]>
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 | ||
bccf2150 AK |
192 | /* |
193 | * Switches to specified vcpu, until a matching vcpu_put() | |
194 | */ | |
ec7660cc | 195 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 196 | { |
ec7660cc | 197 | int cpu = get_cpu(); |
7495e22b PB |
198 | |
199 | __this_cpu_write(kvm_running_vcpu, vcpu); | |
15ad7146 | 200 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 201 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 202 | put_cpu(); |
6aa8b732 | 203 | } |
2f1fe811 | 204 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 205 | |
313a3dc7 | 206 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 207 | { |
15ad7146 | 208 | preempt_disable(); |
313a3dc7 | 209 | kvm_arch_vcpu_put(vcpu); |
15ad7146 | 210 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
7495e22b | 211 | __this_cpu_write(kvm_running_vcpu, NULL); |
15ad7146 | 212 | preempt_enable(); |
6aa8b732 | 213 | } |
2f1fe811 | 214 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 215 | |
7a97cec2 PB |
216 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
217 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
218 | { | |
219 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
220 | ||
221 | /* | |
222 | * We need to wait for the VCPU to reenable interrupts and get out of | |
223 | * READING_SHADOW_PAGE_TABLES mode. | |
224 | */ | |
225 | if (req & KVM_REQUEST_WAIT) | |
226 | return mode != OUTSIDE_GUEST_MODE; | |
227 | ||
228 | /* | |
229 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
230 | */ | |
231 | return mode == IN_GUEST_MODE; | |
232 | } | |
233 | ||
d9e368d6 AK |
234 | static void ack_flush(void *_completed) |
235 | { | |
d9e368d6 AK |
236 | } |
237 | ||
0bbc2ca8 | 238 | static inline bool kvm_kick_many_cpus(cpumask_var_t tmp, bool wait) |
b49defe8 | 239 | { |
0bbc2ca8 SC |
240 | const struct cpumask *cpus; |
241 | ||
242 | if (likely(cpumask_available(tmp))) | |
243 | cpus = tmp; | |
244 | else | |
b49defe8 PB |
245 | cpus = cpu_online_mask; |
246 | ||
247 | if (cpumask_empty(cpus)) | |
248 | return false; | |
249 | ||
250 | smp_call_function_many(cpus, ack_flush, NULL, wait); | |
251 | return true; | |
252 | } | |
253 | ||
7053df4e | 254 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
54163a34 | 255 | struct kvm_vcpu *except, |
7053df4e | 256 | unsigned long *vcpu_bitmap, cpumask_var_t tmp) |
d9e368d6 | 257 | { |
597a5f55 | 258 | int i, cpu, me; |
d9e368d6 | 259 | struct kvm_vcpu *vcpu; |
7053df4e | 260 | bool called; |
6ef7a1bc | 261 | |
3cba4130 | 262 | me = get_cpu(); |
7053df4e | 263 | |
988a2cae | 264 | kvm_for_each_vcpu(i, vcpu, kvm) { |
54163a34 SS |
265 | if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) || |
266 | vcpu == except) | |
7053df4e VK |
267 | continue; |
268 | ||
3cba4130 | 269 | kvm_make_request(req, vcpu); |
6b7e2d09 | 270 | |
178f02ff RK |
271 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) |
272 | continue; | |
6c6e8360 | 273 | |
0bbc2ca8 SC |
274 | /* |
275 | * tmp can be "unavailable" if cpumasks are allocated off stack | |
276 | * as allocation of the mask is deliberately not fatal and is | |
277 | * handled by falling back to kicking all online CPUs. | |
278 | */ | |
279 | if (!cpumask_available(tmp)) | |
280 | continue; | |
281 | ||
85b64045 SC |
282 | /* |
283 | * Note, the vCPU could get migrated to a different pCPU at any | |
284 | * point after kvm_request_needs_ipi(), which could result in | |
285 | * sending an IPI to the previous pCPU. But, that's ok because | |
286 | * the purpose of the IPI is to ensure the vCPU returns to | |
287 | * OUTSIDE_GUEST_MODE, which is satisfied if the vCPU migrates. | |
288 | * Entering READING_SHADOW_PAGE_TABLES after this point is also | |
289 | * ok, as the requirement is only that KVM wait for vCPUs that | |
290 | * were reading SPTEs _before_ any changes were finalized. See | |
291 | * kvm_vcpu_kick() for more details on handling requests. | |
292 | */ | |
0bbc2ca8 | 293 | if (kvm_request_needs_ipi(vcpu, req)) { |
85b64045 SC |
294 | cpu = READ_ONCE(vcpu->cpu); |
295 | if (cpu != -1 && cpu != me) | |
296 | __cpumask_set_cpu(cpu, tmp); | |
297 | } | |
49846896 | 298 | } |
7053df4e VK |
299 | |
300 | called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT)); | |
3cba4130 | 301 | put_cpu(); |
7053df4e VK |
302 | |
303 | return called; | |
304 | } | |
305 | ||
54163a34 SS |
306 | bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req, |
307 | struct kvm_vcpu *except) | |
7053df4e VK |
308 | { |
309 | cpumask_var_t cpus; | |
310 | bool called; | |
7053df4e VK |
311 | |
312 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); | |
313 | ||
54163a34 | 314 | called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus); |
7053df4e | 315 | |
6ef7a1bc | 316 | free_cpumask_var(cpus); |
49846896 | 317 | return called; |
d9e368d6 AK |
318 | } |
319 | ||
54163a34 SS |
320 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
321 | { | |
322 | return kvm_make_all_cpus_request_except(kvm, req, NULL); | |
323 | } | |
a2486020 | 324 | EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request); |
54163a34 | 325 | |
a6d51016 | 326 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 327 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 328 | { |
3cc4e148 | 329 | ++kvm->stat.generic.remote_tlb_flush_requests; |
6bc6db00 | 330 | |
4ae3cb3a LT |
331 | /* |
332 | * We want to publish modifications to the page tables before reading | |
333 | * mode. Pairs with a memory barrier in arch-specific code. | |
334 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
335 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
336 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
337 | * | |
338 | * There is already an smp_mb__after_atomic() before | |
339 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
340 | * barrier here. | |
341 | */ | |
b08660e5 TL |
342 | if (!kvm_arch_flush_remote_tlb(kvm) |
343 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
0193cc90 | 344 | ++kvm->stat.generic.remote_tlb_flush; |
2e53d63a | 345 | } |
2ba9f0d8 | 346 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 347 | #endif |
2e53d63a | 348 | |
49846896 RR |
349 | void kvm_reload_remote_mmus(struct kvm *kvm) |
350 | { | |
445b8236 | 351 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 352 | } |
2e53d63a | 353 | |
6926f95a SC |
354 | #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE |
355 | static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc, | |
356 | gfp_t gfp_flags) | |
357 | { | |
358 | gfp_flags |= mc->gfp_zero; | |
359 | ||
360 | if (mc->kmem_cache) | |
361 | return kmem_cache_alloc(mc->kmem_cache, gfp_flags); | |
362 | else | |
363 | return (void *)__get_free_page(gfp_flags); | |
364 | } | |
365 | ||
366 | int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min) | |
367 | { | |
368 | void *obj; | |
369 | ||
370 | if (mc->nobjs >= min) | |
371 | return 0; | |
372 | while (mc->nobjs < ARRAY_SIZE(mc->objects)) { | |
373 | obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT); | |
374 | if (!obj) | |
375 | return mc->nobjs >= min ? 0 : -ENOMEM; | |
376 | mc->objects[mc->nobjs++] = obj; | |
377 | } | |
378 | return 0; | |
379 | } | |
380 | ||
381 | int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc) | |
382 | { | |
383 | return mc->nobjs; | |
384 | } | |
385 | ||
386 | void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
387 | { | |
388 | while (mc->nobjs) { | |
389 | if (mc->kmem_cache) | |
390 | kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]); | |
391 | else | |
392 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
393 | } | |
394 | } | |
395 | ||
396 | void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
397 | { | |
398 | void *p; | |
399 | ||
400 | if (WARN_ON(!mc->nobjs)) | |
401 | p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT); | |
402 | else | |
403 | p = mc->objects[--mc->nobjs]; | |
404 | BUG_ON(!p); | |
405 | return p; | |
406 | } | |
407 | #endif | |
408 | ||
8bd826d6 | 409 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 410 | { |
fb3f0f51 RR |
411 | mutex_init(&vcpu->mutex); |
412 | vcpu->cpu = -1; | |
fb3f0f51 RR |
413 | vcpu->kvm = kvm; |
414 | vcpu->vcpu_id = id; | |
34bb10b7 | 415 | vcpu->pid = NULL; |
da4ad88c | 416 | rcuwait_init(&vcpu->wait); |
af585b92 | 417 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 418 | |
bf9f6ac8 FW |
419 | vcpu->pre_pcpu = -1; |
420 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
421 | ||
4c088493 R |
422 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
423 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 424 | vcpu->preempted = false; |
d73eb57b | 425 | vcpu->ready = false; |
d5c48deb | 426 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
fe22ed82 | 427 | vcpu->last_used_slot = 0; |
fb3f0f51 | 428 | } |
fb3f0f51 | 429 | |
4543bdc0 SC |
430 | void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
431 | { | |
fb04a1ed | 432 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
4543bdc0 | 433 | kvm_arch_vcpu_destroy(vcpu); |
e529ef66 | 434 | |
9941d224 SC |
435 | /* |
436 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
437 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
438 | * are already gone. | |
439 | */ | |
440 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
441 | ||
8bd826d6 | 442 | free_page((unsigned long)vcpu->run); |
e529ef66 | 443 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 SC |
444 | } |
445 | EXPORT_SYMBOL_GPL(kvm_vcpu_destroy); | |
446 | ||
e930bffe AA |
447 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
448 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
449 | { | |
450 | return container_of(mn, struct kvm, mmu_notifier); | |
451 | } | |
452 | ||
e649b3f0 ET |
453 | static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn, |
454 | struct mm_struct *mm, | |
455 | unsigned long start, unsigned long end) | |
456 | { | |
457 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
458 | int idx; | |
459 | ||
460 | idx = srcu_read_lock(&kvm->srcu); | |
461 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
462 | srcu_read_unlock(&kvm->srcu, idx); | |
463 | } | |
464 | ||
3039bcc7 SC |
465 | typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range); |
466 | ||
f922bd9b SC |
467 | typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start, |
468 | unsigned long end); | |
469 | ||
3039bcc7 SC |
470 | struct kvm_hva_range { |
471 | unsigned long start; | |
472 | unsigned long end; | |
473 | pte_t pte; | |
474 | hva_handler_t handler; | |
f922bd9b | 475 | on_lock_fn_t on_lock; |
3039bcc7 SC |
476 | bool flush_on_ret; |
477 | bool may_block; | |
478 | }; | |
479 | ||
f922bd9b SC |
480 | /* |
481 | * Use a dedicated stub instead of NULL to indicate that there is no callback | |
482 | * function/handler. The compiler technically can't guarantee that a real | |
483 | * function will have a non-zero address, and so it will generate code to | |
484 | * check for !NULL, whereas comparing against a stub will be elided at compile | |
485 | * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9). | |
486 | */ | |
487 | static void kvm_null_fn(void) | |
488 | { | |
489 | ||
490 | } | |
491 | #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn) | |
492 | ||
3039bcc7 SC |
493 | static __always_inline int __kvm_handle_hva_range(struct kvm *kvm, |
494 | const struct kvm_hva_range *range) | |
495 | { | |
8931a454 | 496 | bool ret = false, locked = false; |
f922bd9b | 497 | struct kvm_gfn_range gfn_range; |
3039bcc7 SC |
498 | struct kvm_memory_slot *slot; |
499 | struct kvm_memslots *slots; | |
3039bcc7 SC |
500 | int i, idx; |
501 | ||
f922bd9b SC |
502 | /* A null handler is allowed if and only if on_lock() is provided. */ |
503 | if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) && | |
504 | IS_KVM_NULL_FN(range->handler))) | |
505 | return 0; | |
506 | ||
3039bcc7 SC |
507 | idx = srcu_read_lock(&kvm->srcu); |
508 | ||
509 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
510 | slots = __kvm_memslots(kvm, i); | |
511 | kvm_for_each_memslot(slot, slots) { | |
512 | unsigned long hva_start, hva_end; | |
513 | ||
514 | hva_start = max(range->start, slot->userspace_addr); | |
515 | hva_end = min(range->end, slot->userspace_addr + | |
516 | (slot->npages << PAGE_SHIFT)); | |
517 | if (hva_start >= hva_end) | |
518 | continue; | |
519 | ||
520 | /* | |
521 | * To optimize for the likely case where the address | |
522 | * range is covered by zero or one memslots, don't | |
523 | * bother making these conditional (to avoid writes on | |
524 | * the second or later invocation of the handler). | |
525 | */ | |
526 | gfn_range.pte = range->pte; | |
527 | gfn_range.may_block = range->may_block; | |
528 | ||
529 | /* | |
530 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
531 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
532 | */ | |
533 | gfn_range.start = hva_to_gfn_memslot(hva_start, slot); | |
534 | gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot); | |
535 | gfn_range.slot = slot; | |
536 | ||
8931a454 SC |
537 | if (!locked) { |
538 | locked = true; | |
539 | KVM_MMU_LOCK(kvm); | |
071064f1 PB |
540 | if (!IS_KVM_NULL_FN(range->on_lock)) |
541 | range->on_lock(kvm, range->start, range->end); | |
542 | if (IS_KVM_NULL_FN(range->handler)) | |
543 | break; | |
8931a454 | 544 | } |
3039bcc7 SC |
545 | ret |= range->handler(kvm, &gfn_range); |
546 | } | |
547 | } | |
548 | ||
6bc6db00 | 549 | if (range->flush_on_ret && ret) |
3039bcc7 SC |
550 | kvm_flush_remote_tlbs(kvm); |
551 | ||
8931a454 SC |
552 | if (locked) |
553 | KVM_MMU_UNLOCK(kvm); | |
f922bd9b | 554 | |
3039bcc7 SC |
555 | srcu_read_unlock(&kvm->srcu, idx); |
556 | ||
557 | /* The notifiers are averse to booleans. :-( */ | |
558 | return (int)ret; | |
559 | } | |
560 | ||
561 | static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn, | |
562 | unsigned long start, | |
563 | unsigned long end, | |
564 | pte_t pte, | |
565 | hva_handler_t handler) | |
566 | { | |
567 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
568 | const struct kvm_hva_range range = { | |
569 | .start = start, | |
570 | .end = end, | |
571 | .pte = pte, | |
572 | .handler = handler, | |
f922bd9b | 573 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
574 | .flush_on_ret = true, |
575 | .may_block = false, | |
576 | }; | |
3039bcc7 | 577 | |
f922bd9b | 578 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 SC |
579 | } |
580 | ||
581 | static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn, | |
582 | unsigned long start, | |
583 | unsigned long end, | |
584 | hva_handler_t handler) | |
585 | { | |
586 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
587 | const struct kvm_hva_range range = { | |
588 | .start = start, | |
589 | .end = end, | |
590 | .pte = __pte(0), | |
591 | .handler = handler, | |
f922bd9b | 592 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
593 | .flush_on_ret = false, |
594 | .may_block = false, | |
595 | }; | |
3039bcc7 | 596 | |
f922bd9b | 597 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 | 598 | } |
3da0dd43 IE |
599 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
600 | struct mm_struct *mm, | |
601 | unsigned long address, | |
602 | pte_t pte) | |
603 | { | |
604 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
605 | ||
501b9185 SC |
606 | trace_kvm_set_spte_hva(address); |
607 | ||
c13fda23 | 608 | /* |
52ac8b35 | 609 | * .change_pte() must be surrounded by .invalidate_range_{start,end}(). |
071064f1 PB |
610 | * If mmu_notifier_count is zero, then no in-progress invalidations, |
611 | * including this one, found a relevant memslot at start(); rechecking | |
612 | * memslots here is unnecessary. Note, a false positive (count elevated | |
613 | * by a different invalidation) is sub-optimal but functionally ok. | |
c13fda23 | 614 | */ |
52ac8b35 | 615 | WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count)); |
071064f1 PB |
616 | if (!READ_ONCE(kvm->mmu_notifier_count)) |
617 | return; | |
c13fda23 | 618 | |
3039bcc7 | 619 | kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn); |
3da0dd43 IE |
620 | } |
621 | ||
edb298c6 | 622 | void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 623 | unsigned long end) |
e930bffe | 624 | { |
e930bffe AA |
625 | /* |
626 | * The count increase must become visible at unlock time as no | |
627 | * spte can be established without taking the mmu_lock and | |
628 | * count is also read inside the mmu_lock critical section. | |
629 | */ | |
630 | kvm->mmu_notifier_count++; | |
4a42d848 | 631 | if (likely(kvm->mmu_notifier_count == 1)) { |
f922bd9b SC |
632 | kvm->mmu_notifier_range_start = start; |
633 | kvm->mmu_notifier_range_end = end; | |
4a42d848 DS |
634 | } else { |
635 | /* | |
636 | * Fully tracking multiple concurrent ranges has dimishing | |
637 | * returns. Keep things simple and just find the minimal range | |
638 | * which includes the current and new ranges. As there won't be | |
639 | * enough information to subtract a range after its invalidate | |
640 | * completes, any ranges invalidated concurrently will | |
641 | * accumulate and persist until all outstanding invalidates | |
642 | * complete. | |
643 | */ | |
644 | kvm->mmu_notifier_range_start = | |
f922bd9b | 645 | min(kvm->mmu_notifier_range_start, start); |
4a42d848 | 646 | kvm->mmu_notifier_range_end = |
f922bd9b | 647 | max(kvm->mmu_notifier_range_end, end); |
4a42d848 | 648 | } |
f922bd9b | 649 | } |
3039bcc7 | 650 | |
f922bd9b SC |
651 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
652 | const struct mmu_notifier_range *range) | |
653 | { | |
654 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
655 | const struct kvm_hva_range hva_range = { | |
656 | .start = range->start, | |
657 | .end = range->end, | |
658 | .pte = __pte(0), | |
659 | .handler = kvm_unmap_gfn_range, | |
660 | .on_lock = kvm_inc_notifier_count, | |
661 | .flush_on_ret = true, | |
662 | .may_block = mmu_notifier_range_blockable(range), | |
663 | }; | |
565f3be2 | 664 | |
f922bd9b SC |
665 | trace_kvm_unmap_hva_range(range->start, range->end); |
666 | ||
52ac8b35 PB |
667 | /* |
668 | * Prevent memslot modification between range_start() and range_end() | |
669 | * so that conditionally locking provides the same result in both | |
670 | * functions. Without that guarantee, the mmu_notifier_count | |
671 | * adjustments will be imbalanced. | |
672 | * | |
673 | * Pairs with the decrement in range_end(). | |
674 | */ | |
675 | spin_lock(&kvm->mn_invalidate_lock); | |
676 | kvm->mn_active_invalidate_count++; | |
677 | spin_unlock(&kvm->mn_invalidate_lock); | |
678 | ||
f922bd9b | 679 | __kvm_handle_hva_range(kvm, &hva_range); |
93065ac7 | 680 | |
e649b3f0 | 681 | return 0; |
e930bffe AA |
682 | } |
683 | ||
edb298c6 | 684 | void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 685 | unsigned long end) |
e930bffe | 686 | { |
e930bffe AA |
687 | /* |
688 | * This sequence increase will notify the kvm page fault that | |
689 | * the page that is going to be mapped in the spte could have | |
690 | * been freed. | |
691 | */ | |
692 | kvm->mmu_notifier_seq++; | |
a355aa54 | 693 | smp_wmb(); |
e930bffe AA |
694 | /* |
695 | * The above sequence increase must be visible before the | |
a355aa54 PM |
696 | * below count decrease, which is ensured by the smp_wmb above |
697 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
698 | */ |
699 | kvm->mmu_notifier_count--; | |
f922bd9b SC |
700 | } |
701 | ||
702 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
703 | const struct mmu_notifier_range *range) | |
704 | { | |
705 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
706 | const struct kvm_hva_range hva_range = { | |
707 | .start = range->start, | |
708 | .end = range->end, | |
709 | .pte = __pte(0), | |
710 | .handler = (void *)kvm_null_fn, | |
711 | .on_lock = kvm_dec_notifier_count, | |
712 | .flush_on_ret = false, | |
713 | .may_block = mmu_notifier_range_blockable(range), | |
714 | }; | |
52ac8b35 | 715 | bool wake; |
f922bd9b SC |
716 | |
717 | __kvm_handle_hva_range(kvm, &hva_range); | |
e930bffe | 718 | |
52ac8b35 PB |
719 | /* Pairs with the increment in range_start(). */ |
720 | spin_lock(&kvm->mn_invalidate_lock); | |
721 | wake = (--kvm->mn_active_invalidate_count == 0); | |
722 | spin_unlock(&kvm->mn_invalidate_lock); | |
723 | ||
724 | /* | |
725 | * There can only be one waiter, since the wait happens under | |
726 | * slots_lock. | |
727 | */ | |
728 | if (wake) | |
729 | rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait); | |
730 | ||
e930bffe AA |
731 | BUG_ON(kvm->mmu_notifier_count < 0); |
732 | } | |
733 | ||
734 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
735 | struct mm_struct *mm, | |
57128468 ALC |
736 | unsigned long start, |
737 | unsigned long end) | |
e930bffe | 738 | { |
501b9185 SC |
739 | trace_kvm_age_hva(start, end); |
740 | ||
3039bcc7 | 741 | return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn); |
e930bffe AA |
742 | } |
743 | ||
1d7715c6 VD |
744 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
745 | struct mm_struct *mm, | |
746 | unsigned long start, | |
747 | unsigned long end) | |
748 | { | |
501b9185 SC |
749 | trace_kvm_age_hva(start, end); |
750 | ||
1d7715c6 VD |
751 | /* |
752 | * Even though we do not flush TLB, this will still adversely | |
753 | * affect performance on pre-Haswell Intel EPT, where there is | |
754 | * no EPT Access Bit to clear so that we have to tear down EPT | |
755 | * tables instead. If we find this unacceptable, we can always | |
756 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
757 | * do anything on clear_young. | |
758 | * | |
759 | * Also note that currently we never issue secondary TLB flushes | |
760 | * from clear_young, leaving this job up to the regular system | |
761 | * cadence. If we find this inaccurate, we might come up with a | |
762 | * more sophisticated heuristic later. | |
763 | */ | |
3039bcc7 | 764 | return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn); |
1d7715c6 VD |
765 | } |
766 | ||
8ee53820 AA |
767 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
768 | struct mm_struct *mm, | |
769 | unsigned long address) | |
770 | { | |
501b9185 SC |
771 | trace_kvm_test_age_hva(address); |
772 | ||
3039bcc7 SC |
773 | return kvm_handle_hva_range_no_flush(mn, address, address + 1, |
774 | kvm_test_age_gfn); | |
8ee53820 AA |
775 | } |
776 | ||
85db06e5 MT |
777 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
778 | struct mm_struct *mm) | |
779 | { | |
780 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
781 | int idx; |
782 | ||
783 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 784 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 785 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
786 | } |
787 | ||
e930bffe | 788 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e649b3f0 | 789 | .invalidate_range = kvm_mmu_notifier_invalidate_range, |
e930bffe AA |
790 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
791 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
792 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 793 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 794 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 795 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 796 | .release = kvm_mmu_notifier_release, |
e930bffe | 797 | }; |
4c07b0a4 AK |
798 | |
799 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
800 | { | |
801 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
802 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
803 | } | |
804 | ||
805 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
806 | ||
807 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
808 | { | |
809 | return 0; | |
810 | } | |
811 | ||
e930bffe AA |
812 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
813 | ||
2fdef3a2 SS |
814 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
815 | static int kvm_pm_notifier_call(struct notifier_block *bl, | |
816 | unsigned long state, | |
817 | void *unused) | |
818 | { | |
819 | struct kvm *kvm = container_of(bl, struct kvm, pm_notifier); | |
820 | ||
821 | return kvm_arch_pm_notifier(kvm, state); | |
822 | } | |
823 | ||
824 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
825 | { | |
826 | kvm->pm_notifier.notifier_call = kvm_pm_notifier_call; | |
827 | /* Suspend KVM before we suspend ftrace, RCU, etc. */ | |
828 | kvm->pm_notifier.priority = INT_MAX; | |
829 | register_pm_notifier(&kvm->pm_notifier); | |
830 | } | |
831 | ||
832 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
833 | { | |
834 | unregister_pm_notifier(&kvm->pm_notifier); | |
835 | } | |
836 | #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
837 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
838 | { | |
839 | } | |
840 | ||
841 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
842 | { | |
843 | } | |
844 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
845 | ||
a47d2b07 | 846 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
847 | { |
848 | int i; | |
a47d2b07 | 849 | struct kvm_memslots *slots; |
bf3e05bc | 850 | |
b12ce36a | 851 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
a47d2b07 PB |
852 | if (!slots) |
853 | return NULL; | |
854 | ||
bf3e05bc | 855 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
36947254 | 856 | slots->id_to_index[i] = -1; |
a47d2b07 PB |
857 | |
858 | return slots; | |
859 | } | |
860 | ||
861 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
862 | { | |
863 | if (!memslot->dirty_bitmap) | |
864 | return; | |
865 | ||
866 | kvfree(memslot->dirty_bitmap); | |
867 | memslot->dirty_bitmap = NULL; | |
868 | } | |
869 | ||
e96c81ee | 870 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 871 | { |
e96c81ee | 872 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 873 | |
e96c81ee | 874 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 875 | |
e96c81ee SC |
876 | slot->flags = 0; |
877 | slot->npages = 0; | |
a47d2b07 PB |
878 | } |
879 | ||
880 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
881 | { | |
882 | struct kvm_memory_slot *memslot; | |
883 | ||
884 | if (!slots) | |
885 | return; | |
886 | ||
887 | kvm_for_each_memslot(memslot, slots) | |
e96c81ee | 888 | kvm_free_memslot(kvm, memslot); |
a47d2b07 PB |
889 | |
890 | kvfree(slots); | |
bf3e05bc XG |
891 | } |
892 | ||
bc9e9e67 JZ |
893 | static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc) |
894 | { | |
895 | switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) { | |
896 | case KVM_STATS_TYPE_INSTANT: | |
897 | return 0444; | |
898 | case KVM_STATS_TYPE_CUMULATIVE: | |
899 | case KVM_STATS_TYPE_PEAK: | |
900 | default: | |
901 | return 0644; | |
902 | } | |
903 | } | |
904 | ||
905 | ||
536a6f88 JF |
906 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
907 | { | |
908 | int i; | |
bc9e9e67 JZ |
909 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
910 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
911 | |
912 | if (!kvm->debugfs_dentry) | |
913 | return; | |
914 | ||
915 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
916 | ||
9d5a1dce LC |
917 | if (kvm->debugfs_stat_data) { |
918 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
919 | kfree(kvm->debugfs_stat_data[i]); | |
920 | kfree(kvm->debugfs_stat_data); | |
921 | } | |
536a6f88 JF |
922 | } |
923 | ||
924 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
925 | { | |
85cd39af PB |
926 | static DEFINE_MUTEX(kvm_debugfs_lock); |
927 | struct dentry *dent; | |
536a6f88 JF |
928 | char dir_name[ITOA_MAX_LEN * 2]; |
929 | struct kvm_stat_data *stat_data; | |
bc9e9e67 | 930 | const struct _kvm_stats_desc *pdesc; |
3165af73 | 931 | int i, ret; |
bc9e9e67 JZ |
932 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
933 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
934 | |
935 | if (!debugfs_initialized()) | |
936 | return 0; | |
937 | ||
938 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
85cd39af PB |
939 | mutex_lock(&kvm_debugfs_lock); |
940 | dent = debugfs_lookup(dir_name, kvm_debugfs_dir); | |
941 | if (dent) { | |
942 | pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name); | |
943 | dput(dent); | |
944 | mutex_unlock(&kvm_debugfs_lock); | |
945 | return 0; | |
946 | } | |
947 | dent = debugfs_create_dir(dir_name, kvm_debugfs_dir); | |
948 | mutex_unlock(&kvm_debugfs_lock); | |
949 | if (IS_ERR(dent)) | |
950 | return 0; | |
536a6f88 | 951 | |
85cd39af | 952 | kvm->debugfs_dentry = dent; |
536a6f88 JF |
953 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
954 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 955 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
956 | if (!kvm->debugfs_stat_data) |
957 | return -ENOMEM; | |
958 | ||
bc9e9e67 JZ |
959 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
960 | pdesc = &kvm_vm_stats_desc[i]; | |
b12ce36a | 961 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
962 | if (!stat_data) |
963 | return -ENOMEM; | |
964 | ||
965 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
966 | stat_data->desc = pdesc; |
967 | stat_data->kind = KVM_STAT_VM; | |
968 | kvm->debugfs_stat_data[i] = stat_data; | |
969 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
970 | kvm->debugfs_dentry, stat_data, | |
971 | &stat_fops_per_vm); | |
972 | } | |
973 | ||
974 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
975 | pdesc = &kvm_vcpu_stats_desc[i]; | |
b12ce36a | 976 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
977 | if (!stat_data) |
978 | return -ENOMEM; | |
979 | ||
980 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
981 | stat_data->desc = pdesc; |
982 | stat_data->kind = KVM_STAT_VCPU; | |
004d62eb | 983 | kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data; |
bc9e9e67 | 984 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), |
09cbcef6 MP |
985 | kvm->debugfs_dentry, stat_data, |
986 | &stat_fops_per_vm); | |
536a6f88 | 987 | } |
3165af73 PX |
988 | |
989 | ret = kvm_arch_create_vm_debugfs(kvm); | |
990 | if (ret) { | |
991 | kvm_destroy_vm_debugfs(kvm); | |
992 | return i; | |
993 | } | |
994 | ||
536a6f88 JF |
995 | return 0; |
996 | } | |
997 | ||
1aa9b957 JS |
998 | /* |
999 | * Called after the VM is otherwise initialized, but just before adding it to | |
1000 | * the vm_list. | |
1001 | */ | |
1002 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
1003 | { | |
1004 | return 0; | |
1005 | } | |
1006 | ||
1007 | /* | |
1008 | * Called just after removing the VM from the vm_list, but before doing any | |
1009 | * other destruction. | |
1010 | */ | |
1011 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
1012 | { | |
1013 | } | |
1014 | ||
3165af73 PX |
1015 | /* |
1016 | * Called after per-vm debugfs created. When called kvm->debugfs_dentry should | |
1017 | * be setup already, so we can create arch-specific debugfs entries under it. | |
1018 | * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so | |
1019 | * a per-arch destroy interface is not needed. | |
1020 | */ | |
1021 | int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm) | |
1022 | { | |
1023 | return 0; | |
1024 | } | |
1025 | ||
e08b9637 | 1026 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 1027 | { |
d89f5eff | 1028 | struct kvm *kvm = kvm_arch_alloc_vm(); |
9121923c JM |
1029 | int r = -ENOMEM; |
1030 | int i; | |
6aa8b732 | 1031 | |
d89f5eff JK |
1032 | if (!kvm) |
1033 | return ERR_PTR(-ENOMEM); | |
1034 | ||
531810ca | 1035 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 1036 | mmgrab(current->mm); |
e9ad4ec8 PB |
1037 | kvm->mm = current->mm; |
1038 | kvm_eventfd_init(kvm); | |
1039 | mutex_init(&kvm->lock); | |
1040 | mutex_init(&kvm->irq_lock); | |
1041 | mutex_init(&kvm->slots_lock); | |
b10a038e | 1042 | mutex_init(&kvm->slots_arch_lock); |
52ac8b35 PB |
1043 | spin_lock_init(&kvm->mn_invalidate_lock); |
1044 | rcuwait_init(&kvm->mn_memslots_update_rcuwait); | |
1045 | ||
e9ad4ec8 PB |
1046 | INIT_LIST_HEAD(&kvm->devices); |
1047 | ||
1e702d9a AW |
1048 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
1049 | ||
8a44119a PB |
1050 | if (init_srcu_struct(&kvm->srcu)) |
1051 | goto out_err_no_srcu; | |
1052 | if (init_srcu_struct(&kvm->irq_srcu)) | |
1053 | goto out_err_no_irq_srcu; | |
1054 | ||
e2d3fcaf | 1055 | refcount_set(&kvm->users_count, 1); |
f481b069 | 1056 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4bd518f1 | 1057 | struct kvm_memslots *slots = kvm_alloc_memslots(); |
9121923c | 1058 | |
4bd518f1 | 1059 | if (!slots) |
a97b0e77 | 1060 | goto out_err_no_arch_destroy_vm; |
0e32958e | 1061 | /* Generations must be different for each address space. */ |
164bf7e5 | 1062 | slots->generation = i; |
4bd518f1 | 1063 | rcu_assign_pointer(kvm->memslots[i], slots); |
f481b069 | 1064 | } |
00f034a1 | 1065 | |
e93f8a0f | 1066 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 1067 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 1068 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 1069 | if (!kvm->buses[i]) |
a97b0e77 | 1070 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 1071 | } |
e930bffe | 1072 | |
acd05785 DM |
1073 | kvm->max_halt_poll_ns = halt_poll_ns; |
1074 | ||
e08b9637 | 1075 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 1076 | if (r) |
a97b0e77 | 1077 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
1078 | |
1079 | r = hardware_enable_all(); | |
1080 | if (r) | |
719d93cd | 1081 | goto out_err_no_disable; |
10474ae8 | 1082 | |
c77dcacb | 1083 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 1084 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 1085 | #endif |
6aa8b732 | 1086 | |
74b5c5bf | 1087 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
1088 | if (r) |
1089 | goto out_err_no_mmu_notifier; | |
1090 | ||
1091 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
1092 | if (r) |
1093 | goto out_err; | |
1094 | ||
0d9ce162 | 1095 | mutex_lock(&kvm_lock); |
5e58cfe4 | 1096 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 1097 | mutex_unlock(&kvm_lock); |
d89f5eff | 1098 | |
2ecd9d29 | 1099 | preempt_notifier_inc(); |
2fdef3a2 | 1100 | kvm_init_pm_notifier(kvm); |
2ecd9d29 | 1101 | |
f17abe9a | 1102 | return kvm; |
10474ae8 AG |
1103 | |
1104 | out_err: | |
1aa9b957 JS |
1105 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1106 | if (kvm->mmu_notifier.ops) | |
1107 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
1108 | #endif | |
1109 | out_err_no_mmu_notifier: | |
10474ae8 | 1110 | hardware_disable_all(); |
719d93cd | 1111 | out_err_no_disable: |
a97b0e77 | 1112 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 1113 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 1114 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 1115 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 1116 | kfree(kvm_get_bus(kvm, i)); |
f481b069 | 1117 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 1118 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
8a44119a PB |
1119 | cleanup_srcu_struct(&kvm->irq_srcu); |
1120 | out_err_no_irq_srcu: | |
1121 | cleanup_srcu_struct(&kvm->srcu); | |
1122 | out_err_no_srcu: | |
d89f5eff | 1123 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 1124 | mmdrop(current->mm); |
10474ae8 | 1125 | return ERR_PTR(r); |
f17abe9a AK |
1126 | } |
1127 | ||
07f0a7bd SW |
1128 | static void kvm_destroy_devices(struct kvm *kvm) |
1129 | { | |
e6e3b5a6 | 1130 | struct kvm_device *dev, *tmp; |
07f0a7bd | 1131 | |
a28ebea2 CD |
1132 | /* |
1133 | * We do not need to take the kvm->lock here, because nobody else | |
1134 | * has a reference to the struct kvm at this point and therefore | |
1135 | * cannot access the devices list anyhow. | |
1136 | */ | |
e6e3b5a6 GT |
1137 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
1138 | list_del(&dev->vm_node); | |
07f0a7bd SW |
1139 | dev->ops->destroy(dev); |
1140 | } | |
1141 | } | |
1142 | ||
f17abe9a AK |
1143 | static void kvm_destroy_vm(struct kvm *kvm) |
1144 | { | |
e93f8a0f | 1145 | int i; |
6d4e4c4f AK |
1146 | struct mm_struct *mm = kvm->mm; |
1147 | ||
2fdef3a2 | 1148 | kvm_destroy_pm_notifier(kvm); |
286de8f6 | 1149 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 1150 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 1151 | kvm_arch_sync_events(kvm); |
0d9ce162 | 1152 | mutex_lock(&kvm_lock); |
133de902 | 1153 | list_del(&kvm->vm_list); |
0d9ce162 | 1154 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
1155 | kvm_arch_pre_destroy_vm(kvm); |
1156 | ||
399ec807 | 1157 | kvm_free_irq_routing(kvm); |
df630b8c | 1158 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 1159 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 1160 | |
4a12f951 CB |
1161 | if (bus) |
1162 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
1163 | kvm->buses[i] = NULL; |
1164 | } | |
980da6ce | 1165 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
1166 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1167 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
52ac8b35 PB |
1168 | /* |
1169 | * At this point, pending calls to invalidate_range_start() | |
1170 | * have completed but no more MMU notifiers will run, so | |
1171 | * mn_active_invalidate_count may remain unbalanced. | |
1172 | * No threads can be waiting in install_new_memslots as the | |
1173 | * last reference on KVM has been dropped, but freeing | |
1174 | * memslots would deadlock without this manual intervention. | |
1175 | */ | |
1176 | WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait)); | |
1177 | kvm->mn_active_invalidate_count = 0; | |
f00be0ca | 1178 | #else |
2df72e9b | 1179 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 1180 | #endif |
d19a9cd2 | 1181 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 1182 | kvm_destroy_devices(kvm); |
f481b069 | 1183 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 1184 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
820b3fcd | 1185 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
1186 | cleanup_srcu_struct(&kvm->srcu); |
1187 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 1188 | preempt_notifier_dec(); |
10474ae8 | 1189 | hardware_disable_all(); |
6d4e4c4f | 1190 | mmdrop(mm); |
f17abe9a AK |
1191 | } |
1192 | ||
d39f13b0 IE |
1193 | void kvm_get_kvm(struct kvm *kvm) |
1194 | { | |
e3736c3e | 1195 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
1196 | } |
1197 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
1198 | ||
605c7130 PX |
1199 | /* |
1200 | * Make sure the vm is not during destruction, which is a safe version of | |
1201 | * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise. | |
1202 | */ | |
1203 | bool kvm_get_kvm_safe(struct kvm *kvm) | |
1204 | { | |
1205 | return refcount_inc_not_zero(&kvm->users_count); | |
1206 | } | |
1207 | EXPORT_SYMBOL_GPL(kvm_get_kvm_safe); | |
1208 | ||
d39f13b0 IE |
1209 | void kvm_put_kvm(struct kvm *kvm) |
1210 | { | |
e3736c3e | 1211 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
1212 | kvm_destroy_vm(kvm); |
1213 | } | |
1214 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
1215 | ||
149487bd SC |
1216 | /* |
1217 | * Used to put a reference that was taken on behalf of an object associated | |
1218 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
1219 | * of the new file descriptor fails and the reference cannot be transferred to | |
1220 | * its final owner. In such cases, the caller is still actively using @kvm and | |
1221 | * will fail miserably if the refcount unexpectedly hits zero. | |
1222 | */ | |
1223 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
1224 | { | |
1225 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
1226 | } | |
1227 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 1228 | |
f17abe9a AK |
1229 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
1230 | { | |
1231 | struct kvm *kvm = filp->private_data; | |
1232 | ||
721eecbf GH |
1233 | kvm_irqfd_release(kvm); |
1234 | ||
d39f13b0 | 1235 | kvm_put_kvm(kvm); |
6aa8b732 AK |
1236 | return 0; |
1237 | } | |
1238 | ||
515a0127 TY |
1239 | /* |
1240 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 1241 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 1242 | */ |
3c9bd400 | 1243 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 1244 | { |
515a0127 | 1245 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 1246 | |
b12ce36a | 1247 | memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
1248 | if (!memslot->dirty_bitmap) |
1249 | return -ENOMEM; | |
1250 | ||
a36a57b1 TY |
1251 | return 0; |
1252 | } | |
1253 | ||
bf3e05bc | 1254 | /* |
0577d1ab SC |
1255 | * Delete a memslot by decrementing the number of used slots and shifting all |
1256 | * other entries in the array forward one spot. | |
bf3e05bc | 1257 | */ |
0577d1ab SC |
1258 | static inline void kvm_memslot_delete(struct kvm_memslots *slots, |
1259 | struct kvm_memory_slot *memslot) | |
bf3e05bc | 1260 | { |
063584d4 | 1261 | struct kvm_memory_slot *mslots = slots->memslots; |
0577d1ab | 1262 | int i; |
f85e2cb5 | 1263 | |
0577d1ab SC |
1264 | if (WARN_ON(slots->id_to_index[memslot->id] == -1)) |
1265 | return; | |
0e60b079 | 1266 | |
0577d1ab SC |
1267 | slots->used_slots--; |
1268 | ||
87689270 DM |
1269 | if (atomic_read(&slots->last_used_slot) >= slots->used_slots) |
1270 | atomic_set(&slots->last_used_slot, 0); | |
0774a964 | 1271 | |
0577d1ab | 1272 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) { |
7f379cff IM |
1273 | mslots[i] = mslots[i + 1]; |
1274 | slots->id_to_index[mslots[i].id] = i; | |
7f379cff | 1275 | } |
0577d1ab SC |
1276 | mslots[i] = *memslot; |
1277 | slots->id_to_index[memslot->id] = -1; | |
1278 | } | |
1279 | ||
1280 | /* | |
1281 | * "Insert" a new memslot by incrementing the number of used slots. Returns | |
1282 | * the new slot's initial index into the memslots array. | |
1283 | */ | |
1284 | static inline int kvm_memslot_insert_back(struct kvm_memslots *slots) | |
1285 | { | |
1286 | return slots->used_slots++; | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * Move a changed memslot backwards in the array by shifting existing slots | |
1291 | * with a higher GFN toward the front of the array. Note, the changed memslot | |
1292 | * itself is not preserved in the array, i.e. not swapped at this time, only | |
1293 | * its new index into the array is tracked. Returns the changed memslot's | |
1294 | * current index into the memslots array. | |
1295 | */ | |
1296 | static inline int kvm_memslot_move_backward(struct kvm_memslots *slots, | |
1297 | struct kvm_memory_slot *memslot) | |
1298 | { | |
1299 | struct kvm_memory_slot *mslots = slots->memslots; | |
1300 | int i; | |
1301 | ||
1302 | if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) || | |
1303 | WARN_ON_ONCE(!slots->used_slots)) | |
1304 | return -1; | |
efbeec70 PB |
1305 | |
1306 | /* | |
0577d1ab SC |
1307 | * Move the target memslot backward in the array by shifting existing |
1308 | * memslots with a higher GFN (than the target memslot) towards the | |
1309 | * front of the array. | |
efbeec70 | 1310 | */ |
0577d1ab SC |
1311 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) { |
1312 | if (memslot->base_gfn > mslots[i + 1].base_gfn) | |
1313 | break; | |
1314 | ||
1315 | WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn); | |
f85e2cb5 | 1316 | |
0577d1ab SC |
1317 | /* Shift the next memslot forward one and update its index. */ |
1318 | mslots[i] = mslots[i + 1]; | |
1319 | slots->id_to_index[mslots[i].id] = i; | |
1320 | } | |
1321 | return i; | |
1322 | } | |
1323 | ||
1324 | /* | |
1325 | * Move a changed memslot forwards in the array by shifting existing slots with | |
1326 | * a lower GFN toward the back of the array. Note, the changed memslot itself | |
1327 | * is not preserved in the array, i.e. not swapped at this time, only its new | |
1328 | * index into the array is tracked. Returns the changed memslot's final index | |
1329 | * into the memslots array. | |
1330 | */ | |
1331 | static inline int kvm_memslot_move_forward(struct kvm_memslots *slots, | |
1332 | struct kvm_memory_slot *memslot, | |
1333 | int start) | |
1334 | { | |
1335 | struct kvm_memory_slot *mslots = slots->memslots; | |
1336 | int i; | |
1337 | ||
1338 | for (i = start; i > 0; i--) { | |
1339 | if (memslot->base_gfn < mslots[i - 1].base_gfn) | |
1340 | break; | |
1341 | ||
1342 | WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn); | |
1343 | ||
1344 | /* Shift the next memslot back one and update its index. */ | |
1345 | mslots[i] = mslots[i - 1]; | |
1346 | slots->id_to_index[mslots[i].id] = i; | |
1347 | } | |
1348 | return i; | |
1349 | } | |
1350 | ||
1351 | /* | |
1352 | * Re-sort memslots based on their GFN to account for an added, deleted, or | |
1353 | * moved memslot. Sorting memslots by GFN allows using a binary search during | |
1354 | * memslot lookup. | |
1355 | * | |
1356 | * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry | |
1357 | * at memslots[0] has the highest GFN. | |
1358 | * | |
1359 | * The sorting algorithm takes advantage of having initially sorted memslots | |
1360 | * and knowing the position of the changed memslot. Sorting is also optimized | |
1361 | * by not swapping the updated memslot and instead only shifting other memslots | |
1362 | * and tracking the new index for the update memslot. Only once its final | |
1363 | * index is known is the updated memslot copied into its position in the array. | |
1364 | * | |
1365 | * - When deleting a memslot, the deleted memslot simply needs to be moved to | |
1366 | * the end of the array. | |
1367 | * | |
1368 | * - When creating a memslot, the algorithm "inserts" the new memslot at the | |
1369 | * end of the array and then it forward to its correct location. | |
1370 | * | |
1371 | * - When moving a memslot, the algorithm first moves the updated memslot | |
1372 | * backward to handle the scenario where the memslot's GFN was changed to a | |
1373 | * lower value. update_memslots() then falls through and runs the same flow | |
1374 | * as creating a memslot to move the memslot forward to handle the scenario | |
1375 | * where its GFN was changed to a higher value. | |
1376 | * | |
1377 | * Note, slots are sorted from highest->lowest instead of lowest->highest for | |
1378 | * historical reasons. Originally, invalid memslots where denoted by having | |
1379 | * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots | |
1380 | * to the end of the array. The current algorithm uses dedicated logic to | |
1381 | * delete a memslot and thus does not rely on invalid memslots having GFN=0. | |
1382 | * | |
1383 | * The other historical motiviation for highest->lowest was to improve the | |
1384 | * performance of memslot lookup. KVM originally used a linear search starting | |
1385 | * at memslots[0]. On x86, the largest memslot usually has one of the highest, | |
1386 | * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a | |
1387 | * single memslot above the 4gb boundary. As the largest memslot is also the | |
1388 | * most likely to be referenced, sorting it to the front of the array was | |
1389 | * advantageous. The current binary search starts from the middle of the array | |
1390 | * and uses an LRU pointer to improve performance for all memslots and GFNs. | |
1391 | */ | |
1392 | static void update_memslots(struct kvm_memslots *slots, | |
1393 | struct kvm_memory_slot *memslot, | |
1394 | enum kvm_mr_change change) | |
1395 | { | |
1396 | int i; | |
1397 | ||
1398 | if (change == KVM_MR_DELETE) { | |
1399 | kvm_memslot_delete(slots, memslot); | |
1400 | } else { | |
1401 | if (change == KVM_MR_CREATE) | |
1402 | i = kvm_memslot_insert_back(slots); | |
1403 | else | |
1404 | i = kvm_memslot_move_backward(slots, memslot); | |
1405 | i = kvm_memslot_move_forward(slots, memslot, i); | |
1406 | ||
1407 | /* | |
1408 | * Copy the memslot to its new position in memslots and update | |
1409 | * its index accordingly. | |
1410 | */ | |
1411 | slots->memslots[i] = *memslot; | |
1412 | slots->id_to_index[memslot->id] = i; | |
1413 | } | |
bf3e05bc XG |
1414 | } |
1415 | ||
09170a49 | 1416 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 1417 | { |
4d8b81ab XG |
1418 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1419 | ||
0f8a4de3 | 1420 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
1421 | valid_flags |= KVM_MEM_READONLY; |
1422 | #endif | |
1423 | ||
1424 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1425 | return -EINVAL; |
1426 | ||
1427 | return 0; | |
1428 | } | |
1429 | ||
7ec4fb44 | 1430 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 1431 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 1432 | { |
f481b069 | 1433 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
361209e0 | 1434 | u64 gen = old_memslots->generation; |
7ec4fb44 | 1435 | |
361209e0 SC |
1436 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1437 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1438 | |
52ac8b35 PB |
1439 | /* |
1440 | * Do not store the new memslots while there are invalidations in | |
071064f1 PB |
1441 | * progress, otherwise the locking in invalidate_range_start and |
1442 | * invalidate_range_end will be unbalanced. | |
52ac8b35 PB |
1443 | */ |
1444 | spin_lock(&kvm->mn_invalidate_lock); | |
1445 | prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
1446 | while (kvm->mn_active_invalidate_count) { | |
1447 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1448 | spin_unlock(&kvm->mn_invalidate_lock); | |
1449 | schedule(); | |
1450 | spin_lock(&kvm->mn_invalidate_lock); | |
1451 | } | |
1452 | finish_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
f481b069 | 1453 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
52ac8b35 | 1454 | spin_unlock(&kvm->mn_invalidate_lock); |
b10a038e BG |
1455 | |
1456 | /* | |
1457 | * Acquired in kvm_set_memslot. Must be released before synchronize | |
1458 | * SRCU below in order to avoid deadlock with another thread | |
1459 | * acquiring the slots_arch_lock in an srcu critical section. | |
1460 | */ | |
1461 | mutex_unlock(&kvm->slots_arch_lock); | |
1462 | ||
7ec4fb44 | 1463 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1464 | |
ee3d1570 | 1465 | /* |
361209e0 | 1466 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1467 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1468 | * the number of address spaces. This provides a unique and easily |
1469 | * identifiable generation number while the memslots are in flux. | |
1470 | */ | |
1471 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1472 | ||
1473 | /* | |
4bd518f1 PB |
1474 | * Generations must be unique even across address spaces. We do not need |
1475 | * a global counter for that, instead the generation space is evenly split | |
1476 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1477 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1478 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1479 | */ |
164bf7e5 | 1480 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 1481 | |
15248258 | 1482 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1483 | |
15248258 | 1484 | slots->generation = gen; |
e59dbe09 TY |
1485 | |
1486 | return old_memslots; | |
7ec4fb44 GN |
1487 | } |
1488 | ||
ddc12f2a BG |
1489 | static size_t kvm_memslots_size(int slots) |
1490 | { | |
1491 | return sizeof(struct kvm_memslots) + | |
1492 | (sizeof(struct kvm_memory_slot) * slots); | |
1493 | } | |
1494 | ||
1495 | static void kvm_copy_memslots(struct kvm_memslots *to, | |
1496 | struct kvm_memslots *from) | |
1497 | { | |
1498 | memcpy(to, from, kvm_memslots_size(from->used_slots)); | |
1499 | } | |
1500 | ||
36947254 SC |
1501 | /* |
1502 | * Note, at a minimum, the current number of used slots must be allocated, even | |
1503 | * when deleting a memslot, as we need a complete duplicate of the memslots for | |
1504 | * use when invalidating a memslot prior to deleting/moving the memslot. | |
1505 | */ | |
1506 | static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old, | |
1507 | enum kvm_mr_change change) | |
1508 | { | |
1509 | struct kvm_memslots *slots; | |
ddc12f2a | 1510 | size_t new_size; |
36947254 SC |
1511 | |
1512 | if (change == KVM_MR_CREATE) | |
ddc12f2a | 1513 | new_size = kvm_memslots_size(old->used_slots + 1); |
36947254 | 1514 | else |
ddc12f2a | 1515 | new_size = kvm_memslots_size(old->used_slots); |
36947254 SC |
1516 | |
1517 | slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT); | |
1518 | if (likely(slots)) | |
ddc12f2a | 1519 | kvm_copy_memslots(slots, old); |
36947254 SC |
1520 | |
1521 | return slots; | |
1522 | } | |
1523 | ||
cf47f50b SC |
1524 | static int kvm_set_memslot(struct kvm *kvm, |
1525 | const struct kvm_userspace_memory_region *mem, | |
9d4c197c | 1526 | struct kvm_memory_slot *old, |
cf47f50b SC |
1527 | struct kvm_memory_slot *new, int as_id, |
1528 | enum kvm_mr_change change) | |
1529 | { | |
1530 | struct kvm_memory_slot *slot; | |
1531 | struct kvm_memslots *slots; | |
1532 | int r; | |
1533 | ||
b10a038e BG |
1534 | /* |
1535 | * Released in install_new_memslots. | |
1536 | * | |
1537 | * Must be held from before the current memslots are copied until | |
1538 | * after the new memslots are installed with rcu_assign_pointer, | |
1539 | * then released before the synchronize srcu in install_new_memslots. | |
1540 | * | |
1541 | * When modifying memslots outside of the slots_lock, must be held | |
1542 | * before reading the pointer to the current memslots until after all | |
1543 | * changes to those memslots are complete. | |
1544 | * | |
1545 | * These rules ensure that installing new memslots does not lose | |
1546 | * changes made to the previous memslots. | |
1547 | */ | |
1548 | mutex_lock(&kvm->slots_arch_lock); | |
1549 | ||
36947254 | 1550 | slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change); |
b10a038e BG |
1551 | if (!slots) { |
1552 | mutex_unlock(&kvm->slots_arch_lock); | |
cf47f50b | 1553 | return -ENOMEM; |
b10a038e | 1554 | } |
cf47f50b SC |
1555 | |
1556 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { | |
1557 | /* | |
1558 | * Note, the INVALID flag needs to be in the appropriate entry | |
1559 | * in the freshly allocated memslots, not in @old or @new. | |
1560 | */ | |
1561 | slot = id_to_memslot(slots, old->id); | |
1562 | slot->flags |= KVM_MEMSLOT_INVALID; | |
1563 | ||
1564 | /* | |
b10a038e BG |
1565 | * We can re-use the memory from the old memslots. |
1566 | * It will be overwritten with a copy of the new memslots | |
1567 | * after reacquiring the slots_arch_lock below. | |
cf47f50b SC |
1568 | */ |
1569 | slots = install_new_memslots(kvm, as_id, slots); | |
1570 | ||
1571 | /* From this point no new shadow pages pointing to a deleted, | |
1572 | * or moved, memslot will be created. | |
1573 | * | |
1574 | * validation of sp->gfn happens in: | |
1575 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1576 | * - kvm_is_visible_gfn (mmu_check_root) | |
1577 | */ | |
1578 | kvm_arch_flush_shadow_memslot(kvm, slot); | |
b10a038e BG |
1579 | |
1580 | /* Released in install_new_memslots. */ | |
1581 | mutex_lock(&kvm->slots_arch_lock); | |
1582 | ||
1583 | /* | |
1584 | * The arch-specific fields of the memslots could have changed | |
1585 | * between releasing the slots_arch_lock in | |
1586 | * install_new_memslots and here, so get a fresh copy of the | |
1587 | * slots. | |
1588 | */ | |
1589 | kvm_copy_memslots(slots, __kvm_memslots(kvm, as_id)); | |
cf47f50b SC |
1590 | } |
1591 | ||
1592 | r = kvm_arch_prepare_memory_region(kvm, new, mem, change); | |
1593 | if (r) | |
1594 | goto out_slots; | |
1595 | ||
1596 | update_memslots(slots, new, change); | |
1597 | slots = install_new_memslots(kvm, as_id, slots); | |
1598 | ||
1599 | kvm_arch_commit_memory_region(kvm, mem, old, new, change); | |
1600 | ||
1601 | kvfree(slots); | |
1602 | return 0; | |
1603 | ||
1604 | out_slots: | |
b10a038e BG |
1605 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
1606 | slot = id_to_memslot(slots, old->id); | |
1607 | slot->flags &= ~KVM_MEMSLOT_INVALID; | |
cf47f50b | 1608 | slots = install_new_memslots(kvm, as_id, slots); |
b10a038e BG |
1609 | } else { |
1610 | mutex_unlock(&kvm->slots_arch_lock); | |
1611 | } | |
cf47f50b SC |
1612 | kvfree(slots); |
1613 | return r; | |
1614 | } | |
1615 | ||
5c0b4f3d SC |
1616 | static int kvm_delete_memslot(struct kvm *kvm, |
1617 | const struct kvm_userspace_memory_region *mem, | |
1618 | struct kvm_memory_slot *old, int as_id) | |
1619 | { | |
1620 | struct kvm_memory_slot new; | |
1621 | int r; | |
1622 | ||
1623 | if (!old->npages) | |
1624 | return -EINVAL; | |
1625 | ||
1626 | memset(&new, 0, sizeof(new)); | |
1627 | new.id = old->id; | |
9e9eb226 PX |
1628 | /* |
1629 | * This is only for debugging purpose; it should never be referenced | |
1630 | * for a removed memslot. | |
1631 | */ | |
1632 | new.as_id = as_id; | |
5c0b4f3d SC |
1633 | |
1634 | r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE); | |
1635 | if (r) | |
1636 | return r; | |
1637 | ||
e96c81ee | 1638 | kvm_free_memslot(kvm, old); |
5c0b4f3d SC |
1639 | return 0; |
1640 | } | |
1641 | ||
6aa8b732 AK |
1642 | /* |
1643 | * Allocate some memory and give it an address in the guest physical address | |
1644 | * space. | |
1645 | * | |
1646 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1647 | * |
02d5d55b | 1648 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1649 | */ |
f78e0e2e | 1650 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 1651 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 1652 | { |
6aa8b732 | 1653 | struct kvm_memory_slot old, new; |
163da372 | 1654 | struct kvm_memory_slot *tmp; |
f64c0398 | 1655 | enum kvm_mr_change change; |
163da372 SC |
1656 | int as_id, id; |
1657 | int r; | |
6aa8b732 | 1658 | |
a50d64d6 XG |
1659 | r = check_memory_region_flags(mem); |
1660 | if (r) | |
71a4c30b | 1661 | return r; |
a50d64d6 | 1662 | |
f481b069 PB |
1663 | as_id = mem->slot >> 16; |
1664 | id = (u16)mem->slot; | |
1665 | ||
6aa8b732 AK |
1666 | /* General sanity checks */ |
1667 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
71a4c30b | 1668 | return -EINVAL; |
6aa8b732 | 1669 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 1670 | return -EINVAL; |
fa3d315a | 1671 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 1672 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 1673 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 1674 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 1675 | mem->memory_size)) |
71a4c30b | 1676 | return -EINVAL; |
f481b069 | 1677 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 1678 | return -EINVAL; |
6aa8b732 | 1679 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 1680 | return -EINVAL; |
6aa8b732 | 1681 | |
5c0b4f3d SC |
1682 | /* |
1683 | * Make a full copy of the old memslot, the pointer will become stale | |
1684 | * when the memslots are re-sorted by update_memslots(), and the old | |
1685 | * memslot needs to be referenced after calling update_memslots(), e.g. | |
0dff0846 | 1686 | * to free its resources and for arch specific behavior. |
5c0b4f3d | 1687 | */ |
0577d1ab SC |
1688 | tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
1689 | if (tmp) { | |
1690 | old = *tmp; | |
1691 | tmp = NULL; | |
1692 | } else { | |
1693 | memset(&old, 0, sizeof(old)); | |
1694 | old.id = id; | |
1695 | } | |
163da372 | 1696 | |
5c0b4f3d SC |
1697 | if (!mem->memory_size) |
1698 | return kvm_delete_memslot(kvm, mem, &old, as_id); | |
1699 | ||
9e9eb226 | 1700 | new.as_id = as_id; |
f481b069 | 1701 | new.id = id; |
163da372 SC |
1702 | new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
1703 | new.npages = mem->memory_size >> PAGE_SHIFT; | |
6aa8b732 | 1704 | new.flags = mem->flags; |
414de7ab | 1705 | new.userspace_addr = mem->userspace_addr; |
6aa8b732 | 1706 | |
163da372 SC |
1707 | if (new.npages > KVM_MEM_MAX_NR_PAGES) |
1708 | return -EINVAL; | |
1709 | ||
5c0b4f3d SC |
1710 | if (!old.npages) { |
1711 | change = KVM_MR_CREATE; | |
163da372 SC |
1712 | new.dirty_bitmap = NULL; |
1713 | memset(&new.arch, 0, sizeof(new.arch)); | |
5c0b4f3d SC |
1714 | } else { /* Modify an existing slot. */ |
1715 | if ((new.userspace_addr != old.userspace_addr) || | |
163da372 | 1716 | (new.npages != old.npages) || |
5c0b4f3d | 1717 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) |
71a4c30b | 1718 | return -EINVAL; |
09170a49 | 1719 | |
163da372 | 1720 | if (new.base_gfn != old.base_gfn) |
5c0b4f3d SC |
1721 | change = KVM_MR_MOVE; |
1722 | else if (new.flags != old.flags) | |
1723 | change = KVM_MR_FLAGS_ONLY; | |
1724 | else /* Nothing to change. */ | |
1725 | return 0; | |
163da372 SC |
1726 | |
1727 | /* Copy dirty_bitmap and arch from the current memslot. */ | |
1728 | new.dirty_bitmap = old.dirty_bitmap; | |
1729 | memcpy(&new.arch, &old.arch, sizeof(new.arch)); | |
09170a49 | 1730 | } |
6aa8b732 | 1731 | |
f64c0398 | 1732 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee | 1733 | /* Check for overlaps */ |
163da372 SC |
1734 | kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) { |
1735 | if (tmp->id == id) | |
0a706bee | 1736 | continue; |
163da372 SC |
1737 | if (!((new.base_gfn + new.npages <= tmp->base_gfn) || |
1738 | (new.base_gfn >= tmp->base_gfn + tmp->npages))) | |
71a4c30b | 1739 | return -EEXIST; |
0a706bee | 1740 | } |
6aa8b732 | 1741 | } |
6aa8b732 | 1742 | |
414de7ab SC |
1743 | /* Allocate/free page dirty bitmap as needed */ |
1744 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1745 | new.dirty_bitmap = NULL; | |
044c59c4 | 1746 | else if (!new.dirty_bitmap && !kvm->dirty_ring_size) { |
3c9bd400 | 1747 | r = kvm_alloc_dirty_bitmap(&new); |
71a4c30b SC |
1748 | if (r) |
1749 | return r; | |
3c9bd400 JZ |
1750 | |
1751 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1752 | bitmap_set(new.dirty_bitmap, 0, new.npages); | |
6aa8b732 AK |
1753 | } |
1754 | ||
cf47f50b SC |
1755 | r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change); |
1756 | if (r) | |
1757 | goto out_bitmap; | |
82ce2c96 | 1758 | |
5c0b4f3d SC |
1759 | if (old.dirty_bitmap && !new.dirty_bitmap) |
1760 | kvm_destroy_dirty_bitmap(&old); | |
6aa8b732 AK |
1761 | return 0; |
1762 | ||
bd0e96fd SC |
1763 | out_bitmap: |
1764 | if (new.dirty_bitmap && !old.dirty_bitmap) | |
1765 | kvm_destroy_dirty_bitmap(&new); | |
6aa8b732 | 1766 | return r; |
210c7c4d | 1767 | } |
f78e0e2e SY |
1768 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1769 | ||
1770 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1771 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1772 | { |
1773 | int r; | |
1774 | ||
79fac95e | 1775 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1776 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1777 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1778 | return r; |
1779 | } | |
210c7c4d IE |
1780 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1781 | ||
7940876e SH |
1782 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1783 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1784 | { |
f481b069 | 1785 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1786 | return -EINVAL; |
09170a49 | 1787 | |
47ae31e2 | 1788 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1789 | } |
1790 | ||
0dff0846 | 1791 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
1792 | /** |
1793 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
1794 | * @kvm: pointer to kvm instance | |
1795 | * @log: slot id and address to which we copy the log | |
1796 | * @is_dirty: set to '1' if any dirty pages were found | |
1797 | * @memslot: set to the associated memslot, always valid on success | |
1798 | */ | |
1799 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
1800 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 1801 | { |
9f6b8029 | 1802 | struct kvm_memslots *slots; |
843574a3 | 1803 | int i, as_id, id; |
87bf6e7d | 1804 | unsigned long n; |
6aa8b732 AK |
1805 | unsigned long any = 0; |
1806 | ||
b2cc64c4 PX |
1807 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1808 | if (kvm->dirty_ring_size) | |
1809 | return -ENXIO; | |
1810 | ||
2a49f61d SC |
1811 | *memslot = NULL; |
1812 | *is_dirty = 0; | |
1813 | ||
f481b069 PB |
1814 | as_id = log->slot >> 16; |
1815 | id = (u16)log->slot; | |
1816 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 1817 | return -EINVAL; |
6aa8b732 | 1818 | |
f481b069 | 1819 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 1820 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 1821 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 1822 | return -ENOENT; |
6aa8b732 | 1823 | |
2a49f61d SC |
1824 | kvm_arch_sync_dirty_log(kvm, *memslot); |
1825 | ||
1826 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 1827 | |
cd1a4a98 | 1828 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 1829 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 1830 | |
2a49f61d | 1831 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 1832 | return -EFAULT; |
6aa8b732 | 1833 | |
5bb064dc ZX |
1834 | if (any) |
1835 | *is_dirty = 1; | |
843574a3 | 1836 | return 0; |
6aa8b732 | 1837 | } |
2ba9f0d8 | 1838 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1839 | |
0dff0846 | 1840 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 1841 | /** |
b8b00220 | 1842 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 1843 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
1844 | * @kvm: pointer to kvm instance |
1845 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
1846 | * |
1847 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1848 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1849 | * following order: | |
1850 | * | |
1851 | * 1. Take a snapshot of the bit and clear it if needed. | |
1852 | * 2. Write protect the corresponding page. | |
1853 | * 3. Copy the snapshot to the userspace. | |
1854 | * 4. Upon return caller flushes TLB's if needed. | |
1855 | * | |
1856 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1857 | * entry. This is not a problem because the page is reported dirty using | |
1858 | * the snapshot taken before and step 4 ensures that writes done after | |
1859 | * exiting to userspace will be logged for the next call. | |
1860 | * | |
1861 | */ | |
0dff0846 | 1862 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 1863 | { |
9f6b8029 | 1864 | struct kvm_memslots *slots; |
ba0513b5 | 1865 | struct kvm_memory_slot *memslot; |
58d6db34 | 1866 | int i, as_id, id; |
ba0513b5 MS |
1867 | unsigned long n; |
1868 | unsigned long *dirty_bitmap; | |
1869 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 1870 | bool flush; |
ba0513b5 | 1871 | |
b2cc64c4 PX |
1872 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1873 | if (kvm->dirty_ring_size) | |
1874 | return -ENXIO; | |
1875 | ||
f481b069 PB |
1876 | as_id = log->slot >> 16; |
1877 | id = (u16)log->slot; | |
1878 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 1879 | return -EINVAL; |
ba0513b5 | 1880 | |
f481b069 PB |
1881 | slots = __kvm_memslots(kvm, as_id); |
1882 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
1883 | if (!memslot || !memslot->dirty_bitmap) |
1884 | return -ENOENT; | |
ba0513b5 MS |
1885 | |
1886 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 1887 | |
0dff0846 SC |
1888 | kvm_arch_sync_dirty_log(kvm, memslot); |
1889 | ||
ba0513b5 | 1890 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 1891 | flush = false; |
2a31b9db PB |
1892 | if (kvm->manual_dirty_log_protect) { |
1893 | /* | |
1894 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
1895 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
1896 | * is some code duplication between this function and | |
1897 | * kvm_get_dirty_log, but hopefully all architecture | |
1898 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
1899 | * can be eliminated. | |
1900 | */ | |
1901 | dirty_bitmap_buffer = dirty_bitmap; | |
1902 | } else { | |
1903 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
1904 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 1905 | |
531810ca | 1906 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
1907 | for (i = 0; i < n / sizeof(long); i++) { |
1908 | unsigned long mask; | |
1909 | gfn_t offset; | |
ba0513b5 | 1910 | |
2a31b9db PB |
1911 | if (!dirty_bitmap[i]) |
1912 | continue; | |
1913 | ||
0dff0846 | 1914 | flush = true; |
2a31b9db PB |
1915 | mask = xchg(&dirty_bitmap[i], 0); |
1916 | dirty_bitmap_buffer[i] = mask; | |
1917 | ||
a67794ca LT |
1918 | offset = i * BITS_PER_LONG; |
1919 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1920 | offset, mask); | |
2a31b9db | 1921 | } |
531810ca | 1922 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
1923 | } |
1924 | ||
0dff0846 SC |
1925 | if (flush) |
1926 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
1927 | ||
2a31b9db PB |
1928 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
1929 | return -EFAULT; | |
1930 | return 0; | |
1931 | } | |
0dff0846 SC |
1932 | |
1933 | ||
1934 | /** | |
1935 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
1936 | * @kvm: kvm instance | |
1937 | * @log: slot id and address to which we copy the log | |
1938 | * | |
1939 | * Steps 1-4 below provide general overview of dirty page logging. See | |
1940 | * kvm_get_dirty_log_protect() function description for additional details. | |
1941 | * | |
1942 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
1943 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
1944 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
1945 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
1946 | * writes will be marked dirty for next log read. | |
1947 | * | |
1948 | * 1. Take a snapshot of the bit and clear it if needed. | |
1949 | * 2. Write protect the corresponding page. | |
1950 | * 3. Copy the snapshot to the userspace. | |
1951 | * 4. Flush TLB's if needed. | |
1952 | */ | |
1953 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
1954 | struct kvm_dirty_log *log) | |
1955 | { | |
1956 | int r; | |
1957 | ||
1958 | mutex_lock(&kvm->slots_lock); | |
1959 | ||
1960 | r = kvm_get_dirty_log_protect(kvm, log); | |
1961 | ||
1962 | mutex_unlock(&kvm->slots_lock); | |
1963 | return r; | |
1964 | } | |
2a31b9db PB |
1965 | |
1966 | /** | |
1967 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
1968 | * and reenable dirty page tracking for the corresponding pages. | |
1969 | * @kvm: pointer to kvm instance | |
1970 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
1971 | */ | |
0dff0846 SC |
1972 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
1973 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
1974 | { |
1975 | struct kvm_memslots *slots; | |
1976 | struct kvm_memory_slot *memslot; | |
98938aa8 | 1977 | int as_id, id; |
2a31b9db | 1978 | gfn_t offset; |
98938aa8 | 1979 | unsigned long i, n; |
2a31b9db PB |
1980 | unsigned long *dirty_bitmap; |
1981 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 1982 | bool flush; |
2a31b9db | 1983 | |
b2cc64c4 PX |
1984 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1985 | if (kvm->dirty_ring_size) | |
1986 | return -ENXIO; | |
1987 | ||
2a31b9db PB |
1988 | as_id = log->slot >> 16; |
1989 | id = (u16)log->slot; | |
1990 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
1991 | return -EINVAL; | |
1992 | ||
76d58e0f | 1993 | if (log->first_page & 63) |
2a31b9db PB |
1994 | return -EINVAL; |
1995 | ||
1996 | slots = __kvm_memslots(kvm, as_id); | |
1997 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
1998 | if (!memslot || !memslot->dirty_bitmap) |
1999 | return -ENOENT; | |
2a31b9db PB |
2000 | |
2001 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 2002 | |
4ddc9204 | 2003 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
2004 | |
2005 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
2006 | log->num_pages > memslot->npages - log->first_page || |
2007 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
2008 | return -EINVAL; | |
98938aa8 | 2009 | |
0dff0846 SC |
2010 | kvm_arch_sync_dirty_log(kvm, memslot); |
2011 | ||
2012 | flush = false; | |
2a31b9db PB |
2013 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
2014 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
2015 | return -EFAULT; | |
ba0513b5 | 2016 | |
531810ca | 2017 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
2018 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
2019 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
2020 | i++, offset += BITS_PER_LONG) { |
2021 | unsigned long mask = *dirty_bitmap_buffer++; | |
2022 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
2023 | if (!mask) | |
ba0513b5 MS |
2024 | continue; |
2025 | ||
2a31b9db | 2026 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 2027 | |
2a31b9db PB |
2028 | /* |
2029 | * mask contains the bits that really have been cleared. This | |
2030 | * never includes any bits beyond the length of the memslot (if | |
2031 | * the length is not aligned to 64 pages), therefore it is not | |
2032 | * a problem if userspace sets them in log->dirty_bitmap. | |
2033 | */ | |
58d2930f | 2034 | if (mask) { |
0dff0846 | 2035 | flush = true; |
58d2930f TY |
2036 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
2037 | offset, mask); | |
2038 | } | |
ba0513b5 | 2039 | } |
531810ca | 2040 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 2041 | |
0dff0846 SC |
2042 | if (flush) |
2043 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2044 | ||
58d6db34 | 2045 | return 0; |
ba0513b5 | 2046 | } |
0dff0846 SC |
2047 | |
2048 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
2049 | struct kvm_clear_dirty_log *log) | |
2050 | { | |
2051 | int r; | |
2052 | ||
2053 | mutex_lock(&kvm->slots_lock); | |
2054 | ||
2055 | r = kvm_clear_dirty_log_protect(kvm, log); | |
2056 | ||
2057 | mutex_unlock(&kvm->slots_lock); | |
2058 | return r; | |
2059 | } | |
2060 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 2061 | |
49c7754c GN |
2062 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
2063 | { | |
2064 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
2065 | } | |
a1f4d395 | 2066 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 2067 | |
8e73485c PB |
2068 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
2069 | { | |
fe22ed82 DM |
2070 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
2071 | struct kvm_memory_slot *slot; | |
2072 | int slot_index; | |
2073 | ||
2074 | slot = try_get_memslot(slots, vcpu->last_used_slot, gfn); | |
2075 | if (slot) | |
2076 | return slot; | |
2077 | ||
2078 | /* | |
2079 | * Fall back to searching all memslots. We purposely use | |
2080 | * search_memslots() instead of __gfn_to_memslot() to avoid | |
2081 | * thrashing the VM-wide last_used_index in kvm_memslots. | |
2082 | */ | |
2083 | slot = search_memslots(slots, gfn, &slot_index); | |
2084 | if (slot) { | |
2085 | vcpu->last_used_slot = slot_index; | |
2086 | return slot; | |
2087 | } | |
2088 | ||
2089 | return NULL; | |
8e73485c | 2090 | } |
e72436bc | 2091 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot); |
8e73485c | 2092 | |
33e94154 | 2093 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 2094 | { |
bf3e05bc | 2095 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 2096 | |
c36b7150 | 2097 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
2098 | } |
2099 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
2100 | ||
995decb6 VK |
2101 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
2102 | { | |
2103 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2104 | ||
2105 | return kvm_is_visible_memslot(memslot); | |
2106 | } | |
2107 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
2108 | ||
f9b84e19 | 2109 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
2110 | { |
2111 | struct vm_area_struct *vma; | |
2112 | unsigned long addr, size; | |
2113 | ||
2114 | size = PAGE_SIZE; | |
2115 | ||
42cde48b | 2116 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
2117 | if (kvm_is_error_hva(addr)) |
2118 | return PAGE_SIZE; | |
2119 | ||
d8ed45c5 | 2120 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
2121 | vma = find_vma(current->mm, addr); |
2122 | if (!vma) | |
2123 | goto out; | |
2124 | ||
2125 | size = vma_kernel_pagesize(vma); | |
2126 | ||
2127 | out: | |
d8ed45c5 | 2128 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
2129 | |
2130 | return size; | |
2131 | } | |
2132 | ||
4d8b81ab XG |
2133 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
2134 | { | |
2135 | return slot->flags & KVM_MEM_READONLY; | |
2136 | } | |
2137 | ||
4d8b81ab XG |
2138 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2139 | gfn_t *nr_pages, bool write) | |
539cb660 | 2140 | { |
bc6678a3 | 2141 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 2142 | return KVM_HVA_ERR_BAD; |
48987781 | 2143 | |
4d8b81ab XG |
2144 | if (memslot_is_readonly(slot) && write) |
2145 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
2146 | |
2147 | if (nr_pages) | |
2148 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
2149 | ||
4d8b81ab | 2150 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 2151 | } |
48987781 | 2152 | |
4d8b81ab XG |
2153 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2154 | gfn_t *nr_pages) | |
2155 | { | |
2156 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 2157 | } |
48987781 | 2158 | |
4d8b81ab | 2159 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 2160 | gfn_t gfn) |
4d8b81ab XG |
2161 | { |
2162 | return gfn_to_hva_many(slot, gfn, NULL); | |
2163 | } | |
2164 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
2165 | ||
48987781 XG |
2166 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
2167 | { | |
49c7754c | 2168 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 2169 | } |
0d150298 | 2170 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 2171 | |
8e73485c PB |
2172 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
2173 | { | |
2174 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
2175 | } | |
2176 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
2177 | ||
86ab8cff | 2178 | /* |
970c0d4b WY |
2179 | * Return the hva of a @gfn and the R/W attribute if possible. |
2180 | * | |
2181 | * @slot: the kvm_memory_slot which contains @gfn | |
2182 | * @gfn: the gfn to be translated | |
2183 | * @writable: used to return the read/write attribute of the @slot if the hva | |
2184 | * is valid and @writable is not NULL | |
86ab8cff | 2185 | */ |
64d83126 CD |
2186 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
2187 | gfn_t gfn, bool *writable) | |
86ab8cff | 2188 | { |
a2ac07fe GN |
2189 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
2190 | ||
2191 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
2192 | *writable = !memslot_is_readonly(slot); |
2193 | ||
a2ac07fe | 2194 | return hva; |
86ab8cff XG |
2195 | } |
2196 | ||
64d83126 CD |
2197 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
2198 | { | |
2199 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2200 | ||
2201 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2202 | } | |
2203 | ||
8e73485c PB |
2204 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
2205 | { | |
2206 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2207 | ||
2208 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2209 | } | |
2210 | ||
fafc3dba HY |
2211 | static inline int check_user_page_hwpoison(unsigned long addr) |
2212 | { | |
0d731759 | 2213 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 2214 | |
0d731759 | 2215 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
2216 | return rc == -EHWPOISON; |
2217 | } | |
2218 | ||
2fc84311 | 2219 | /* |
b9b33da2 PB |
2220 | * The fast path to get the writable pfn which will be stored in @pfn, |
2221 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 2222 | * only part that runs if we can in atomic context. |
2fc84311 | 2223 | */ |
b9b33da2 PB |
2224 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
2225 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 2226 | { |
8d4e1288 | 2227 | struct page *page[1]; |
954bbbc2 | 2228 | |
12ce13fe XG |
2229 | /* |
2230 | * Fast pin a writable pfn only if it is a write fault request | |
2231 | * or the caller allows to map a writable pfn for a read fault | |
2232 | * request. | |
2233 | */ | |
2234 | if (!(write_fault || writable)) | |
2235 | return false; | |
612819c3 | 2236 | |
dadbb612 | 2237 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 2238 | *pfn = page_to_pfn(page[0]); |
612819c3 | 2239 | |
2fc84311 XG |
2240 | if (writable) |
2241 | *writable = true; | |
2242 | return true; | |
2243 | } | |
af585b92 | 2244 | |
2fc84311 XG |
2245 | return false; |
2246 | } | |
612819c3 | 2247 | |
2fc84311 XG |
2248 | /* |
2249 | * The slow path to get the pfn of the specified host virtual address, | |
2250 | * 1 indicates success, -errno is returned if error is detected. | |
2251 | */ | |
2252 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 2253 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 2254 | { |
ce53053c AV |
2255 | unsigned int flags = FOLL_HWPOISON; |
2256 | struct page *page; | |
2fc84311 | 2257 | int npages = 0; |
612819c3 | 2258 | |
2fc84311 XG |
2259 | might_sleep(); |
2260 | ||
2261 | if (writable) | |
2262 | *writable = write_fault; | |
2263 | ||
ce53053c AV |
2264 | if (write_fault) |
2265 | flags |= FOLL_WRITE; | |
2266 | if (async) | |
2267 | flags |= FOLL_NOWAIT; | |
d4944b0e | 2268 | |
ce53053c | 2269 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
2270 | if (npages != 1) |
2271 | return npages; | |
2272 | ||
2273 | /* map read fault as writable if possible */ | |
12ce13fe | 2274 | if (unlikely(!write_fault) && writable) { |
ce53053c | 2275 | struct page *wpage; |
2fc84311 | 2276 | |
dadbb612 | 2277 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 2278 | *writable = true; |
ce53053c AV |
2279 | put_page(page); |
2280 | page = wpage; | |
612819c3 | 2281 | } |
887c08ac | 2282 | } |
ce53053c | 2283 | *pfn = page_to_pfn(page); |
2fc84311 XG |
2284 | return npages; |
2285 | } | |
539cb660 | 2286 | |
4d8b81ab XG |
2287 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
2288 | { | |
2289 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
2290 | return false; | |
2e2e3738 | 2291 | |
4d8b81ab XG |
2292 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
2293 | return false; | |
887c08ac | 2294 | |
4d8b81ab XG |
2295 | return true; |
2296 | } | |
bf998156 | 2297 | |
f8be156b NP |
2298 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
2299 | { | |
2300 | if (kvm_is_reserved_pfn(pfn)) | |
2301 | return 1; | |
2302 | return get_page_unless_zero(pfn_to_page(pfn)); | |
2303 | } | |
2304 | ||
92176a8e PB |
2305 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
2306 | unsigned long addr, bool *async, | |
a340b3e2 KA |
2307 | bool write_fault, bool *writable, |
2308 | kvm_pfn_t *p_pfn) | |
92176a8e | 2309 | { |
a9545779 | 2310 | kvm_pfn_t pfn; |
bd2fae8d PB |
2311 | pte_t *ptep; |
2312 | spinlock_t *ptl; | |
add6a0cd PB |
2313 | int r; |
2314 | ||
9fd6dad1 | 2315 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2316 | if (r) { |
2317 | /* | |
2318 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
2319 | * not call the fault handler, so do it here. | |
2320 | */ | |
2321 | bool unlocked = false; | |
64019a2e | 2322 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
2323 | (write_fault ? FAULT_FLAG_WRITE : 0), |
2324 | &unlocked); | |
a8387d0b PB |
2325 | if (unlocked) |
2326 | return -EAGAIN; | |
add6a0cd PB |
2327 | if (r) |
2328 | return r; | |
2329 | ||
9fd6dad1 | 2330 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2331 | if (r) |
2332 | return r; | |
bd2fae8d | 2333 | } |
add6a0cd | 2334 | |
bd2fae8d PB |
2335 | if (write_fault && !pte_write(*ptep)) { |
2336 | pfn = KVM_PFN_ERR_RO_FAULT; | |
2337 | goto out; | |
add6a0cd PB |
2338 | } |
2339 | ||
a340b3e2 | 2340 | if (writable) |
bd2fae8d PB |
2341 | *writable = pte_write(*ptep); |
2342 | pfn = pte_pfn(*ptep); | |
add6a0cd PB |
2343 | |
2344 | /* | |
2345 | * Get a reference here because callers of *hva_to_pfn* and | |
2346 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
2347 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
36c3ce6c | 2348 | * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will |
add6a0cd PB |
2349 | * simply do nothing for reserved pfns. |
2350 | * | |
2351 | * Whoever called remap_pfn_range is also going to call e.g. | |
2352 | * unmap_mapping_range before the underlying pages are freed, | |
2353 | * causing a call to our MMU notifier. | |
f8be156b NP |
2354 | * |
2355 | * Certain IO or PFNMAP mappings can be backed with valid | |
2356 | * struct pages, but be allocated without refcounting e.g., | |
2357 | * tail pages of non-compound higher order allocations, which | |
2358 | * would then underflow the refcount when the caller does the | |
2359 | * required put_page. Don't allow those pages here. | |
add6a0cd | 2360 | */ |
f8be156b NP |
2361 | if (!kvm_try_get_pfn(pfn)) |
2362 | r = -EFAULT; | |
add6a0cd | 2363 | |
bd2fae8d PB |
2364 | out: |
2365 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 2366 | *p_pfn = pfn; |
f8be156b NP |
2367 | |
2368 | return r; | |
92176a8e PB |
2369 | } |
2370 | ||
12ce13fe XG |
2371 | /* |
2372 | * Pin guest page in memory and return its pfn. | |
2373 | * @addr: host virtual address which maps memory to the guest | |
2374 | * @atomic: whether this function can sleep | |
2375 | * @async: whether this function need to wait IO complete if the | |
2376 | * host page is not in the memory | |
2377 | * @write_fault: whether we should get a writable host page | |
2378 | * @writable: whether it allows to map a writable host page for !@write_fault | |
2379 | * | |
2380 | * The function will map a writable host page for these two cases: | |
2381 | * 1): @write_fault = true | |
2382 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
2383 | * whether the mapping is writable. | |
2384 | */ | |
ba049e93 | 2385 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
2386 | bool write_fault, bool *writable) |
2387 | { | |
2388 | struct vm_area_struct *vma; | |
ba049e93 | 2389 | kvm_pfn_t pfn = 0; |
92176a8e | 2390 | int npages, r; |
2e2e3738 | 2391 | |
2fc84311 XG |
2392 | /* we can do it either atomically or asynchronously, not both */ |
2393 | BUG_ON(atomic && async); | |
8d4e1288 | 2394 | |
b9b33da2 | 2395 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
2396 | return pfn; |
2397 | ||
2398 | if (atomic) | |
2399 | return KVM_PFN_ERR_FAULT; | |
2400 | ||
2401 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
2402 | if (npages == 1) | |
2403 | return pfn; | |
8d4e1288 | 2404 | |
d8ed45c5 | 2405 | mmap_read_lock(current->mm); |
2fc84311 XG |
2406 | if (npages == -EHWPOISON || |
2407 | (!async && check_user_page_hwpoison(addr))) { | |
2408 | pfn = KVM_PFN_ERR_HWPOISON; | |
2409 | goto exit; | |
2410 | } | |
2411 | ||
a8387d0b | 2412 | retry: |
fc98c03b | 2413 | vma = vma_lookup(current->mm, addr); |
2fc84311 XG |
2414 | |
2415 | if (vma == NULL) | |
2416 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 2417 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
a340b3e2 | 2418 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn); |
a8387d0b PB |
2419 | if (r == -EAGAIN) |
2420 | goto retry; | |
92176a8e PB |
2421 | if (r < 0) |
2422 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 2423 | } else { |
4d8b81ab | 2424 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
2425 | *async = true; |
2426 | pfn = KVM_PFN_ERR_FAULT; | |
2427 | } | |
2428 | exit: | |
d8ed45c5 | 2429 | mmap_read_unlock(current->mm); |
2e2e3738 | 2430 | return pfn; |
35149e21 AL |
2431 | } |
2432 | ||
ba049e93 DW |
2433 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
2434 | bool atomic, bool *async, bool write_fault, | |
4a42d848 | 2435 | bool *writable, hva_t *hva) |
887c08ac | 2436 | { |
4d8b81ab XG |
2437 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
2438 | ||
4a42d848 DS |
2439 | if (hva) |
2440 | *hva = addr; | |
2441 | ||
b2740d35 PB |
2442 | if (addr == KVM_HVA_ERR_RO_BAD) { |
2443 | if (writable) | |
2444 | *writable = false; | |
4d8b81ab | 2445 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 2446 | } |
4d8b81ab | 2447 | |
b2740d35 PB |
2448 | if (kvm_is_error_hva(addr)) { |
2449 | if (writable) | |
2450 | *writable = false; | |
81c52c56 | 2451 | return KVM_PFN_NOSLOT; |
b2740d35 | 2452 | } |
4d8b81ab XG |
2453 | |
2454 | /* Do not map writable pfn in the readonly memslot. */ | |
2455 | if (writable && memslot_is_readonly(slot)) { | |
2456 | *writable = false; | |
2457 | writable = NULL; | |
2458 | } | |
2459 | ||
2460 | return hva_to_pfn(addr, atomic, async, write_fault, | |
2461 | writable); | |
887c08ac | 2462 | } |
3520469d | 2463 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 2464 | |
ba049e93 | 2465 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
2466 | bool *writable) |
2467 | { | |
e37afc6e | 2468 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
4a42d848 | 2469 | write_fault, writable, NULL); |
612819c3 MT |
2470 | } |
2471 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
2472 | ||
ba049e93 | 2473 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2474 | { |
4a42d848 | 2475 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL); |
506f0d6f | 2476 | } |
e37afc6e | 2477 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 2478 | |
ba049e93 | 2479 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2480 | { |
4a42d848 | 2481 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL); |
506f0d6f | 2482 | } |
037d92dc | 2483 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 2484 | |
ba049e93 | 2485 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2486 | { |
2487 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2488 | } | |
2489 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
2490 | ||
ba049e93 | 2491 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
2492 | { |
2493 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
2494 | } | |
2495 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
2496 | ||
ba049e93 | 2497 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2498 | { |
2499 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2500 | } | |
2501 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
2502 | ||
d9ef13c2 PB |
2503 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2504 | struct page **pages, int nr_pages) | |
48987781 XG |
2505 | { |
2506 | unsigned long addr; | |
076b925d | 2507 | gfn_t entry = 0; |
48987781 | 2508 | |
d9ef13c2 | 2509 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
2510 | if (kvm_is_error_hva(addr)) |
2511 | return -1; | |
2512 | ||
2513 | if (entry < nr_pages) | |
2514 | return 0; | |
2515 | ||
dadbb612 | 2516 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
2517 | } |
2518 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
2519 | ||
ba049e93 | 2520 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 2521 | { |
81c52c56 | 2522 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 2523 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2524 | |
bf4bea8e | 2525 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 2526 | WARN_ON(1); |
6cede2e6 | 2527 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 2528 | } |
a2766325 XG |
2529 | |
2530 | return pfn_to_page(pfn); | |
2531 | } | |
2532 | ||
35149e21 AL |
2533 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
2534 | { | |
ba049e93 | 2535 | kvm_pfn_t pfn; |
2e2e3738 AL |
2536 | |
2537 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 2538 | |
a2766325 | 2539 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
2540 | } |
2541 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
2542 | ||
91724814 BO |
2543 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache) |
2544 | { | |
2545 | if (pfn == 0) | |
2546 | return; | |
2547 | ||
2548 | if (cache) | |
2549 | cache->pfn = cache->gfn = 0; | |
2550 | ||
2551 | if (dirty) | |
2552 | kvm_release_pfn_dirty(pfn); | |
2553 | else | |
2554 | kvm_release_pfn_clean(pfn); | |
2555 | } | |
2556 | ||
2557 | static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn, | |
2558 | struct gfn_to_pfn_cache *cache, u64 gen) | |
2559 | { | |
2560 | kvm_release_pfn(cache->pfn, cache->dirty, cache); | |
2561 | ||
2562 | cache->pfn = gfn_to_pfn_memslot(slot, gfn); | |
2563 | cache->gfn = gfn; | |
2564 | cache->dirty = false; | |
2565 | cache->generation = gen; | |
2566 | } | |
2567 | ||
1eff70a9 | 2568 | static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn, |
91724814 BO |
2569 | struct kvm_host_map *map, |
2570 | struct gfn_to_pfn_cache *cache, | |
2571 | bool atomic) | |
e45adf66 KA |
2572 | { |
2573 | kvm_pfn_t pfn; | |
2574 | void *hva = NULL; | |
2575 | struct page *page = KVM_UNMAPPED_PAGE; | |
1eff70a9 | 2576 | struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn); |
91724814 | 2577 | u64 gen = slots->generation; |
e45adf66 KA |
2578 | |
2579 | if (!map) | |
2580 | return -EINVAL; | |
2581 | ||
91724814 BO |
2582 | if (cache) { |
2583 | if (!cache->pfn || cache->gfn != gfn || | |
2584 | cache->generation != gen) { | |
2585 | if (atomic) | |
2586 | return -EAGAIN; | |
2587 | kvm_cache_gfn_to_pfn(slot, gfn, cache, gen); | |
2588 | } | |
2589 | pfn = cache->pfn; | |
2590 | } else { | |
2591 | if (atomic) | |
2592 | return -EAGAIN; | |
2593 | pfn = gfn_to_pfn_memslot(slot, gfn); | |
2594 | } | |
e45adf66 KA |
2595 | if (is_error_noslot_pfn(pfn)) |
2596 | return -EINVAL; | |
2597 | ||
2598 | if (pfn_valid(pfn)) { | |
2599 | page = pfn_to_page(pfn); | |
91724814 BO |
2600 | if (atomic) |
2601 | hva = kmap_atomic(page); | |
2602 | else | |
2603 | hva = kmap(page); | |
d30b214d | 2604 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2605 | } else if (!atomic) { |
e45adf66 | 2606 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
91724814 BO |
2607 | } else { |
2608 | return -EINVAL; | |
d30b214d | 2609 | #endif |
e45adf66 KA |
2610 | } |
2611 | ||
2612 | if (!hva) | |
2613 | return -EFAULT; | |
2614 | ||
2615 | map->page = page; | |
2616 | map->hva = hva; | |
2617 | map->pfn = pfn; | |
2618 | map->gfn = gfn; | |
2619 | ||
2620 | return 0; | |
2621 | } | |
2622 | ||
91724814 BO |
2623 | int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map, |
2624 | struct gfn_to_pfn_cache *cache, bool atomic) | |
1eff70a9 | 2625 | { |
91724814 BO |
2626 | return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map, |
2627 | cache, atomic); | |
1eff70a9 BO |
2628 | } |
2629 | EXPORT_SYMBOL_GPL(kvm_map_gfn); | |
2630 | ||
e45adf66 KA |
2631 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
2632 | { | |
91724814 BO |
2633 | return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map, |
2634 | NULL, false); | |
e45adf66 KA |
2635 | } |
2636 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); | |
2637 | ||
28bd726a PX |
2638 | static void __kvm_unmap_gfn(struct kvm *kvm, |
2639 | struct kvm_memory_slot *memslot, | |
91724814 BO |
2640 | struct kvm_host_map *map, |
2641 | struct gfn_to_pfn_cache *cache, | |
2642 | bool dirty, bool atomic) | |
e45adf66 KA |
2643 | { |
2644 | if (!map) | |
2645 | return; | |
2646 | ||
2647 | if (!map->hva) | |
2648 | return; | |
2649 | ||
91724814 BO |
2650 | if (map->page != KVM_UNMAPPED_PAGE) { |
2651 | if (atomic) | |
2652 | kunmap_atomic(map->hva); | |
2653 | else | |
2654 | kunmap(map->page); | |
2655 | } | |
eb1f2f38 | 2656 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2657 | else if (!atomic) |
e45adf66 | 2658 | memunmap(map->hva); |
91724814 BO |
2659 | else |
2660 | WARN_ONCE(1, "Unexpected unmapping in atomic context"); | |
eb1f2f38 | 2661 | #endif |
e45adf66 | 2662 | |
91724814 | 2663 | if (dirty) |
28bd726a | 2664 | mark_page_dirty_in_slot(kvm, memslot, map->gfn); |
91724814 BO |
2665 | |
2666 | if (cache) | |
2667 | cache->dirty |= dirty; | |
2668 | else | |
2669 | kvm_release_pfn(map->pfn, dirty, NULL); | |
e45adf66 KA |
2670 | |
2671 | map->hva = NULL; | |
2672 | map->page = NULL; | |
2673 | } | |
1eff70a9 | 2674 | |
91724814 BO |
2675 | int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map, |
2676 | struct gfn_to_pfn_cache *cache, bool dirty, bool atomic) | |
1eff70a9 | 2677 | { |
28bd726a | 2678 | __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map, |
91724814 | 2679 | cache, dirty, atomic); |
1eff70a9 BO |
2680 | return 0; |
2681 | } | |
2682 | EXPORT_SYMBOL_GPL(kvm_unmap_gfn); | |
2683 | ||
2684 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) | |
2685 | { | |
28bd726a PX |
2686 | __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), |
2687 | map, NULL, dirty, false); | |
1eff70a9 | 2688 | } |
e45adf66 KA |
2689 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); |
2690 | ||
8e73485c PB |
2691 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
2692 | { | |
ba049e93 | 2693 | kvm_pfn_t pfn; |
8e73485c PB |
2694 | |
2695 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
2696 | ||
2697 | return kvm_pfn_to_page(pfn); | |
2698 | } | |
2699 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
2700 | ||
b4231d61 IE |
2701 | void kvm_release_page_clean(struct page *page) |
2702 | { | |
32cad84f XG |
2703 | WARN_ON(is_error_page(page)); |
2704 | ||
35149e21 | 2705 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
2706 | } |
2707 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
2708 | ||
ba049e93 | 2709 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 2710 | { |
bf4bea8e | 2711 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 2712 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
2713 | } |
2714 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
2715 | ||
b4231d61 | 2716 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 2717 | { |
a2766325 XG |
2718 | WARN_ON(is_error_page(page)); |
2719 | ||
35149e21 AL |
2720 | kvm_release_pfn_dirty(page_to_pfn(page)); |
2721 | } | |
2722 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
2723 | ||
f7a6509f | 2724 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
2725 | { |
2726 | kvm_set_pfn_dirty(pfn); | |
2727 | kvm_release_pfn_clean(pfn); | |
2728 | } | |
f7a6509f | 2729 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 2730 | |
ba049e93 | 2731 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2732 | { |
d29c03a5 ML |
2733 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2734 | SetPageDirty(pfn_to_page(pfn)); | |
8a7ae055 | 2735 | } |
35149e21 AL |
2736 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
2737 | ||
ba049e93 | 2738 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 2739 | { |
a78986aa | 2740 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2e2e3738 | 2741 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
2742 | } |
2743 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
2744 | ||
195aefde IE |
2745 | static int next_segment(unsigned long len, int offset) |
2746 | { | |
2747 | if (len > PAGE_SIZE - offset) | |
2748 | return PAGE_SIZE - offset; | |
2749 | else | |
2750 | return len; | |
2751 | } | |
2752 | ||
8e73485c PB |
2753 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
2754 | void *data, int offset, int len) | |
195aefde | 2755 | { |
e0506bcb IE |
2756 | int r; |
2757 | unsigned long addr; | |
195aefde | 2758 | |
8e73485c | 2759 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
2760 | if (kvm_is_error_hva(addr)) |
2761 | return -EFAULT; | |
3180a7fc | 2762 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 2763 | if (r) |
195aefde | 2764 | return -EFAULT; |
195aefde IE |
2765 | return 0; |
2766 | } | |
8e73485c PB |
2767 | |
2768 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
2769 | int len) | |
2770 | { | |
2771 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2772 | ||
2773 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2774 | } | |
195aefde IE |
2775 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
2776 | ||
8e73485c PB |
2777 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
2778 | int offset, int len) | |
2779 | { | |
2780 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2781 | ||
2782 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2783 | } | |
2784 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
2785 | ||
195aefde IE |
2786 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
2787 | { | |
2788 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2789 | int seg; | |
2790 | int offset = offset_in_page(gpa); | |
2791 | int ret; | |
2792 | ||
2793 | while ((seg = next_segment(len, offset)) != 0) { | |
2794 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
2795 | if (ret < 0) | |
2796 | return ret; | |
2797 | offset = 0; | |
2798 | len -= seg; | |
2799 | data += seg; | |
2800 | ++gfn; | |
2801 | } | |
2802 | return 0; | |
2803 | } | |
2804 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
2805 | ||
8e73485c | 2806 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 2807 | { |
7ec54588 | 2808 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 2809 | int seg; |
7ec54588 | 2810 | int offset = offset_in_page(gpa); |
8e73485c PB |
2811 | int ret; |
2812 | ||
2813 | while ((seg = next_segment(len, offset)) != 0) { | |
2814 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
2815 | if (ret < 0) | |
2816 | return ret; | |
2817 | offset = 0; | |
2818 | len -= seg; | |
2819 | data += seg; | |
2820 | ++gfn; | |
2821 | } | |
2822 | return 0; | |
2823 | } | |
2824 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 2825 | |
8e73485c PB |
2826 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2827 | void *data, int offset, unsigned long len) | |
2828 | { | |
2829 | int r; | |
2830 | unsigned long addr; | |
2831 | ||
2832 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
2833 | if (kvm_is_error_hva(addr)) |
2834 | return -EFAULT; | |
0aac03f0 | 2835 | pagefault_disable(); |
3180a7fc | 2836 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 2837 | pagefault_enable(); |
7ec54588 MT |
2838 | if (r) |
2839 | return -EFAULT; | |
2840 | return 0; | |
2841 | } | |
7ec54588 | 2842 | |
8e73485c PB |
2843 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
2844 | void *data, unsigned long len) | |
2845 | { | |
2846 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2847 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2848 | int offset = offset_in_page(gpa); | |
2849 | ||
2850 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2851 | } | |
2852 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
2853 | ||
28bd726a PX |
2854 | static int __kvm_write_guest_page(struct kvm *kvm, |
2855 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 2856 | const void *data, int offset, int len) |
195aefde | 2857 | { |
e0506bcb IE |
2858 | int r; |
2859 | unsigned long addr; | |
195aefde | 2860 | |
251eb841 | 2861 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
2862 | if (kvm_is_error_hva(addr)) |
2863 | return -EFAULT; | |
8b0cedff | 2864 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 2865 | if (r) |
195aefde | 2866 | return -EFAULT; |
28bd726a | 2867 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
2868 | return 0; |
2869 | } | |
8e73485c PB |
2870 | |
2871 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
2872 | const void *data, int offset, int len) | |
2873 | { | |
2874 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2875 | ||
28bd726a | 2876 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 2877 | } |
195aefde IE |
2878 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
2879 | ||
8e73485c PB |
2880 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
2881 | const void *data, int offset, int len) | |
2882 | { | |
2883 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2884 | ||
28bd726a | 2885 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
2886 | } |
2887 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
2888 | ||
195aefde IE |
2889 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
2890 | unsigned long len) | |
2891 | { | |
2892 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2893 | int seg; | |
2894 | int offset = offset_in_page(gpa); | |
2895 | int ret; | |
2896 | ||
2897 | while ((seg = next_segment(len, offset)) != 0) { | |
2898 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
2899 | if (ret < 0) | |
2900 | return ret; | |
2901 | offset = 0; | |
2902 | len -= seg; | |
2903 | data += seg; | |
2904 | ++gfn; | |
2905 | } | |
2906 | return 0; | |
2907 | } | |
ff651cb6 | 2908 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 2909 | |
8e73485c PB |
2910 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
2911 | unsigned long len) | |
2912 | { | |
2913 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2914 | int seg; | |
2915 | int offset = offset_in_page(gpa); | |
2916 | int ret; | |
2917 | ||
2918 | while ((seg = next_segment(len, offset)) != 0) { | |
2919 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
2920 | if (ret < 0) | |
2921 | return ret; | |
2922 | offset = 0; | |
2923 | len -= seg; | |
2924 | data += seg; | |
2925 | ++gfn; | |
2926 | } | |
2927 | return 0; | |
2928 | } | |
2929 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
2930 | ||
5a2d4365 PB |
2931 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
2932 | struct gfn_to_hva_cache *ghc, | |
2933 | gpa_t gpa, unsigned long len) | |
49c7754c | 2934 | { |
49c7754c | 2935 | int offset = offset_in_page(gpa); |
8f964525 AH |
2936 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
2937 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
2938 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
2939 | gfn_t nr_pages_avail; | |
49c7754c | 2940 | |
6ad1e29f | 2941 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 2942 | ghc->generation = slots->generation; |
6ad1e29f SC |
2943 | |
2944 | if (start_gfn > end_gfn) { | |
2945 | ghc->hva = KVM_HVA_ERR_BAD; | |
2946 | return -EINVAL; | |
2947 | } | |
f1b9dd5e JM |
2948 | |
2949 | /* | |
2950 | * If the requested region crosses two memslots, we still | |
2951 | * verify that the entire region is valid here. | |
2952 | */ | |
6ad1e29f | 2953 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
2954 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
2955 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
2956 | &nr_pages_avail); | |
2957 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 2958 | return -EFAULT; |
f1b9dd5e JM |
2959 | } |
2960 | ||
2961 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 2962 | if (nr_pages_needed == 1) |
49c7754c | 2963 | ghc->hva += offset; |
f1b9dd5e | 2964 | else |
8f964525 | 2965 | ghc->memslot = NULL; |
f1b9dd5e | 2966 | |
6ad1e29f SC |
2967 | ghc->gpa = gpa; |
2968 | ghc->len = len; | |
2969 | return 0; | |
49c7754c | 2970 | } |
5a2d4365 | 2971 | |
4e335d9e | 2972 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
2973 | gpa_t gpa, unsigned long len) |
2974 | { | |
4e335d9e | 2975 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
2976 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
2977 | } | |
4e335d9e | 2978 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 2979 | |
4e335d9e | 2980 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
2981 | void *data, unsigned int offset, |
2982 | unsigned long len) | |
49c7754c | 2983 | { |
4e335d9e | 2984 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 2985 | int r; |
4ec6e863 | 2986 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 2987 | |
4ec6e863 | 2988 | BUG_ON(len + offset > ghc->len); |
8f964525 | 2989 | |
dc9ce71e SC |
2990 | if (slots->generation != ghc->generation) { |
2991 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
2992 | return -EFAULT; | |
2993 | } | |
8f964525 | 2994 | |
49c7754c GN |
2995 | if (kvm_is_error_hva(ghc->hva)) |
2996 | return -EFAULT; | |
2997 | ||
fcfbc617 SC |
2998 | if (unlikely(!ghc->memslot)) |
2999 | return kvm_write_guest(kvm, gpa, data, len); | |
3000 | ||
4ec6e863 | 3001 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
3002 | if (r) |
3003 | return -EFAULT; | |
28bd726a | 3004 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
3005 | |
3006 | return 0; | |
3007 | } | |
4e335d9e | 3008 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 3009 | |
4e335d9e PB |
3010 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3011 | void *data, unsigned long len) | |
4ec6e863 | 3012 | { |
4e335d9e | 3013 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 3014 | } |
4e335d9e | 3015 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 3016 | |
0958f0ce VK |
3017 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3018 | void *data, unsigned int offset, | |
3019 | unsigned long len) | |
e03b644f | 3020 | { |
4e335d9e | 3021 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 3022 | int r; |
0958f0ce | 3023 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 3024 | |
0958f0ce | 3025 | BUG_ON(len + offset > ghc->len); |
8f964525 | 3026 | |
dc9ce71e SC |
3027 | if (slots->generation != ghc->generation) { |
3028 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3029 | return -EFAULT; | |
3030 | } | |
8f964525 | 3031 | |
e03b644f GN |
3032 | if (kvm_is_error_hva(ghc->hva)) |
3033 | return -EFAULT; | |
3034 | ||
fcfbc617 | 3035 | if (unlikely(!ghc->memslot)) |
0958f0ce | 3036 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 3037 | |
0958f0ce | 3038 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
3039 | if (r) |
3040 | return -EFAULT; | |
3041 | ||
3042 | return 0; | |
3043 | } | |
0958f0ce VK |
3044 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
3045 | ||
3046 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
3047 | void *data, unsigned long len) | |
3048 | { | |
3049 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
3050 | } | |
4e335d9e | 3051 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 3052 | |
195aefde IE |
3053 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
3054 | { | |
2f541442 | 3055 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
3056 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3057 | int seg; | |
3058 | int offset = offset_in_page(gpa); | |
3059 | int ret; | |
3060 | ||
bfda0e84 | 3061 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 3062 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
3063 | if (ret < 0) |
3064 | return ret; | |
3065 | offset = 0; | |
3066 | len -= seg; | |
3067 | ++gfn; | |
3068 | } | |
3069 | return 0; | |
3070 | } | |
3071 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
3072 | ||
28bd726a PX |
3073 | void mark_page_dirty_in_slot(struct kvm *kvm, |
3074 | struct kvm_memory_slot *memslot, | |
3075 | gfn_t gfn) | |
6aa8b732 | 3076 | { |
044c59c4 | 3077 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 3078 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 3079 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 3080 | |
fb04a1ed PX |
3081 | if (kvm->dirty_ring_size) |
3082 | kvm_dirty_ring_push(kvm_dirty_ring_get(kvm), | |
3083 | slot, rel_gfn); | |
3084 | else | |
3085 | set_bit_le(rel_gfn, memslot->dirty_bitmap); | |
6aa8b732 AK |
3086 | } |
3087 | } | |
a6a0b05d | 3088 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 3089 | |
49c7754c GN |
3090 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
3091 | { | |
3092 | struct kvm_memory_slot *memslot; | |
3093 | ||
3094 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 3095 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 3096 | } |
2ba9f0d8 | 3097 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 3098 | |
8e73485c PB |
3099 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
3100 | { | |
3101 | struct kvm_memory_slot *memslot; | |
3102 | ||
3103 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 3104 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
3105 | } |
3106 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
3107 | ||
20b7035c JS |
3108 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
3109 | { | |
3110 | if (!vcpu->sigset_active) | |
3111 | return; | |
3112 | ||
3113 | /* | |
3114 | * This does a lockless modification of ->real_blocked, which is fine | |
3115 | * because, only current can change ->real_blocked and all readers of | |
3116 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
3117 | * of ->blocked. | |
3118 | */ | |
3119 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
3120 | } | |
3121 | ||
3122 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
3123 | { | |
3124 | if (!vcpu->sigset_active) | |
3125 | return; | |
3126 | ||
3127 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
3128 | sigemptyset(¤t->real_blocked); | |
3129 | } | |
3130 | ||
aca6ff29 WL |
3131 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
3132 | { | |
dee339b5 | 3133 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 3134 | |
2cbd7824 | 3135 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 3136 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3137 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
3138 | if (!grow) |
3139 | goto out; | |
3140 | ||
dee339b5 NW |
3141 | val *= grow; |
3142 | if (val < grow_start) | |
3143 | val = grow_start; | |
aca6ff29 | 3144 | |
258785ef DM |
3145 | if (val > vcpu->kvm->max_halt_poll_ns) |
3146 | val = vcpu->kvm->max_halt_poll_ns; | |
313f636d | 3147 | |
aca6ff29 | 3148 | vcpu->halt_poll_ns = val; |
7fa08e71 | 3149 | out: |
2cbd7824 | 3150 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3151 | } |
3152 | ||
3153 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
3154 | { | |
ae232ea4 | 3155 | unsigned int old, val, shrink, grow_start; |
aca6ff29 | 3156 | |
2cbd7824 | 3157 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 3158 | shrink = READ_ONCE(halt_poll_ns_shrink); |
ae232ea4 | 3159 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3160 | if (shrink == 0) |
aca6ff29 WL |
3161 | val = 0; |
3162 | else | |
6b6de68c | 3163 | val /= shrink; |
aca6ff29 | 3164 | |
ae232ea4 SS |
3165 | if (val < grow_start) |
3166 | val = 0; | |
3167 | ||
aca6ff29 | 3168 | vcpu->halt_poll_ns = val; |
2cbd7824 | 3169 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3170 | } |
3171 | ||
f7819512 PB |
3172 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
3173 | { | |
50c28f21 JS |
3174 | int ret = -EINTR; |
3175 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
3176 | ||
f7819512 PB |
3177 | if (kvm_arch_vcpu_runnable(vcpu)) { |
3178 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 3179 | goto out; |
f7819512 PB |
3180 | } |
3181 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 3182 | goto out; |
f7819512 | 3183 | if (signal_pending(current)) |
50c28f21 | 3184 | goto out; |
084071d5 MT |
3185 | if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu)) |
3186 | goto out; | |
f7819512 | 3187 | |
50c28f21 JS |
3188 | ret = 0; |
3189 | out: | |
3190 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
3191 | return ret; | |
f7819512 PB |
3192 | } |
3193 | ||
cb953129 DM |
3194 | static inline void |
3195 | update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited) | |
3196 | { | |
3197 | if (waited) | |
0193cc90 | 3198 | vcpu->stat.generic.halt_poll_fail_ns += poll_ns; |
cb953129 | 3199 | else |
0193cc90 | 3200 | vcpu->stat.generic.halt_poll_success_ns += poll_ns; |
cb953129 DM |
3201 | } |
3202 | ||
b6958ce4 ED |
3203 | /* |
3204 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
3205 | */ | |
8776e519 | 3206 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 3207 | { |
cb953129 | 3208 | ktime_t start, cur, poll_end; |
f7819512 | 3209 | bool waited = false; |
aca6ff29 | 3210 | u64 block_ns; |
f7819512 | 3211 | |
07ab0f8d MZ |
3212 | kvm_arch_vcpu_blocking(vcpu); |
3213 | ||
cb953129 | 3214 | start = cur = poll_end = ktime_get(); |
cdd6ad3a | 3215 | if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) { |
19020f8a | 3216 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); |
f95ef0cd | 3217 | |
0193cc90 | 3218 | ++vcpu->stat.generic.halt_attempted_poll; |
f7819512 PB |
3219 | do { |
3220 | /* | |
3221 | * This sets KVM_REQ_UNHALT if an interrupt | |
3222 | * arrives. | |
3223 | */ | |
3224 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
0193cc90 | 3225 | ++vcpu->stat.generic.halt_successful_poll; |
3491caf2 | 3226 | if (!vcpu_valid_wakeup(vcpu)) |
0193cc90 | 3227 | ++vcpu->stat.generic.halt_poll_invalid; |
8ccba534 JZ |
3228 | |
3229 | KVM_STATS_LOG_HIST_UPDATE( | |
3230 | vcpu->stat.generic.halt_poll_success_hist, | |
3231 | ktime_to_ns(ktime_get()) - | |
3232 | ktime_to_ns(start)); | |
f7819512 PB |
3233 | goto out; |
3234 | } | |
74775654 | 3235 | cpu_relax(); |
cb953129 | 3236 | poll_end = cur = ktime_get(); |
6bd5b743 | 3237 | } while (kvm_vcpu_can_poll(cur, stop)); |
8ccba534 JZ |
3238 | |
3239 | KVM_STATS_LOG_HIST_UPDATE( | |
3240 | vcpu->stat.generic.halt_poll_fail_hist, | |
3241 | ktime_to_ns(ktime_get()) - ktime_to_ns(start)); | |
f7819512 | 3242 | } |
e5c239cf | 3243 | |
8ccba534 | 3244 | |
da4ad88c | 3245 | prepare_to_rcuwait(&vcpu->wait); |
e5c239cf | 3246 | for (;;) { |
da4ad88c | 3247 | set_current_state(TASK_INTERRUPTIBLE); |
e5c239cf | 3248 | |
f7819512 | 3249 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
3250 | break; |
3251 | ||
f7819512 | 3252 | waited = true; |
b6958ce4 | 3253 | schedule(); |
b6958ce4 | 3254 | } |
da4ad88c | 3255 | finish_rcuwait(&vcpu->wait); |
f7819512 | 3256 | cur = ktime_get(); |
87bcc5fa JZ |
3257 | if (waited) { |
3258 | vcpu->stat.generic.halt_wait_ns += | |
3259 | ktime_to_ns(cur) - ktime_to_ns(poll_end); | |
8ccba534 JZ |
3260 | KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist, |
3261 | ktime_to_ns(cur) - ktime_to_ns(poll_end)); | |
87bcc5fa | 3262 | } |
f7819512 | 3263 | out: |
07ab0f8d | 3264 | kvm_arch_vcpu_unblocking(vcpu); |
aca6ff29 WL |
3265 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
3266 | ||
cb953129 DM |
3267 | update_halt_poll_stats( |
3268 | vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited); | |
3269 | ||
44551b2f WL |
3270 | if (!kvm_arch_no_poll(vcpu)) { |
3271 | if (!vcpu_valid_wakeup(vcpu)) { | |
aca6ff29 | 3272 | shrink_halt_poll_ns(vcpu); |
acd05785 | 3273 | } else if (vcpu->kvm->max_halt_poll_ns) { |
44551b2f WL |
3274 | if (block_ns <= vcpu->halt_poll_ns) |
3275 | ; | |
3276 | /* we had a long block, shrink polling */ | |
acd05785 DM |
3277 | else if (vcpu->halt_poll_ns && |
3278 | block_ns > vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
3279 | shrink_halt_poll_ns(vcpu); |
3280 | /* we had a short halt and our poll time is too small */ | |
acd05785 DM |
3281 | else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && |
3282 | block_ns < vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
3283 | grow_halt_poll_ns(vcpu); |
3284 | } else { | |
3285 | vcpu->halt_poll_ns = 0; | |
3286 | } | |
3287 | } | |
aca6ff29 | 3288 | |
3491caf2 CB |
3289 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
3290 | kvm_arch_vcpu_block_finish(vcpu); | |
b6958ce4 | 3291 | } |
2ba9f0d8 | 3292 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 3293 | |
178f02ff | 3294 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 3295 | { |
da4ad88c | 3296 | struct rcuwait *waitp; |
b6d33834 | 3297 | |
da4ad88c DB |
3298 | waitp = kvm_arch_vcpu_get_wait(vcpu); |
3299 | if (rcuwait_wake_up(waitp)) { | |
d73eb57b | 3300 | WRITE_ONCE(vcpu->ready, true); |
0193cc90 | 3301 | ++vcpu->stat.generic.halt_wakeup; |
178f02ff | 3302 | return true; |
b6d33834 CD |
3303 | } |
3304 | ||
178f02ff | 3305 | return false; |
dd1a4cc1 RK |
3306 | } |
3307 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
3308 | ||
0266c894 | 3309 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
3310 | /* |
3311 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
3312 | */ | |
3313 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
3314 | { | |
85b64045 | 3315 | int me, cpu; |
dd1a4cc1 | 3316 | |
178f02ff RK |
3317 | if (kvm_vcpu_wake_up(vcpu)) |
3318 | return; | |
3319 | ||
85b64045 SC |
3320 | /* |
3321 | * Note, the vCPU could get migrated to a different pCPU at any point | |
3322 | * after kvm_arch_vcpu_should_kick(), which could result in sending an | |
3323 | * IPI to the previous pCPU. But, that's ok because the purpose of the | |
3324 | * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the | |
3325 | * vCPU also requires it to leave IN_GUEST_MODE. | |
3326 | */ | |
b6d33834 | 3327 | me = get_cpu(); |
85b64045 SC |
3328 | if (kvm_arch_vcpu_should_kick(vcpu)) { |
3329 | cpu = READ_ONCE(vcpu->cpu); | |
3330 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
b6d33834 | 3331 | smp_send_reschedule(cpu); |
85b64045 | 3332 | } |
b6d33834 CD |
3333 | put_cpu(); |
3334 | } | |
a20ed54d | 3335 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 3336 | #endif /* !CONFIG_S390 */ |
b6d33834 | 3337 | |
fa93384f | 3338 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
3339 | { |
3340 | struct pid *pid; | |
3341 | struct task_struct *task = NULL; | |
fa93384f | 3342 | int ret = 0; |
41628d33 KW |
3343 | |
3344 | rcu_read_lock(); | |
3345 | pid = rcu_dereference(target->pid); | |
3346 | if (pid) | |
27fbe64b | 3347 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
3348 | rcu_read_unlock(); |
3349 | if (!task) | |
c45c528e | 3350 | return ret; |
c45c528e | 3351 | ret = yield_to(task, 1); |
41628d33 | 3352 | put_task_struct(task); |
c45c528e R |
3353 | |
3354 | return ret; | |
41628d33 KW |
3355 | } |
3356 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
3357 | ||
06e48c51 R |
3358 | /* |
3359 | * Helper that checks whether a VCPU is eligible for directed yield. | |
3360 | * Most eligible candidate to yield is decided by following heuristics: | |
3361 | * | |
3362 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
3363 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 3364 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
3365 | * |
3366 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
3367 | * chance last time (mostly it has become eligible now since we have probably | |
3368 | * yielded to lockholder in last iteration. This is done by toggling | |
3369 | * @dy_eligible each time a VCPU checked for eligibility.) | |
3370 | * | |
3371 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
3372 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
3373 | * burning. Giving priority for a potential lock-holder increases lock | |
3374 | * progress. | |
3375 | * | |
3376 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
3377 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
3378 | * and continue with next VCPU and so on. | |
3379 | */ | |
7940876e | 3380 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 3381 | { |
4a55dd72 | 3382 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
3383 | bool eligible; |
3384 | ||
3385 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 3386 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
3387 | |
3388 | if (vcpu->spin_loop.in_spin_loop) | |
3389 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
3390 | ||
3391 | return eligible; | |
4a55dd72 SW |
3392 | #else |
3393 | return true; | |
06e48c51 | 3394 | #endif |
4a55dd72 | 3395 | } |
c45c528e | 3396 | |
17e433b5 WL |
3397 | /* |
3398 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
3399 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
3400 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
3401 | */ | |
3402 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
3403 | { | |
3404 | return kvm_arch_vcpu_runnable(vcpu); | |
3405 | } | |
3406 | ||
3407 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
3408 | { | |
3409 | if (kvm_arch_dy_runnable(vcpu)) | |
3410 | return true; | |
3411 | ||
3412 | #ifdef CONFIG_KVM_ASYNC_PF | |
3413 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
3414 | return true; | |
3415 | #endif | |
3416 | ||
3417 | return false; | |
3418 | } | |
3419 | ||
52acd22f WL |
3420 | bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
3421 | { | |
3422 | return false; | |
3423 | } | |
3424 | ||
199b5763 | 3425 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 3426 | { |
217ece61 RR |
3427 | struct kvm *kvm = me->kvm; |
3428 | struct kvm_vcpu *vcpu; | |
3429 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
3430 | int yielded = 0; | |
c45c528e | 3431 | int try = 3; |
217ece61 RR |
3432 | int pass; |
3433 | int i; | |
d255f4f2 | 3434 | |
4c088493 | 3435 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
3436 | /* |
3437 | * We boost the priority of a VCPU that is runnable but not | |
3438 | * currently running, because it got preempted by something | |
3439 | * else and called schedule in __vcpu_run. Hopefully that | |
3440 | * VCPU is holding the lock that we need and will release it. | |
3441 | * We approximate round-robin by starting at the last boosted VCPU. | |
3442 | */ | |
c45c528e | 3443 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 3444 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 3445 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
3446 | i = last_boosted_vcpu; |
3447 | continue; | |
3448 | } else if (pass && i > last_boosted_vcpu) | |
3449 | break; | |
d73eb57b | 3450 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 3451 | continue; |
217ece61 RR |
3452 | if (vcpu == me) |
3453 | continue; | |
da4ad88c DB |
3454 | if (rcuwait_active(&vcpu->wait) && |
3455 | !vcpu_dy_runnable(vcpu)) | |
217ece61 | 3456 | continue; |
046ddeed | 3457 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
52acd22f WL |
3458 | !kvm_arch_dy_has_pending_interrupt(vcpu) && |
3459 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 3460 | continue; |
06e48c51 R |
3461 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
3462 | continue; | |
c45c528e R |
3463 | |
3464 | yielded = kvm_vcpu_yield_to(vcpu); | |
3465 | if (yielded > 0) { | |
217ece61 | 3466 | kvm->last_boosted_vcpu = i; |
217ece61 | 3467 | break; |
c45c528e R |
3468 | } else if (yielded < 0) { |
3469 | try--; | |
3470 | if (!try) | |
3471 | break; | |
217ece61 | 3472 | } |
217ece61 RR |
3473 | } |
3474 | } | |
4c088493 | 3475 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
3476 | |
3477 | /* Ensure vcpu is not eligible during next spinloop */ | |
3478 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
3479 | } |
3480 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
3481 | ||
fb04a1ed PX |
3482 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
3483 | { | |
3484 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
3485 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && | |
3486 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
3487 | kvm->dirty_ring_size / PAGE_SIZE); | |
3488 | #else | |
3489 | return false; | |
3490 | #endif | |
3491 | } | |
3492 | ||
1499fa80 | 3493 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 3494 | { |
11bac800 | 3495 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
3496 | struct page *page; |
3497 | ||
e4a533a4 | 3498 | if (vmf->pgoff == 0) |
039576c0 | 3499 | page = virt_to_page(vcpu->run); |
09566765 | 3500 | #ifdef CONFIG_X86 |
e4a533a4 | 3501 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 3502 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 3503 | #endif |
4b4357e0 | 3504 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3505 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
3506 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 3507 | #endif |
fb04a1ed PX |
3508 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
3509 | page = kvm_dirty_ring_get_page( | |
3510 | &vcpu->dirty_ring, | |
3511 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 3512 | else |
5b1c1493 | 3513 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 3514 | get_page(page); |
e4a533a4 NP |
3515 | vmf->page = page; |
3516 | return 0; | |
9a2bb7f4 AK |
3517 | } |
3518 | ||
f0f37e2f | 3519 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 3520 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
3521 | }; |
3522 | ||
3523 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
3524 | { | |
fb04a1ed | 3525 | struct kvm_vcpu *vcpu = file->private_data; |
11476d27 | 3526 | unsigned long pages = vma_pages(vma); |
fb04a1ed PX |
3527 | |
3528 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
3529 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
3530 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
3531 | return -EINVAL; | |
3532 | ||
9a2bb7f4 AK |
3533 | vma->vm_ops = &kvm_vcpu_vm_ops; |
3534 | return 0; | |
3535 | } | |
3536 | ||
bccf2150 AK |
3537 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
3538 | { | |
3539 | struct kvm_vcpu *vcpu = filp->private_data; | |
3540 | ||
66c0b394 | 3541 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
3542 | return 0; |
3543 | } | |
3544 | ||
3d3aab1b | 3545 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
3546 | .release = kvm_vcpu_release, |
3547 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 3548 | .mmap = kvm_vcpu_mmap, |
6038f373 | 3549 | .llseek = noop_llseek, |
7ddfd3e0 | 3550 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
3551 | }; |
3552 | ||
3553 | /* | |
3554 | * Allocates an inode for the vcpu. | |
3555 | */ | |
3556 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
3557 | { | |
e46b4692 MY |
3558 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
3559 | ||
3560 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
3561 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
3562 | } |
3563 | ||
3e7093d0 | 3564 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 3565 | { |
741cbbae | 3566 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
d56f5136 | 3567 | struct dentry *debugfs_dentry; |
45b5939e | 3568 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 3569 | |
45b5939e | 3570 | if (!debugfs_initialized()) |
3e7093d0 | 3571 | return; |
45b5939e LC |
3572 | |
3573 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
3574 | debugfs_dentry = debugfs_create_dir(dir_name, |
3575 | vcpu->kvm->debugfs_dentry); | |
45b5939e | 3576 | |
d56f5136 | 3577 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
741cbbae | 3578 | #endif |
45b5939e LC |
3579 | } |
3580 | ||
c5ea7660 AK |
3581 | /* |
3582 | * Creates some virtual cpus. Good luck creating more than one. | |
3583 | */ | |
73880c80 | 3584 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
3585 | { |
3586 | int r; | |
e09fefde | 3587 | struct kvm_vcpu *vcpu; |
8bd826d6 | 3588 | struct page *page; |
c5ea7660 | 3589 | |
0b1b1dfd | 3590 | if (id >= KVM_MAX_VCPU_ID) |
338c7dba AH |
3591 | return -EINVAL; |
3592 | ||
6c7caebc PB |
3593 | mutex_lock(&kvm->lock); |
3594 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
3595 | mutex_unlock(&kvm->lock); | |
3596 | return -EINVAL; | |
3597 | } | |
3598 | ||
3599 | kvm->created_vcpus++; | |
3600 | mutex_unlock(&kvm->lock); | |
3601 | ||
897cc38e SC |
3602 | r = kvm_arch_vcpu_precreate(kvm, id); |
3603 | if (r) | |
3604 | goto vcpu_decrement; | |
3605 | ||
85f47930 | 3606 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
e529ef66 SC |
3607 | if (!vcpu) { |
3608 | r = -ENOMEM; | |
6c7caebc PB |
3609 | goto vcpu_decrement; |
3610 | } | |
c5ea7660 | 3611 | |
fcd97ad5 | 3612 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 3613 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
3614 | if (!page) { |
3615 | r = -ENOMEM; | |
e529ef66 | 3616 | goto vcpu_free; |
8bd826d6 SC |
3617 | } |
3618 | vcpu->run = page_address(page); | |
3619 | ||
3620 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
3621 | |
3622 | r = kvm_arch_vcpu_create(vcpu); | |
3623 | if (r) | |
8bd826d6 | 3624 | goto vcpu_free_run_page; |
e529ef66 | 3625 | |
fb04a1ed PX |
3626 | if (kvm->dirty_ring_size) { |
3627 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
3628 | id, kvm->dirty_ring_size); | |
3629 | if (r) | |
3630 | goto arch_vcpu_destroy; | |
3631 | } | |
3632 | ||
11ec2804 | 3633 | mutex_lock(&kvm->lock); |
e09fefde DH |
3634 | if (kvm_get_vcpu_by_id(kvm, id)) { |
3635 | r = -EEXIST; | |
3636 | goto unlock_vcpu_destroy; | |
3637 | } | |
73880c80 | 3638 | |
8750e72a RK |
3639 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
3640 | BUG_ON(kvm->vcpus[vcpu->vcpu_idx]); | |
c5ea7660 | 3641 | |
ce55c049 JZ |
3642 | /* Fill the stats id string for the vcpu */ |
3643 | snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d", | |
3644 | task_pid_nr(current), id); | |
3645 | ||
fb3f0f51 | 3646 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 3647 | kvm_get_kvm(kvm); |
bccf2150 | 3648 | r = create_vcpu_fd(vcpu); |
73880c80 | 3649 | if (r < 0) { |
149487bd | 3650 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 3651 | goto unlock_vcpu_destroy; |
73880c80 GN |
3652 | } |
3653 | ||
8750e72a | 3654 | kvm->vcpus[vcpu->vcpu_idx] = vcpu; |
dd489240 PB |
3655 | |
3656 | /* | |
3657 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
3658 | * before kvm->online_vcpu's incremented value. | |
3659 | */ | |
73880c80 GN |
3660 | smp_wmb(); |
3661 | atomic_inc(&kvm->online_vcpus); | |
3662 | ||
73880c80 | 3663 | mutex_unlock(&kvm->lock); |
42897d86 | 3664 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 3665 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 3666 | return r; |
39c3b86e | 3667 | |
d780592b | 3668 | unlock_vcpu_destroy: |
7d8fece6 | 3669 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
3670 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
3671 | arch_vcpu_destroy: | |
d40ccc62 | 3672 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
3673 | vcpu_free_run_page: |
3674 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
3675 | vcpu_free: |
3676 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
3677 | vcpu_decrement: |
3678 | mutex_lock(&kvm->lock); | |
3679 | kvm->created_vcpus--; | |
3680 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
3681 | return r; |
3682 | } | |
3683 | ||
1961d276 AK |
3684 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
3685 | { | |
3686 | if (sigset) { | |
3687 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
3688 | vcpu->sigset_active = 1; | |
3689 | vcpu->sigset = *sigset; | |
3690 | } else | |
3691 | vcpu->sigset_active = 0; | |
3692 | return 0; | |
3693 | } | |
3694 | ||
ce55c049 JZ |
3695 | static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer, |
3696 | size_t size, loff_t *offset) | |
3697 | { | |
3698 | struct kvm_vcpu *vcpu = file->private_data; | |
3699 | ||
3700 | return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header, | |
3701 | &kvm_vcpu_stats_desc[0], &vcpu->stat, | |
3702 | sizeof(vcpu->stat), user_buffer, size, offset); | |
3703 | } | |
3704 | ||
3705 | static const struct file_operations kvm_vcpu_stats_fops = { | |
3706 | .read = kvm_vcpu_stats_read, | |
3707 | .llseek = noop_llseek, | |
3708 | }; | |
3709 | ||
3710 | static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu) | |
3711 | { | |
3712 | int fd; | |
3713 | struct file *file; | |
3714 | char name[15 + ITOA_MAX_LEN + 1]; | |
3715 | ||
3716 | snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id); | |
3717 | ||
3718 | fd = get_unused_fd_flags(O_CLOEXEC); | |
3719 | if (fd < 0) | |
3720 | return fd; | |
3721 | ||
3722 | file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY); | |
3723 | if (IS_ERR(file)) { | |
3724 | put_unused_fd(fd); | |
3725 | return PTR_ERR(file); | |
3726 | } | |
3727 | file->f_mode |= FMODE_PREAD; | |
3728 | fd_install(fd, file); | |
3729 | ||
3730 | return fd; | |
3731 | } | |
3732 | ||
bccf2150 AK |
3733 | static long kvm_vcpu_ioctl(struct file *filp, |
3734 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 3735 | { |
bccf2150 | 3736 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 3737 | void __user *argp = (void __user *)arg; |
313a3dc7 | 3738 | int r; |
fa3795a7 DH |
3739 | struct kvm_fpu *fpu = NULL; |
3740 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 3741 | |
0b8f1173 | 3742 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged) |
6d4e4c4f | 3743 | return -EIO; |
2122ff5e | 3744 | |
2ea75be3 DM |
3745 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
3746 | return -EINVAL; | |
3747 | ||
2122ff5e | 3748 | /* |
5cb0944c PB |
3749 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
3750 | * execution; mutex_lock() would break them. | |
2122ff5e | 3751 | */ |
5cb0944c PB |
3752 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
3753 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 3754 | return r; |
2122ff5e | 3755 | |
ec7660cc CD |
3756 | if (mutex_lock_killable(&vcpu->mutex)) |
3757 | return -EINTR; | |
6aa8b732 | 3758 | switch (ioctl) { |
0e4524a5 CB |
3759 | case KVM_RUN: { |
3760 | struct pid *oldpid; | |
f0fe5108 AK |
3761 | r = -EINVAL; |
3762 | if (arg) | |
3763 | goto out; | |
0e4524a5 | 3764 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 3765 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 3766 | /* The thread running this VCPU changed. */ |
bd2a6394 | 3767 | struct pid *newpid; |
f95ef0cd | 3768 | |
bd2a6394 CD |
3769 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
3770 | if (r) | |
3771 | break; | |
3772 | ||
3773 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
3774 | rcu_assign_pointer(vcpu->pid, newpid); |
3775 | if (oldpid) | |
3776 | synchronize_rcu(); | |
3777 | put_pid(oldpid); | |
3778 | } | |
1b94f6f8 | 3779 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 3780 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 3781 | break; |
0e4524a5 | 3782 | } |
6aa8b732 | 3783 | case KVM_GET_REGS: { |
3e4bb3ac | 3784 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3785 | |
3e4bb3ac | 3786 | r = -ENOMEM; |
b12ce36a | 3787 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 3788 | if (!kvm_regs) |
6aa8b732 | 3789 | goto out; |
3e4bb3ac XZ |
3790 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
3791 | if (r) | |
3792 | goto out_free1; | |
6aa8b732 | 3793 | r = -EFAULT; |
3e4bb3ac XZ |
3794 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
3795 | goto out_free1; | |
6aa8b732 | 3796 | r = 0; |
3e4bb3ac XZ |
3797 | out_free1: |
3798 | kfree(kvm_regs); | |
6aa8b732 AK |
3799 | break; |
3800 | } | |
3801 | case KVM_SET_REGS: { | |
3e4bb3ac | 3802 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3803 | |
ff5c2c03 SL |
3804 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
3805 | if (IS_ERR(kvm_regs)) { | |
3806 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 3807 | goto out; |
ff5c2c03 | 3808 | } |
3e4bb3ac | 3809 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 3810 | kfree(kvm_regs); |
6aa8b732 AK |
3811 | break; |
3812 | } | |
3813 | case KVM_GET_SREGS: { | |
b12ce36a BG |
3814 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
3815 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
3816 | r = -ENOMEM; |
3817 | if (!kvm_sregs) | |
3818 | goto out; | |
3819 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
3820 | if (r) |
3821 | goto out; | |
3822 | r = -EFAULT; | |
fa3795a7 | 3823 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
3824 | goto out; |
3825 | r = 0; | |
3826 | break; | |
3827 | } | |
3828 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
3829 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
3830 | if (IS_ERR(kvm_sregs)) { | |
3831 | r = PTR_ERR(kvm_sregs); | |
18595411 | 3832 | kvm_sregs = NULL; |
6aa8b732 | 3833 | goto out; |
ff5c2c03 | 3834 | } |
fa3795a7 | 3835 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
3836 | break; |
3837 | } | |
62d9f0db MT |
3838 | case KVM_GET_MP_STATE: { |
3839 | struct kvm_mp_state mp_state; | |
3840 | ||
3841 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
3842 | if (r) | |
3843 | goto out; | |
3844 | r = -EFAULT; | |
893bdbf1 | 3845 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
3846 | goto out; |
3847 | r = 0; | |
3848 | break; | |
3849 | } | |
3850 | case KVM_SET_MP_STATE: { | |
3851 | struct kvm_mp_state mp_state; | |
3852 | ||
3853 | r = -EFAULT; | |
893bdbf1 | 3854 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
3855 | goto out; |
3856 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
3857 | break; |
3858 | } | |
6aa8b732 AK |
3859 | case KVM_TRANSLATE: { |
3860 | struct kvm_translation tr; | |
3861 | ||
3862 | r = -EFAULT; | |
893bdbf1 | 3863 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 3864 | goto out; |
8b006791 | 3865 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
3866 | if (r) |
3867 | goto out; | |
3868 | r = -EFAULT; | |
893bdbf1 | 3869 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
3870 | goto out; |
3871 | r = 0; | |
3872 | break; | |
3873 | } | |
d0bfb940 JK |
3874 | case KVM_SET_GUEST_DEBUG: { |
3875 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
3876 | |
3877 | r = -EFAULT; | |
893bdbf1 | 3878 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 3879 | goto out; |
d0bfb940 | 3880 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
3881 | break; |
3882 | } | |
1961d276 AK |
3883 | case KVM_SET_SIGNAL_MASK: { |
3884 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3885 | struct kvm_signal_mask kvm_sigmask; | |
3886 | sigset_t sigset, *p; | |
3887 | ||
3888 | p = NULL; | |
3889 | if (argp) { | |
3890 | r = -EFAULT; | |
3891 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3892 | sizeof(kvm_sigmask))) |
1961d276 AK |
3893 | goto out; |
3894 | r = -EINVAL; | |
893bdbf1 | 3895 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
3896 | goto out; |
3897 | r = -EFAULT; | |
3898 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 3899 | sizeof(sigset))) |
1961d276 AK |
3900 | goto out; |
3901 | p = &sigset; | |
3902 | } | |
376d41ff | 3903 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
3904 | break; |
3905 | } | |
b8836737 | 3906 | case KVM_GET_FPU: { |
b12ce36a | 3907 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
3908 | r = -ENOMEM; |
3909 | if (!fpu) | |
3910 | goto out; | |
3911 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
3912 | if (r) |
3913 | goto out; | |
3914 | r = -EFAULT; | |
fa3795a7 | 3915 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
3916 | goto out; |
3917 | r = 0; | |
3918 | break; | |
3919 | } | |
3920 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
3921 | fpu = memdup_user(argp, sizeof(*fpu)); |
3922 | if (IS_ERR(fpu)) { | |
3923 | r = PTR_ERR(fpu); | |
18595411 | 3924 | fpu = NULL; |
b8836737 | 3925 | goto out; |
ff5c2c03 | 3926 | } |
fa3795a7 | 3927 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
3928 | break; |
3929 | } | |
ce55c049 JZ |
3930 | case KVM_GET_STATS_FD: { |
3931 | r = kvm_vcpu_ioctl_get_stats_fd(vcpu); | |
3932 | break; | |
3933 | } | |
bccf2150 | 3934 | default: |
313a3dc7 | 3935 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
3936 | } |
3937 | out: | |
ec7660cc | 3938 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
3939 | kfree(fpu); |
3940 | kfree(kvm_sregs); | |
bccf2150 AK |
3941 | return r; |
3942 | } | |
3943 | ||
de8e5d74 | 3944 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
3945 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
3946 | unsigned int ioctl, unsigned long arg) | |
3947 | { | |
3948 | struct kvm_vcpu *vcpu = filp->private_data; | |
3949 | void __user *argp = compat_ptr(arg); | |
3950 | int r; | |
3951 | ||
0b8f1173 | 3952 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged) |
1dda606c AG |
3953 | return -EIO; |
3954 | ||
3955 | switch (ioctl) { | |
3956 | case KVM_SET_SIGNAL_MASK: { | |
3957 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3958 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
3959 | sigset_t sigset; |
3960 | ||
3961 | if (argp) { | |
3962 | r = -EFAULT; | |
3963 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3964 | sizeof(kvm_sigmask))) |
1dda606c AG |
3965 | goto out; |
3966 | r = -EINVAL; | |
3968cf62 | 3967 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
3968 | goto out; |
3969 | r = -EFAULT; | |
1393b4aa PB |
3970 | if (get_compat_sigset(&sigset, |
3971 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 3972 | goto out; |
760a9a30 AC |
3973 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
3974 | } else | |
3975 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
3976 | break; |
3977 | } | |
3978 | default: | |
3979 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
3980 | } | |
3981 | ||
3982 | out: | |
3983 | return r; | |
3984 | } | |
3985 | #endif | |
3986 | ||
a1cd3f08 CLG |
3987 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
3988 | { | |
3989 | struct kvm_device *dev = filp->private_data; | |
3990 | ||
3991 | if (dev->ops->mmap) | |
3992 | return dev->ops->mmap(dev, vma); | |
3993 | ||
3994 | return -ENODEV; | |
3995 | } | |
3996 | ||
852b6d57 SW |
3997 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
3998 | int (*accessor)(struct kvm_device *dev, | |
3999 | struct kvm_device_attr *attr), | |
4000 | unsigned long arg) | |
4001 | { | |
4002 | struct kvm_device_attr attr; | |
4003 | ||
4004 | if (!accessor) | |
4005 | return -EPERM; | |
4006 | ||
4007 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
4008 | return -EFAULT; | |
4009 | ||
4010 | return accessor(dev, &attr); | |
4011 | } | |
4012 | ||
4013 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
4014 | unsigned long arg) | |
4015 | { | |
4016 | struct kvm_device *dev = filp->private_data; | |
4017 | ||
0b8f1173 | 4018 | if (dev->kvm->mm != current->mm || dev->kvm->vm_bugged) |
ddba9180 SC |
4019 | return -EIO; |
4020 | ||
852b6d57 SW |
4021 | switch (ioctl) { |
4022 | case KVM_SET_DEVICE_ATTR: | |
4023 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
4024 | case KVM_GET_DEVICE_ATTR: | |
4025 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
4026 | case KVM_HAS_DEVICE_ATTR: | |
4027 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
4028 | default: | |
4029 | if (dev->ops->ioctl) | |
4030 | return dev->ops->ioctl(dev, ioctl, arg); | |
4031 | ||
4032 | return -ENOTTY; | |
4033 | } | |
4034 | } | |
4035 | ||
852b6d57 SW |
4036 | static int kvm_device_release(struct inode *inode, struct file *filp) |
4037 | { | |
4038 | struct kvm_device *dev = filp->private_data; | |
4039 | struct kvm *kvm = dev->kvm; | |
4040 | ||
2bde9b3e CLG |
4041 | if (dev->ops->release) { |
4042 | mutex_lock(&kvm->lock); | |
4043 | list_del(&dev->vm_node); | |
4044 | dev->ops->release(dev); | |
4045 | mutex_unlock(&kvm->lock); | |
4046 | } | |
4047 | ||
852b6d57 SW |
4048 | kvm_put_kvm(kvm); |
4049 | return 0; | |
4050 | } | |
4051 | ||
4052 | static const struct file_operations kvm_device_fops = { | |
4053 | .unlocked_ioctl = kvm_device_ioctl, | |
4054 | .release = kvm_device_release, | |
7ddfd3e0 | 4055 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 4056 | .mmap = kvm_device_mmap, |
852b6d57 SW |
4057 | }; |
4058 | ||
4059 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
4060 | { | |
4061 | if (filp->f_op != &kvm_device_fops) | |
4062 | return NULL; | |
4063 | ||
4064 | return filp->private_data; | |
4065 | } | |
4066 | ||
8538cb22 | 4067 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 4068 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
4069 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
4070 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 4071 | #endif |
d60eacb0 WD |
4072 | }; |
4073 | ||
8538cb22 | 4074 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
4075 | { |
4076 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
4077 | return -ENOSPC; | |
4078 | ||
4079 | if (kvm_device_ops_table[type] != NULL) | |
4080 | return -EEXIST; | |
4081 | ||
4082 | kvm_device_ops_table[type] = ops; | |
4083 | return 0; | |
4084 | } | |
4085 | ||
571ee1b6 WL |
4086 | void kvm_unregister_device_ops(u32 type) |
4087 | { | |
4088 | if (kvm_device_ops_table[type] != NULL) | |
4089 | kvm_device_ops_table[type] = NULL; | |
4090 | } | |
4091 | ||
852b6d57 SW |
4092 | static int kvm_ioctl_create_device(struct kvm *kvm, |
4093 | struct kvm_create_device *cd) | |
4094 | { | |
8538cb22 | 4095 | const struct kvm_device_ops *ops = NULL; |
852b6d57 SW |
4096 | struct kvm_device *dev; |
4097 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 4098 | int type; |
852b6d57 SW |
4099 | int ret; |
4100 | ||
d60eacb0 WD |
4101 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
4102 | return -ENODEV; | |
4103 | ||
1d487e9b PB |
4104 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
4105 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 4106 | if (ops == NULL) |
852b6d57 | 4107 | return -ENODEV; |
852b6d57 SW |
4108 | |
4109 | if (test) | |
4110 | return 0; | |
4111 | ||
b12ce36a | 4112 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
4113 | if (!dev) |
4114 | return -ENOMEM; | |
4115 | ||
4116 | dev->ops = ops; | |
4117 | dev->kvm = kvm; | |
852b6d57 | 4118 | |
a28ebea2 | 4119 | mutex_lock(&kvm->lock); |
1d487e9b | 4120 | ret = ops->create(dev, type); |
852b6d57 | 4121 | if (ret < 0) { |
a28ebea2 | 4122 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
4123 | kfree(dev); |
4124 | return ret; | |
4125 | } | |
a28ebea2 CD |
4126 | list_add(&dev->vm_node, &kvm->devices); |
4127 | mutex_unlock(&kvm->lock); | |
852b6d57 | 4128 | |
023e9fdd CD |
4129 | if (ops->init) |
4130 | ops->init(dev); | |
4131 | ||
cfa39381 | 4132 | kvm_get_kvm(kvm); |
24009b05 | 4133 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 4134 | if (ret < 0) { |
149487bd | 4135 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
4136 | mutex_lock(&kvm->lock); |
4137 | list_del(&dev->vm_node); | |
4138 | mutex_unlock(&kvm->lock); | |
a0f1d21c | 4139 | ops->destroy(dev); |
852b6d57 SW |
4140 | return ret; |
4141 | } | |
4142 | ||
852b6d57 SW |
4143 | cd->fd = ret; |
4144 | return 0; | |
4145 | } | |
4146 | ||
92b591a4 AG |
4147 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
4148 | { | |
4149 | switch (arg) { | |
4150 | case KVM_CAP_USER_MEMORY: | |
4151 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
4152 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
4153 | case KVM_CAP_INTERNAL_ERROR_DATA: |
4154 | #ifdef CONFIG_HAVE_KVM_MSI | |
4155 | case KVM_CAP_SIGNAL_MSI: | |
4156 | #endif | |
297e2105 | 4157 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 4158 | case KVM_CAP_IRQFD: |
92b591a4 AG |
4159 | case KVM_CAP_IRQFD_RESAMPLE: |
4160 | #endif | |
e9ea5069 | 4161 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 4162 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 4163 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 4164 | case KVM_CAP_HALT_POLL: |
92b591a4 | 4165 | return 1; |
4b4357e0 | 4166 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
4167 | case KVM_CAP_COALESCED_MMIO: |
4168 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
4169 | case KVM_CAP_COALESCED_PIO: |
4170 | return 1; | |
30422558 | 4171 | #endif |
3c9bd400 JZ |
4172 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4173 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
4174 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
4175 | #endif | |
92b591a4 AG |
4176 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4177 | case KVM_CAP_IRQ_ROUTING: | |
4178 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
4179 | #endif |
4180 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
4181 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
4182 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 4183 | #endif |
c110ae57 PB |
4184 | case KVM_CAP_NR_MEMSLOTS: |
4185 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed PX |
4186 | case KVM_CAP_DIRTY_LOG_RING: |
4187 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
4188 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); | |
4189 | #else | |
4190 | return 0; | |
4191 | #endif | |
ce55c049 JZ |
4192 | case KVM_CAP_BINARY_STATS_FD: |
4193 | return 1; | |
92b591a4 AG |
4194 | default: |
4195 | break; | |
4196 | } | |
4197 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
4198 | } | |
4199 | ||
fb04a1ed PX |
4200 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
4201 | { | |
4202 | int r; | |
4203 | ||
4204 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
4205 | return -EINVAL; | |
4206 | ||
4207 | /* the size should be power of 2 */ | |
4208 | if (!size || (size & (size - 1))) | |
4209 | return -EINVAL; | |
4210 | ||
4211 | /* Should be bigger to keep the reserved entries, or a page */ | |
4212 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
4213 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
4214 | return -EINVAL; | |
4215 | ||
4216 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
4217 | sizeof(struct kvm_dirty_gfn)) | |
4218 | return -E2BIG; | |
4219 | ||
4220 | /* We only allow it to set once */ | |
4221 | if (kvm->dirty_ring_size) | |
4222 | return -EINVAL; | |
4223 | ||
4224 | mutex_lock(&kvm->lock); | |
4225 | ||
4226 | if (kvm->created_vcpus) { | |
4227 | /* We don't allow to change this value after vcpu created */ | |
4228 | r = -EINVAL; | |
4229 | } else { | |
4230 | kvm->dirty_ring_size = size; | |
4231 | r = 0; | |
4232 | } | |
4233 | ||
4234 | mutex_unlock(&kvm->lock); | |
4235 | return r; | |
4236 | } | |
4237 | ||
4238 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
4239 | { | |
4240 | int i; | |
4241 | struct kvm_vcpu *vcpu; | |
4242 | int cleared = 0; | |
4243 | ||
4244 | if (!kvm->dirty_ring_size) | |
4245 | return -EINVAL; | |
4246 | ||
4247 | mutex_lock(&kvm->slots_lock); | |
4248 | ||
4249 | kvm_for_each_vcpu(i, vcpu, kvm) | |
4250 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
4251 | ||
4252 | mutex_unlock(&kvm->slots_lock); | |
4253 | ||
4254 | if (cleared) | |
4255 | kvm_flush_remote_tlbs(kvm); | |
4256 | ||
4257 | return cleared; | |
4258 | } | |
4259 | ||
e5d83c74 PB |
4260 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4261 | struct kvm_enable_cap *cap) | |
4262 | { | |
4263 | return -EINVAL; | |
4264 | } | |
4265 | ||
4266 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
4267 | struct kvm_enable_cap *cap) | |
4268 | { | |
4269 | switch (cap->cap) { | |
2a31b9db | 4270 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
4271 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
4272 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
4273 | ||
4274 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
4275 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
4276 | ||
4277 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
4278 | return -EINVAL; |
4279 | kvm->manual_dirty_log_protect = cap->args[0]; | |
4280 | return 0; | |
3c9bd400 | 4281 | } |
2a31b9db | 4282 | #endif |
acd05785 DM |
4283 | case KVM_CAP_HALT_POLL: { |
4284 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
4285 | return -EINVAL; | |
4286 | ||
4287 | kvm->max_halt_poll_ns = cap->args[0]; | |
4288 | return 0; | |
4289 | } | |
fb04a1ed PX |
4290 | case KVM_CAP_DIRTY_LOG_RING: |
4291 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); | |
e5d83c74 PB |
4292 | default: |
4293 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
4294 | } | |
4295 | } | |
4296 | ||
fcfe1bae JZ |
4297 | static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer, |
4298 | size_t size, loff_t *offset) | |
4299 | { | |
4300 | struct kvm *kvm = file->private_data; | |
4301 | ||
4302 | return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header, | |
4303 | &kvm_vm_stats_desc[0], &kvm->stat, | |
4304 | sizeof(kvm->stat), user_buffer, size, offset); | |
4305 | } | |
4306 | ||
4307 | static const struct file_operations kvm_vm_stats_fops = { | |
4308 | .read = kvm_vm_stats_read, | |
4309 | .llseek = noop_llseek, | |
4310 | }; | |
4311 | ||
4312 | static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm) | |
4313 | { | |
4314 | int fd; | |
4315 | struct file *file; | |
4316 | ||
4317 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4318 | if (fd < 0) | |
4319 | return fd; | |
4320 | ||
4321 | file = anon_inode_getfile("kvm-vm-stats", | |
4322 | &kvm_vm_stats_fops, kvm, O_RDONLY); | |
4323 | if (IS_ERR(file)) { | |
4324 | put_unused_fd(fd); | |
4325 | return PTR_ERR(file); | |
4326 | } | |
4327 | file->f_mode |= FMODE_PREAD; | |
4328 | fd_install(fd, file); | |
4329 | ||
4330 | return fd; | |
4331 | } | |
4332 | ||
bccf2150 AK |
4333 | static long kvm_vm_ioctl(struct file *filp, |
4334 | unsigned int ioctl, unsigned long arg) | |
4335 | { | |
4336 | struct kvm *kvm = filp->private_data; | |
4337 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 4338 | int r; |
bccf2150 | 4339 | |
0b8f1173 | 4340 | if (kvm->mm != current->mm || kvm->vm_bugged) |
6d4e4c4f | 4341 | return -EIO; |
bccf2150 AK |
4342 | switch (ioctl) { |
4343 | case KVM_CREATE_VCPU: | |
4344 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 4345 | break; |
e5d83c74 PB |
4346 | case KVM_ENABLE_CAP: { |
4347 | struct kvm_enable_cap cap; | |
4348 | ||
4349 | r = -EFAULT; | |
4350 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4351 | goto out; | |
4352 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
4353 | break; | |
4354 | } | |
6fc138d2 IE |
4355 | case KVM_SET_USER_MEMORY_REGION: { |
4356 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
4357 | ||
4358 | r = -EFAULT; | |
4359 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 4360 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
4361 | goto out; |
4362 | ||
47ae31e2 | 4363 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
4364 | break; |
4365 | } | |
4366 | case KVM_GET_DIRTY_LOG: { | |
4367 | struct kvm_dirty_log log; | |
4368 | ||
4369 | r = -EFAULT; | |
893bdbf1 | 4370 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 4371 | goto out; |
2c6f5df9 | 4372 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
4373 | break; |
4374 | } | |
2a31b9db PB |
4375 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4376 | case KVM_CLEAR_DIRTY_LOG: { | |
4377 | struct kvm_clear_dirty_log log; | |
4378 | ||
4379 | r = -EFAULT; | |
4380 | if (copy_from_user(&log, argp, sizeof(log))) | |
4381 | goto out; | |
4382 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4383 | break; | |
4384 | } | |
4385 | #endif | |
4b4357e0 | 4386 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
4387 | case KVM_REGISTER_COALESCED_MMIO: { |
4388 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4389 | |
5f94c174 | 4390 | r = -EFAULT; |
893bdbf1 | 4391 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4392 | goto out; |
5f94c174 | 4393 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4394 | break; |
4395 | } | |
4396 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
4397 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4398 | |
5f94c174 | 4399 | r = -EFAULT; |
893bdbf1 | 4400 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4401 | goto out; |
5f94c174 | 4402 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4403 | break; |
4404 | } | |
4405 | #endif | |
721eecbf GH |
4406 | case KVM_IRQFD: { |
4407 | struct kvm_irqfd data; | |
4408 | ||
4409 | r = -EFAULT; | |
893bdbf1 | 4410 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 4411 | goto out; |
d4db2935 | 4412 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
4413 | break; |
4414 | } | |
d34e6b17 GH |
4415 | case KVM_IOEVENTFD: { |
4416 | struct kvm_ioeventfd data; | |
4417 | ||
4418 | r = -EFAULT; | |
893bdbf1 | 4419 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
4420 | goto out; |
4421 | r = kvm_ioeventfd(kvm, &data); | |
4422 | break; | |
4423 | } | |
07975ad3 JK |
4424 | #ifdef CONFIG_HAVE_KVM_MSI |
4425 | case KVM_SIGNAL_MSI: { | |
4426 | struct kvm_msi msi; | |
4427 | ||
4428 | r = -EFAULT; | |
893bdbf1 | 4429 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
4430 | goto out; |
4431 | r = kvm_send_userspace_msi(kvm, &msi); | |
4432 | break; | |
4433 | } | |
23d43cf9 CD |
4434 | #endif |
4435 | #ifdef __KVM_HAVE_IRQ_LINE | |
4436 | case KVM_IRQ_LINE_STATUS: | |
4437 | case KVM_IRQ_LINE: { | |
4438 | struct kvm_irq_level irq_event; | |
4439 | ||
4440 | r = -EFAULT; | |
893bdbf1 | 4441 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
4442 | goto out; |
4443 | ||
aa2fbe6d YZ |
4444 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
4445 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
4446 | if (r) |
4447 | goto out; | |
4448 | ||
4449 | r = -EFAULT; | |
4450 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 4451 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
4452 | goto out; |
4453 | } | |
4454 | ||
4455 | r = 0; | |
4456 | break; | |
4457 | } | |
73880c80 | 4458 | #endif |
aa8d5944 AG |
4459 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4460 | case KVM_SET_GSI_ROUTING: { | |
4461 | struct kvm_irq_routing routing; | |
4462 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 4463 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
4464 | |
4465 | r = -EFAULT; | |
4466 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
4467 | goto out; | |
4468 | r = -EINVAL; | |
5c0aea0e DH |
4469 | if (!kvm_arch_can_set_irq_routing(kvm)) |
4470 | goto out; | |
caf1ff26 | 4471 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
4472 | goto out; |
4473 | if (routing.flags) | |
4474 | goto out; | |
f8c1b85b | 4475 | if (routing.nr) { |
f8c1b85b | 4476 | urouting = argp; |
7ec28e26 DE |
4477 | entries = vmemdup_user(urouting->entries, |
4478 | array_size(sizeof(*entries), | |
4479 | routing.nr)); | |
4480 | if (IS_ERR(entries)) { | |
4481 | r = PTR_ERR(entries); | |
4482 | goto out; | |
4483 | } | |
f8c1b85b | 4484 | } |
aa8d5944 AG |
4485 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
4486 | routing.flags); | |
7ec28e26 | 4487 | kvfree(entries); |
aa8d5944 AG |
4488 | break; |
4489 | } | |
4490 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
4491 | case KVM_CREATE_DEVICE: { |
4492 | struct kvm_create_device cd; | |
4493 | ||
4494 | r = -EFAULT; | |
4495 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
4496 | goto out; | |
4497 | ||
4498 | r = kvm_ioctl_create_device(kvm, &cd); | |
4499 | if (r) | |
4500 | goto out; | |
4501 | ||
4502 | r = -EFAULT; | |
4503 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
4504 | goto out; | |
4505 | ||
4506 | r = 0; | |
4507 | break; | |
4508 | } | |
92b591a4 AG |
4509 | case KVM_CHECK_EXTENSION: |
4510 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
4511 | break; | |
fb04a1ed PX |
4512 | case KVM_RESET_DIRTY_RINGS: |
4513 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
4514 | break; | |
fcfe1bae JZ |
4515 | case KVM_GET_STATS_FD: |
4516 | r = kvm_vm_ioctl_get_stats_fd(kvm); | |
4517 | break; | |
f17abe9a | 4518 | default: |
1fe779f8 | 4519 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
4520 | } |
4521 | out: | |
4522 | return r; | |
4523 | } | |
4524 | ||
de8e5d74 | 4525 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
4526 | struct compat_kvm_dirty_log { |
4527 | __u32 slot; | |
4528 | __u32 padding1; | |
4529 | union { | |
4530 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4531 | __u64 padding2; | |
4532 | }; | |
4533 | }; | |
4534 | ||
8750f9bb PB |
4535 | struct compat_kvm_clear_dirty_log { |
4536 | __u32 slot; | |
4537 | __u32 num_pages; | |
4538 | __u64 first_page; | |
4539 | union { | |
4540 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4541 | __u64 padding2; | |
4542 | }; | |
4543 | }; | |
4544 | ||
6ff5894c AB |
4545 | static long kvm_vm_compat_ioctl(struct file *filp, |
4546 | unsigned int ioctl, unsigned long arg) | |
4547 | { | |
4548 | struct kvm *kvm = filp->private_data; | |
4549 | int r; | |
4550 | ||
0b8f1173 | 4551 | if (kvm->mm != current->mm || kvm->vm_bugged) |
6ff5894c AB |
4552 | return -EIO; |
4553 | switch (ioctl) { | |
8750f9bb PB |
4554 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4555 | case KVM_CLEAR_DIRTY_LOG: { | |
4556 | struct compat_kvm_clear_dirty_log compat_log; | |
4557 | struct kvm_clear_dirty_log log; | |
4558 | ||
4559 | if (copy_from_user(&compat_log, (void __user *)arg, | |
4560 | sizeof(compat_log))) | |
4561 | return -EFAULT; | |
4562 | log.slot = compat_log.slot; | |
4563 | log.num_pages = compat_log.num_pages; | |
4564 | log.first_page = compat_log.first_page; | |
4565 | log.padding2 = compat_log.padding2; | |
4566 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4567 | ||
4568 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4569 | break; | |
4570 | } | |
4571 | #endif | |
6ff5894c AB |
4572 | case KVM_GET_DIRTY_LOG: { |
4573 | struct compat_kvm_dirty_log compat_log; | |
4574 | struct kvm_dirty_log log; | |
4575 | ||
6ff5894c AB |
4576 | if (copy_from_user(&compat_log, (void __user *)arg, |
4577 | sizeof(compat_log))) | |
f6a3b168 | 4578 | return -EFAULT; |
6ff5894c AB |
4579 | log.slot = compat_log.slot; |
4580 | log.padding1 = compat_log.padding1; | |
4581 | log.padding2 = compat_log.padding2; | |
4582 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4583 | ||
4584 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
4585 | break; |
4586 | } | |
4587 | default: | |
4588 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
4589 | } | |
6ff5894c AB |
4590 | return r; |
4591 | } | |
4592 | #endif | |
4593 | ||
3d3aab1b | 4594 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
4595 | .release = kvm_vm_release, |
4596 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 4597 | .llseek = noop_llseek, |
7ddfd3e0 | 4598 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
4599 | }; |
4600 | ||
54526d1f NT |
4601 | bool file_is_kvm(struct file *file) |
4602 | { | |
4603 | return file && file->f_op == &kvm_vm_fops; | |
4604 | } | |
4605 | EXPORT_SYMBOL_GPL(file_is_kvm); | |
4606 | ||
e08b9637 | 4607 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 4608 | { |
aac87636 | 4609 | int r; |
f17abe9a | 4610 | struct kvm *kvm; |
506cfba9 | 4611 | struct file *file; |
f17abe9a | 4612 | |
e08b9637 | 4613 | kvm = kvm_create_vm(type); |
d6d28168 AK |
4614 | if (IS_ERR(kvm)) |
4615 | return PTR_ERR(kvm); | |
4b4357e0 | 4616 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 4617 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
4618 | if (r < 0) |
4619 | goto put_kvm; | |
6ce5a090 | 4620 | #endif |
506cfba9 | 4621 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
4622 | if (r < 0) |
4623 | goto put_kvm; | |
4624 | ||
fcfe1bae JZ |
4625 | snprintf(kvm->stats_id, sizeof(kvm->stats_id), |
4626 | "kvm-%d", task_pid_nr(current)); | |
4627 | ||
506cfba9 AV |
4628 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
4629 | if (IS_ERR(file)) { | |
4630 | put_unused_fd(r); | |
78588335 ME |
4631 | r = PTR_ERR(file); |
4632 | goto put_kvm; | |
506cfba9 | 4633 | } |
536a6f88 | 4634 | |
525df861 PB |
4635 | /* |
4636 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
4637 | * already set, with ->release() being kvm_vm_release(). In error | |
4638 | * cases it will be called by the final fput(file) and will take | |
4639 | * care of doing kvm_put_kvm(kvm). | |
4640 | */ | |
536a6f88 | 4641 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
4642 | put_unused_fd(r); |
4643 | fput(file); | |
536a6f88 JF |
4644 | return -ENOMEM; |
4645 | } | |
286de8f6 | 4646 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 4647 | |
506cfba9 | 4648 | fd_install(r, file); |
aac87636 | 4649 | return r; |
78588335 ME |
4650 | |
4651 | put_kvm: | |
4652 | kvm_put_kvm(kvm); | |
4653 | return r; | |
f17abe9a AK |
4654 | } |
4655 | ||
4656 | static long kvm_dev_ioctl(struct file *filp, | |
4657 | unsigned int ioctl, unsigned long arg) | |
4658 | { | |
07c45a36 | 4659 | long r = -EINVAL; |
f17abe9a AK |
4660 | |
4661 | switch (ioctl) { | |
4662 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
4663 | if (arg) |
4664 | goto out; | |
f17abe9a AK |
4665 | r = KVM_API_VERSION; |
4666 | break; | |
4667 | case KVM_CREATE_VM: | |
e08b9637 | 4668 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 4669 | break; |
018d00d2 | 4670 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 4671 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 4672 | break; |
07c45a36 | 4673 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
4674 | if (arg) |
4675 | goto out; | |
adb1ff46 AK |
4676 | r = PAGE_SIZE; /* struct kvm_run */ |
4677 | #ifdef CONFIG_X86 | |
4678 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 4679 | #endif |
4b4357e0 | 4680 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 4681 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 4682 | #endif |
07c45a36 | 4683 | break; |
d4c9ff2d FEL |
4684 | case KVM_TRACE_ENABLE: |
4685 | case KVM_TRACE_PAUSE: | |
4686 | case KVM_TRACE_DISABLE: | |
2023a29c | 4687 | r = -EOPNOTSUPP; |
d4c9ff2d | 4688 | break; |
6aa8b732 | 4689 | default: |
043405e1 | 4690 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
4691 | } |
4692 | out: | |
4693 | return r; | |
4694 | } | |
4695 | ||
6aa8b732 | 4696 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 4697 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 4698 | .llseek = noop_llseek, |
7ddfd3e0 | 4699 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
4700 | }; |
4701 | ||
4702 | static struct miscdevice kvm_dev = { | |
bbe4432e | 4703 | KVM_MINOR, |
6aa8b732 AK |
4704 | "kvm", |
4705 | &kvm_chardev_ops, | |
4706 | }; | |
4707 | ||
75b7127c | 4708 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
4709 | { |
4710 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 4711 | int r; |
1b6c0168 | 4712 | |
7f59f492 | 4713 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4714 | return; |
10474ae8 | 4715 | |
7f59f492 | 4716 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 4717 | |
13a34e06 | 4718 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
4719 | |
4720 | if (r) { | |
4721 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
4722 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 4723 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 4724 | } |
1b6c0168 AK |
4725 | } |
4726 | ||
8c18b2d2 | 4727 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 4728 | { |
4a937f96 | 4729 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4730 | if (kvm_usage_count) |
4731 | hardware_enable_nolock(NULL); | |
4a937f96 | 4732 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4733 | return 0; |
75b7127c TY |
4734 | } |
4735 | ||
4736 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
4737 | { |
4738 | int cpu = raw_smp_processor_id(); | |
4739 | ||
7f59f492 | 4740 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4741 | return; |
7f59f492 | 4742 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 4743 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
4744 | } |
4745 | ||
8c18b2d2 | 4746 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 4747 | { |
4a937f96 | 4748 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4749 | if (kvm_usage_count) |
4750 | hardware_disable_nolock(NULL); | |
4a937f96 | 4751 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4752 | return 0; |
75b7127c TY |
4753 | } |
4754 | ||
10474ae8 AG |
4755 | static void hardware_disable_all_nolock(void) |
4756 | { | |
4757 | BUG_ON(!kvm_usage_count); | |
4758 | ||
4759 | kvm_usage_count--; | |
4760 | if (!kvm_usage_count) | |
75b7127c | 4761 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
4762 | } |
4763 | ||
4764 | static void hardware_disable_all(void) | |
4765 | { | |
4a937f96 | 4766 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 4767 | hardware_disable_all_nolock(); |
4a937f96 | 4768 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4769 | } |
4770 | ||
4771 | static int hardware_enable_all(void) | |
4772 | { | |
4773 | int r = 0; | |
4774 | ||
4a937f96 | 4775 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
4776 | |
4777 | kvm_usage_count++; | |
4778 | if (kvm_usage_count == 1) { | |
4779 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 4780 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
4781 | |
4782 | if (atomic_read(&hardware_enable_failed)) { | |
4783 | hardware_disable_all_nolock(); | |
4784 | r = -EBUSY; | |
4785 | } | |
4786 | } | |
4787 | ||
4a937f96 | 4788 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4789 | |
4790 | return r; | |
4791 | } | |
4792 | ||
9a2b85c6 | 4793 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 4794 | void *v) |
9a2b85c6 | 4795 | { |
8e1c1815 SY |
4796 | /* |
4797 | * Some (well, at least mine) BIOSes hang on reboot if | |
4798 | * in vmx root mode. | |
4799 | * | |
4800 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
4801 | */ | |
1170adc6 | 4802 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 4803 | kvm_rebooting = true; |
75b7127c | 4804 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
4805 | return NOTIFY_OK; |
4806 | } | |
4807 | ||
4808 | static struct notifier_block kvm_reboot_notifier = { | |
4809 | .notifier_call = kvm_reboot, | |
4810 | .priority = 0, | |
4811 | }; | |
4812 | ||
e93f8a0f | 4813 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
4814 | { |
4815 | int i; | |
4816 | ||
4817 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 4818 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
4819 | |
4820 | kvm_iodevice_destructor(pos); | |
4821 | } | |
e93f8a0f | 4822 | kfree(bus); |
2eeb2e94 GH |
4823 | } |
4824 | ||
c21fbff1 | 4825 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 4826 | const struct kvm_io_range *r2) |
743eeb0b | 4827 | { |
8f4216c7 JW |
4828 | gpa_t addr1 = r1->addr; |
4829 | gpa_t addr2 = r2->addr; | |
4830 | ||
4831 | if (addr1 < addr2) | |
743eeb0b | 4832 | return -1; |
8f4216c7 JW |
4833 | |
4834 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
4835 | * accept any overlapping write. Any order is acceptable for | |
4836 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
4837 | * we process all of them. | |
4838 | */ | |
4839 | if (r2->len) { | |
4840 | addr1 += r1->len; | |
4841 | addr2 += r2->len; | |
4842 | } | |
4843 | ||
4844 | if (addr1 > addr2) | |
743eeb0b | 4845 | return 1; |
8f4216c7 | 4846 | |
743eeb0b SL |
4847 | return 0; |
4848 | } | |
4849 | ||
a343c9b7 PB |
4850 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
4851 | { | |
c21fbff1 | 4852 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
4853 | } |
4854 | ||
39369f7a | 4855 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
4856 | gpa_t addr, int len) |
4857 | { | |
4858 | struct kvm_io_range *range, key; | |
4859 | int off; | |
4860 | ||
4861 | key = (struct kvm_io_range) { | |
4862 | .addr = addr, | |
4863 | .len = len, | |
4864 | }; | |
4865 | ||
4866 | range = bsearch(&key, bus->range, bus->dev_count, | |
4867 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
4868 | if (range == NULL) | |
4869 | return -ENOENT; | |
4870 | ||
4871 | off = range - bus->range; | |
4872 | ||
c21fbff1 | 4873 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
4874 | off--; |
4875 | ||
4876 | return off; | |
4877 | } | |
4878 | ||
e32edf4f | 4879 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
4880 | struct kvm_io_range *range, const void *val) |
4881 | { | |
4882 | int idx; | |
4883 | ||
4884 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
4885 | if (idx < 0) | |
4886 | return -EOPNOTSUPP; | |
4887 | ||
4888 | while (idx < bus->dev_count && | |
c21fbff1 | 4889 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4890 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4891 | range->len, val)) |
4892 | return idx; | |
4893 | idx++; | |
4894 | } | |
4895 | ||
4896 | return -EOPNOTSUPP; | |
4897 | } | |
4898 | ||
bda9020e | 4899 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 4900 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 4901 | int len, const void *val) |
2eeb2e94 | 4902 | { |
90d83dc3 | 4903 | struct kvm_io_bus *bus; |
743eeb0b | 4904 | struct kvm_io_range range; |
126a5af5 | 4905 | int r; |
743eeb0b SL |
4906 | |
4907 | range = (struct kvm_io_range) { | |
4908 | .addr = addr, | |
4909 | .len = len, | |
4910 | }; | |
90d83dc3 | 4911 | |
e32edf4f | 4912 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4913 | if (!bus) |
4914 | return -ENOMEM; | |
e32edf4f | 4915 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4916 | return r < 0 ? r : 0; |
4917 | } | |
a2420107 | 4918 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
4919 | |
4920 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
4921 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
4922 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
4923 | { |
4924 | struct kvm_io_bus *bus; | |
4925 | struct kvm_io_range range; | |
4926 | ||
4927 | range = (struct kvm_io_range) { | |
4928 | .addr = addr, | |
4929 | .len = len, | |
4930 | }; | |
4931 | ||
e32edf4f | 4932 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4933 | if (!bus) |
4934 | return -ENOMEM; | |
126a5af5 CH |
4935 | |
4936 | /* First try the device referenced by cookie. */ | |
4937 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 4938 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 4939 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
4940 | val)) |
4941 | return cookie; | |
4942 | ||
4943 | /* | |
4944 | * cookie contained garbage; fall back to search and return the | |
4945 | * correct cookie value. | |
4946 | */ | |
e32edf4f | 4947 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4948 | } |
4949 | ||
e32edf4f NN |
4950 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
4951 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
4952 | { |
4953 | int idx; | |
4954 | ||
4955 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
4956 | if (idx < 0) |
4957 | return -EOPNOTSUPP; | |
4958 | ||
4959 | while (idx < bus->dev_count && | |
c21fbff1 | 4960 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4961 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4962 | range->len, val)) |
4963 | return idx; | |
743eeb0b SL |
4964 | idx++; |
4965 | } | |
4966 | ||
bda9020e MT |
4967 | return -EOPNOTSUPP; |
4968 | } | |
2eeb2e94 | 4969 | |
bda9020e | 4970 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 4971 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 4972 | int len, void *val) |
bda9020e | 4973 | { |
90d83dc3 | 4974 | struct kvm_io_bus *bus; |
743eeb0b | 4975 | struct kvm_io_range range; |
126a5af5 | 4976 | int r; |
743eeb0b SL |
4977 | |
4978 | range = (struct kvm_io_range) { | |
4979 | .addr = addr, | |
4980 | .len = len, | |
4981 | }; | |
e93f8a0f | 4982 | |
e32edf4f | 4983 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4984 | if (!bus) |
4985 | return -ENOMEM; | |
e32edf4f | 4986 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
4987 | return r < 0 ? r : 0; |
4988 | } | |
743eeb0b | 4989 | |
79fac95e | 4990 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
4991 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
4992 | int len, struct kvm_io_device *dev) | |
6c474694 | 4993 | { |
d4c67a7a | 4994 | int i; |
e93f8a0f | 4995 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 4996 | struct kvm_io_range range; |
090b7aff | 4997 | |
4a12f951 | 4998 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
4999 | if (!bus) |
5000 | return -ENOMEM; | |
5001 | ||
6ea34c9b AK |
5002 | /* exclude ioeventfd which is limited by maximum fd */ |
5003 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 5004 | return -ENOSPC; |
2eeb2e94 | 5005 | |
90952cd3 | 5006 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 5007 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
5008 | if (!new_bus) |
5009 | return -ENOMEM; | |
d4c67a7a GH |
5010 | |
5011 | range = (struct kvm_io_range) { | |
5012 | .addr = addr, | |
5013 | .len = len, | |
5014 | .dev = dev, | |
5015 | }; | |
5016 | ||
5017 | for (i = 0; i < bus->dev_count; i++) | |
5018 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
5019 | break; | |
5020 | ||
5021 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
5022 | new_bus->dev_count++; | |
5023 | new_bus->range[i] = range; | |
5024 | memcpy(new_bus->range + i + 1, bus->range + i, | |
5025 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
5026 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
5027 | synchronize_srcu_expedited(&kvm->srcu); | |
5028 | kfree(bus); | |
090b7aff GH |
5029 | |
5030 | return 0; | |
5031 | } | |
5032 | ||
5d3c4c79 SC |
5033 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5034 | struct kvm_io_device *dev) | |
090b7aff | 5035 | { |
f6588660 | 5036 | int i, j; |
e93f8a0f | 5037 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 5038 | |
7c896d37 SC |
5039 | lockdep_assert_held(&kvm->slots_lock); |
5040 | ||
4a12f951 | 5041 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 5042 | if (!bus) |
5d3c4c79 | 5043 | return 0; |
df630b8c | 5044 | |
7c896d37 | 5045 | for (i = 0; i < bus->dev_count; i++) { |
a1300716 | 5046 | if (bus->range[i].dev == dev) { |
090b7aff GH |
5047 | break; |
5048 | } | |
7c896d37 | 5049 | } |
e93f8a0f | 5050 | |
90db1043 | 5051 | if (i == bus->dev_count) |
5d3c4c79 | 5052 | return 0; |
a1300716 | 5053 | |
90952cd3 | 5054 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 5055 | GFP_KERNEL_ACCOUNT); |
f6588660 | 5056 | if (new_bus) { |
871c433b | 5057 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
5058 | new_bus->dev_count--; |
5059 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 5060 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
2ee37574 SC |
5061 | } |
5062 | ||
5063 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
5064 | synchronize_srcu_expedited(&kvm->srcu); | |
5065 | ||
5066 | /* Destroy the old bus _after_ installing the (null) bus. */ | |
5067 | if (!new_bus) { | |
90db1043 | 5068 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
f6588660 RK |
5069 | for (j = 0; j < bus->dev_count; j++) { |
5070 | if (j == i) | |
5071 | continue; | |
5072 | kvm_iodevice_destructor(bus->range[j].dev); | |
5073 | } | |
90db1043 | 5074 | } |
a1300716 | 5075 | |
e93f8a0f | 5076 | kfree(bus); |
5d3c4c79 | 5077 | return new_bus ? 0 : -ENOMEM; |
2eeb2e94 GH |
5078 | } |
5079 | ||
8a39d006 AP |
5080 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5081 | gpa_t addr) | |
5082 | { | |
5083 | struct kvm_io_bus *bus; | |
5084 | int dev_idx, srcu_idx; | |
5085 | struct kvm_io_device *iodev = NULL; | |
5086 | ||
5087 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
5088 | ||
5089 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
5090 | if (!bus) |
5091 | goto out_unlock; | |
8a39d006 AP |
5092 | |
5093 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
5094 | if (dev_idx < 0) | |
5095 | goto out_unlock; | |
5096 | ||
5097 | iodev = bus->range[dev_idx].dev; | |
5098 | ||
5099 | out_unlock: | |
5100 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
5101 | ||
5102 | return iodev; | |
5103 | } | |
5104 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
5105 | ||
536a6f88 JF |
5106 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
5107 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
5108 | const char *fmt) | |
5109 | { | |
5110 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5111 | inode->i_private; | |
5112 | ||
605c7130 PX |
5113 | /* |
5114 | * The debugfs files are a reference to the kvm struct which | |
5115 | * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe | |
5116 | * avoids the race between open and the removal of the debugfs directory. | |
536a6f88 | 5117 | */ |
605c7130 | 5118 | if (!kvm_get_kvm_safe(stat_data->kvm)) |
536a6f88 JF |
5119 | return -ENOENT; |
5120 | ||
833b45de | 5121 | if (simple_attr_open(inode, file, get, |
bc9e9e67 | 5122 | kvm_stats_debugfs_mode(stat_data->desc) & 0222 |
09cbcef6 MP |
5123 | ? set : NULL, |
5124 | fmt)) { | |
536a6f88 JF |
5125 | kvm_put_kvm(stat_data->kvm); |
5126 | return -ENOMEM; | |
5127 | } | |
5128 | ||
5129 | return 0; | |
5130 | } | |
5131 | ||
5132 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
5133 | { | |
5134 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5135 | inode->i_private; | |
5136 | ||
5137 | simple_attr_release(inode, file); | |
5138 | kvm_put_kvm(stat_data->kvm); | |
5139 | ||
5140 | return 0; | |
5141 | } | |
5142 | ||
09cbcef6 | 5143 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 5144 | { |
bc9e9e67 | 5145 | *val = *(u64 *)((void *)(&kvm->stat) + offset); |
536a6f88 | 5146 | |
09cbcef6 MP |
5147 | return 0; |
5148 | } | |
5149 | ||
5150 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
5151 | { | |
bc9e9e67 | 5152 | *(u64 *)((void *)(&kvm->stat) + offset) = 0; |
536a6f88 JF |
5153 | |
5154 | return 0; | |
5155 | } | |
5156 | ||
09cbcef6 | 5157 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 5158 | { |
09cbcef6 MP |
5159 | int i; |
5160 | struct kvm_vcpu *vcpu; | |
ce35ef27 | 5161 | |
09cbcef6 | 5162 | *val = 0; |
ce35ef27 | 5163 | |
09cbcef6 | 5164 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5165 | *val += *(u64 *)((void *)(&vcpu->stat) + offset); |
ce35ef27 SJS |
5166 | |
5167 | return 0; | |
5168 | } | |
5169 | ||
09cbcef6 | 5170 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 5171 | { |
09cbcef6 MP |
5172 | int i; |
5173 | struct kvm_vcpu *vcpu; | |
536a6f88 | 5174 | |
09cbcef6 | 5175 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5176 | *(u64 *)((void *)(&vcpu->stat) + offset) = 0; |
09cbcef6 MP |
5177 | |
5178 | return 0; | |
5179 | } | |
536a6f88 | 5180 | |
09cbcef6 | 5181 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 5182 | { |
09cbcef6 | 5183 | int r = -EFAULT; |
536a6f88 | 5184 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
536a6f88 | 5185 | |
bc9e9e67 | 5186 | switch (stat_data->kind) { |
09cbcef6 MP |
5187 | case KVM_STAT_VM: |
5188 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5189 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5190 | break; |
5191 | case KVM_STAT_VCPU: | |
5192 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5193 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5194 | break; |
5195 | } | |
536a6f88 | 5196 | |
09cbcef6 | 5197 | return r; |
536a6f88 JF |
5198 | } |
5199 | ||
09cbcef6 | 5200 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 5201 | { |
09cbcef6 | 5202 | int r = -EFAULT; |
ce35ef27 | 5203 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
ce35ef27 SJS |
5204 | |
5205 | if (val) | |
5206 | return -EINVAL; | |
5207 | ||
bc9e9e67 | 5208 | switch (stat_data->kind) { |
09cbcef6 MP |
5209 | case KVM_STAT_VM: |
5210 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5211 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5212 | break; |
5213 | case KVM_STAT_VCPU: | |
5214 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5215 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5216 | break; |
5217 | } | |
ce35ef27 | 5218 | |
09cbcef6 | 5219 | return r; |
ce35ef27 SJS |
5220 | } |
5221 | ||
09cbcef6 | 5222 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
5223 | { |
5224 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
5225 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
5226 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
5227 | } |
5228 | ||
09cbcef6 MP |
5229 | static const struct file_operations stat_fops_per_vm = { |
5230 | .owner = THIS_MODULE, | |
5231 | .open = kvm_stat_data_open, | |
536a6f88 | 5232 | .release = kvm_debugfs_release, |
09cbcef6 MP |
5233 | .read = simple_attr_read, |
5234 | .write = simple_attr_write, | |
5235 | .llseek = no_llseek, | |
536a6f88 JF |
5236 | }; |
5237 | ||
8b88b099 | 5238 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
5239 | { |
5240 | unsigned offset = (long)_offset; | |
ba1389b7 | 5241 | struct kvm *kvm; |
536a6f88 | 5242 | u64 tmp_val; |
ba1389b7 | 5243 | |
8b88b099 | 5244 | *val = 0; |
0d9ce162 | 5245 | mutex_lock(&kvm_lock); |
536a6f88 | 5246 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5247 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
5248 | *val += tmp_val; |
5249 | } | |
0d9ce162 | 5250 | mutex_unlock(&kvm_lock); |
8b88b099 | 5251 | return 0; |
ba1389b7 AK |
5252 | } |
5253 | ||
ce35ef27 SJS |
5254 | static int vm_stat_clear(void *_offset, u64 val) |
5255 | { | |
5256 | unsigned offset = (long)_offset; | |
5257 | struct kvm *kvm; | |
ce35ef27 SJS |
5258 | |
5259 | if (val) | |
5260 | return -EINVAL; | |
5261 | ||
0d9ce162 | 5262 | mutex_lock(&kvm_lock); |
ce35ef27 | 5263 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5264 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 5265 | } |
0d9ce162 | 5266 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5267 | |
5268 | return 0; | |
5269 | } | |
5270 | ||
5271 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
bc9e9e67 | 5272 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n"); |
ba1389b7 | 5273 | |
8b88b099 | 5274 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
5275 | { |
5276 | unsigned offset = (long)_offset; | |
1165f5fe | 5277 | struct kvm *kvm; |
536a6f88 | 5278 | u64 tmp_val; |
1165f5fe | 5279 | |
8b88b099 | 5280 | *val = 0; |
0d9ce162 | 5281 | mutex_lock(&kvm_lock); |
536a6f88 | 5282 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5283 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
5284 | *val += tmp_val; |
5285 | } | |
0d9ce162 | 5286 | mutex_unlock(&kvm_lock); |
8b88b099 | 5287 | return 0; |
1165f5fe AK |
5288 | } |
5289 | ||
ce35ef27 SJS |
5290 | static int vcpu_stat_clear(void *_offset, u64 val) |
5291 | { | |
5292 | unsigned offset = (long)_offset; | |
5293 | struct kvm *kvm; | |
ce35ef27 SJS |
5294 | |
5295 | if (val) | |
5296 | return -EINVAL; | |
5297 | ||
0d9ce162 | 5298 | mutex_lock(&kvm_lock); |
ce35ef27 | 5299 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5300 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 5301 | } |
0d9ce162 | 5302 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5303 | |
5304 | return 0; | |
5305 | } | |
5306 | ||
5307 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
5308 | "%llu\n"); | |
bc9e9e67 | 5309 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n"); |
1165f5fe | 5310 | |
286de8f6 CI |
5311 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
5312 | { | |
5313 | struct kobj_uevent_env *env; | |
286de8f6 CI |
5314 | unsigned long long created, active; |
5315 | ||
5316 | if (!kvm_dev.this_device || !kvm) | |
5317 | return; | |
5318 | ||
0d9ce162 | 5319 | mutex_lock(&kvm_lock); |
286de8f6 CI |
5320 | if (type == KVM_EVENT_CREATE_VM) { |
5321 | kvm_createvm_count++; | |
5322 | kvm_active_vms++; | |
5323 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
5324 | kvm_active_vms--; | |
5325 | } | |
5326 | created = kvm_createvm_count; | |
5327 | active = kvm_active_vms; | |
0d9ce162 | 5328 | mutex_unlock(&kvm_lock); |
286de8f6 | 5329 | |
b12ce36a | 5330 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
5331 | if (!env) |
5332 | return; | |
5333 | ||
5334 | add_uevent_var(env, "CREATED=%llu", created); | |
5335 | add_uevent_var(env, "COUNT=%llu", active); | |
5336 | ||
fdeaf7e3 | 5337 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 5338 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
5339 | kvm->userspace_pid = task_pid_nr(current); |
5340 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 5341 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
5342 | } |
5343 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 5344 | |
85cd39af | 5345 | if (kvm->debugfs_dentry) { |
b12ce36a | 5346 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
5347 | |
5348 | if (p) { | |
5349 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
5350 | if (!IS_ERR(tmp)) | |
5351 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
5352 | kfree(p); | |
286de8f6 CI |
5353 | } |
5354 | } | |
5355 | /* no need for checks, since we are adding at most only 5 keys */ | |
5356 | env->envp[env->envp_idx++] = NULL; | |
5357 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
5358 | kfree(env); | |
286de8f6 CI |
5359 | } |
5360 | ||
929f45e3 | 5361 | static void kvm_init_debug(void) |
6aa8b732 | 5362 | { |
bc9e9e67 JZ |
5363 | const struct file_operations *fops; |
5364 | const struct _kvm_stats_desc *pdesc; | |
5365 | int i; | |
6aa8b732 | 5366 | |
76f7c879 | 5367 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 5368 | |
bc9e9e67 JZ |
5369 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
5370 | pdesc = &kvm_vm_stats_desc[i]; | |
5371 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5372 | fops = &vm_stat_fops; | |
5373 | else | |
5374 | fops = &vm_stat_readonly_fops; | |
5375 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5376 | kvm_debugfs_dir, | |
5377 | (void *)(long)pdesc->desc.offset, fops); | |
5378 | } | |
5379 | ||
5380 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
5381 | pdesc = &kvm_vcpu_stats_desc[i]; | |
5382 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5383 | fops = &vcpu_stat_fops; | |
5384 | else | |
5385 | fops = &vcpu_stat_readonly_fops; | |
5386 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5387 | kvm_debugfs_dir, | |
5388 | (void *)(long)pdesc->desc.offset, fops); | |
4f69b680 | 5389 | } |
6aa8b732 AK |
5390 | } |
5391 | ||
fb3600cc | 5392 | static int kvm_suspend(void) |
59ae6c6b | 5393 | { |
10474ae8 | 5394 | if (kvm_usage_count) |
75b7127c | 5395 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
5396 | return 0; |
5397 | } | |
5398 | ||
fb3600cc | 5399 | static void kvm_resume(void) |
59ae6c6b | 5400 | { |
ca84d1a2 | 5401 | if (kvm_usage_count) { |
2eb06c30 WL |
5402 | #ifdef CONFIG_LOCKDEP |
5403 | WARN_ON(lockdep_is_held(&kvm_count_lock)); | |
5404 | #endif | |
75b7127c | 5405 | hardware_enable_nolock(NULL); |
ca84d1a2 | 5406 | } |
59ae6c6b AK |
5407 | } |
5408 | ||
fb3600cc | 5409 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
5410 | .suspend = kvm_suspend, |
5411 | .resume = kvm_resume, | |
5412 | }; | |
5413 | ||
15ad7146 AK |
5414 | static inline |
5415 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
5416 | { | |
5417 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
5418 | } | |
5419 | ||
5420 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
5421 | { | |
5422 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 5423 | |
046ddeed | 5424 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 5425 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 5426 | |
7495e22b | 5427 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 5428 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 5429 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
5430 | } |
5431 | ||
5432 | static void kvm_sched_out(struct preempt_notifier *pn, | |
5433 | struct task_struct *next) | |
5434 | { | |
5435 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
5436 | ||
3ba9f93b | 5437 | if (current->on_rq) { |
046ddeed | 5438 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
5439 | WRITE_ONCE(vcpu->ready, true); |
5440 | } | |
e9b11c17 | 5441 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
5442 | __this_cpu_write(kvm_running_vcpu, NULL); |
5443 | } | |
5444 | ||
5445 | /** | |
5446 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
5447 | * |
5448 | * We can disable preemption locally around accessing the per-CPU variable, | |
5449 | * and use the resolved vcpu pointer after enabling preemption again, | |
5450 | * because even if the current thread is migrated to another CPU, reading | |
5451 | * the per-CPU value later will give us the same value as we update the | |
5452 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
5453 | */ |
5454 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
5455 | { | |
1f03b2bc MZ |
5456 | struct kvm_vcpu *vcpu; |
5457 | ||
5458 | preempt_disable(); | |
5459 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
5460 | preempt_enable(); | |
5461 | ||
5462 | return vcpu; | |
7495e22b | 5463 | } |
379a3c8e | 5464 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
5465 | |
5466 | /** | |
5467 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
5468 | */ | |
5469 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
5470 | { | |
5471 | return &kvm_running_vcpu; | |
15ad7146 AK |
5472 | } |
5473 | ||
b9904085 SC |
5474 | struct kvm_cpu_compat_check { |
5475 | void *opaque; | |
5476 | int *ret; | |
5477 | }; | |
5478 | ||
5479 | static void check_processor_compat(void *data) | |
f257d6dc | 5480 | { |
b9904085 SC |
5481 | struct kvm_cpu_compat_check *c = data; |
5482 | ||
5483 | *c->ret = kvm_arch_check_processor_compat(c->opaque); | |
f257d6dc SC |
5484 | } |
5485 | ||
0ee75bea | 5486 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 5487 | struct module *module) |
6aa8b732 | 5488 | { |
b9904085 | 5489 | struct kvm_cpu_compat_check c; |
6aa8b732 | 5490 | int r; |
002c7f7c | 5491 | int cpu; |
6aa8b732 | 5492 | |
f8c16bba ZX |
5493 | r = kvm_arch_init(opaque); |
5494 | if (r) | |
d2308784 | 5495 | goto out_fail; |
cb498ea2 | 5496 | |
7dac16c3 AH |
5497 | /* |
5498 | * kvm_arch_init makes sure there's at most one caller | |
5499 | * for architectures that support multiple implementations, | |
5500 | * like intel and amd on x86. | |
36343f6e PB |
5501 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
5502 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 5503 | */ |
36343f6e PB |
5504 | r = kvm_irqfd_init(); |
5505 | if (r) | |
5506 | goto out_irqfd; | |
7dac16c3 | 5507 | |
8437a617 | 5508 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
5509 | r = -ENOMEM; |
5510 | goto out_free_0; | |
5511 | } | |
5512 | ||
b9904085 | 5513 | r = kvm_arch_hardware_setup(opaque); |
6aa8b732 | 5514 | if (r < 0) |
faf0be22 | 5515 | goto out_free_1; |
6aa8b732 | 5516 | |
b9904085 SC |
5517 | c.ret = &r; |
5518 | c.opaque = opaque; | |
002c7f7c | 5519 | for_each_online_cpu(cpu) { |
b9904085 | 5520 | smp_call_function_single(cpu, check_processor_compat, &c, 1); |
002c7f7c | 5521 | if (r < 0) |
faf0be22 | 5522 | goto out_free_2; |
002c7f7c YS |
5523 | } |
5524 | ||
73c1b41e | 5525 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 5526 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 5527 | if (r) |
d2308784 | 5528 | goto out_free_2; |
6aa8b732 AK |
5529 | register_reboot_notifier(&kvm_reboot_notifier); |
5530 | ||
c16f862d | 5531 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
5532 | if (!vcpu_align) |
5533 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
5534 | kvm_vcpu_cache = |
5535 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
5536 | SLAB_ACCOUNT, | |
5537 | offsetof(struct kvm_vcpu, arch), | |
ce55c049 JZ |
5538 | offsetofend(struct kvm_vcpu, stats_id) |
5539 | - offsetof(struct kvm_vcpu, arch), | |
46515736 | 5540 | NULL); |
c16f862d RR |
5541 | if (!kvm_vcpu_cache) { |
5542 | r = -ENOMEM; | |
fb3600cc | 5543 | goto out_free_3; |
c16f862d RR |
5544 | } |
5545 | ||
af585b92 GN |
5546 | r = kvm_async_pf_init(); |
5547 | if (r) | |
5548 | goto out_free; | |
5549 | ||
6aa8b732 | 5550 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
5551 | kvm_vm_fops.owner = module; |
5552 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
5553 | |
5554 | r = misc_register(&kvm_dev); | |
5555 | if (r) { | |
1170adc6 | 5556 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 5557 | goto out_unreg; |
6aa8b732 AK |
5558 | } |
5559 | ||
fb3600cc RW |
5560 | register_syscore_ops(&kvm_syscore_ops); |
5561 | ||
15ad7146 AK |
5562 | kvm_preempt_ops.sched_in = kvm_sched_in; |
5563 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
5564 | ||
929f45e3 | 5565 | kvm_init_debug(); |
0ea4ed8e | 5566 | |
3c3c29fd PB |
5567 | r = kvm_vfio_ops_init(); |
5568 | WARN_ON(r); | |
5569 | ||
c7addb90 | 5570 | return 0; |
6aa8b732 | 5571 | |
af585b92 GN |
5572 | out_unreg: |
5573 | kvm_async_pf_deinit(); | |
6aa8b732 | 5574 | out_free: |
c16f862d | 5575 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 5576 | out_free_3: |
6aa8b732 | 5577 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5578 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 5579 | out_free_2: |
e9b11c17 | 5580 | kvm_arch_hardware_unsetup(); |
faf0be22 | 5581 | out_free_1: |
7f59f492 | 5582 | free_cpumask_var(cpus_hardware_enabled); |
d2308784 | 5583 | out_free_0: |
a0f155e9 | 5584 | kvm_irqfd_exit(); |
36343f6e | 5585 | out_irqfd: |
7dac16c3 AH |
5586 | kvm_arch_exit(); |
5587 | out_fail: | |
6aa8b732 AK |
5588 | return r; |
5589 | } | |
cb498ea2 | 5590 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 5591 | |
cb498ea2 | 5592 | void kvm_exit(void) |
6aa8b732 | 5593 | { |
4bd33b56 | 5594 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 5595 | misc_deregister(&kvm_dev); |
c16f862d | 5596 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 5597 | kvm_async_pf_deinit(); |
fb3600cc | 5598 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 5599 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5600 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 5601 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 5602 | kvm_arch_hardware_unsetup(); |
f8c16bba | 5603 | kvm_arch_exit(); |
a0f155e9 | 5604 | kvm_irqfd_exit(); |
7f59f492 | 5605 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 5606 | kvm_vfio_ops_exit(); |
6aa8b732 | 5607 | } |
cb498ea2 | 5608 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
5609 | |
5610 | struct kvm_vm_worker_thread_context { | |
5611 | struct kvm *kvm; | |
5612 | struct task_struct *parent; | |
5613 | struct completion init_done; | |
5614 | kvm_vm_thread_fn_t thread_fn; | |
5615 | uintptr_t data; | |
5616 | int err; | |
5617 | }; | |
5618 | ||
5619 | static int kvm_vm_worker_thread(void *context) | |
5620 | { | |
5621 | /* | |
5622 | * The init_context is allocated on the stack of the parent thread, so | |
5623 | * we have to locally copy anything that is needed beyond initialization | |
5624 | */ | |
5625 | struct kvm_vm_worker_thread_context *init_context = context; | |
5626 | struct kvm *kvm = init_context->kvm; | |
5627 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
5628 | uintptr_t data = init_context->data; | |
5629 | int err; | |
5630 | ||
5631 | err = kthread_park(current); | |
5632 | /* kthread_park(current) is never supposed to return an error */ | |
5633 | WARN_ON(err != 0); | |
5634 | if (err) | |
5635 | goto init_complete; | |
5636 | ||
5637 | err = cgroup_attach_task_all(init_context->parent, current); | |
5638 | if (err) { | |
5639 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
5640 | __func__, err); | |
5641 | goto init_complete; | |
5642 | } | |
5643 | ||
5644 | set_user_nice(current, task_nice(init_context->parent)); | |
5645 | ||
5646 | init_complete: | |
5647 | init_context->err = err; | |
5648 | complete(&init_context->init_done); | |
5649 | init_context = NULL; | |
5650 | ||
5651 | if (err) | |
5652 | return err; | |
5653 | ||
5654 | /* Wait to be woken up by the spawner before proceeding. */ | |
5655 | kthread_parkme(); | |
5656 | ||
5657 | if (!kthread_should_stop()) | |
5658 | err = thread_fn(kvm, data); | |
5659 | ||
5660 | return err; | |
5661 | } | |
5662 | ||
5663 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
5664 | uintptr_t data, const char *name, | |
5665 | struct task_struct **thread_ptr) | |
5666 | { | |
5667 | struct kvm_vm_worker_thread_context init_context = {}; | |
5668 | struct task_struct *thread; | |
5669 | ||
5670 | *thread_ptr = NULL; | |
5671 | init_context.kvm = kvm; | |
5672 | init_context.parent = current; | |
5673 | init_context.thread_fn = thread_fn; | |
5674 | init_context.data = data; | |
5675 | init_completion(&init_context.init_done); | |
5676 | ||
5677 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
5678 | "%s-%d", name, task_pid_nr(current)); | |
5679 | if (IS_ERR(thread)) | |
5680 | return PTR_ERR(thread); | |
5681 | ||
5682 | /* kthread_run is never supposed to return NULL */ | |
5683 | WARN_ON(thread == NULL); | |
5684 | ||
5685 | wait_for_completion(&init_context.init_done); | |
5686 | ||
5687 | if (!init_context.err) | |
5688 | *thread_ptr = thread; | |
5689 | ||
5690 | return init_context.err; | |
5691 | } |