1399 | ||
36947254 SC |
1400 | /* |
1401 | * Note, at a minimum, the current number of used slots must be allocated, even | |
1402 | * when deleting a memslot, as we need a complete duplicate of the memslots for | |
1403 | * use when invalidating a memslot prior to deleting/moving the memslot. | |
1404 | */ | |
1405 | static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old, | |
1406 | enum kvm_mr_change change) | |
1407 | { | |
1408 | struct kvm_memslots *slots; | |
ddc12f2a | 1409 | size_t new_size; |
36947254 SC |
1410 | |
1411 | if (change == KVM_MR_CREATE) | |
ddc12f2a | 1412 | new_size = kvm_memslots_size(old->used_slots + 1); |
36947254 | 1413 | else |
ddc12f2a | 1414 | new_size = kvm_memslots_size(old->used_slots); |
36947254 SC |
1415 | |
1416 | slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT); | |
1417 | if (likely(slots)) | |
ddc12f2a | 1418 | kvm_copy_memslots(slots, old); |
36947254 SC |
1419 | |
1420 | return slots; | |
1421 | } | |
1422 | ||
cf47f50b SC |
1423 | static int kvm_set_memslot(struct kvm *kvm, |
1424 | const struct kvm_userspace_memory_region *mem, | |
9d4c197c | 1425 | struct kvm_memory_slot *old, |
cf47f50b SC |
1426 | struct kvm_memory_slot *new, int as_id, |
1427 | enum kvm_mr_change change) | |
1428 | { | |
1429 | struct kvm_memory_slot *slot; | |
1430 | struct kvm_memslots *slots; | |
1431 | int r; | |
1432 | ||
b10a038e BG |
1433 | /* |
1434 | * Released in install_new_memslots. | |
1435 | * | |
1436 | * Must be held from before the current memslots are copied until | |
1437 | * after the new memslots are installed with rcu_assign_pointer, | |
1438 | * then released before the synchronize srcu in install_new_memslots. | |
1439 | * | |
1440 | * When modifying memslots outside of the slots_lock, must be held | |
1441 | * before reading the pointer to the current memslots until after all | |
1442 | * changes to those memslots are complete. | |
1443 | * | |
1444 | * These rules ensure that installing new memslots does not lose | |
1445 | * changes made to the previous memslots. | |
1446 | */ | |
1447 | mutex_lock(&kvm->slots_arch_lock); | |
1448 | ||
36947254 | 1449 | slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change); |
b10a038e BG |
1450 | if (!slots) { |
1451 | mutex_unlock(&kvm->slots_arch_lock); | |
cf47f50b | 1452 | return -ENOMEM; |
b10a038e | 1453 | } |
cf47f50b SC |
1454 | |
1455 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { | |
1456 | /* | |
1457 | * Note, the INVALID flag needs to be in the appropriate entry | |
1458 | * in the freshly allocated memslots, not in @old or @new. | |
1459 | */ | |
1460 | slot = id_to_memslot(slots, old->id); | |
1461 | slot->flags |= KVM_MEMSLOT_INVALID; | |
1462 | ||
1463 | /* | |
b10a038e BG |
1464 | * We can re-use the memory from the old memslots. |
1465 | * It will be overwritten with a copy of the new memslots | |
1466 | * after reacquiring the slots_arch_lock below. | |
cf47f50b SC |
1467 | */ |
1468 | slots = install_new_memslots(kvm, as_id, slots); | |
1469 | ||
1470 | /* From this point no new shadow pages pointing to a deleted, | |
1471 | * or moved, memslot will be created. | |
1472 | * | |
1473 | * validation of sp->gfn happens in: | |
1474 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1475 | * - kvm_is_visible_gfn (mmu_check_root) | |
1476 | */ | |
1477 | kvm_arch_flush_shadow_memslot(kvm, slot); | |
b10a038e BG |
1478 | |
1479 | /* Released in install_new_memslots. */ | |
1480 | mutex_lock(&kvm->slots_arch_lock); | |
1481 | ||
1482 | /* | |
1483 | * The arch-specific fields of the memslots could have changed | |
1484 | * between releasing the slots_arch_lock in | |
1485 | * install_new_memslots and here, so get a fresh copy of the | |
1486 | * slots. | |
1487 | */ | |
1488 | kvm_copy_memslots(slots, __kvm_memslots(kvm, as_id)); | |
cf47f50b SC |
1489 | } |
1490 | ||
1491 | r = kvm_arch_prepare_memory_region(kvm, new, mem, change); | |
1492 | if (r) | |
1493 | goto out_slots; | |
1494 | ||
1495 | update_memslots(slots, new, change); | |
1496 | slots = install_new_memslots(kvm, as_id, slots); | |
1497 | ||
1498 | kvm_arch_commit_memory_region(kvm, mem, old, new, change); | |
1499 | ||
1500 | kvfree(slots); | |
1501 | return 0; | |
1502 | ||
1503 | out_slots: | |
b10a038e BG |
1504 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
1505 | slot = id_to_memslot(slots, old->id); | |
1506 | slot->flags &= ~KVM_MEMSLOT_INVALID; | |
cf47f50b | 1507 | slots = install_new_memslots(kvm, as_id, slots); |
b10a038e BG |
1508 | } else { |
1509 | mutex_unlock(&kvm->slots_arch_lock); | |
1510 | } | |
cf47f50b SC |
1511 | kvfree(slots); |
1512 | return r; | |
1513 | } | |
1514 | ||
5c0b4f3d SC |
1515 | static int kvm_delete_memslot(struct kvm *kvm, |
1516 | const struct kvm_userspace_memory_region *mem, | |
1517 | struct kvm_memory_slot *old, int as_id) | |
1518 | { | |
1519 | struct kvm_memory_slot new; | |
1520 | int r; | |
1521 | ||
1522 | if (!old->npages) | |
1523 | return -EINVAL; | |
1524 | ||
1525 | memset(&new, 0, sizeof(new)); | |
1526 | new.id = old->id; | |
9e9eb226 PX |
1527 | /* |
1528 | * This is only for debugging purpose; it should never be referenced | |
1529 | * for a removed memslot. | |
1530 | */ | |
1531 | new.as_id = as_id; | |
5c0b4f3d SC |
1532 | |
1533 | r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE); | |
1534 | if (r) | |
1535 | return r; | |
1536 | ||
e96c81ee | 1537 | kvm_free_memslot(kvm, old); |
5c0b4f3d SC |
1538 | return 0; |
1539 | } | |
1540 | ||
6aa8b732 AK |
1541 | /* |
1542 | * Allocate some memory and give it an address in the guest physical address | |
1543 | * space. | |
1544 | * | |
1545 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1546 | * |
02d5d55b | 1547 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1548 | */ |
f78e0e2e | 1549 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 1550 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 1551 | { |
6aa8b732 | 1552 | struct kvm_memory_slot old, new; |
163da372 | 1553 | struct kvm_memory_slot *tmp; |
f64c0398 | 1554 | enum kvm_mr_change change; |
163da372 SC |
1555 | int as_id, id; |
1556 | int r; | |
6aa8b732 | 1557 | |
a50d64d6 XG |
1558 | r = check_memory_region_flags(mem); |
1559 | if (r) | |
71a4c30b | 1560 | return r; |
a50d64d6 | 1561 | |
f481b069 PB |
1562 | as_id = mem->slot >> 16; |
1563 | id = (u16)mem->slot; | |
1564 | ||
6aa8b732 AK |
1565 | /* General sanity checks */ |
1566 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
71a4c30b | 1567 | return -EINVAL; |
6aa8b732 | 1568 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 1569 | return -EINVAL; |
fa3d315a | 1570 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 1571 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 1572 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 1573 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 1574 | mem->memory_size)) |
71a4c30b | 1575 | return -EINVAL; |
f481b069 | 1576 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 1577 | return -EINVAL; |
6aa8b732 | 1578 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 1579 | return -EINVAL; |
6aa8b732 | 1580 | |
5c0b4f3d SC |
1581 | /* |
1582 | * Make a full copy of the old memslot, the pointer will become stale | |
1583 | * when the memslots are re-sorted by update_memslots(), and the old | |
1584 | * memslot needs to be referenced after calling update_memslots(), e.g. | |
0dff0846 | 1585 | * to free its resources and for arch specific behavior. |
5c0b4f3d | 1586 | */ |
0577d1ab SC |
1587 | tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
1588 | if (tmp) { | |
1589 | old = *tmp; | |
1590 | tmp = NULL; | |
1591 | } else { | |
1592 | memset(&old, 0, sizeof(old)); | |
1593 | old.id = id; | |
1594 | } | |
163da372 | 1595 | |
5c0b4f3d SC |
1596 | if (!mem->memory_size) |
1597 | return kvm_delete_memslot(kvm, mem, &old, as_id); | |
1598 | ||
9e9eb226 | 1599 | new.as_id = as_id; |
f481b069 | 1600 | new.id = id; |
163da372 SC |
1601 | new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
1602 | new.npages = mem->memory_size >> PAGE_SHIFT; | |
6aa8b732 | 1603 | new.flags = mem->flags; |
414de7ab | 1604 | new.userspace_addr = mem->userspace_addr; |
6aa8b732 | 1605 | |
163da372 SC |
1606 | if (new.npages > KVM_MEM_MAX_NR_PAGES) |
1607 | return -EINVAL; | |
1608 | ||
5c0b4f3d SC |
1609 | if (!old.npages) { |
1610 | change = KVM_MR_CREATE; | |
163da372 SC |
1611 | new.dirty_bitmap = NULL; |
1612 | memset(&new.arch, 0, sizeof(new.arch)); | |
5c0b4f3d SC |
1613 | } else { /* Modify an existing slot. */ |
1614 | if ((new.userspace_addr != old.userspace_addr) || | |
163da372 | 1615 | (new.npages != old.npages) || |
5c0b4f3d | 1616 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) |
71a4c30b | 1617 | return -EINVAL; |
09170a49 | 1618 | |
163da372 | 1619 | if (new.base_gfn != old.base_gfn) |
5c0b4f3d SC |
1620 | change = KVM_MR_MOVE; |
1621 | else if (new.flags != old.flags) | |
1622 | change = KVM_MR_FLAGS_ONLY; | |
1623 | else /* Nothing to change. */ | |
1624 | return 0; | |
163da372 SC |
1625 | |
1626 | /* Copy dirty_bitmap and arch from the current memslot. */ | |
1627 | new.dirty_bitmap = old.dirty_bitmap; | |
1628 | memcpy(&new.arch, &old.arch, sizeof(new.arch)); | |
09170a49 | 1629 | } |
6aa8b732 | 1630 | |
f64c0398 | 1631 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee | 1632 | /* Check for overlaps */ |
163da372 SC |
1633 | kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) { |
1634 | if (tmp->id == id) | |
0a706bee | 1635 | continue; |
163da372 SC |
1636 | if (!((new.base_gfn + new.npages <= tmp->base_gfn) || |
1637 | (new.base_gfn >= tmp->base_gfn + tmp->npages))) | |
71a4c30b | 1638 | return -EEXIST; |
0a706bee | 1639 | } |
6aa8b732 | 1640 | } |
6aa8b732 | 1641 | |
414de7ab SC |
1642 | /* Allocate/free page dirty bitmap as needed */ |
1643 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1644 | new.dirty_bitmap = NULL; | |
044c59c4 | 1645 | else if (!new.dirty_bitmap && !kvm->dirty_ring_size) { |
3c9bd400 | 1646 | r = kvm_alloc_dirty_bitmap(&new); |
71a4c30b SC |
1647 | if (r) |
1648 | return r; | |
3c9bd400 JZ |
1649 | |
1650 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1651 | bitmap_set(new.dirty_bitmap, 0, new.npages); | |
6aa8b732 AK |
1652 | } |
1653 | ||
cf47f50b SC |
1654 | r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change); |
1655 | if (r) | |
1656 | goto out_bitmap; | |
82ce2c96 | 1657 | |
5c0b4f3d SC |
1658 | if (old.dirty_bitmap && !new.dirty_bitmap) |
1659 | kvm_destroy_dirty_bitmap(&old); | |
6aa8b732 AK |
1660 | return 0; |
1661 | ||
bd0e96fd SC |
1662 | out_bitmap: |
1663 | if (new.dirty_bitmap && !old.dirty_bitmap) | |
1664 | kvm_destroy_dirty_bitmap(&new); | |
6aa8b732 | 1665 | return r; |
210c7c4d | 1666 | } |
f78e0e2e SY |
1667 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1668 | ||
1669 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1670 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1671 | { |
1672 | int r; | |
1673 | ||
79fac95e | 1674 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1675 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1676 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1677 | return r; |
1678 | } | |
210c7c4d IE |
1679 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1680 | ||
7940876e SH |
1681 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1682 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1683 | { |
f481b069 | 1684 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1685 | return -EINVAL; |
09170a49 | 1686 | |
47ae31e2 | 1687 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1688 | } |
1689 | ||
0dff0846 | 1690 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
1691 | /** |
1692 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
1693 | * @kvm: pointer to kvm instance | |
1694 | * @log: slot id and address to which we copy the log | |
1695 | * @is_dirty: set to '1' if any dirty pages were found | |
1696 | * @memslot: set to the associated memslot, always valid on success | |
1697 | */ | |
1698 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
1699 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 1700 | { |
9f6b8029 | 1701 | struct kvm_memslots *slots; |
843574a3 | 1702 | int i, as_id, id; |
87bf6e7d | 1703 | unsigned long n; |
6aa8b732 AK |
1704 | unsigned long any = 0; |
1705 | ||
b2cc64c4 PX |
1706 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1707 | if (kvm->dirty_ring_size) | |
1708 | return -ENXIO; | |
1709 | ||
2a49f61d SC |
1710 | *memslot = NULL; |
1711 | *is_dirty = 0; | |
1712 | ||
f481b069 PB |
1713 | as_id = log->slot >> 16; |
1714 | id = (u16)log->slot; | |
1715 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 1716 | return -EINVAL; |
6aa8b732 | 1717 | |
f481b069 | 1718 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 1719 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 1720 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 1721 | return -ENOENT; |
6aa8b732 | 1722 | |
2a49f61d SC |
1723 | kvm_arch_sync_dirty_log(kvm, *memslot); |
1724 | ||
1725 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 1726 | |
cd1a4a98 | 1727 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 1728 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 1729 | |
2a49f61d | 1730 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 1731 | return -EFAULT; |
6aa8b732 | 1732 | |
5bb064dc ZX |
1733 | if (any) |
1734 | *is_dirty = 1; | |
843574a3 | 1735 | return 0; |
6aa8b732 | 1736 | } |
2ba9f0d8 | 1737 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1738 | |
0dff0846 | 1739 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 1740 | /** |
b8b00220 | 1741 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 1742 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
1743 | * @kvm: pointer to kvm instance |
1744 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
1745 | * |
1746 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1747 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1748 | * following order: | |
1749 | * | |
1750 | * 1. Take a snapshot of the bit and clear it if needed. | |
1751 | * 2. Write protect the corresponding page. | |
1752 | * 3. Copy the snapshot to the userspace. | |
1753 | * 4. Upon return caller flushes TLB's if needed. | |
1754 | * | |
1755 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1756 | * entry. This is not a problem because the page is reported dirty using | |
1757 | * the snapshot taken before and step 4 ensures that writes done after | |
1758 | * exiting to userspace will be logged for the next call. | |
1759 | * | |
1760 | */ | |
0dff0846 | 1761 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 1762 | { |
9f6b8029 | 1763 | struct kvm_memslots *slots; |
ba0513b5 | 1764 | struct kvm_memory_slot *memslot; |
58d6db34 | 1765 | int i, as_id, id; |
ba0513b5 MS |
1766 | unsigned long n; |
1767 | unsigned long *dirty_bitmap; | |
1768 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 1769 | bool flush; |
ba0513b5 | 1770 | |
b2cc64c4 PX |
1771 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1772 | if (kvm->dirty_ring_size) | |
1773 | return -ENXIO; | |
1774 | ||
f481b069 PB |
1775 | as_id = log->slot >> 16; |
1776 | id = (u16)log->slot; | |
1777 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 1778 | return -EINVAL; |
ba0513b5 | 1779 | |
f481b069 PB |
1780 | slots = __kvm_memslots(kvm, as_id); |
1781 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
1782 | if (!memslot || !memslot->dirty_bitmap) |
1783 | return -ENOENT; | |
ba0513b5 MS |
1784 | |
1785 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 1786 | |
0dff0846 SC |
1787 | kvm_arch_sync_dirty_log(kvm, memslot); |
1788 | ||
ba0513b5 | 1789 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 1790 | flush = false; |
2a31b9db PB |
1791 | if (kvm->manual_dirty_log_protect) { |
1792 | /* | |
1793 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
1794 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
1795 | * is some code duplication between this function and | |
1796 | * kvm_get_dirty_log, but hopefully all architecture | |
1797 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
1798 | * can be eliminated. | |
1799 | */ | |
1800 | dirty_bitmap_buffer = dirty_bitmap; | |
1801 | } else { | |
1802 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
1803 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 1804 | |
531810ca | 1805 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
1806 | for (i = 0; i < n / sizeof(long); i++) { |
1807 | unsigned long mask; | |
1808 | gfn_t offset; | |
ba0513b5 | 1809 | |
2a31b9db PB |
1810 | if (!dirty_bitmap[i]) |
1811 | continue; | |
1812 | ||
0dff0846 | 1813 | flush = true; |
2a31b9db PB |
1814 | mask = xchg(&dirty_bitmap[i], 0); |
1815 | dirty_bitmap_buffer[i] = mask; | |
1816 | ||
a67794ca LT |
1817 | offset = i * BITS_PER_LONG; |
1818 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1819 | offset, mask); | |
2a31b9db | 1820 | } |
531810ca | 1821 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
1822 | } |
1823 | ||
0dff0846 SC |
1824 | if (flush) |
1825 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
1826 | ||
2a31b9db PB |
1827 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
1828 | return -EFAULT; | |
1829 | return 0; | |
1830 | } | |
0dff0846 SC |
1831 | |
1832 | ||
1833 | /** | |
1834 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
1835 | * @kvm: kvm instance | |
1836 | * @log: slot id and address to which we copy the log | |
1837 | * | |
1838 | * Steps 1-4 below provide general overview of dirty page logging. See | |
1839 | * kvm_get_dirty_log_protect() function description for additional details. | |
1840 | * | |
1841 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
1842 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
1843 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
1844 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
1845 | * writes will be marked dirty for next log read. | |
1846 | * | |
1847 | * 1. Take a snapshot of the bit and clear it if needed. | |
1848 | * 2. Write protect the corresponding page. | |
1849 | * 3. Copy the snapshot to the userspace. | |
1850 | * 4. Flush TLB's if needed. | |
1851 | */ | |
1852 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
1853 | struct kvm_dirty_log *log) | |
1854 | { | |
1855 | int r; | |
1856 | ||
1857 | mutex_lock(&kvm->slots_lock); | |
1858 | ||
1859 | r = kvm_get_dirty_log_protect(kvm, log); | |
1860 | ||
1861 | mutex_unlock(&kvm->slots_lock); | |
1862 | return r; | |
1863 | } | |
2a31b9db PB |
1864 | |
1865 | /** | |
1866 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
1867 | * and reenable dirty page tracking for the corresponding pages. | |
1868 | * @kvm: pointer to kvm instance | |
1869 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
1870 | */ | |
0dff0846 SC |
1871 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
1872 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
1873 | { |
1874 | struct kvm_memslots *slots; | |
1875 | struct kvm_memory_slot *memslot; | |
98938aa8 | 1876 | int as_id, id; |
2a31b9db | 1877 | gfn_t offset; |
98938aa8 | 1878 | unsigned long i, n; |
2a31b9db PB |
1879 | unsigned long *dirty_bitmap; |
1880 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 1881 | bool flush; |
2a31b9db | 1882 | |
b2cc64c4 PX |
1883 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1884 | if (kvm->dirty_ring_size) | |
1885 | return -ENXIO; | |
1886 | ||
2a31b9db PB |
1887 | as_id = log->slot >> 16; |
1888 | id = (u16)log->slot; | |
1889 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
1890 | return -EINVAL; | |
1891 | ||
76d58e0f | 1892 | if (log->first_page & 63) |
2a31b9db PB |
1893 | return -EINVAL; |
1894 | ||
1895 | slots = __kvm_memslots(kvm, as_id); | |
1896 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
1897 | if (!memslot || !memslot->dirty_bitmap) |
1898 | return -ENOENT; | |
2a31b9db PB |
1899 | |
1900 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 1901 | |
4ddc9204 | 1902 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
1903 | |
1904 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
1905 | log->num_pages > memslot->npages - log->first_page || |
1906 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
1907 | return -EINVAL; | |
98938aa8 | 1908 | |
0dff0846 SC |
1909 | kvm_arch_sync_dirty_log(kvm, memslot); |
1910 | ||
1911 | flush = false; | |
2a31b9db PB |
1912 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
1913 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
1914 | return -EFAULT; | |
ba0513b5 | 1915 | |
531810ca | 1916 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
1917 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
1918 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
1919 | i++, offset += BITS_PER_LONG) { |
1920 | unsigned long mask = *dirty_bitmap_buffer++; | |
1921 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
1922 | if (!mask) | |
ba0513b5 MS |
1923 | continue; |
1924 | ||
2a31b9db | 1925 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 1926 | |
2a31b9db PB |
1927 | /* |
1928 | * mask contains the bits that really have been cleared. This | |
1929 | * never includes any bits beyond the length of the memslot (if | |
1930 | * the length is not aligned to 64 pages), therefore it is not | |
1931 | * a problem if userspace sets them in log->dirty_bitmap. | |
1932 | */ | |
58d2930f | 1933 | if (mask) { |
0dff0846 | 1934 | flush = true; |
58d2930f TY |
1935 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
1936 | offset, mask); | |
1937 | } | |
ba0513b5 | 1938 | } |
531810ca | 1939 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 1940 | |
0dff0846 SC |
1941 | if (flush) |
1942 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
1943 | ||
58d6db34 | 1944 | return 0; |
ba0513b5 | 1945 | } |
0dff0846 SC |
1946 | |
1947 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
1948 | struct kvm_clear_dirty_log *log) | |
1949 | { | |
1950 | int r; | |
1951 | ||
1952 | mutex_lock(&kvm->slots_lock); | |
1953 | ||
1954 | r = kvm_clear_dirty_log_protect(kvm, log); | |
1955 | ||
1956 | mutex_unlock(&kvm->slots_lock); | |
1957 | return r; | |
1958 | } | |
1959 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 1960 | |
49c7754c GN |
1961 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
1962 | { | |
1963 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1964 | } | |
a1f4d395 | 1965 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 1966 | |
8e73485c PB |
1967 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
1968 | { | |
1969 | return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn); | |
1970 | } | |
e72436bc | 1971 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot); |
8e73485c | 1972 | |
33e94154 | 1973 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 1974 | { |
bf3e05bc | 1975 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 1976 | |
c36b7150 | 1977 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
1978 | } |
1979 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
1980 | ||
995decb6 VK |
1981 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
1982 | { | |
1983 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1984 | ||
1985 | return kvm_is_visible_memslot(memslot); | |
1986 | } | |
1987 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
1988 | ||
f9b84e19 | 1989 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
1990 | { |
1991 | struct vm_area_struct *vma; | |
1992 | unsigned long addr, size; | |
1993 | ||
1994 | size = PAGE_SIZE; | |
1995 | ||
42cde48b | 1996 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
1997 | if (kvm_is_error_hva(addr)) |
1998 | return PAGE_SIZE; | |
1999 | ||
d8ed45c5 | 2000 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
2001 | vma = find_vma(current->mm, addr); |
2002 | if (!vma) | |
2003 | goto out; | |
2004 | ||
2005 | size = vma_kernel_pagesize(vma); | |
2006 | ||
2007 | out: | |
d8ed45c5 | 2008 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
2009 | |
2010 | return size; | |
2011 | } | |
2012 | ||
4d8b81ab XG |
2013 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
2014 | { | |
2015 | return slot->flags & KVM_MEM_READONLY; | |
2016 | } | |
2017 | ||
4d8b81ab XG |
2018 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2019 | gfn_t *nr_pages, bool write) | |
539cb660 | 2020 | { |
bc6678a3 | 2021 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 2022 | return KVM_HVA_ERR_BAD; |
48987781 | 2023 | |
4d8b81ab XG |
2024 | if (memslot_is_readonly(slot) && write) |
2025 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
2026 | |
2027 | if (nr_pages) | |
2028 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
2029 | ||
4d8b81ab | 2030 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 2031 | } |
48987781 | 2032 | |
4d8b81ab XG |
2033 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2034 | gfn_t *nr_pages) | |
2035 | { | |
2036 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 2037 | } |
48987781 | 2038 | |
4d8b81ab | 2039 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 2040 | gfn_t gfn) |
4d8b81ab XG |
2041 | { |
2042 | return gfn_to_hva_many(slot, gfn, NULL); | |
2043 | } | |
2044 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
2045 | ||
48987781 XG |
2046 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
2047 | { | |
49c7754c | 2048 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 2049 | } |
0d150298 | 2050 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 2051 | |
8e73485c PB |
2052 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
2053 | { | |
2054 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
2055 | } | |
2056 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
2057 | ||
86ab8cff | 2058 | /* |
970c0d4b WY |
2059 | * Return the hva of a @gfn and the R/W attribute if possible. |
2060 | * | |
2061 | * @slot: the kvm_memory_slot which contains @gfn | |
2062 | * @gfn: the gfn to be translated | |
2063 | * @writable: used to return the read/write attribute of the @slot if the hva | |
2064 | * is valid and @writable is not NULL | |
86ab8cff | 2065 | */ |
64d83126 CD |
2066 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
2067 | gfn_t gfn, bool *writable) | |
86ab8cff | 2068 | { |
a2ac07fe GN |
2069 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
2070 | ||
2071 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
2072 | *writable = !memslot_is_readonly(slot); |
2073 | ||
a2ac07fe | 2074 | return hva; |
86ab8cff XG |
2075 | } |
2076 | ||
64d83126 CD |
2077 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
2078 | { | |
2079 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2080 | ||
2081 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2082 | } | |
2083 | ||
8e73485c PB |
2084 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
2085 | { | |
2086 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2087 | ||
2088 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2089 | } | |
2090 | ||
fafc3dba HY |
2091 | static inline int check_user_page_hwpoison(unsigned long addr) |
2092 | { | |
0d731759 | 2093 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 2094 | |
0d731759 | 2095 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
2096 | return rc == -EHWPOISON; |
2097 | } | |
2098 | ||
2fc84311 | 2099 | /* |
b9b33da2 PB |
2100 | * The fast path to get the writable pfn which will be stored in @pfn, |
2101 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 2102 | * only part that runs if we can in atomic context. |
2fc84311 | 2103 | */ |
b9b33da2 PB |
2104 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
2105 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 2106 | { |
8d4e1288 | 2107 | struct page *page[1]; |
954bbbc2 | 2108 | |
12ce13fe XG |
2109 | /* |
2110 | * Fast pin a writable pfn only if it is a write fault request | |
2111 | * or the caller allows to map a writable pfn for a read fault | |
2112 | * request. | |
2113 | */ | |
2114 | if (!(write_fault || writable)) | |
2115 | return false; | |
612819c3 | 2116 | |
dadbb612 | 2117 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 2118 | *pfn = page_to_pfn(page[0]); |
612819c3 | 2119 | |
2fc84311 XG |
2120 | if (writable) |
2121 | *writable = true; | |
2122 | return true; | |
2123 | } | |
af585b92 | 2124 | |
2fc84311 XG |
2125 | return false; |
2126 | } | |
612819c3 | 2127 | |
2fc84311 XG |
2128 | /* |
2129 | * The slow path to get the pfn of the specified host virtual address, | |
2130 | * 1 indicates success, -errno is returned if error is detected. | |
2131 | */ | |
2132 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 2133 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 2134 | { |
ce53053c AV |
2135 | unsigned int flags = FOLL_HWPOISON; |
2136 | struct page *page; | |
2fc84311 | 2137 | int npages = 0; |
612819c3 | 2138 | |
2fc84311 XG |
2139 | might_sleep(); |
2140 | ||
2141 | if (writable) | |
2142 | *writable = write_fault; | |
2143 | ||
ce53053c AV |
2144 | if (write_fault) |
2145 | flags |= FOLL_WRITE; | |
2146 | if (async) | |
2147 | flags |= FOLL_NOWAIT; | |
d4944b0e | 2148 | |
ce53053c | 2149 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
2150 | if (npages != 1) |
2151 | return npages; | |
2152 | ||
2153 | /* map read fault as writable if possible */ | |
12ce13fe | 2154 | if (unlikely(!write_fault) && writable) { |
ce53053c | 2155 | struct page *wpage; |
2fc84311 | 2156 | |
dadbb612 | 2157 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 2158 | *writable = true; |
ce53053c AV |
2159 | put_page(page); |
2160 | page = wpage; | |
612819c3 | 2161 | } |
887c08ac | 2162 | } |
ce53053c | 2163 | *pfn = page_to_pfn(page); |
2fc84311 XG |
2164 | return npages; |
2165 | } | |
539cb660 | 2166 | |
4d8b81ab XG |
2167 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
2168 | { | |
2169 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
2170 | return false; | |
2e2e3738 | 2171 | |
4d8b81ab XG |
2172 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
2173 | return false; | |
887c08ac | 2174 | |
4d8b81ab XG |
2175 | return true; |
2176 | } | |
bf998156 | 2177 | |
f8be156b NP |
2178 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
2179 | { | |
2180 | if (kvm_is_reserved_pfn(pfn)) | |
2181 | return 1; | |
2182 | return get_page_unless_zero(pfn_to_page(pfn)); | |
2183 | } | |
2184 | ||
92176a8e PB |
2185 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
2186 | unsigned long addr, bool *async, | |
a340b3e2 KA |
2187 | bool write_fault, bool *writable, |
2188 | kvm_pfn_t *p_pfn) | |
92176a8e | 2189 | { |
a9545779 | 2190 | kvm_pfn_t pfn; |
bd2fae8d PB |
2191 | pte_t *ptep; |
2192 | spinlock_t *ptl; | |
add6a0cd PB |
2193 | int r; |
2194 | ||
9fd6dad1 | 2195 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2196 | if (r) { |
2197 | /* | |
2198 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
2199 | * not call the fault handler, so do it here. | |
2200 | */ | |
2201 | bool unlocked = false; | |
64019a2e | 2202 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
2203 | (write_fault ? FAULT_FLAG_WRITE : 0), |
2204 | &unlocked); | |
a8387d0b PB |
2205 | if (unlocked) |
2206 | return -EAGAIN; | |
add6a0cd PB |
2207 | if (r) |
2208 | return r; | |
2209 | ||
9fd6dad1 | 2210 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2211 | if (r) |
2212 | return r; | |
bd2fae8d | 2213 | } |
add6a0cd | 2214 | |
bd2fae8d PB |
2215 | if (write_fault && !pte_write(*ptep)) { |
2216 | pfn = KVM_PFN_ERR_RO_FAULT; | |
2217 | goto out; | |
add6a0cd PB |
2218 | } |
2219 | ||
a340b3e2 | 2220 | if (writable) |
bd2fae8d PB |
2221 | *writable = pte_write(*ptep); |
2222 | pfn = pte_pfn(*ptep); | |
add6a0cd PB |
2223 | |
2224 | /* | |
2225 | * Get a reference here because callers of *hva_to_pfn* and | |
2226 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
2227 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
2228 | * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will | |
2229 | * simply do nothing for reserved pfns. | |
2230 | * | |
2231 | * Whoever called remap_pfn_range is also going to call e.g. | |
2232 | * unmap_mapping_range before the underlying pages are freed, | |
2233 | * causing a call to our MMU notifier. | |
f8be156b NP |
2234 | * |
2235 | * Certain IO or PFNMAP mappings can be backed with valid | |
2236 | * struct pages, but be allocated without refcounting e.g., | |
2237 | * tail pages of non-compound higher order allocations, which | |
2238 | * would then underflow the refcount when the caller does the | |
2239 | * required put_page. Don't allow those pages here. | |
add6a0cd | 2240 | */ |
f8be156b NP |
2241 | if (!kvm_try_get_pfn(pfn)) |
2242 | r = -EFAULT; | |
add6a0cd | 2243 | |
bd2fae8d PB |
2244 | out: |
2245 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 2246 | *p_pfn = pfn; |
f8be156b NP |
2247 | |
2248 | return r; | |
92176a8e PB |
2249 | } |
2250 | ||
12ce13fe XG |
2251 | /* |
2252 | * Pin guest page in memory and return its pfn. | |
2253 | * @addr: host virtual address which maps memory to the guest | |
2254 | * @atomic: whether this function can sleep | |
2255 | * @async: whether this function need to wait IO complete if the | |
2256 | * host page is not in the memory | |
2257 | * @write_fault: whether we should get a writable host page | |
2258 | * @writable: whether it allows to map a writable host page for !@write_fault | |
2259 | * | |
2260 | * The function will map a writable host page for these two cases: | |
2261 | * 1): @write_fault = true | |
2262 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
2263 | * whether the mapping is writable. | |
2264 | */ | |
ba049e93 | 2265 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
2266 | bool write_fault, bool *writable) |
2267 | { | |
2268 | struct vm_area_struct *vma; | |
ba049e93 | 2269 | kvm_pfn_t pfn = 0; |
92176a8e | 2270 | int npages, r; |
2e2e3738 | 2271 | |
2fc84311 XG |
2272 | /* we can do it either atomically or asynchronously, not both */ |
2273 | BUG_ON(atomic && async); | |
8d4e1288 | 2274 | |
b9b33da2 | 2275 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
2276 | return pfn; |
2277 | ||
2278 | if (atomic) | |
2279 | return KVM_PFN_ERR_FAULT; | |
2280 | ||
2281 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
2282 | if (npages == 1) | |
2283 | return pfn; | |
8d4e1288 | 2284 | |
d8ed45c5 | 2285 | mmap_read_lock(current->mm); |
2fc84311 XG |
2286 | if (npages == -EHWPOISON || |
2287 | (!async && check_user_page_hwpoison(addr))) { | |
2288 | pfn = KVM_PFN_ERR_HWPOISON; | |
2289 | goto exit; | |
2290 | } | |
2291 | ||
a8387d0b | 2292 | retry: |
fc98c03b | 2293 | vma = vma_lookup(current->mm, addr); |
2fc84311 XG |
2294 | |
2295 | if (vma == NULL) | |
2296 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 2297 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
a340b3e2 | 2298 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn); |
a8387d0b PB |
2299 | if (r == -EAGAIN) |
2300 | goto retry; | |
92176a8e PB |
2301 | if (r < 0) |
2302 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 2303 | } else { |
4d8b81ab | 2304 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
2305 | *async = true; |
2306 | pfn = KVM_PFN_ERR_FAULT; | |
2307 | } | |
2308 | exit: | |
d8ed45c5 | 2309 | mmap_read_unlock(current->mm); |
2e2e3738 | 2310 | return pfn; |
35149e21 AL |
2311 | } |
2312 | ||
ba049e93 DW |
2313 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
2314 | bool atomic, bool *async, bool write_fault, | |
4a42d848 | 2315 | bool *writable, hva_t *hva) |
887c08ac | 2316 | { |
4d8b81ab XG |
2317 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
2318 | ||
4a42d848 DS |
2319 | if (hva) |
2320 | *hva = addr; | |
2321 | ||
b2740d35 PB |
2322 | if (addr == KVM_HVA_ERR_RO_BAD) { |
2323 | if (writable) | |
2324 | *writable = false; | |
4d8b81ab | 2325 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 2326 | } |
4d8b81ab | 2327 | |
b2740d35 PB |
2328 | if (kvm_is_error_hva(addr)) { |
2329 | if (writable) | |
2330 | *writable = false; | |
81c52c56 | 2331 | return KVM_PFN_NOSLOT; |
b2740d35 | 2332 | } |
4d8b81ab XG |
2333 | |
2334 | /* Do not map writable pfn in the readonly memslot. */ | |
2335 | if (writable && memslot_is_readonly(slot)) { | |
2336 | *writable = false; | |
2337 | writable = NULL; | |
2338 | } | |
2339 | ||
2340 | return hva_to_pfn(addr, atomic, async, write_fault, | |
2341 | writable); | |
887c08ac | 2342 | } |
3520469d | 2343 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 2344 | |
ba049e93 | 2345 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
2346 | bool *writable) |
2347 | { | |
e37afc6e | 2348 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
4a42d848 | 2349 | write_fault, writable, NULL); |
612819c3 MT |
2350 | } |
2351 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
2352 | ||
ba049e93 | 2353 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2354 | { |
4a42d848 | 2355 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL); |
506f0d6f | 2356 | } |
e37afc6e | 2357 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 2358 | |
ba049e93 | 2359 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2360 | { |
4a42d848 | 2361 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL); |
506f0d6f | 2362 | } |
037d92dc | 2363 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 2364 | |
ba049e93 | 2365 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2366 | { |
2367 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2368 | } | |
2369 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
2370 | ||
ba049e93 | 2371 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
2372 | { |
2373 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
2374 | } | |
2375 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
2376 | ||
ba049e93 | 2377 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2378 | { |
2379 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2380 | } | |
2381 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
2382 | ||
d9ef13c2 PB |
2383 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2384 | struct page **pages, int nr_pages) | |
48987781 XG |
2385 | { |
2386 | unsigned long addr; | |
076b925d | 2387 | gfn_t entry = 0; |
48987781 | 2388 | |
d9ef13c2 | 2389 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
2390 | if (kvm_is_error_hva(addr)) |
2391 | return -1; | |
2392 | ||
2393 | if (entry < nr_pages) | |
2394 | return 0; | |
2395 | ||
dadbb612 | 2396 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
2397 | } |
2398 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
2399 | ||
ba049e93 | 2400 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 2401 | { |
81c52c56 | 2402 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 2403 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2404 | |
bf4bea8e | 2405 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 2406 | WARN_ON(1); |
6cede2e6 | 2407 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 2408 | } |
a2766325 XG |
2409 | |
2410 | return pfn_to_page(pfn); | |
2411 | } | |
2412 | ||
35149e21 AL |
2413 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
2414 | { | |
ba049e93 | 2415 | kvm_pfn_t pfn; |
2e2e3738 AL |
2416 | |
2417 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 2418 | |
a2766325 | 2419 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
2420 | } |
2421 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
2422 | ||
91724814 BO |
2423 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache) |
2424 | { | |
2425 | if (pfn == 0) | |
2426 | return; | |
2427 | ||
2428 | if (cache) | |
2429 | cache->pfn = cache->gfn = 0; | |
2430 | ||
2431 | if (dirty) | |
2432 | kvm_release_pfn_dirty(pfn); | |
2433 | else | |
2434 | kvm_release_pfn_clean(pfn); | |
2435 | } | |
2436 | ||
2437 | static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn, | |
2438 | struct gfn_to_pfn_cache *cache, u64 gen) | |
2439 | { | |
2440 | kvm_release_pfn(cache->pfn, cache->dirty, cache); | |
2441 | ||
2442 | cache->pfn = gfn_to_pfn_memslot(slot, gfn); | |
2443 | cache->gfn = gfn; | |
2444 | cache->dirty = false; | |
2445 | cache->generation = gen; | |
2446 | } | |
2447 | ||
1eff70a9 | 2448 | static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn, |
91724814 BO |
2449 | struct kvm_host_map *map, |
2450 | struct gfn_to_pfn_cache *cache, | |
2451 | bool atomic) | |
e45adf66 KA |
2452 | { |
2453 | kvm_pfn_t pfn; | |
2454 | void *hva = NULL; | |
2455 | struct page *page = KVM_UNMAPPED_PAGE; | |
1eff70a9 | 2456 | struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn); |
91724814 | 2457 | u64 gen = slots->generation; |
e45adf66 KA |
2458 | |
2459 | if (!map) | |
2460 | return -EINVAL; | |
2461 | ||
91724814 BO |
2462 | if (cache) { |
2463 | if (!cache->pfn || cache->gfn != gfn || | |
2464 | cache->generation != gen) { | |
2465 | if (atomic) | |
2466 | return -EAGAIN; | |
2467 | kvm_cache_gfn_to_pfn(slot, gfn, cache, gen); | |
2468 | } | |
2469 | pfn = cache->pfn; | |
2470 | } else { | |
2471 | if (atomic) | |
2472 | return -EAGAIN; | |
2473 | pfn = gfn_to_pfn_memslot(slot, gfn); | |
2474 | } | |
e45adf66 KA |
2475 | if (is_error_noslot_pfn(pfn)) |
2476 | return -EINVAL; | |
2477 | ||
2478 | if (pfn_valid(pfn)) { | |
2479 | page = pfn_to_page(pfn); | |
91724814 BO |
2480 | if (atomic) |
2481 | hva = kmap_atomic(page); | |
2482 | else | |
2483 | hva = kmap(page); | |
d30b214d | 2484 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2485 | } else if (!atomic) { |
e45adf66 | 2486 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
91724814 BO |
2487 | } else { |
2488 | return -EINVAL; | |
d30b214d | 2489 | #endif |
e45adf66 KA |
2490 | } |
2491 | ||
2492 | if (!hva) | |
2493 | return -EFAULT; | |
2494 | ||
2495 | map->page = page; | |
2496 | map->hva = hva; | |
2497 | map->pfn = pfn; | |
2498 | map->gfn = gfn; | |
2499 | ||
2500 | return 0; | |
2501 | } | |
2502 | ||
91724814 BO |
2503 | int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map, |
2504 | struct gfn_to_pfn_cache *cache, bool atomic) | |
1eff70a9 | 2505 | { |
91724814 BO |
2506 | return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map, |
2507 | cache, atomic); | |
1eff70a9 BO |
2508 | } |
2509 | EXPORT_SYMBOL_GPL(kvm_map_gfn); | |
2510 | ||
e45adf66 KA |
2511 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
2512 | { | |
91724814 BO |
2513 | return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map, |
2514 | NULL, false); | |
e45adf66 KA |
2515 | } |
2516 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); | |
2517 | ||
28bd726a PX |
2518 | static void __kvm_unmap_gfn(struct kvm *kvm, |
2519 | struct kvm_memory_slot *memslot, | |
91724814 BO |
2520 | struct kvm_host_map *map, |
2521 | struct gfn_to_pfn_cache *cache, | |
2522 | bool dirty, bool atomic) | |
e45adf66 KA |
2523 | { |
2524 | if (!map) | |
2525 | return; | |
2526 | ||
2527 | if (!map->hva) | |
2528 | return; | |
2529 | ||
91724814 BO |
2530 | if (map->page != KVM_UNMAPPED_PAGE) { |
2531 | if (atomic) | |
2532 | kunmap_atomic(map->hva); | |
2533 | else | |
2534 | kunmap(map->page); | |
2535 | } | |
eb1f2f38 | 2536 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2537 | else if (!atomic) |
e45adf66 | 2538 | memunmap(map->hva); |
91724814 BO |
2539 | else |
2540 | WARN_ONCE(1, "Unexpected unmapping in atomic context"); | |
eb1f2f38 | 2541 | #endif |
e45adf66 | 2542 | |
91724814 | 2543 | if (dirty) |
28bd726a | 2544 | mark_page_dirty_in_slot(kvm, memslot, map->gfn); |
91724814 BO |
2545 | |
2546 | if (cache) | |
2547 | cache->dirty |= dirty; | |
2548 | else | |
2549 | kvm_release_pfn(map->pfn, dirty, NULL); | |
e45adf66 KA |
2550 | |
2551 | map->hva = NULL; | |
2552 | map->page = NULL; | |
2553 | } | |
1eff70a9 | 2554 | |
91724814 BO |
2555 | int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map, |
2556 | struct gfn_to_pfn_cache *cache, bool dirty, bool atomic) | |
1eff70a9 | 2557 | { |
28bd726a | 2558 | __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map, |
91724814 | 2559 | cache, dirty, atomic); |
1eff70a9 BO |
2560 | return 0; |
2561 | } | |
2562 | EXPORT_SYMBOL_GPL(kvm_unmap_gfn); | |
2563 | ||
2564 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) | |
2565 | { | |
28bd726a PX |
2566 | __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), |
2567 | map, NULL, dirty, false); | |
1eff70a9 | 2568 | } |
e45adf66 KA |
2569 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); |
2570 | ||
8e73485c PB |
2571 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
2572 | { | |
ba049e93 | 2573 | kvm_pfn_t pfn; |
8e73485c PB |
2574 | |
2575 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
2576 | ||
2577 | return kvm_pfn_to_page(pfn); | |
2578 | } | |
2579 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
2580 | ||
b4231d61 IE |
2581 | void kvm_release_page_clean(struct page *page) |
2582 | { | |
32cad84f XG |
2583 | WARN_ON(is_error_page(page)); |
2584 | ||
35149e21 | 2585 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
2586 | } |
2587 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
2588 | ||
ba049e93 | 2589 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 2590 | { |
bf4bea8e | 2591 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 2592 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
2593 | } |
2594 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
2595 | ||
b4231d61 | 2596 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 2597 | { |
a2766325 XG |
2598 | WARN_ON(is_error_page(page)); |
2599 | ||
35149e21 AL |
2600 | kvm_release_pfn_dirty(page_to_pfn(page)); |
2601 | } | |
2602 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
2603 | ||
f7a6509f | 2604 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
2605 | { |
2606 | kvm_set_pfn_dirty(pfn); | |
2607 | kvm_release_pfn_clean(pfn); | |
2608 | } | |
f7a6509f | 2609 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 2610 | |
ba049e93 | 2611 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2612 | { |
d29c03a5 ML |
2613 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2614 | SetPageDirty(pfn_to_page(pfn)); | |
8a7ae055 | 2615 | } |
35149e21 AL |
2616 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
2617 | ||
ba049e93 | 2618 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 2619 | { |
a78986aa | 2620 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2e2e3738 | 2621 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
2622 | } |
2623 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
2624 | ||
ba049e93 | 2625 | void kvm_get_pfn(kvm_pfn_t pfn) |
35149e21 | 2626 | { |
bf4bea8e | 2627 | if (!kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 2628 | get_page(pfn_to_page(pfn)); |
35149e21 AL |
2629 | } |
2630 | EXPORT_SYMBOL_GPL(kvm_get_pfn); | |
8a7ae055 | 2631 | |
195aefde IE |
2632 | static int next_segment(unsigned long len, int offset) |
2633 | { | |
2634 | if (len > PAGE_SIZE - offset) | |
2635 | return PAGE_SIZE - offset; | |
2636 | else | |
2637 | return len; | |
2638 | } | |
2639 | ||
8e73485c PB |
2640 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
2641 | void *data, int offset, int len) | |
195aefde | 2642 | { |
e0506bcb IE |
2643 | int r; |
2644 | unsigned long addr; | |
195aefde | 2645 | |
8e73485c | 2646 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
2647 | if (kvm_is_error_hva(addr)) |
2648 | return -EFAULT; | |
3180a7fc | 2649 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 2650 | if (r) |
195aefde | 2651 | return -EFAULT; |
195aefde IE |
2652 | return 0; |
2653 | } | |
8e73485c PB |
2654 | |
2655 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
2656 | int len) | |
2657 | { | |
2658 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2659 | ||
2660 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2661 | } | |
195aefde IE |
2662 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
2663 | ||
8e73485c PB |
2664 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
2665 | int offset, int len) | |
2666 | { | |
2667 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2668 | ||
2669 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2670 | } | |
2671 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
2672 | ||
195aefde IE |
2673 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
2674 | { | |
2675 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2676 | int seg; | |
2677 | int offset = offset_in_page(gpa); | |
2678 | int ret; | |
2679 | ||
2680 | while ((seg = next_segment(len, offset)) != 0) { | |
2681 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
2682 | if (ret < 0) | |
2683 | return ret; | |
2684 | offset = 0; | |
2685 | len -= seg; | |
2686 | data += seg; | |
2687 | ++gfn; | |
2688 | } | |
2689 | return 0; | |
2690 | } | |
2691 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
2692 | ||
8e73485c | 2693 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 2694 | { |
7ec54588 | 2695 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 2696 | int seg; |
7ec54588 | 2697 | int offset = offset_in_page(gpa); |
8e73485c PB |
2698 | int ret; |
2699 | ||
2700 | while ((seg = next_segment(len, offset)) != 0) { | |
2701 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
2702 | if (ret < 0) | |
2703 | return ret; | |
2704 | offset = 0; | |
2705 | len -= seg; | |
2706 | data += seg; | |
2707 | ++gfn; | |
2708 | } | |
2709 | return 0; | |
2710 | } | |
2711 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 2712 | |
8e73485c PB |
2713 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2714 | void *data, int offset, unsigned long len) | |
2715 | { | |
2716 | int r; | |
2717 | unsigned long addr; | |
2718 | ||
2719 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
2720 | if (kvm_is_error_hva(addr)) |
2721 | return -EFAULT; | |
0aac03f0 | 2722 | pagefault_disable(); |
3180a7fc | 2723 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 2724 | pagefault_enable(); |
7ec54588 MT |
2725 | if (r) |
2726 | return -EFAULT; | |
2727 | return 0; | |
2728 | } | |
7ec54588 | 2729 | |
8e73485c PB |
2730 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
2731 | void *data, unsigned long len) | |
2732 | { | |
2733 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2734 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2735 | int offset = offset_in_page(gpa); | |
2736 | ||
2737 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2738 | } | |
2739 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
2740 | ||
28bd726a PX |
2741 | static int __kvm_write_guest_page(struct kvm *kvm, |
2742 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 2743 | const void *data, int offset, int len) |
195aefde | 2744 | { |
e0506bcb IE |
2745 | int r; |
2746 | unsigned long addr; | |
195aefde | 2747 | |
251eb841 | 2748 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
2749 | if (kvm_is_error_hva(addr)) |
2750 | return -EFAULT; | |
8b0cedff | 2751 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 2752 | if (r) |
195aefde | 2753 | return -EFAULT; |
28bd726a | 2754 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
2755 | return 0; |
2756 | } | |
8e73485c PB |
2757 | |
2758 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
2759 | const void *data, int offset, int len) | |
2760 | { | |
2761 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2762 | ||
28bd726a | 2763 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 2764 | } |
195aefde IE |
2765 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
2766 | ||
8e73485c PB |
2767 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
2768 | const void *data, int offset, int len) | |
2769 | { | |
2770 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2771 | ||
28bd726a | 2772 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
2773 | } |
2774 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
2775 | ||
195aefde IE |
2776 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
2777 | unsigned long len) | |
2778 | { | |
2779 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2780 | int seg; | |
2781 | int offset = offset_in_page(gpa); | |
2782 | int ret; | |
2783 | ||
2784 | while ((seg = next_segment(len, offset)) != 0) { | |
2785 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
2786 | if (ret < 0) | |
2787 | return ret; | |
2788 | offset = 0; | |
2789 | len -= seg; | |
2790 | data += seg; | |
2791 | ++gfn; | |
2792 | } | |
2793 | return 0; | |
2794 | } | |
ff651cb6 | 2795 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 2796 | |
8e73485c PB |
2797 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
2798 | unsigned long len) | |
2799 | { | |
2800 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2801 | int seg; | |
2802 | int offset = offset_in_page(gpa); | |
2803 | int ret; | |
2804 | ||
2805 | while ((seg = next_segment(len, offset)) != 0) { | |
2806 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
2807 | if (ret < 0) | |
2808 | return ret; | |
2809 | offset = 0; | |
2810 | len -= seg; | |
2811 | data += seg; | |
2812 | ++gfn; | |
2813 | } | |
2814 | return 0; | |
2815 | } | |
2816 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
2817 | ||
5a2d4365 PB |
2818 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
2819 | struct gfn_to_hva_cache *ghc, | |
2820 | gpa_t gpa, unsigned long len) | |
49c7754c | 2821 | { |
49c7754c | 2822 | int offset = offset_in_page(gpa); |
8f964525 AH |
2823 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
2824 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
2825 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
2826 | gfn_t nr_pages_avail; | |
49c7754c | 2827 | |
6ad1e29f | 2828 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 2829 | ghc->generation = slots->generation; |
6ad1e29f SC |
2830 | |
2831 | if (start_gfn > end_gfn) { | |
2832 | ghc->hva = KVM_HVA_ERR_BAD; | |
2833 | return -EINVAL; | |
2834 | } | |
f1b9dd5e JM |
2835 | |
2836 | /* | |
2837 | * If the requested region crosses two memslots, we still | |
2838 | * verify that the entire region is valid here. | |
2839 | */ | |
6ad1e29f | 2840 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
2841 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
2842 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
2843 | &nr_pages_avail); | |
2844 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 2845 | return -EFAULT; |
f1b9dd5e JM |
2846 | } |
2847 | ||
2848 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 2849 | if (nr_pages_needed == 1) |
49c7754c | 2850 | ghc->hva += offset; |
f1b9dd5e | 2851 | else |
8f964525 | 2852 | ghc->memslot = NULL; |
f1b9dd5e | 2853 | |
6ad1e29f SC |
2854 | ghc->gpa = gpa; |
2855 | ghc->len = len; | |
2856 | return 0; | |
49c7754c | 2857 | } |
5a2d4365 | 2858 | |
4e335d9e | 2859 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
2860 | gpa_t gpa, unsigned long len) |
2861 | { | |
4e335d9e | 2862 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
2863 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
2864 | } | |
4e335d9e | 2865 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 2866 | |
4e335d9e | 2867 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
2868 | void *data, unsigned int offset, |
2869 | unsigned long len) | |
49c7754c | 2870 | { |
4e335d9e | 2871 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 2872 | int r; |
4ec6e863 | 2873 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 2874 | |
4ec6e863 | 2875 | BUG_ON(len + offset > ghc->len); |
8f964525 | 2876 | |
dc9ce71e SC |
2877 | if (slots->generation != ghc->generation) { |
2878 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
2879 | return -EFAULT; | |
2880 | } | |
8f964525 | 2881 | |
49c7754c GN |
2882 | if (kvm_is_error_hva(ghc->hva)) |
2883 | return -EFAULT; | |
2884 | ||
fcfbc617 SC |
2885 | if (unlikely(!ghc->memslot)) |
2886 | return kvm_write_guest(kvm, gpa, data, len); | |
2887 | ||
4ec6e863 | 2888 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
2889 | if (r) |
2890 | return -EFAULT; | |
28bd726a | 2891 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
2892 | |
2893 | return 0; | |
2894 | } | |
4e335d9e | 2895 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 2896 | |
4e335d9e PB |
2897 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2898 | void *data, unsigned long len) | |
4ec6e863 | 2899 | { |
4e335d9e | 2900 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 2901 | } |
4e335d9e | 2902 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 2903 | |
0958f0ce VK |
2904 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
2905 | void *data, unsigned int offset, | |
2906 | unsigned long len) | |
e03b644f | 2907 | { |
4e335d9e | 2908 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 2909 | int r; |
0958f0ce | 2910 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 2911 | |
0958f0ce | 2912 | BUG_ON(len + offset > ghc->len); |
8f964525 | 2913 | |
dc9ce71e SC |
2914 | if (slots->generation != ghc->generation) { |
2915 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
2916 | return -EFAULT; | |
2917 | } | |
8f964525 | 2918 | |
e03b644f GN |
2919 | if (kvm_is_error_hva(ghc->hva)) |
2920 | return -EFAULT; | |
2921 | ||
fcfbc617 | 2922 | if (unlikely(!ghc->memslot)) |
0958f0ce | 2923 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 2924 | |
0958f0ce | 2925 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
2926 | if (r) |
2927 | return -EFAULT; | |
2928 | ||
2929 | return 0; | |
2930 | } | |
0958f0ce VK |
2931 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
2932 | ||
2933 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
2934 | void *data, unsigned long len) | |
2935 | { | |
2936 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
2937 | } | |
4e335d9e | 2938 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 2939 | |
195aefde IE |
2940 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
2941 | { | |
2f541442 | 2942 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
2943 | gfn_t gfn = gpa >> PAGE_SHIFT; |
2944 | int seg; | |
2945 | int offset = offset_in_page(gpa); | |
2946 | int ret; | |
2947 | ||
bfda0e84 | 2948 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 2949 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
2950 | if (ret < 0) |
2951 | return ret; | |
2952 | offset = 0; | |
2953 | len -= seg; | |
2954 | ++gfn; | |
2955 | } | |
2956 | return 0; | |
2957 | } | |
2958 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
2959 | ||
28bd726a PX |
2960 | void mark_page_dirty_in_slot(struct kvm *kvm, |
2961 | struct kvm_memory_slot *memslot, | |
2962 | gfn_t gfn) | |
6aa8b732 | 2963 | { |
044c59c4 | 2964 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 2965 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 2966 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 2967 | |
fb04a1ed PX |
2968 | if (kvm->dirty_ring_size) |
2969 | kvm_dirty_ring_push(kvm_dirty_ring_get(kvm), | |
2970 | slot, rel_gfn); | |
2971 | else | |
2972 | set_bit_le(rel_gfn, memslot->dirty_bitmap); | |
6aa8b732 AK |
2973 | } |
2974 | } | |
a6a0b05d | 2975 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 2976 | |
49c7754c GN |
2977 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
2978 | { | |
2979 | struct kvm_memory_slot *memslot; | |
2980 | ||
2981 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 2982 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 2983 | } |
2ba9f0d8 | 2984 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 2985 | |
8e73485c PB |
2986 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
2987 | { | |
2988 | struct kvm_memory_slot *memslot; | |
2989 | ||
2990 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 2991 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
2992 | } |
2993 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
2994 | ||
20b7035c JS |
2995 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
2996 | { | |
2997 | if (!vcpu->sigset_active) | |
2998 | return; | |
2999 | ||
3000 | /* | |
3001 | * This does a lockless modification of ->real_blocked, which is fine | |
3002 | * because, only current can change ->real_blocked and all readers of | |
3003 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
3004 | * of ->blocked. | |
3005 | */ | |
3006 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
3007 | } | |
3008 | ||
3009 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
3010 | { | |
3011 | if (!vcpu->sigset_active) | |
3012 | return; | |
3013 | ||
3014 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
3015 | sigemptyset(¤t->real_blocked); | |
3016 | } | |
3017 | ||
aca6ff29 WL |
3018 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
3019 | { | |
dee339b5 | 3020 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 3021 | |
2cbd7824 | 3022 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 3023 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3024 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
3025 | if (!grow) |
3026 | goto out; | |
3027 | ||
dee339b5 NW |
3028 | val *= grow; |
3029 | if (val < grow_start) | |
3030 | val = grow_start; | |
aca6ff29 | 3031 | |
258785ef DM |
3032 | if (val > vcpu->kvm->max_halt_poll_ns) |
3033 | val = vcpu->kvm->max_halt_poll_ns; | |
313f636d | 3034 | |
aca6ff29 | 3035 | vcpu->halt_poll_ns = val; |
7fa08e71 | 3036 | out: |
2cbd7824 | 3037 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3038 | } |
3039 | ||
3040 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
3041 | { | |
6b6de68c | 3042 | unsigned int old, val, shrink; |
aca6ff29 | 3043 | |
2cbd7824 | 3044 | old = val = vcpu->halt_poll_ns; |
6b6de68c CB |
3045 | shrink = READ_ONCE(halt_poll_ns_shrink); |
3046 | if (shrink == 0) | |
aca6ff29 WL |
3047 | val = 0; |
3048 | else | |
6b6de68c | 3049 | val /= shrink; |
aca6ff29 WL |
3050 | |
3051 | vcpu->halt_poll_ns = val; | |
2cbd7824 | 3052 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3053 | } |
3054 | ||
f7819512 PB |
3055 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
3056 | { | |
50c28f21 JS |
3057 | int ret = -EINTR; |
3058 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
3059 | ||
f7819512 PB |
3060 | if (kvm_arch_vcpu_runnable(vcpu)) { |
3061 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 3062 | goto out; |
f7819512 PB |
3063 | } |
3064 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 3065 | goto out; |
f7819512 | 3066 | if (signal_pending(current)) |
50c28f21 | 3067 | goto out; |
084071d5 MT |
3068 | if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu)) |
3069 | goto out; | |
f7819512 | 3070 | |
50c28f21 JS |
3071 | ret = 0; |
3072 | out: | |
3073 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
3074 | return ret; | |
f7819512 PB |
3075 | } |
3076 | ||
cb953129 DM |
3077 | static inline void |
3078 | update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited) | |
3079 | { | |
3080 | if (waited) | |
0193cc90 | 3081 | vcpu->stat.generic.halt_poll_fail_ns += poll_ns; |
cb953129 | 3082 | else |
0193cc90 | 3083 | vcpu->stat.generic.halt_poll_success_ns += poll_ns; |
cb953129 DM |
3084 | } |
3085 | ||
b6958ce4 ED |
3086 | /* |
3087 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
3088 | */ | |
8776e519 | 3089 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 3090 | { |
cb953129 | 3091 | ktime_t start, cur, poll_end; |
f7819512 | 3092 | bool waited = false; |
aca6ff29 | 3093 | u64 block_ns; |
f7819512 | 3094 | |
07ab0f8d MZ |
3095 | kvm_arch_vcpu_blocking(vcpu); |
3096 | ||
cb953129 | 3097 | start = cur = poll_end = ktime_get(); |
cdd6ad3a | 3098 | if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) { |
19020f8a | 3099 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); |
f95ef0cd | 3100 | |
0193cc90 | 3101 | ++vcpu->stat.generic.halt_attempted_poll; |
f7819512 PB |
3102 | do { |
3103 | /* | |
3104 | * This sets KVM_REQ_UNHALT if an interrupt | |
3105 | * arrives. | |
3106 | */ | |
3107 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
0193cc90 | 3108 | ++vcpu->stat.generic.halt_successful_poll; |
3491caf2 | 3109 | if (!vcpu_valid_wakeup(vcpu)) |
0193cc90 | 3110 | ++vcpu->stat.generic.halt_poll_invalid; |
f7819512 PB |
3111 | goto out; |
3112 | } | |
cb953129 | 3113 | poll_end = cur = ktime_get(); |
6bd5b743 | 3114 | } while (kvm_vcpu_can_poll(cur, stop)); |
f7819512 | 3115 | } |
e5c239cf | 3116 | |
da4ad88c | 3117 | prepare_to_rcuwait(&vcpu->wait); |
e5c239cf | 3118 | for (;;) { |
da4ad88c | 3119 | set_current_state(TASK_INTERRUPTIBLE); |
e5c239cf | 3120 | |
f7819512 | 3121 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
3122 | break; |
3123 | ||
f7819512 | 3124 | waited = true; |
b6958ce4 | 3125 | schedule(); |
b6958ce4 | 3126 | } |
da4ad88c | 3127 | finish_rcuwait(&vcpu->wait); |
f7819512 | 3128 | cur = ktime_get(); |
f7819512 | 3129 | out: |
07ab0f8d | 3130 | kvm_arch_vcpu_unblocking(vcpu); |
aca6ff29 WL |
3131 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
3132 | ||
cb953129 DM |
3133 | update_halt_poll_stats( |
3134 | vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited); | |
3135 | ||
44551b2f WL |
3136 | if (!kvm_arch_no_poll(vcpu)) { |
3137 | if (!vcpu_valid_wakeup(vcpu)) { | |
aca6ff29 | 3138 | shrink_halt_poll_ns(vcpu); |
acd05785 | 3139 | } else if (vcpu->kvm->max_halt_poll_ns) { |
44551b2f WL |
3140 | if (block_ns <= vcpu->halt_poll_ns) |
3141 | ; | |
3142 | /* we had a long block, shrink polling */ | |
acd05785 DM |
3143 | else if (vcpu->halt_poll_ns && |
3144 | block_ns > vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
3145 | shrink_halt_poll_ns(vcpu); |
3146 | /* we had a short halt and our poll time is too small */ | |
acd05785 DM |
3147 | else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && |
3148 | block_ns < vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
3149 | grow_halt_poll_ns(vcpu); |
3150 | } else { | |
3151 | vcpu->halt_poll_ns = 0; | |
3152 | } | |
3153 | } | |
aca6ff29 | 3154 | |
3491caf2 CB |
3155 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
3156 | kvm_arch_vcpu_block_finish(vcpu); | |
b6958ce4 | 3157 | } |
2ba9f0d8 | 3158 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 3159 | |
178f02ff | 3160 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 3161 | { |
da4ad88c | 3162 | struct rcuwait *waitp; |
b6d33834 | 3163 | |
da4ad88c DB |
3164 | waitp = kvm_arch_vcpu_get_wait(vcpu); |
3165 | if (rcuwait_wake_up(waitp)) { | |
d73eb57b | 3166 | WRITE_ONCE(vcpu->ready, true); |
0193cc90 | 3167 | ++vcpu->stat.generic.halt_wakeup; |
178f02ff | 3168 | return true; |
b6d33834 CD |
3169 | } |
3170 | ||
178f02ff | 3171 | return false; |
dd1a4cc1 RK |
3172 | } |
3173 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
3174 | ||
0266c894 | 3175 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
3176 | /* |
3177 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
3178 | */ | |
3179 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
3180 | { | |
3181 | int me; | |
3182 | int cpu = vcpu->cpu; | |
3183 | ||
178f02ff RK |
3184 | if (kvm_vcpu_wake_up(vcpu)) |
3185 | return; | |
3186 | ||
b6d33834 CD |
3187 | me = get_cpu(); |
3188 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
3189 | if (kvm_arch_vcpu_should_kick(vcpu)) | |
3190 | smp_send_reschedule(cpu); | |
3191 | put_cpu(); | |
3192 | } | |
a20ed54d | 3193 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 3194 | #endif /* !CONFIG_S390 */ |
b6d33834 | 3195 | |
fa93384f | 3196 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
3197 | { |
3198 | struct pid *pid; | |
3199 | struct task_struct *task = NULL; | |
fa93384f | 3200 | int ret = 0; |
41628d33 KW |
3201 | |
3202 | rcu_read_lock(); | |
3203 | pid = rcu_dereference(target->pid); | |
3204 | if (pid) | |
27fbe64b | 3205 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
3206 | rcu_read_unlock(); |
3207 | if (!task) | |
c45c528e | 3208 | return ret; |
c45c528e | 3209 | ret = yield_to(task, 1); |
41628d33 | 3210 | put_task_struct(task); |
c45c528e R |
3211 | |
3212 | return ret; | |
41628d33 KW |
3213 | } |
3214 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
3215 | ||
06e48c51 R |
3216 | /* |
3217 | * Helper that checks whether a VCPU is eligible for directed yield. | |
3218 | * Most eligible candidate to yield is decided by following heuristics: | |
3219 | * | |
3220 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
3221 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 3222 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
3223 | * |
3224 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
3225 | * chance last time (mostly it has become eligible now since we have probably | |
3226 | * yielded to lockholder in last iteration. This is done by toggling | |
3227 | * @dy_eligible each time a VCPU checked for eligibility.) | |
3228 | * | |
3229 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
3230 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
3231 | * burning. Giving priority for a potential lock-holder increases lock | |
3232 | * progress. | |
3233 | * | |
3234 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
3235 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
3236 | * and continue with next VCPU and so on. | |
3237 | */ | |
7940876e | 3238 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 3239 | { |
4a55dd72 | 3240 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
3241 | bool eligible; |
3242 | ||
3243 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 3244 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
3245 | |
3246 | if (vcpu->spin_loop.in_spin_loop) | |
3247 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
3248 | ||
3249 | return eligible; | |
4a55dd72 SW |
3250 | #else |
3251 | return true; | |
06e48c51 | 3252 | #endif |
4a55dd72 | 3253 | } |
c45c528e | 3254 | |
17e433b5 WL |
3255 | /* |
3256 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
3257 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
3258 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
3259 | */ | |
3260 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
3261 | { | |
3262 | return kvm_arch_vcpu_runnable(vcpu); | |
3263 | } | |
3264 | ||
3265 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
3266 | { | |
3267 | if (kvm_arch_dy_runnable(vcpu)) | |
3268 | return true; | |
3269 | ||
3270 | #ifdef CONFIG_KVM_ASYNC_PF | |
3271 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
3272 | return true; | |
3273 | #endif | |
3274 | ||
3275 | return false; | |
3276 | } | |
3277 | ||
52acd22f WL |
3278 | bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
3279 | { | |
3280 | return false; | |
3281 | } | |
3282 | ||
199b5763 | 3283 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 3284 | { |
217ece61 RR |
3285 | struct kvm *kvm = me->kvm; |
3286 | struct kvm_vcpu *vcpu; | |
3287 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
3288 | int yielded = 0; | |
c45c528e | 3289 | int try = 3; |
217ece61 RR |
3290 | int pass; |
3291 | int i; | |
d255f4f2 | 3292 | |
4c088493 | 3293 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
3294 | /* |
3295 | * We boost the priority of a VCPU that is runnable but not | |
3296 | * currently running, because it got preempted by something | |
3297 | * else and called schedule in __vcpu_run. Hopefully that | |
3298 | * VCPU is holding the lock that we need and will release it. | |
3299 | * We approximate round-robin by starting at the last boosted VCPU. | |
3300 | */ | |
c45c528e | 3301 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 3302 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 3303 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
3304 | i = last_boosted_vcpu; |
3305 | continue; | |
3306 | } else if (pass && i > last_boosted_vcpu) | |
3307 | break; | |
d73eb57b | 3308 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 3309 | continue; |
217ece61 RR |
3310 | if (vcpu == me) |
3311 | continue; | |
da4ad88c DB |
3312 | if (rcuwait_active(&vcpu->wait) && |
3313 | !vcpu_dy_runnable(vcpu)) | |
217ece61 | 3314 | continue; |
046ddeed | 3315 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
52acd22f WL |
3316 | !kvm_arch_dy_has_pending_interrupt(vcpu) && |
3317 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 3318 | continue; |
06e48c51 R |
3319 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
3320 | continue; | |
c45c528e R |
3321 | |
3322 | yielded = kvm_vcpu_yield_to(vcpu); | |
3323 | if (yielded > 0) { | |
217ece61 | 3324 | kvm->last_boosted_vcpu = i; |
217ece61 | 3325 | break; |
c45c528e R |
3326 | } else if (yielded < 0) { |
3327 | try--; | |
3328 | if (!try) | |
3329 | break; | |
217ece61 | 3330 | } |
217ece61 RR |
3331 | } |
3332 | } | |
4c088493 | 3333 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
3334 | |
3335 | /* Ensure vcpu is not eligible during next spinloop */ | |
3336 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
3337 | } |
3338 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
3339 | ||
fb04a1ed PX |
3340 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
3341 | { | |
3342 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
3343 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && | |
3344 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
3345 | kvm->dirty_ring_size / PAGE_SIZE); | |
3346 | #else | |
3347 | return false; | |
3348 | #endif | |
3349 | } | |
3350 | ||
1499fa80 | 3351 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 3352 | { |
11bac800 | 3353 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
3354 | struct page *page; |
3355 | ||
e4a533a4 | 3356 | if (vmf->pgoff == 0) |
039576c0 | 3357 | page = virt_to_page(vcpu->run); |
09566765 | 3358 | #ifdef CONFIG_X86 |
e4a533a4 | 3359 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 3360 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 3361 | #endif |
4b4357e0 | 3362 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3363 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
3364 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 3365 | #endif |
fb04a1ed PX |
3366 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
3367 | page = kvm_dirty_ring_get_page( | |
3368 | &vcpu->dirty_ring, | |
3369 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 3370 | else |
5b1c1493 | 3371 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 3372 | get_page(page); |
e4a533a4 | 3373 | vmf->page = page; |
3374 | return 0; | |
9a2bb7f4 AK |
3375 | } |
3376 | ||
f0f37e2f | 3377 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 3378 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
3379 | }; |
3380 | ||
3381 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
3382 | { | |
fb04a1ed PX |
3383 | struct kvm_vcpu *vcpu = file->private_data; |
3384 | unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; | |
3385 | ||
3386 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
3387 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
3388 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
3389 | return -EINVAL; | |
3390 | ||
9a2bb7f4 AK |
3391 | vma->vm_ops = &kvm_vcpu_vm_ops; |
3392 | return 0; | |
3393 | } | |
3394 | ||
bccf2150 AK |
3395 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
3396 | { | |
3397 | struct kvm_vcpu *vcpu = filp->private_data; | |
3398 | ||
66c0b394 | 3399 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
3400 | return 0; |
3401 | } | |
3402 | ||
3d3aab1b | 3403 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
3404 | .release = kvm_vcpu_release, |
3405 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 3406 | .mmap = kvm_vcpu_mmap, |
6038f373 | 3407 | .llseek = noop_llseek, |
7ddfd3e0 | 3408 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
3409 | }; |
3410 | ||
3411 | /* | |
3412 | * Allocates an inode for the vcpu. | |
3413 | */ | |
3414 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
3415 | { | |
e46b4692 MY |
3416 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
3417 | ||
3418 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
3419 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
3420 | } |
3421 | ||
3e7093d0 | 3422 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 3423 | { |
741cbbae | 3424 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
d56f5136 | 3425 | struct dentry *debugfs_dentry; |
45b5939e | 3426 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 3427 | |
45b5939e | 3428 | if (!debugfs_initialized()) |
3e7093d0 | 3429 | return; |
45b5939e LC |
3430 | |
3431 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
3432 | debugfs_dentry = debugfs_create_dir(dir_name, |
3433 | vcpu->kvm->debugfs_dentry); | |
45b5939e | 3434 | |
d56f5136 | 3435 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
741cbbae | 3436 | #endif |
45b5939e LC |
3437 | } |
3438 | ||
c5ea7660 AK |
3439 | /* |
3440 | * Creates some virtual cpus. Good luck creating more than one. | |
3441 | */ | |
73880c80 | 3442 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
3443 | { |
3444 | int r; | |
e09fefde | 3445 | struct kvm_vcpu *vcpu; |
8bd826d6 | 3446 | struct page *page; |
c5ea7660 | 3447 | |
0b1b1dfd | 3448 | if (id >= KVM_MAX_VCPU_ID) |
338c7dba AH |
3449 | return -EINVAL; |
3450 | ||
6c7caebc PB |
3451 | mutex_lock(&kvm->lock); |
3452 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
3453 | mutex_unlock(&kvm->lock); | |
3454 | return -EINVAL; | |
3455 | } | |
3456 | ||
3457 | kvm->created_vcpus++; | |
3458 | mutex_unlock(&kvm->lock); | |
3459 | ||
897cc38e SC |
3460 | r = kvm_arch_vcpu_precreate(kvm, id); |
3461 | if (r) | |
3462 | goto vcpu_decrement; | |
3463 | ||
85f47930 | 3464 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
e529ef66 SC |
3465 | if (!vcpu) { |
3466 | r = -ENOMEM; | |
6c7caebc PB |
3467 | goto vcpu_decrement; |
3468 | } | |
c5ea7660 | 3469 | |
fcd97ad5 | 3470 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 3471 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
3472 | if (!page) { |
3473 | r = -ENOMEM; | |
e529ef66 | 3474 | goto vcpu_free; |
8bd826d6 SC |
3475 | } |
3476 | vcpu->run = page_address(page); | |
3477 | ||
3478 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
3479 | |
3480 | r = kvm_arch_vcpu_create(vcpu); | |
3481 | if (r) | |
8bd826d6 | 3482 | goto vcpu_free_run_page; |
e529ef66 | 3483 | |
fb04a1ed PX |
3484 | if (kvm->dirty_ring_size) { |
3485 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
3486 | id, kvm->dirty_ring_size); | |
3487 | if (r) | |
3488 | goto arch_vcpu_destroy; | |
3489 | } | |
3490 | ||
11ec2804 | 3491 | mutex_lock(&kvm->lock); |
e09fefde DH |
3492 | if (kvm_get_vcpu_by_id(kvm, id)) { |
3493 | r = -EEXIST; | |
3494 | goto unlock_vcpu_destroy; | |
3495 | } | |
73880c80 | 3496 | |
8750e72a RK |
3497 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
3498 | BUG_ON(kvm->vcpus[vcpu->vcpu_idx]); | |
c5ea7660 | 3499 | |
ce55c049 JZ |
3500 | /* Fill the stats id string for the vcpu */ |
3501 | snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d", | |
3502 | task_pid_nr(current), id); | |
3503 | ||
fb3f0f51 | 3504 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 3505 | kvm_get_kvm(kvm); |
bccf2150 | 3506 | r = create_vcpu_fd(vcpu); |
73880c80 | 3507 | if (r < 0) { |
149487bd | 3508 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 3509 | goto unlock_vcpu_destroy; |
73880c80 GN |
3510 | } |
3511 | ||
8750e72a | 3512 | kvm->vcpus[vcpu->vcpu_idx] = vcpu; |
dd489240 PB |
3513 | |
3514 | /* | |
3515 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
3516 | * before kvm->online_vcpu's incremented value. | |
3517 | */ | |
73880c80 GN |
3518 | smp_wmb(); |
3519 | atomic_inc(&kvm->online_vcpus); | |
3520 | ||
73880c80 | 3521 | mutex_unlock(&kvm->lock); |
42897d86 | 3522 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 3523 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 3524 | return r; |
39c3b86e | 3525 | |
d780592b | 3526 | unlock_vcpu_destroy: |
7d8fece6 | 3527 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
3528 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
3529 | arch_vcpu_destroy: | |
d40ccc62 | 3530 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
3531 | vcpu_free_run_page: |
3532 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
3533 | vcpu_free: |
3534 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
3535 | vcpu_decrement: |
3536 | mutex_lock(&kvm->lock); | |
3537 | kvm->created_vcpus--; | |
3538 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
3539 | return r; |
3540 | } | |
3541 | ||
1961d276 AK |
3542 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
3543 | { | |
3544 | if (sigset) { | |
3545 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
3546 | vcpu->sigset_active = 1; | |
3547 | vcpu->sigset = *sigset; | |
3548 | } else | |
3549 | vcpu->sigset_active = 0; | |
3550 | return 0; | |
3551 | } | |
3552 | ||
ce55c049 JZ |
3553 | static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer, |
3554 | size_t size, loff_t *offset) | |
3555 | { | |
3556 | struct kvm_vcpu *vcpu = file->private_data; | |
3557 | ||
3558 | return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header, | |
3559 | &kvm_vcpu_stats_desc[0], &vcpu->stat, | |
3560 | sizeof(vcpu->stat), user_buffer, size, offset); | |
3561 | } | |
3562 | ||
3563 | static const struct file_operations kvm_vcpu_stats_fops = { | |
3564 | .read = kvm_vcpu_stats_read, | |
3565 | .llseek = noop_llseek, | |
3566 | }; | |
3567 | ||
3568 | static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu) | |
3569 | { | |
3570 | int fd; | |
3571 | struct file *file; | |
3572 | char name[15 + ITOA_MAX_LEN + 1]; | |
3573 | ||
3574 | snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id); | |
3575 | ||
3576 | fd = get_unused_fd_flags(O_CLOEXEC); | |
3577 | if (fd < 0) | |
3578 | return fd; | |
3579 | ||
3580 | file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY); | |
3581 | if (IS_ERR(file)) { | |
3582 | put_unused_fd(fd); | |
3583 | return PTR_ERR(file); | |
3584 | } | |
3585 | file->f_mode |= FMODE_PREAD; | |
3586 | fd_install(fd, file); | |
3587 | ||
3588 | return fd; | |
3589 | } | |
3590 | ||
bccf2150 AK |
3591 | static long kvm_vcpu_ioctl(struct file *filp, |
3592 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 3593 | { |
bccf2150 | 3594 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 3595 | void __user *argp = (void __user *)arg; |
313a3dc7 | 3596 | int r; |
fa3795a7 DH |
3597 | struct kvm_fpu *fpu = NULL; |
3598 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 3599 | |
6d4e4c4f AK |
3600 | if (vcpu->kvm->mm != current->mm) |
3601 | return -EIO; | |
2122ff5e | 3602 | |
2ea75be3 DM |
3603 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
3604 | return -EINVAL; | |
3605 | ||
2122ff5e | 3606 | /* |
5cb0944c PB |
3607 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
3608 | * execution; mutex_lock() would break them. | |
2122ff5e | 3609 | */ |
5cb0944c PB |
3610 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
3611 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 3612 | return r; |
2122ff5e | 3613 | |
ec7660cc CD |
3614 | if (mutex_lock_killable(&vcpu->mutex)) |
3615 | return -EINTR; | |
6aa8b732 | 3616 | switch (ioctl) { |
0e4524a5 CB |
3617 | case KVM_RUN: { |
3618 | struct pid *oldpid; | |
f0fe5108 AK |
3619 | r = -EINVAL; |
3620 | if (arg) | |
3621 | goto out; | |
0e4524a5 | 3622 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 3623 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 3624 | /* The thread running this VCPU changed. */ |
bd2a6394 | 3625 | struct pid *newpid; |
f95ef0cd | 3626 | |
bd2a6394 CD |
3627 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
3628 | if (r) | |
3629 | break; | |
3630 | ||
3631 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
3632 | rcu_assign_pointer(vcpu->pid, newpid); |
3633 | if (oldpid) | |
3634 | synchronize_rcu(); | |
3635 | put_pid(oldpid); | |
3636 | } | |
1b94f6f8 | 3637 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 3638 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 3639 | break; |
0e4524a5 | 3640 | } |
6aa8b732 | 3641 | case KVM_GET_REGS: { |
3e4bb3ac | 3642 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3643 | |
3e4bb3ac | 3644 | r = -ENOMEM; |
b12ce36a | 3645 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 3646 | if (!kvm_regs) |
6aa8b732 | 3647 | goto out; |
3e4bb3ac XZ |
3648 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
3649 | if (r) | |
3650 | goto out_free1; | |
6aa8b732 | 3651 | r = -EFAULT; |
3e4bb3ac XZ |
3652 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
3653 | goto out_free1; | |
6aa8b732 | 3654 | r = 0; |
3e4bb3ac XZ |
3655 | out_free1: |
3656 | kfree(kvm_regs); | |
6aa8b732 AK |
3657 | break; |
3658 | } | |
3659 | case KVM_SET_REGS: { | |
3e4bb3ac | 3660 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3661 | |
ff5c2c03 SL |
3662 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
3663 | if (IS_ERR(kvm_regs)) { | |
3664 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 3665 | goto out; |
ff5c2c03 | 3666 | } |
3e4bb3ac | 3667 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 3668 | kfree(kvm_regs); |
6aa8b732 AK |
3669 | break; |
3670 | } | |
3671 | case KVM_GET_SREGS: { | |
b12ce36a BG |
3672 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
3673 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
3674 | r = -ENOMEM; |
3675 | if (!kvm_sregs) | |
3676 | goto out; | |
3677 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
3678 | if (r) |
3679 | goto out; | |
3680 | r = -EFAULT; | |
fa3795a7 | 3681 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
3682 | goto out; |
3683 | r = 0; | |
3684 | break; | |
3685 | } | |
3686 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
3687 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
3688 | if (IS_ERR(kvm_sregs)) { | |
3689 | r = PTR_ERR(kvm_sregs); | |
18595411 | 3690 | kvm_sregs = NULL; |
6aa8b732 | 3691 | goto out; |
ff5c2c03 | 3692 | } |
fa3795a7 | 3693 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
3694 | break; |
3695 | } | |
62d9f0db MT |
3696 | case KVM_GET_MP_STATE: { |
3697 | struct kvm_mp_state mp_state; | |
3698 | ||
3699 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
3700 | if (r) | |
3701 | goto out; | |
3702 | r = -EFAULT; | |
893bdbf1 | 3703 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
3704 | goto out; |
3705 | r = 0; | |
3706 | break; | |
3707 | } | |
3708 | case KVM_SET_MP_STATE: { | |
3709 | struct kvm_mp_state mp_state; | |
3710 | ||
3711 | r = -EFAULT; | |
893bdbf1 | 3712 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
3713 | goto out; |
3714 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
3715 | break; |
3716 | } | |
6aa8b732 AK |
3717 | case KVM_TRANSLATE: { |
3718 | struct kvm_translation tr; | |
3719 | ||
3720 | r = -EFAULT; | |
893bdbf1 | 3721 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 3722 | goto out; |
8b006791 | 3723 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
3724 | if (r) |
3725 | goto out; | |
3726 | r = -EFAULT; | |
893bdbf1 | 3727 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
3728 | goto out; |
3729 | r = 0; | |
3730 | break; | |
3731 | } | |
d0bfb940 JK |
3732 | case KVM_SET_GUEST_DEBUG: { |
3733 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
3734 | |
3735 | r = -EFAULT; | |
893bdbf1 | 3736 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 3737 | goto out; |
d0bfb940 | 3738 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
3739 | break; |
3740 | } | |
1961d276 AK |
3741 | case KVM_SET_SIGNAL_MASK: { |
3742 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3743 | struct kvm_signal_mask kvm_sigmask; | |
3744 | sigset_t sigset, *p; | |
3745 | ||
3746 | p = NULL; | |
3747 | if (argp) { | |
3748 | r = -EFAULT; | |
3749 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3750 | sizeof(kvm_sigmask))) |
1961d276 AK |
3751 | goto out; |
3752 | r = -EINVAL; | |
893bdbf1 | 3753 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
3754 | goto out; |
3755 | r = -EFAULT; | |
3756 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 3757 | sizeof(sigset))) |
1961d276 AK |
3758 | goto out; |
3759 | p = &sigset; | |
3760 | } | |
376d41ff | 3761 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
3762 | break; |
3763 | } | |
b8836737 | 3764 | case KVM_GET_FPU: { |
b12ce36a | 3765 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
3766 | r = -ENOMEM; |
3767 | if (!fpu) | |
3768 | goto out; | |
3769 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
3770 | if (r) |
3771 | goto out; | |
3772 | r = -EFAULT; | |
fa3795a7 | 3773 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
3774 | goto out; |
3775 | r = 0; | |
3776 | break; | |
3777 | } | |
3778 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
3779 | fpu = memdup_user(argp, sizeof(*fpu)); |
3780 | if (IS_ERR(fpu)) { | |
3781 | r = PTR_ERR(fpu); | |
18595411 | 3782 | fpu = NULL; |
b8836737 | 3783 | goto out; |
ff5c2c03 | 3784 | } |
fa3795a7 | 3785 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
3786 | break; |
3787 | } | |
ce55c049 JZ |
3788 | case KVM_GET_STATS_FD: { |
3789 | r = kvm_vcpu_ioctl_get_stats_fd(vcpu); | |
3790 | break; | |
3791 | } | |
bccf2150 | 3792 | default: |
313a3dc7 | 3793 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
3794 | } |
3795 | out: | |
ec7660cc | 3796 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
3797 | kfree(fpu); |
3798 | kfree(kvm_sregs); | |
bccf2150 AK |
3799 | return r; |
3800 | } | |
3801 | ||
de8e5d74 | 3802 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
3803 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
3804 | unsigned int ioctl, unsigned long arg) | |
3805 | { | |
3806 | struct kvm_vcpu *vcpu = filp->private_data; | |
3807 | void __user *argp = compat_ptr(arg); | |
3808 | int r; | |
3809 | ||
3810 | if (vcpu->kvm->mm != current->mm) | |
3811 | return -EIO; | |
3812 | ||
3813 | switch (ioctl) { | |
3814 | case KVM_SET_SIGNAL_MASK: { | |
3815 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3816 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
3817 | sigset_t sigset; |
3818 | ||
3819 | if (argp) { | |
3820 | r = -EFAULT; | |
3821 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3822 | sizeof(kvm_sigmask))) |
1dda606c AG |
3823 | goto out; |
3824 | r = -EINVAL; | |
3968cf62 | 3825 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
3826 | goto out; |
3827 | r = -EFAULT; | |
1393b4aa PB |
3828 | if (get_compat_sigset(&sigset, |
3829 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 3830 | goto out; |
760a9a30 AC |
3831 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
3832 | } else | |
3833 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
3834 | break; |
3835 | } | |
3836 | default: | |
3837 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
3838 | } | |
3839 | ||
3840 | out: | |
3841 | return r; | |
3842 | } | |
3843 | #endif | |
3844 | ||
a1cd3f08 CLG |
3845 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
3846 | { | |
3847 | struct kvm_device *dev = filp->private_data; | |
3848 | ||
3849 | if (dev->ops->mmap) | |
3850 | return dev->ops->mmap(dev, vma); | |
3851 | ||
3852 | return -ENODEV; | |
3853 | } | |
3854 | ||
852b6d57 SW |
3855 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
3856 | int (*accessor)(struct kvm_device *dev, | |
3857 | struct kvm_device_attr *attr), | |
3858 | unsigned long arg) | |
3859 | { | |
3860 | struct kvm_device_attr attr; | |
3861 | ||
3862 | if (!accessor) | |
3863 | return -EPERM; | |
3864 | ||
3865 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
3866 | return -EFAULT; | |
3867 | ||
3868 | return accessor(dev, &attr); | |
3869 | } | |
3870 | ||
3871 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
3872 | unsigned long arg) | |
3873 | { | |
3874 | struct kvm_device *dev = filp->private_data; | |
3875 | ||
ddba9180 SC |
3876 | if (dev->kvm->mm != current->mm) |
3877 | return -EIO; | |
3878 | ||
852b6d57 SW |
3879 | switch (ioctl) { |
3880 | case KVM_SET_DEVICE_ATTR: | |
3881 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
3882 | case KVM_GET_DEVICE_ATTR: | |
3883 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
3884 | case KVM_HAS_DEVICE_ATTR: | |
3885 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
3886 | default: | |
3887 | if (dev->ops->ioctl) | |
3888 | return dev->ops->ioctl(dev, ioctl, arg); | |
3889 | ||
3890 | return -ENOTTY; | |
3891 | } | |
3892 | } | |
3893 | ||
852b6d57 SW |
3894 | static int kvm_device_release(struct inode *inode, struct file *filp) |
3895 | { | |
3896 | struct kvm_device *dev = filp->private_data; | |
3897 | struct kvm *kvm = dev->kvm; | |
3898 | ||
2bde9b3e CLG |
3899 | if (dev->ops->release) { |
3900 | mutex_lock(&kvm->lock); | |
3901 | list_del(&dev->vm_node); | |
3902 | dev->ops->release(dev); | |
3903 | mutex_unlock(&kvm->lock); | |
3904 | } | |
3905 | ||
852b6d57 SW |
3906 | kvm_put_kvm(kvm); |
3907 | return 0; | |
3908 | } | |
3909 | ||
3910 | static const struct file_operations kvm_device_fops = { | |
3911 | .unlocked_ioctl = kvm_device_ioctl, | |
3912 | .release = kvm_device_release, | |
7ddfd3e0 | 3913 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 3914 | .mmap = kvm_device_mmap, |
852b6d57 SW |
3915 | }; |
3916 | ||
3917 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
3918 | { | |
3919 | if (filp->f_op != &kvm_device_fops) | |
3920 | return NULL; | |
3921 | ||
3922 | return filp->private_data; | |
3923 | } | |
3924 | ||
8538cb22 | 3925 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 3926 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
3927 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
3928 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 3929 | #endif |
d60eacb0 WD |
3930 | }; |
3931 | ||
8538cb22 | 3932 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
3933 | { |
3934 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
3935 | return -ENOSPC; | |
3936 | ||
3937 | if (kvm_device_ops_table[type] != NULL) | |
3938 | return -EEXIST; | |
3939 | ||
3940 | kvm_device_ops_table[type] = ops; | |
3941 | return 0; | |
3942 | } | |
3943 | ||
571ee1b6 WL |
3944 | void kvm_unregister_device_ops(u32 type) |
3945 | { | |
3946 | if (kvm_device_ops_table[type] != NULL) | |
3947 | kvm_device_ops_table[type] = NULL; | |
3948 | } | |
3949 | ||
852b6d57 SW |
3950 | static int kvm_ioctl_create_device(struct kvm *kvm, |
3951 | struct kvm_create_device *cd) | |
3952 | { | |
8538cb22 | 3953 | const struct kvm_device_ops *ops = NULL; |
852b6d57 SW |
3954 | struct kvm_device *dev; |
3955 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 3956 | int type; |
852b6d57 SW |
3957 | int ret; |
3958 | ||
d60eacb0 WD |
3959 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
3960 | return -ENODEV; | |
3961 | ||
1d487e9b PB |
3962 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
3963 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 3964 | if (ops == NULL) |
852b6d57 | 3965 | return -ENODEV; |
852b6d57 SW |
3966 | |
3967 | if (test) | |
3968 | return 0; | |
3969 | ||
b12ce36a | 3970 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
3971 | if (!dev) |
3972 | return -ENOMEM; | |
3973 | ||
3974 | dev->ops = ops; | |
3975 | dev->kvm = kvm; | |
852b6d57 | 3976 | |
a28ebea2 | 3977 | mutex_lock(&kvm->lock); |
1d487e9b | 3978 | ret = ops->create(dev, type); |
852b6d57 | 3979 | if (ret < 0) { |
a28ebea2 | 3980 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
3981 | kfree(dev); |
3982 | return ret; | |
3983 | } | |
a28ebea2 CD |
3984 | list_add(&dev->vm_node, &kvm->devices); |
3985 | mutex_unlock(&kvm->lock); | |
852b6d57 | 3986 | |
023e9fdd CD |
3987 | if (ops->init) |
3988 | ops->init(dev); | |
3989 | ||
cfa39381 | 3990 | kvm_get_kvm(kvm); |
24009b05 | 3991 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 3992 | if (ret < 0) { |
149487bd | 3993 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
3994 | mutex_lock(&kvm->lock); |
3995 | list_del(&dev->vm_node); | |
3996 | mutex_unlock(&kvm->lock); | |
a0f1d21c | 3997 | ops->destroy(dev); |
852b6d57 SW |
3998 | return ret; |
3999 | } | |
4000 | ||
852b6d57 SW |
4001 | cd->fd = ret; |
4002 | return 0; | |
4003 | } | |
4004 | ||
92b591a4 AG |
4005 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
4006 | { | |
4007 | switch (arg) { | |
4008 | case KVM_CAP_USER_MEMORY: | |
4009 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
4010 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
4011 | case KVM_CAP_INTERNAL_ERROR_DATA: |
4012 | #ifdef CONFIG_HAVE_KVM_MSI | |
4013 | case KVM_CAP_SIGNAL_MSI: | |
4014 | #endif | |
297e2105 | 4015 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 4016 | case KVM_CAP_IRQFD: |
92b591a4 AG |
4017 | case KVM_CAP_IRQFD_RESAMPLE: |
4018 | #endif | |
e9ea5069 | 4019 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 4020 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 4021 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 4022 | case KVM_CAP_HALT_POLL: |
92b591a4 | 4023 | return 1; |
4b4357e0 | 4024 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
4025 | case KVM_CAP_COALESCED_MMIO: |
4026 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
4027 | case KVM_CAP_COALESCED_PIO: |
4028 | return 1; | |
30422558 | 4029 | #endif |
3c9bd400 JZ |
4030 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4031 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
4032 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
4033 | #endif | |
92b591a4 AG |
4034 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4035 | case KVM_CAP_IRQ_ROUTING: | |
4036 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
4037 | #endif |
4038 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
4039 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
4040 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 4041 | #endif |
c110ae57 PB |
4042 | case KVM_CAP_NR_MEMSLOTS: |
4043 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed PX |
4044 | case KVM_CAP_DIRTY_LOG_RING: |
4045 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
4046 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); | |
4047 | #else | |
4048 | return 0; | |
4049 | #endif | |
ce55c049 JZ |
4050 | case KVM_CAP_BINARY_STATS_FD: |
4051 | return 1; | |
92b591a4 AG |
4052 | default: |
4053 | break; | |
4054 | } | |
4055 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
4056 | } | |
4057 | ||
fb04a1ed PX |
4058 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
4059 | { | |
4060 | int r; | |
4061 | ||
4062 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
4063 | return -EINVAL; | |
4064 | ||
4065 | /* the size should be power of 2 */ | |
4066 | if (!size || (size & (size - 1))) | |
4067 | return -EINVAL; | |
4068 | ||
4069 | /* Should be bigger to keep the reserved entries, or a page */ | |
4070 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
4071 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
4072 | return -EINVAL; | |
4073 | ||
4074 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
4075 | sizeof(struct kvm_dirty_gfn)) | |
4076 | return -E2BIG; | |
4077 | ||
4078 | /* We only allow it to set once */ | |
4079 | if (kvm->dirty_ring_size) | |
4080 | return -EINVAL; | |
4081 | ||
4082 | mutex_lock(&kvm->lock); | |
4083 | ||
4084 | if (kvm->created_vcpus) { | |
4085 | /* We don't allow to change this value after vcpu created */ | |
4086 | r = -EINVAL; | |
4087 | } else { | |
4088 | kvm->dirty_ring_size = size; | |
4089 | r = 0; | |
4090 | } | |
4091 | ||
4092 | mutex_unlock(&kvm->lock); | |
4093 | return r; | |
4094 | } | |
4095 | ||
4096 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
4097 | { | |
4098 | int i; | |
4099 | struct kvm_vcpu *vcpu; | |
4100 | int cleared = 0; | |
4101 | ||
4102 | if (!kvm->dirty_ring_size) | |
4103 | return -EINVAL; | |
4104 | ||
4105 | mutex_lock(&kvm->slots_lock); | |
4106 | ||
4107 | kvm_for_each_vcpu(i, vcpu, kvm) | |
4108 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
4109 | ||
4110 | mutex_unlock(&kvm->slots_lock); | |
4111 | ||
4112 | if (cleared) | |
4113 | kvm_flush_remote_tlbs(kvm); | |
4114 | ||
4115 | return cleared; | |
4116 | } | |
4117 | ||
e5d83c74 PB |
4118 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4119 | struct kvm_enable_cap *cap) | |
4120 | { | |
4121 | return -EINVAL; | |
4122 | } | |
4123 | ||
4124 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
4125 | struct kvm_enable_cap *cap) | |
4126 | { | |
4127 | switch (cap->cap) { | |
2a31b9db | 4128 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
4129 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
4130 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
4131 | ||
4132 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
4133 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
4134 | ||
4135 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
4136 | return -EINVAL; |
4137 | kvm->manual_dirty_log_protect = cap->args[0]; | |
4138 | return 0; | |
3c9bd400 | 4139 | } |
2a31b9db | 4140 | #endif |
acd05785 DM |
4141 | case KVM_CAP_HALT_POLL: { |
4142 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
4143 | return -EINVAL; | |
4144 | ||
4145 | kvm->max_halt_poll_ns = cap->args[0]; | |
4146 | return 0; | |
4147 | } | |
fb04a1ed PX |
4148 | case KVM_CAP_DIRTY_LOG_RING: |
4149 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); | |
e5d83c74 PB |
4150 | default: |
4151 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
4152 | } | |
4153 | } | |
4154 | ||
fcfe1bae JZ |
4155 | static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer, |
4156 | size_t size, loff_t *offset) | |
4157 | { | |
4158 | struct kvm *kvm = file->private_data; | |
4159 | ||
4160 | return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header, | |
4161 | &kvm_vm_stats_desc[0], &kvm->stat, | |
4162 | sizeof(kvm->stat), user_buffer, size, offset); | |
4163 | } | |
4164 | ||
4165 | static const struct file_operations kvm_vm_stats_fops = { | |
4166 | .read = kvm_vm_stats_read, | |
4167 | .llseek = noop_llseek, | |
4168 | }; | |
4169 | ||
4170 | static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm) | |
4171 | { | |
4172 | int fd; | |
4173 | struct file *file; | |
4174 | ||
4175 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4176 | if (fd < 0) | |
4177 | return fd; | |
4178 | ||
4179 | file = anon_inode_getfile("kvm-vm-stats", | |
4180 | &kvm_vm_stats_fops, kvm, O_RDONLY); | |
4181 | if (IS_ERR(file)) { | |
4182 | put_unused_fd(fd); | |
4183 | return PTR_ERR(file); | |
4184 | } | |
4185 | file->f_mode |= FMODE_PREAD; | |
4186 | fd_install(fd, file); | |
4187 | ||
4188 | return fd; | |
4189 | } | |
4190 | ||
bccf2150 AK |
4191 | static long kvm_vm_ioctl(struct file *filp, |
4192 | unsigned int ioctl, unsigned long arg) | |
4193 | { | |
4194 | struct kvm *kvm = filp->private_data; | |
4195 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 4196 | int r; |
bccf2150 | 4197 | |
6d4e4c4f AK |
4198 | if (kvm->mm != current->mm) |
4199 | return -EIO; | |
bccf2150 AK |
4200 | switch (ioctl) { |
4201 | case KVM_CREATE_VCPU: | |
4202 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 4203 | break; |
e5d83c74 PB |
4204 | case KVM_ENABLE_CAP: { |
4205 | struct kvm_enable_cap cap; | |
4206 | ||
4207 | r = -EFAULT; | |
4208 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4209 | goto out; | |
4210 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
4211 | break; | |
4212 | } | |
6fc138d2 IE |
4213 | case KVM_SET_USER_MEMORY_REGION: { |
4214 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
4215 | ||
4216 | r = -EFAULT; | |
4217 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 4218 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
4219 | goto out; |
4220 | ||
47ae31e2 | 4221 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
4222 | break; |
4223 | } | |
4224 | case KVM_GET_DIRTY_LOG: { | |
4225 | struct kvm_dirty_log log; | |
4226 | ||
4227 | r = -EFAULT; | |
893bdbf1 | 4228 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 4229 | goto out; |
2c6f5df9 | 4230 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
4231 | break; |
4232 | } | |
2a31b9db PB |
4233 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4234 | case KVM_CLEAR_DIRTY_LOG: { | |
4235 | struct kvm_clear_dirty_log log; | |
4236 | ||
4237 | r = -EFAULT; | |
4238 | if (copy_from_user(&log, argp, sizeof(log))) | |
4239 | goto out; | |
4240 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4241 | break; | |
4242 | } | |
4243 | #endif | |
4b4357e0 | 4244 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
4245 | case KVM_REGISTER_COALESCED_MMIO: { |
4246 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4247 | |
5f94c174 | 4248 | r = -EFAULT; |
893bdbf1 | 4249 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4250 | goto out; |
5f94c174 | 4251 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4252 | break; |
4253 | } | |
4254 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
4255 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4256 | |
5f94c174 | 4257 | r = -EFAULT; |
893bdbf1 | 4258 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4259 | goto out; |
5f94c174 | 4260 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4261 | break; |
4262 | } | |
4263 | #endif | |
721eecbf GH |
4264 | case KVM_IRQFD: { |
4265 | struct kvm_irqfd data; | |
4266 | ||
4267 | r = -EFAULT; | |
893bdbf1 | 4268 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 4269 | goto out; |
d4db2935 | 4270 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
4271 | break; |
4272 | } | |
d34e6b17 GH |
4273 | case KVM_IOEVENTFD: { |
4274 | struct kvm_ioeventfd data; | |
4275 | ||
4276 | r = -EFAULT; | |
893bdbf1 | 4277 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
4278 | goto out; |
4279 | r = kvm_ioeventfd(kvm, &data); | |
4280 | break; | |
4281 | } | |
07975ad3 JK |
4282 | #ifdef CONFIG_HAVE_KVM_MSI |
4283 | case KVM_SIGNAL_MSI: { | |
4284 | struct kvm_msi msi; | |
4285 | ||
4286 | r = -EFAULT; | |
893bdbf1 | 4287 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
4288 | goto out; |
4289 | r = kvm_send_userspace_msi(kvm, &msi); | |
4290 | break; | |
4291 | } | |
23d43cf9 CD |
4292 | #endif |
4293 | #ifdef __KVM_HAVE_IRQ_LINE | |
4294 | case KVM_IRQ_LINE_STATUS: | |
4295 | case KVM_IRQ_LINE: { | |
4296 | struct kvm_irq_level irq_event; | |
4297 | ||
4298 | r = -EFAULT; | |
893bdbf1 | 4299 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
4300 | goto out; |
4301 | ||
aa2fbe6d YZ |
4302 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
4303 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
4304 | if (r) |
4305 | goto out; | |
4306 | ||
4307 | r = -EFAULT; | |
4308 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 4309 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
4310 | goto out; |
4311 | } | |
4312 | ||
4313 | r = 0; | |
4314 | break; | |
4315 | } | |
73880c80 | 4316 | #endif |
aa8d5944 AG |
4317 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4318 | case KVM_SET_GSI_ROUTING: { | |
4319 | struct kvm_irq_routing routing; | |
4320 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 4321 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
4322 | |
4323 | r = -EFAULT; | |
4324 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
4325 | goto out; | |
4326 | r = -EINVAL; | |
5c0aea0e DH |
4327 | if (!kvm_arch_can_set_irq_routing(kvm)) |
4328 | goto out; | |
caf1ff26 | 4329 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
4330 | goto out; |
4331 | if (routing.flags) | |
4332 | goto out; | |
f8c1b85b | 4333 | if (routing.nr) { |
f8c1b85b | 4334 | urouting = argp; |
7ec28e26 DE |
4335 | entries = vmemdup_user(urouting->entries, |
4336 | array_size(sizeof(*entries), | |
4337 | routing.nr)); | |
4338 | if (IS_ERR(entries)) { | |
4339 | r = PTR_ERR(entries); | |
4340 | goto out; | |
4341 | } | |
f8c1b85b | 4342 | } |
aa8d5944 AG |
4343 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
4344 | routing.flags); | |
7ec28e26 | 4345 | kvfree(entries); |
aa8d5944 AG |
4346 | break; |
4347 | } | |
4348 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
4349 | case KVM_CREATE_DEVICE: { |
4350 | struct kvm_create_device cd; | |
4351 | ||
4352 | r = -EFAULT; | |
4353 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
4354 | goto out; | |
4355 | ||
4356 | r = kvm_ioctl_create_device(kvm, &cd); | |
4357 | if (r) | |
4358 | goto out; | |
4359 | ||
4360 | r = -EFAULT; | |
4361 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
4362 | goto out; | |
4363 | ||
4364 | r = 0; | |
4365 | break; | |
4366 | } | |
92b591a4 AG |
4367 | case KVM_CHECK_EXTENSION: |
4368 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
4369 | break; | |
fb04a1ed PX |
4370 | case KVM_RESET_DIRTY_RINGS: |
4371 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
4372 | break; | |
fcfe1bae JZ |
4373 | case KVM_GET_STATS_FD: |
4374 | r = kvm_vm_ioctl_get_stats_fd(kvm); | |
4375 | break; | |
f17abe9a | 4376 | default: |
1fe779f8 | 4377 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
4378 | } |
4379 | out: | |
4380 | return r; | |
4381 | } | |
4382 | ||
de8e5d74 | 4383 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
4384 | struct compat_kvm_dirty_log { |
4385 | __u32 slot; | |
4386 | __u32 padding1; | |
4387 | union { | |
4388 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4389 | __u64 padding2; | |
4390 | }; | |
4391 | }; | |
4392 | ||
4393 | static long kvm_vm_compat_ioctl(struct file *filp, | |
4394 | unsigned int ioctl, unsigned long arg) | |
4395 | { | |
4396 | struct kvm *kvm = filp->private_data; | |
4397 | int r; | |
4398 | ||
4399 | if (kvm->mm != current->mm) | |
4400 | return -EIO; | |
4401 | switch (ioctl) { | |
4402 | case KVM_GET_DIRTY_LOG: { | |
4403 | struct compat_kvm_dirty_log compat_log; | |
4404 | struct kvm_dirty_log log; | |
4405 | ||
6ff5894c AB |
4406 | if (copy_from_user(&compat_log, (void __user *)arg, |
4407 | sizeof(compat_log))) | |
f6a3b168 | 4408 | return -EFAULT; |
6ff5894c AB |
4409 | log.slot = compat_log.slot; |
4410 | log.padding1 = compat_log.padding1; | |
4411 | log.padding2 = compat_log.padding2; | |
4412 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4413 | ||
4414 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
4415 | break; |
4416 | } | |
4417 | default: | |
4418 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
4419 | } | |
6ff5894c AB |
4420 | return r; |
4421 | } | |
4422 | #endif | |
4423 | ||
3d3aab1b | 4424 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
4425 | .release = kvm_vm_release, |
4426 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 4427 | .llseek = noop_llseek, |
7ddfd3e0 | 4428 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
4429 | }; |
4430 | ||
54526d1f NT |
4431 | bool file_is_kvm(struct file *file) |
4432 | { | |
4433 | return file && file->f_op == &kvm_vm_fops; | |
4434 | } | |
4435 | EXPORT_SYMBOL_GPL(file_is_kvm); | |
4436 | ||
e08b9637 | 4437 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 4438 | { |
aac87636 | 4439 | int r; |
f17abe9a | 4440 | struct kvm *kvm; |
506cfba9 | 4441 | struct file *file; |
f17abe9a | 4442 | |
e08b9637 | 4443 | kvm = kvm_create_vm(type); |
d6d28168 AK |
4444 | if (IS_ERR(kvm)) |
4445 | return PTR_ERR(kvm); | |
4b4357e0 | 4446 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 4447 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
4448 | if (r < 0) |
4449 | goto put_kvm; | |
6ce5a090 | 4450 | #endif |
506cfba9 | 4451 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
4452 | if (r < 0) |
4453 | goto put_kvm; | |
4454 | ||
fcfe1bae JZ |
4455 | snprintf(kvm->stats_id, sizeof(kvm->stats_id), |
4456 | "kvm-%d", task_pid_nr(current)); | |
4457 | ||
506cfba9 AV |
4458 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
4459 | if (IS_ERR(file)) { | |
4460 | put_unused_fd(r); | |
78588335 ME |
4461 | r = PTR_ERR(file); |
4462 | goto put_kvm; | |
506cfba9 | 4463 | } |
536a6f88 | 4464 | |
525df861 PB |
4465 | /* |
4466 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
4467 | * already set, with ->release() being kvm_vm_release(). In error | |
4468 | * cases it will be called by the final fput(file) and will take | |
4469 | * care of doing kvm_put_kvm(kvm). | |
4470 | */ | |
536a6f88 | 4471 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
4472 | put_unused_fd(r); |
4473 | fput(file); | |
536a6f88 JF |
4474 | return -ENOMEM; |
4475 | } | |
286de8f6 | 4476 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 4477 | |
506cfba9 | 4478 | fd_install(r, file); |
aac87636 | 4479 | return r; |
78588335 ME |
4480 | |
4481 | put_kvm: | |
4482 | kvm_put_kvm(kvm); | |
4483 | return r; | |
f17abe9a AK |
4484 | } |
4485 | ||
4486 | static long kvm_dev_ioctl(struct file *filp, | |
4487 | unsigned int ioctl, unsigned long arg) | |
4488 | { | |
07c45a36 | 4489 | long r = -EINVAL; |
f17abe9a AK |
4490 | |
4491 | switch (ioctl) { | |
4492 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
4493 | if (arg) |
4494 | goto out; | |
f17abe9a AK |
4495 | r = KVM_API_VERSION; |
4496 | break; | |
4497 | case KVM_CREATE_VM: | |
e08b9637 | 4498 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 4499 | break; |
018d00d2 | 4500 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 4501 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 4502 | break; |
07c45a36 | 4503 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
4504 | if (arg) |
4505 | goto out; | |
adb1ff46 AK |
4506 | r = PAGE_SIZE; /* struct kvm_run */ |
4507 | #ifdef CONFIG_X86 | |
4508 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 4509 | #endif |
4b4357e0 | 4510 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 4511 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 4512 | #endif |
07c45a36 | 4513 | break; |
d4c9ff2d FEL |
4514 | case KVM_TRACE_ENABLE: |
4515 | case KVM_TRACE_PAUSE: | |
4516 | case KVM_TRACE_DISABLE: | |
2023a29c | 4517 | r = -EOPNOTSUPP; |
d4c9ff2d | 4518 | break; |
6aa8b732 | 4519 | default: |
043405e1 | 4520 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
4521 | } |
4522 | out: | |
4523 | return r; | |
4524 | } | |
4525 | ||
6aa8b732 | 4526 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 4527 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 4528 | .llseek = noop_llseek, |
7ddfd3e0 | 4529 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
4530 | }; |
4531 | ||
4532 | static struct miscdevice kvm_dev = { | |
bbe4432e | 4533 | KVM_MINOR, |
6aa8b732 AK |
4534 | "kvm", |
4535 | &kvm_chardev_ops, | |
4536 | }; | |
4537 | ||
75b7127c | 4538 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
4539 | { |
4540 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 4541 | int r; |
1b6c0168 | 4542 | |
7f59f492 | 4543 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4544 | return; |
10474ae8 | 4545 | |
7f59f492 | 4546 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 4547 | |
13a34e06 | 4548 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
4549 | |
4550 | if (r) { | |
4551 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
4552 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 4553 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 4554 | } |
1b6c0168 AK |
4555 | } |
4556 | ||
8c18b2d2 | 4557 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 4558 | { |
4a937f96 | 4559 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4560 | if (kvm_usage_count) |
4561 | hardware_enable_nolock(NULL); | |
4a937f96 | 4562 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4563 | return 0; |
75b7127c TY |
4564 | } |
4565 | ||
4566 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
4567 | { |
4568 | int cpu = raw_smp_processor_id(); | |
4569 | ||
7f59f492 | 4570 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4571 | return; |
7f59f492 | 4572 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 4573 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
4574 | } |
4575 | ||
8c18b2d2 | 4576 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 4577 | { |
4a937f96 | 4578 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4579 | if (kvm_usage_count) |
4580 | hardware_disable_nolock(NULL); | |
4a937f96 | 4581 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4582 | return 0; |
75b7127c TY |
4583 | } |
4584 | ||
10474ae8 AG |
4585 | static void hardware_disable_all_nolock(void) |
4586 | { | |
4587 | BUG_ON(!kvm_usage_count); | |
4588 | ||
4589 | kvm_usage_count--; | |
4590 | if (!kvm_usage_count) | |
75b7127c | 4591 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
4592 | } |
4593 | ||
4594 | static void hardware_disable_all(void) | |
4595 | { | |
4a937f96 | 4596 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 4597 | hardware_disable_all_nolock(); |
4a937f96 | 4598 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4599 | } |
4600 | ||
4601 | static int hardware_enable_all(void) | |
4602 | { | |
4603 | int r = 0; | |
4604 | ||
4a937f96 | 4605 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
4606 | |
4607 | kvm_usage_count++; | |
4608 | if (kvm_usage_count == 1) { | |
4609 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 4610 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
4611 | |
4612 | if (atomic_read(&hardware_enable_failed)) { | |
4613 | hardware_disable_all_nolock(); | |
4614 | r = -EBUSY; | |
4615 | } | |
4616 | } | |
4617 | ||
4a937f96 | 4618 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4619 | |
4620 | return r; | |
4621 | } | |
4622 | ||
9a2b85c6 | 4623 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 4624 | void *v) |
9a2b85c6 | 4625 | { |
8e1c1815 SY |
4626 | /* |
4627 | * Some (well, at least mine) BIOSes hang on reboot if | |
4628 | * in vmx root mode. | |
4629 | * | |
4630 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
4631 | */ | |
1170adc6 | 4632 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 4633 | kvm_rebooting = true; |
75b7127c | 4634 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
4635 | return NOTIFY_OK; |
4636 | } | |
4637 | ||
4638 | static struct notifier_block kvm_reboot_notifier = { | |
4639 | .notifier_call = kvm_reboot, | |
4640 | .priority = 0, | |
4641 | }; | |
4642 | ||
e93f8a0f | 4643 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
4644 | { |
4645 | int i; | |
4646 | ||
4647 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 4648 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
4649 | |
4650 | kvm_iodevice_destructor(pos); | |
4651 | } | |
e93f8a0f | 4652 | kfree(bus); |
2eeb2e94 GH |
4653 | } |
4654 | ||
c21fbff1 | 4655 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 4656 | const struct kvm_io_range *r2) |
743eeb0b | 4657 | { |
8f4216c7 JW |
4658 | gpa_t addr1 = r1->addr; |
4659 | gpa_t addr2 = r2->addr; | |
4660 | ||
4661 | if (addr1 < addr2) | |
743eeb0b | 4662 | return -1; |
8f4216c7 JW |
4663 | |
4664 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
4665 | * accept any overlapping write. Any order is acceptable for | |
4666 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
4667 | * we process all of them. | |
4668 | */ | |
4669 | if (r2->len) { | |
4670 | addr1 += r1->len; | |
4671 | addr2 += r2->len; | |
4672 | } | |
4673 | ||
4674 | if (addr1 > addr2) | |
743eeb0b | 4675 | return 1; |
8f4216c7 | 4676 | |
743eeb0b SL |
4677 | return 0; |
4678 | } | |
4679 | ||
a343c9b7 PB |
4680 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
4681 | { | |
c21fbff1 | 4682 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
4683 | } |
4684 | ||
39369f7a | 4685 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
4686 | gpa_t addr, int len) |
4687 | { | |
4688 | struct kvm_io_range *range, key; | |
4689 | int off; | |
4690 | ||
4691 | key = (struct kvm_io_range) { | |
4692 | .addr = addr, | |
4693 | .len = len, | |
4694 | }; | |
4695 | ||
4696 | range = bsearch(&key, bus->range, bus->dev_count, | |
4697 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
4698 | if (range == NULL) | |
4699 | return -ENOENT; | |
4700 | ||
4701 | off = range - bus->range; | |
4702 | ||
c21fbff1 | 4703 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
4704 | off--; |
4705 | ||
4706 | return off; | |
4707 | } | |
4708 | ||
e32edf4f | 4709 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
4710 | struct kvm_io_range *range, const void *val) |
4711 | { | |
4712 | int idx; | |
4713 | ||
4714 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
4715 | if (idx < 0) | |
4716 | return -EOPNOTSUPP; | |
4717 | ||
4718 | while (idx < bus->dev_count && | |
c21fbff1 | 4719 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4720 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4721 | range->len, val)) |
4722 | return idx; | |
4723 | idx++; | |
4724 | } | |
4725 | ||
4726 | return -EOPNOTSUPP; | |
4727 | } | |
4728 | ||
bda9020e | 4729 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 4730 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 4731 | int len, const void *val) |
2eeb2e94 | 4732 | { |
90d83dc3 | 4733 | struct kvm_io_bus *bus; |
743eeb0b | 4734 | struct kvm_io_range range; |
126a5af5 | 4735 | int r; |
743eeb0b SL |
4736 | |
4737 | range = (struct kvm_io_range) { | |
4738 | .addr = addr, | |
4739 | .len = len, | |
4740 | }; | |
90d83dc3 | 4741 | |
e32edf4f | 4742 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4743 | if (!bus) |
4744 | return -ENOMEM; | |
e32edf4f | 4745 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4746 | return r < 0 ? r : 0; |
4747 | } | |
a2420107 | 4748 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
4749 | |
4750 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
4751 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
4752 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
4753 | { |
4754 | struct kvm_io_bus *bus; | |
4755 | struct kvm_io_range range; | |
4756 | ||
4757 | range = (struct kvm_io_range) { | |
4758 | .addr = addr, | |
4759 | .len = len, | |
4760 | }; | |
4761 | ||
e32edf4f | 4762 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4763 | if (!bus) |
4764 | return -ENOMEM; | |
126a5af5 CH |
4765 | |
4766 | /* First try the device referenced by cookie. */ | |
4767 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 4768 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 4769 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
4770 | val)) |
4771 | return cookie; | |
4772 | ||
4773 | /* | |
4774 | * cookie contained garbage; fall back to search and return the | |
4775 | * correct cookie value. | |
4776 | */ | |
e32edf4f | 4777 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4778 | } |
4779 | ||
e32edf4f NN |
4780 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
4781 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
4782 | { |
4783 | int idx; | |
4784 | ||
4785 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
4786 | if (idx < 0) |
4787 | return -EOPNOTSUPP; | |
4788 | ||
4789 | while (idx < bus->dev_count && | |
c21fbff1 | 4790 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4791 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4792 | range->len, val)) |
4793 | return idx; | |
743eeb0b SL |
4794 | idx++; |
4795 | } | |
4796 | ||
bda9020e MT |
4797 | return -EOPNOTSUPP; |
4798 | } | |
2eeb2e94 | 4799 | |
bda9020e | 4800 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 4801 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 4802 | int len, void *val) |
bda9020e | 4803 | { |
90d83dc3 | 4804 | struct kvm_io_bus *bus; |
743eeb0b | 4805 | struct kvm_io_range range; |
126a5af5 | 4806 | int r; |
743eeb0b SL |
4807 | |
4808 | range = (struct kvm_io_range) { | |
4809 | .addr = addr, | |
4810 | .len = len, | |
4811 | }; | |
e93f8a0f | 4812 | |
e32edf4f | 4813 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4814 | if (!bus) |
4815 | return -ENOMEM; | |
e32edf4f | 4816 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
4817 | return r < 0 ? r : 0; |
4818 | } | |
743eeb0b | 4819 | |
79fac95e | 4820 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
4821 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
4822 | int len, struct kvm_io_device *dev) | |
6c474694 | 4823 | { |
d4c67a7a | 4824 | int i; |
e93f8a0f | 4825 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 4826 | struct kvm_io_range range; |
090b7aff | 4827 | |
4a12f951 | 4828 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
4829 | if (!bus) |
4830 | return -ENOMEM; | |
4831 | ||
6ea34c9b AK |
4832 | /* exclude ioeventfd which is limited by maximum fd */ |
4833 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 4834 | return -ENOSPC; |
2eeb2e94 | 4835 | |
90952cd3 | 4836 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 4837 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
4838 | if (!new_bus) |
4839 | return -ENOMEM; | |
d4c67a7a GH |
4840 | |
4841 | range = (struct kvm_io_range) { | |
4842 | .addr = addr, | |
4843 | .len = len, | |
4844 | .dev = dev, | |
4845 | }; | |
4846 | ||
4847 | for (i = 0; i < bus->dev_count; i++) | |
4848 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
4849 | break; | |
4850 | ||
4851 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
4852 | new_bus->dev_count++; | |
4853 | new_bus->range[i] = range; | |
4854 | memcpy(new_bus->range + i + 1, bus->range + i, | |
4855 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
4856 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
4857 | synchronize_srcu_expedited(&kvm->srcu); | |
4858 | kfree(bus); | |
090b7aff GH |
4859 | |
4860 | return 0; | |
4861 | } | |
4862 | ||
5d3c4c79 SC |
4863 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
4864 | struct kvm_io_device *dev) | |
090b7aff | 4865 | { |
f6588660 | 4866 | int i, j; |
e93f8a0f | 4867 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 4868 | |
7c896d37 SC |
4869 | lockdep_assert_held(&kvm->slots_lock); |
4870 | ||
4a12f951 | 4871 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 4872 | if (!bus) |
5d3c4c79 | 4873 | return 0; |
df630b8c | 4874 | |
7c896d37 | 4875 | for (i = 0; i < bus->dev_count; i++) { |
a1300716 | 4876 | if (bus->range[i].dev == dev) { |
090b7aff GH |
4877 | break; |
4878 | } | |
7c896d37 | 4879 | } |
e93f8a0f | 4880 | |
90db1043 | 4881 | if (i == bus->dev_count) |
5d3c4c79 | 4882 | return 0; |
a1300716 | 4883 | |
90952cd3 | 4884 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 4885 | GFP_KERNEL_ACCOUNT); |
f6588660 | 4886 | if (new_bus) { |
871c433b | 4887 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
4888 | new_bus->dev_count--; |
4889 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 4890 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
2ee37574 SC |
4891 | } |
4892 | ||
4893 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
4894 | synchronize_srcu_expedited(&kvm->srcu); | |
4895 | ||
4896 | /* Destroy the old bus _after_ installing the (null) bus. */ | |
4897 | if (!new_bus) { | |
90db1043 | 4898 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
f6588660 RK |
4899 | for (j = 0; j < bus->dev_count; j++) { |
4900 | if (j == i) | |
4901 | continue; | |
4902 | kvm_iodevice_destructor(bus->range[j].dev); | |
4903 | } | |
90db1043 | 4904 | } |
a1300716 | 4905 | |
e93f8a0f | 4906 | kfree(bus); |
5d3c4c79 | 4907 | return new_bus ? 0 : -ENOMEM; |
2eeb2e94 GH |
4908 | } |
4909 | ||
8a39d006 AP |
4910 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
4911 | gpa_t addr) | |
4912 | { | |
4913 | struct kvm_io_bus *bus; | |
4914 | int dev_idx, srcu_idx; | |
4915 | struct kvm_io_device *iodev = NULL; | |
4916 | ||
4917 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
4918 | ||
4919 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
4920 | if (!bus) |
4921 | goto out_unlock; | |
8a39d006 AP |
4922 | |
4923 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
4924 | if (dev_idx < 0) | |
4925 | goto out_unlock; | |
4926 | ||
4927 | iodev = bus->range[dev_idx].dev; | |
4928 | ||
4929 | out_unlock: | |
4930 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
4931 | ||
4932 | return iodev; | |
4933 | } | |
4934 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
4935 | ||
536a6f88 JF |
4936 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
4937 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
4938 | const char *fmt) | |
4939 | { | |
4940 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
4941 | inode->i_private; | |
4942 | ||
4943 | /* The debugfs files are a reference to the kvm struct which | |
4944 | * is still valid when kvm_destroy_vm is called. | |
4945 | * To avoid the race between open and the removal of the debugfs | |
4946 | * directory we test against the users count. | |
4947 | */ | |
e3736c3e | 4948 | if (!refcount_inc_not_zero(&stat_data->kvm->users_count)) |
536a6f88 JF |
4949 | return -ENOENT; |
4950 | ||
833b45de | 4951 | if (simple_attr_open(inode, file, get, |
bc9e9e67 | 4952 | kvm_stats_debugfs_mode(stat_data->desc) & 0222 |
09cbcef6 MP |
4953 | ? set : NULL, |
4954 | fmt)) { | |
536a6f88 JF |
4955 | kvm_put_kvm(stat_data->kvm); |
4956 | return -ENOMEM; | |
4957 | } | |
4958 | ||
4959 | return 0; | |
4960 | } | |
4961 | ||
4962 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
4963 | { | |
4964 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
4965 | inode->i_private; | |
4966 | ||
4967 | simple_attr_release(inode, file); | |
4968 | kvm_put_kvm(stat_data->kvm); | |
4969 | ||
4970 | return 0; | |
4971 | } | |
4972 | ||
09cbcef6 | 4973 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 4974 | { |
bc9e9e67 | 4975 | *val = *(u64 *)((void *)(&kvm->stat) + offset); |
536a6f88 | 4976 | |
09cbcef6 MP |
4977 | return 0; |
4978 | } | |
4979 | ||
4980 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
4981 | { | |
bc9e9e67 | 4982 | *(u64 *)((void *)(&kvm->stat) + offset) = 0; |
536a6f88 JF |
4983 | |
4984 | return 0; | |
4985 | } | |
4986 | ||
09cbcef6 | 4987 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 4988 | { |
09cbcef6 MP |
4989 | int i; |
4990 | struct kvm_vcpu *vcpu; | |
ce35ef27 | 4991 | |
09cbcef6 | 4992 | *val = 0; |
ce35ef27 | 4993 | |
09cbcef6 | 4994 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 4995 | *val += *(u64 *)((void *)(&vcpu->stat) + offset); |
ce35ef27 SJS |
4996 | |
4997 | return 0; | |
4998 | } | |
4999 | ||
09cbcef6 | 5000 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 5001 | { |
09cbcef6 MP |
5002 | int i; |
5003 | struct kvm_vcpu *vcpu; | |
536a6f88 | 5004 | |
09cbcef6 | 5005 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5006 | *(u64 *)((void *)(&vcpu->stat) + offset) = 0; |
09cbcef6 MP |
5007 | |
5008 | return 0; | |
5009 | } | |
536a6f88 | 5010 | |
09cbcef6 | 5011 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 5012 | { |
09cbcef6 | 5013 | int r = -EFAULT; |
536a6f88 | 5014 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
536a6f88 | 5015 | |
bc9e9e67 | 5016 | switch (stat_data->kind) { |
09cbcef6 MP |
5017 | case KVM_STAT_VM: |
5018 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5019 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5020 | break; |
5021 | case KVM_STAT_VCPU: | |
5022 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5023 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5024 | break; |
5025 | } | |
536a6f88 | 5026 | |
09cbcef6 | 5027 | return r; |
536a6f88 JF |
5028 | } |
5029 | ||
09cbcef6 | 5030 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 5031 | { |
09cbcef6 | 5032 | int r = -EFAULT; |
ce35ef27 | 5033 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
ce35ef27 SJS |
5034 | |
5035 | if (val) | |
5036 | return -EINVAL; | |
5037 | ||
bc9e9e67 | 5038 | switch (stat_data->kind) { |
09cbcef6 MP |
5039 | case KVM_STAT_VM: |
5040 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5041 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5042 | break; |
5043 | case KVM_STAT_VCPU: | |
5044 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5045 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5046 | break; |
5047 | } | |
ce35ef27 | 5048 | |
09cbcef6 | 5049 | return r; |
ce35ef27 SJS |
5050 | } |
5051 | ||
09cbcef6 | 5052 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
5053 | { |
5054 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
5055 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
5056 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
5057 | } |
5058 | ||
09cbcef6 MP |
5059 | static const struct file_operations stat_fops_per_vm = { |
5060 | .owner = THIS_MODULE, | |
5061 | .open = kvm_stat_data_open, | |
536a6f88 | 5062 | .release = kvm_debugfs_release, |
09cbcef6 MP |
5063 | .read = simple_attr_read, |
5064 | .write = simple_attr_write, | |
5065 | .llseek = no_llseek, | |
536a6f88 JF |
5066 | }; |
5067 | ||
8b88b099 | 5068 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
5069 | { |
5070 | unsigned offset = (long)_offset; | |
ba1389b7 | 5071 | struct kvm *kvm; |
536a6f88 | 5072 | u64 tmp_val; |
ba1389b7 | 5073 | |
8b88b099 | 5074 | *val = 0; |
0d9ce162 | 5075 | mutex_lock(&kvm_lock); |
536a6f88 | 5076 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5077 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
5078 | *val += tmp_val; |
5079 | } | |
0d9ce162 | 5080 | mutex_unlock(&kvm_lock); |
8b88b099 | 5081 | return 0; |
ba1389b7 AK |
5082 | } |
5083 | ||
ce35ef27 SJS |
5084 | static int vm_stat_clear(void *_offset, u64 val) |
5085 | { | |
5086 | unsigned offset = (long)_offset; | |
5087 | struct kvm *kvm; | |
ce35ef27 SJS |
5088 | |
5089 | if (val) | |
5090 | return -EINVAL; | |
5091 | ||
0d9ce162 | 5092 | mutex_lock(&kvm_lock); |
ce35ef27 | 5093 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5094 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 5095 | } |
0d9ce162 | 5096 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5097 | |
5098 | return 0; | |
5099 | } | |
5100 | ||
5101 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
bc9e9e67 | 5102 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n"); |
ba1389b7 | 5103 | |
8b88b099 | 5104 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
5105 | { |
5106 | unsigned offset = (long)_offset; | |
1165f5fe | 5107 | struct kvm *kvm; |
536a6f88 | 5108 | u64 tmp_val; |
1165f5fe | 5109 | |
8b88b099 | 5110 | *val = 0; |
0d9ce162 | 5111 | mutex_lock(&kvm_lock); |
536a6f88 | 5112 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5113 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
5114 | *val += tmp_val; |
5115 | } | |
0d9ce162 | 5116 | mutex_unlock(&kvm_lock); |
8b88b099 | 5117 | return 0; |
1165f5fe AK |
5118 | } |
5119 | ||
ce35ef27 SJS |
5120 | static int vcpu_stat_clear(void *_offset, u64 val) |
5121 | { | |
5122 | unsigned offset = (long)_offset; | |
5123 | struct kvm *kvm; | |
ce35ef27 SJS |
5124 | |
5125 | if (val) | |
5126 | return -EINVAL; | |
5127 | ||
0d9ce162 | 5128 | mutex_lock(&kvm_lock); |
ce35ef27 | 5129 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5130 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 5131 | } |
0d9ce162 | 5132 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5133 | |
5134 | return 0; | |
5135 | } | |
5136 | ||
5137 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
5138 | "%llu\n"); | |
bc9e9e67 | 5139 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n"); |
1165f5fe | 5140 | |
286de8f6 CI |
5141 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
5142 | { | |
5143 | struct kobj_uevent_env *env; | |
286de8f6 CI |
5144 | unsigned long long created, active; |
5145 | ||
5146 | if (!kvm_dev.this_device || !kvm) | |
5147 | return; | |
5148 | ||
0d9ce162 | 5149 | mutex_lock(&kvm_lock); |
286de8f6 CI |
5150 | if (type == KVM_EVENT_CREATE_VM) { |
5151 | kvm_createvm_count++; | |
5152 | kvm_active_vms++; | |
5153 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
5154 | kvm_active_vms--; | |
5155 | } | |
5156 | created = kvm_createvm_count; | |
5157 | active = kvm_active_vms; | |
0d9ce162 | 5158 | mutex_unlock(&kvm_lock); |
286de8f6 | 5159 | |
b12ce36a | 5160 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
5161 | if (!env) |
5162 | return; | |
5163 | ||
5164 | add_uevent_var(env, "CREATED=%llu", created); | |
5165 | add_uevent_var(env, "COUNT=%llu", active); | |
5166 | ||
fdeaf7e3 | 5167 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 5168 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
5169 | kvm->userspace_pid = task_pid_nr(current); |
5170 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 5171 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
5172 | } |
5173 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 5174 | |
8ed0579c | 5175 | if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) { |
b12ce36a | 5176 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
5177 | |
5178 | if (p) { | |
5179 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
5180 | if (!IS_ERR(tmp)) | |
5181 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
5182 | kfree(p); | |
286de8f6 CI |
5183 | } |
5184 | } | |
5185 | /* no need for checks, since we are adding at most only 5 keys */ | |
5186 | env->envp[env->envp_idx++] = NULL; | |
5187 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
5188 | kfree(env); | |
286de8f6 CI |
5189 | } |
5190 | ||
929f45e3 | 5191 | static void kvm_init_debug(void) |
6aa8b732 | 5192 | { |
bc9e9e67 JZ |
5193 | const struct file_operations *fops; |
5194 | const struct _kvm_stats_desc *pdesc; | |
5195 | int i; | |
6aa8b732 | 5196 | |
76f7c879 | 5197 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 5198 | |
bc9e9e67 JZ |
5199 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
5200 | pdesc = &kvm_vm_stats_desc[i]; | |
5201 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5202 | fops = &vm_stat_fops; | |
5203 | else | |
5204 | fops = &vm_stat_readonly_fops; | |
5205 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5206 | kvm_debugfs_dir, | |
5207 | (void *)(long)pdesc->desc.offset, fops); | |
5208 | } | |
5209 | ||
5210 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
5211 | pdesc = &kvm_vcpu_stats_desc[i]; | |
5212 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5213 | fops = &vcpu_stat_fops; | |
5214 | else | |
5215 | fops = &vcpu_stat_readonly_fops; | |
5216 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5217 | kvm_debugfs_dir, | |
5218 | (void *)(long)pdesc->desc.offset, fops); | |
4f69b680 | 5219 | } |
6aa8b732 AK |
5220 | } |
5221 | ||
fb3600cc | 5222 | static int kvm_suspend(void) |
59ae6c6b | 5223 | { |
10474ae8 | 5224 | if (kvm_usage_count) |
75b7127c | 5225 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
5226 | return 0; |
5227 | } | |
5228 | ||
fb3600cc | 5229 | static void kvm_resume(void) |
59ae6c6b | 5230 | { |
ca84d1a2 | 5231 | if (kvm_usage_count) { |
2eb06c30 WL |
5232 | #ifdef CONFIG_LOCKDEP |
5233 | WARN_ON(lockdep_is_held(&kvm_count_lock)); | |
5234 | #endif | |
75b7127c | 5235 | hardware_enable_nolock(NULL); |
ca84d1a2 | 5236 | } |
59ae6c6b AK |
5237 | } |
5238 | ||
fb3600cc | 5239 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
5240 | .suspend = kvm_suspend, |
5241 | .resume = kvm_resume, | |
5242 | }; | |
5243 | ||
15ad7146 AK |
5244 | static inline |
5245 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
5246 | { | |
5247 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
5248 | } | |
5249 | ||
5250 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
5251 | { | |
5252 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 5253 | |
046ddeed | 5254 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 5255 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 5256 | |
7495e22b | 5257 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 5258 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 5259 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
5260 | } |
5261 | ||
5262 | static void kvm_sched_out(struct preempt_notifier *pn, | |
5263 | struct task_struct *next) | |
5264 | { | |
5265 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
5266 | ||
3ba9f93b | 5267 | if (current->on_rq) { |
046ddeed | 5268 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
5269 | WRITE_ONCE(vcpu->ready, true); |
5270 | } | |
e9b11c17 | 5271 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
5272 | __this_cpu_write(kvm_running_vcpu, NULL); |
5273 | } | |
5274 | ||
5275 | /** | |
5276 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
5277 | * |
5278 | * We can disable preemption locally around accessing the per-CPU variable, | |
5279 | * and use the resolved vcpu pointer after enabling preemption again, | |
5280 | * because even if the current thread is migrated to another CPU, reading | |
5281 | * the per-CPU value later will give us the same value as we update the | |
5282 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
5283 | */ |
5284 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
5285 | { | |
1f03b2bc MZ |
5286 | struct kvm_vcpu *vcpu; |
5287 | ||
5288 | preempt_disable(); | |
5289 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
5290 | preempt_enable(); | |
5291 | ||
5292 | return vcpu; | |
7495e22b | 5293 | } |
379a3c8e | 5294 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
5295 | |
5296 | /** | |
5297 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
5298 | */ | |
5299 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
5300 | { | |
5301 | return &kvm_running_vcpu; | |
15ad7146 AK |
5302 | } |
5303 | ||
b9904085 SC |
5304 | struct kvm_cpu_compat_check { |
5305 | void *opaque; | |
5306 | int *ret; | |
5307 | }; | |
5308 | ||
5309 | static void check_processor_compat(void *data) | |
f257d6dc | 5310 | { |
b9904085 SC |
5311 | struct kvm_cpu_compat_check *c = data; |
5312 | ||
5313 | *c->ret = kvm_arch_check_processor_compat(c->opaque); | |
f257d6dc SC |
5314 | } |
5315 | ||
0ee75bea | 5316 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 5317 | struct module *module) |
6aa8b732 | 5318 | { |
b9904085 | 5319 | struct kvm_cpu_compat_check c; |
6aa8b732 | 5320 | int r; |
002c7f7c | 5321 | int cpu; |
6aa8b732 | 5322 | |
f8c16bba ZX |
5323 | r = kvm_arch_init(opaque); |
5324 | if (r) | |
d2308784 | 5325 | goto out_fail; |
cb498ea2 | 5326 | |
7dac16c3 AH |
5327 | /* |
5328 | * kvm_arch_init makes sure there's at most one caller | |
5329 | * for architectures that support multiple implementations, | |
5330 | * like intel and amd on x86. | |
36343f6e PB |
5331 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
5332 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 5333 | */ |
36343f6e PB |
5334 | r = kvm_irqfd_init(); |
5335 | if (r) | |
5336 | goto out_irqfd; | |
7dac16c3 | 5337 | |
8437a617 | 5338 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
5339 | r = -ENOMEM; |
5340 | goto out_free_0; | |
5341 | } | |
5342 | ||
b9904085 | 5343 | r = kvm_arch_hardware_setup(opaque); |
6aa8b732 | 5344 | if (r < 0) |
faf0be22 | 5345 | goto out_free_1; |
6aa8b732 | 5346 | |
b9904085 SC |
5347 | c.ret = &r; |
5348 | c.opaque = opaque; | |
002c7f7c | 5349 | for_each_online_cpu(cpu) { |
b9904085 | 5350 | smp_call_function_single(cpu, check_processor_compat, &c, 1); |
002c7f7c | 5351 | if (r < 0) |
faf0be22 | 5352 | goto out_free_2; |
002c7f7c YS |
5353 | } |
5354 | ||
73c1b41e | 5355 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 5356 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 5357 | if (r) |
d2308784 | 5358 | goto out_free_2; |
6aa8b732 AK |
5359 | register_reboot_notifier(&kvm_reboot_notifier); |
5360 | ||
c16f862d | 5361 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
5362 | if (!vcpu_align) |
5363 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
5364 | kvm_vcpu_cache = |
5365 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
5366 | SLAB_ACCOUNT, | |
5367 | offsetof(struct kvm_vcpu, arch), | |
ce55c049 JZ |
5368 | offsetofend(struct kvm_vcpu, stats_id) |
5369 | - offsetof(struct kvm_vcpu, arch), | |
46515736 | 5370 | NULL); |
c16f862d RR |
5371 | if (!kvm_vcpu_cache) { |
5372 | r = -ENOMEM; | |
fb3600cc | 5373 | goto out_free_3; |
c16f862d RR |
5374 | } |
5375 | ||
af585b92 GN |
5376 | r = kvm_async_pf_init(); |
5377 | if (r) | |
5378 | goto out_free; | |
5379 | ||
6aa8b732 | 5380 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
5381 | kvm_vm_fops.owner = module; |
5382 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
5383 | |
5384 | r = misc_register(&kvm_dev); | |
5385 | if (r) { | |
1170adc6 | 5386 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 5387 | goto out_unreg; |
6aa8b732 AK |
5388 | } |
5389 | ||
fb3600cc RW |
5390 | register_syscore_ops(&kvm_syscore_ops); |
5391 | ||
15ad7146 AK |
5392 | kvm_preempt_ops.sched_in = kvm_sched_in; |
5393 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
5394 | ||
929f45e3 | 5395 | kvm_init_debug(); |
0ea4ed8e | 5396 | |
3c3c29fd PB |
5397 | r = kvm_vfio_ops_init(); |
5398 | WARN_ON(r); | |
5399 | ||
c7addb90 | 5400 | return 0; |
6aa8b732 | 5401 | |
af585b92 GN |
5402 | out_unreg: |
5403 | kvm_async_pf_deinit(); | |
6aa8b732 | 5404 | out_free: |
c16f862d | 5405 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 5406 | out_free_3: |
6aa8b732 | 5407 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5408 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 5409 | out_free_2: |
e9b11c17 | 5410 | kvm_arch_hardware_unsetup(); |
faf0be22 | 5411 | out_free_1: |
7f59f492 | 5412 | free_cpumask_var(cpus_hardware_enabled); |
d2308784 | 5413 | out_free_0: |
a0f155e9 | 5414 | kvm_irqfd_exit(); |
36343f6e | 5415 | out_irqfd: |
7dac16c3 AH |
5416 | kvm_arch_exit(); |
5417 | out_fail: | |
6aa8b732 AK |
5418 | return r; |
5419 | } | |
cb498ea2 | 5420 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 5421 | |
cb498ea2 | 5422 | void kvm_exit(void) |
6aa8b732 | 5423 | { |
4bd33b56 | 5424 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 5425 | misc_deregister(&kvm_dev); |
c16f862d | 5426 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 5427 | kvm_async_pf_deinit(); |
fb3600cc | 5428 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 5429 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5430 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 5431 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 5432 | kvm_arch_hardware_unsetup(); |
f8c16bba | 5433 | kvm_arch_exit(); |
a0f155e9 | 5434 | kvm_irqfd_exit(); |
7f59f492 | 5435 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 5436 | kvm_vfio_ops_exit(); |
6aa8b732 | 5437 | } |
cb498ea2 | 5438 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
5439 | |
5440 | struct kvm_vm_worker_thread_context { | |
5441 | struct kvm *kvm; | |
5442 | struct task_struct *parent; | |
5443 | struct completion init_done; | |
5444 | kvm_vm_thread_fn_t thread_fn; | |
5445 | uintptr_t data; | |
5446 | int err; | |
5447 | }; | |
5448 | ||
5449 | static int kvm_vm_worker_thread(void *context) | |
5450 | { | |
5451 | /* | |
5452 | * The init_context is allocated on the stack of the parent thread, so | |
5453 | * we have to locally copy anything that is needed beyond initialization | |
5454 | */ | |
5455 | struct kvm_vm_worker_thread_context *init_context = context; | |
5456 | struct kvm *kvm = init_context->kvm; | |
5457 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
5458 | uintptr_t data = init_context->data; | |
5459 | int err; | |
5460 | ||
5461 | err = kthread_park(current); | |
5462 | /* kthread_park(current) is never supposed to return an error */ | |
5463 | WARN_ON(err != 0); | |
5464 | if (err) | |
5465 | goto init_complete; | |
5466 | ||
5467 | err = cgroup_attach_task_all(init_context->parent, current); | |
5468 | if (err) { | |
5469 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
5470 | __func__, err); | |
5471 | goto init_complete; | |
5472 | } | |
5473 | ||
5474 | set_user_nice(current, task_nice(init_context->parent)); | |
5475 | ||
5476 | init_complete: | |
5477 | init_context->err = err; | |
5478 | complete(&init_context->init_done); | |
5479 | init_context = NULL; | |
5480 | ||
5481 | if (err) | |
5482 | return err; | |
5483 | ||
5484 | /* Wait to be woken up by the spawner before proceeding. */ | |
5485 | kthread_parkme(); | |
5486 | ||
5487 | if (!kthread_should_stop()) | |
5488 | err = thread_fn(kvm, data); | |
5489 | ||
5490 | return err; | |
5491 | } | |
5492 | ||
5493 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
5494 | uintptr_t data, const char *name, | |
5495 | struct task_struct **thread_ptr) | |
5496 | { | |
5497 | struct kvm_vm_worker_thread_context init_context = {}; | |
5498 | struct task_struct *thread; | |
5499 | ||
5500 | *thread_ptr = NULL; | |
5501 | init_context.kvm = kvm; | |
5502 | init_context.parent = current; | |
5503 | init_context.thread_fn = thread_fn; | |
5504 | init_context.data = data; | |
5505 | init_completion(&init_context.init_done); | |
5506 | ||
5507 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
5508 | "%s-%d", name, task_pid_nr(current)); | |
5509 | if (IS_ERR(thread)) | |
5510 | return PTR_ERR(thread); | |
5511 | ||
5512 | /* kthread_run is never supposed to return NULL */ | |
5513 | WARN_ON(thread == NULL); | |
5514 | ||
5515 | wait_for_completion(&init_context.init_done); | |
5516 | ||
5517 | if (!init_context.err) | |
5518 | *thread_ptr = thread; | |
5519 | ||
5520 | return init_context.err; | |
5521 | } |