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 | ||
baff59cc VK |
158 | static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask); |
159 | ||
e649b3f0 ET |
160 | __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
161 | unsigned long start, unsigned long end) | |
b1394e74 RK |
162 | { |
163 | } | |
164 | ||
a78986aa SC |
165 | bool kvm_is_zone_device_pfn(kvm_pfn_t pfn) |
166 | { | |
167 | /* | |
168 | * The metadata used by is_zone_device_page() to determine whether or | |
169 | * not a page is ZONE_DEVICE is guaranteed to be valid if and only if | |
170 | * the device has been pinned, e.g. by get_user_pages(). WARN if the | |
171 | * page_count() is zero to help detect bad usage of this helper. | |
172 | */ | |
173 | if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn)))) | |
174 | return false; | |
175 | ||
176 | return is_zone_device_page(pfn_to_page(pfn)); | |
177 | } | |
178 | ||
ba049e93 | 179 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
cbff90a7 | 180 | { |
a78986aa SC |
181 | /* |
182 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting | |
183 | * perspective they are "normal" pages, albeit with slightly different | |
184 | * usage rules. | |
185 | */ | |
11feeb49 | 186 | if (pfn_valid(pfn)) |
a78986aa | 187 | return PageReserved(pfn_to_page(pfn)) && |
7df003c8 | 188 | !is_zero_pfn(pfn) && |
a78986aa | 189 | !kvm_is_zone_device_pfn(pfn); |
cbff90a7 BAY |
190 | |
191 | return true; | |
192 | } | |
193 | ||
bccf2150 AK |
194 | /* |
195 | * Switches to specified vcpu, until a matching vcpu_put() | |
196 | */ | |
ec7660cc | 197 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 198 | { |
ec7660cc | 199 | int cpu = get_cpu(); |
7495e22b PB |
200 | |
201 | __this_cpu_write(kvm_running_vcpu, vcpu); | |
15ad7146 | 202 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 203 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 204 | put_cpu(); |
6aa8b732 | 205 | } |
2f1fe811 | 206 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 207 | |
313a3dc7 | 208 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 209 | { |
15ad7146 | 210 | preempt_disable(); |
313a3dc7 | 211 | kvm_arch_vcpu_put(vcpu); |
15ad7146 | 212 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
7495e22b | 213 | __this_cpu_write(kvm_running_vcpu, NULL); |
15ad7146 | 214 | preempt_enable(); |
6aa8b732 | 215 | } |
2f1fe811 | 216 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 217 | |
7a97cec2 PB |
218 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
219 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
220 | { | |
221 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
222 | ||
223 | /* | |
224 | * We need to wait for the VCPU to reenable interrupts and get out of | |
225 | * READING_SHADOW_PAGE_TABLES mode. | |
226 | */ | |
227 | if (req & KVM_REQUEST_WAIT) | |
228 | return mode != OUTSIDE_GUEST_MODE; | |
229 | ||
230 | /* | |
231 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
232 | */ | |
233 | return mode == IN_GUEST_MODE; | |
234 | } | |
235 | ||
d9e368d6 AK |
236 | static void ack_flush(void *_completed) |
237 | { | |
d9e368d6 AK |
238 | } |
239 | ||
620b2438 | 240 | static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait) |
b49defe8 | 241 | { |
b49defe8 PB |
242 | if (cpumask_empty(cpus)) |
243 | return false; | |
244 | ||
245 | smp_call_function_many(cpus, ack_flush, NULL, wait); | |
246 | return true; | |
247 | } | |
248 | ||
ae0946cd | 249 | static void kvm_make_vcpu_request(struct kvm *kvm, struct kvm_vcpu *vcpu, |
620b2438 | 250 | unsigned int req, struct cpumask *tmp, |
ae0946cd VK |
251 | int current_cpu) |
252 | { | |
253 | int cpu; | |
254 | ||
255 | kvm_make_request(req, vcpu); | |
256 | ||
257 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) | |
258 | return; | |
259 | ||
ae0946cd VK |
260 | /* |
261 | * Note, the vCPU could get migrated to a different pCPU at any point | |
262 | * after kvm_request_needs_ipi(), which could result in sending an IPI | |
263 | * to the previous pCPU. But, that's OK because the purpose of the IPI | |
264 | * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is | |
265 | * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES | |
266 | * after this point is also OK, as the requirement is only that KVM wait | |
267 | * for vCPUs that were reading SPTEs _before_ any changes were | |
268 | * finalized. See kvm_vcpu_kick() for more details on handling requests. | |
269 | */ | |
270 | if (kvm_request_needs_ipi(vcpu, req)) { | |
271 | cpu = READ_ONCE(vcpu->cpu); | |
272 | if (cpu != -1 && cpu != current_cpu) | |
273 | __cpumask_set_cpu(cpu, tmp); | |
274 | } | |
275 | } | |
276 | ||
7053df4e | 277 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
620b2438 | 278 | unsigned long *vcpu_bitmap) |
d9e368d6 | 279 | { |
d9e368d6 | 280 | struct kvm_vcpu *vcpu; |
620b2438 | 281 | struct cpumask *cpus; |
ae0946cd | 282 | int i, me; |
7053df4e | 283 | bool called; |
6ef7a1bc | 284 | |
3cba4130 | 285 | me = get_cpu(); |
7053df4e | 286 | |
620b2438 VK |
287 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
288 | cpumask_clear(cpus); | |
289 | ||
ae0946cd VK |
290 | for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) { |
291 | vcpu = kvm_get_vcpu(kvm, i); | |
381cecc5 | 292 | if (!vcpu) |
7053df4e | 293 | continue; |
620b2438 | 294 | kvm_make_vcpu_request(kvm, vcpu, req, cpus, me); |
49846896 | 295 | } |
7053df4e | 296 | |
620b2438 | 297 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); |
3cba4130 | 298 | put_cpu(); |
7053df4e VK |
299 | |
300 | return called; | |
301 | } | |
302 | ||
54163a34 SS |
303 | bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req, |
304 | struct kvm_vcpu *except) | |
7053df4e | 305 | { |
ae0946cd | 306 | struct kvm_vcpu *vcpu; |
baff59cc | 307 | struct cpumask *cpus; |
7053df4e | 308 | bool called; |
ae0946cd | 309 | int i, me; |
7053df4e | 310 | |
ae0946cd VK |
311 | me = get_cpu(); |
312 | ||
baff59cc VK |
313 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
314 | cpumask_clear(cpus); | |
315 | ||
ae0946cd VK |
316 | kvm_for_each_vcpu(i, vcpu, kvm) { |
317 | if (vcpu == except) | |
318 | continue; | |
319 | kvm_make_vcpu_request(kvm, vcpu, req, cpus, me); | |
320 | } | |
321 | ||
322 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); | |
323 | put_cpu(); | |
7053df4e | 324 | |
49846896 | 325 | return called; |
d9e368d6 AK |
326 | } |
327 | ||
54163a34 SS |
328 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
329 | { | |
330 | return kvm_make_all_cpus_request_except(kvm, req, NULL); | |
331 | } | |
a2486020 | 332 | EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request); |
54163a34 | 333 | |
a6d51016 | 334 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 335 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 336 | { |
3cc4e148 | 337 | ++kvm->stat.generic.remote_tlb_flush_requests; |
6bc6db00 | 338 | |
4ae3cb3a LT |
339 | /* |
340 | * We want to publish modifications to the page tables before reading | |
341 | * mode. Pairs with a memory barrier in arch-specific code. | |
342 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
343 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
344 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
345 | * | |
346 | * There is already an smp_mb__after_atomic() before | |
347 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
348 | * barrier here. | |
349 | */ | |
b08660e5 TL |
350 | if (!kvm_arch_flush_remote_tlb(kvm) |
351 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
0193cc90 | 352 | ++kvm->stat.generic.remote_tlb_flush; |
2e53d63a | 353 | } |
2ba9f0d8 | 354 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 355 | #endif |
2e53d63a | 356 | |
49846896 RR |
357 | void kvm_reload_remote_mmus(struct kvm *kvm) |
358 | { | |
445b8236 | 359 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 360 | } |
2e53d63a | 361 | |
6926f95a SC |
362 | #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE |
363 | static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc, | |
364 | gfp_t gfp_flags) | |
365 | { | |
366 | gfp_flags |= mc->gfp_zero; | |
367 | ||
368 | if (mc->kmem_cache) | |
369 | return kmem_cache_alloc(mc->kmem_cache, gfp_flags); | |
370 | else | |
371 | return (void *)__get_free_page(gfp_flags); | |
372 | } | |
373 | ||
374 | int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min) | |
375 | { | |
376 | void *obj; | |
377 | ||
378 | if (mc->nobjs >= min) | |
379 | return 0; | |
380 | while (mc->nobjs < ARRAY_SIZE(mc->objects)) { | |
381 | obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT); | |
382 | if (!obj) | |
383 | return mc->nobjs >= min ? 0 : -ENOMEM; | |
384 | mc->objects[mc->nobjs++] = obj; | |
385 | } | |
386 | return 0; | |
387 | } | |
388 | ||
389 | int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc) | |
390 | { | |
391 | return mc->nobjs; | |
392 | } | |
393 | ||
394 | void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
395 | { | |
396 | while (mc->nobjs) { | |
397 | if (mc->kmem_cache) | |
398 | kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]); | |
399 | else | |
400 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
401 | } | |
402 | } | |
403 | ||
404 | void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
405 | { | |
406 | void *p; | |
407 | ||
408 | if (WARN_ON(!mc->nobjs)) | |
409 | p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT); | |
410 | else | |
411 | p = mc->objects[--mc->nobjs]; | |
412 | BUG_ON(!p); | |
413 | return p; | |
414 | } | |
415 | #endif | |
416 | ||
8bd826d6 | 417 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 418 | { |
fb3f0f51 RR |
419 | mutex_init(&vcpu->mutex); |
420 | vcpu->cpu = -1; | |
fb3f0f51 RR |
421 | vcpu->kvm = kvm; |
422 | vcpu->vcpu_id = id; | |
34bb10b7 | 423 | vcpu->pid = NULL; |
da4ad88c | 424 | rcuwait_init(&vcpu->wait); |
af585b92 | 425 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 426 | |
bf9f6ac8 FW |
427 | vcpu->pre_pcpu = -1; |
428 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
429 | ||
4c088493 R |
430 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
431 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 432 | vcpu->preempted = false; |
d73eb57b | 433 | vcpu->ready = false; |
d5c48deb | 434 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
fe22ed82 | 435 | vcpu->last_used_slot = 0; |
fb3f0f51 | 436 | } |
fb3f0f51 | 437 | |
4543bdc0 SC |
438 | void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
439 | { | |
fb04a1ed | 440 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
4543bdc0 | 441 | kvm_arch_vcpu_destroy(vcpu); |
e529ef66 | 442 | |
9941d224 SC |
443 | /* |
444 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
445 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
446 | * are already gone. | |
447 | */ | |
448 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
449 | ||
8bd826d6 | 450 | free_page((unsigned long)vcpu->run); |
e529ef66 | 451 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 SC |
452 | } |
453 | EXPORT_SYMBOL_GPL(kvm_vcpu_destroy); | |
454 | ||
e930bffe AA |
455 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
456 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
457 | { | |
458 | return container_of(mn, struct kvm, mmu_notifier); | |
459 | } | |
460 | ||
e649b3f0 ET |
461 | static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn, |
462 | struct mm_struct *mm, | |
463 | unsigned long start, unsigned long end) | |
464 | { | |
465 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
466 | int idx; | |
467 | ||
468 | idx = srcu_read_lock(&kvm->srcu); | |
469 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
470 | srcu_read_unlock(&kvm->srcu, idx); | |
471 | } | |
472 | ||
3039bcc7 SC |
473 | typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range); |
474 | ||
f922bd9b SC |
475 | typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start, |
476 | unsigned long end); | |
477 | ||
3039bcc7 SC |
478 | struct kvm_hva_range { |
479 | unsigned long start; | |
480 | unsigned long end; | |
481 | pte_t pte; | |
482 | hva_handler_t handler; | |
f922bd9b | 483 | on_lock_fn_t on_lock; |
3039bcc7 SC |
484 | bool flush_on_ret; |
485 | bool may_block; | |
486 | }; | |
487 | ||
f922bd9b SC |
488 | /* |
489 | * Use a dedicated stub instead of NULL to indicate that there is no callback | |
490 | * function/handler. The compiler technically can't guarantee that a real | |
491 | * function will have a non-zero address, and so it will generate code to | |
492 | * check for !NULL, whereas comparing against a stub will be elided at compile | |
493 | * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9). | |
494 | */ | |
495 | static void kvm_null_fn(void) | |
496 | { | |
497 | ||
498 | } | |
499 | #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn) | |
500 | ||
3039bcc7 SC |
501 | static __always_inline int __kvm_handle_hva_range(struct kvm *kvm, |
502 | const struct kvm_hva_range *range) | |
503 | { | |
8931a454 | 504 | bool ret = false, locked = false; |
f922bd9b | 505 | struct kvm_gfn_range gfn_range; |
3039bcc7 SC |
506 | struct kvm_memory_slot *slot; |
507 | struct kvm_memslots *slots; | |
3039bcc7 SC |
508 | int i, idx; |
509 | ||
f922bd9b SC |
510 | /* A null handler is allowed if and only if on_lock() is provided. */ |
511 | if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) && | |
512 | IS_KVM_NULL_FN(range->handler))) | |
513 | return 0; | |
514 | ||
3039bcc7 SC |
515 | idx = srcu_read_lock(&kvm->srcu); |
516 | ||
517 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
518 | slots = __kvm_memslots(kvm, i); | |
519 | kvm_for_each_memslot(slot, slots) { | |
520 | unsigned long hva_start, hva_end; | |
521 | ||
522 | hva_start = max(range->start, slot->userspace_addr); | |
523 | hva_end = min(range->end, slot->userspace_addr + | |
524 | (slot->npages << PAGE_SHIFT)); | |
525 | if (hva_start >= hva_end) | |
526 | continue; | |
527 | ||
528 | /* | |
529 | * To optimize for the likely case where the address | |
530 | * range is covered by zero or one memslots, don't | |
531 | * bother making these conditional (to avoid writes on | |
532 | * the second or later invocation of the handler). | |
533 | */ | |
534 | gfn_range.pte = range->pte; | |
535 | gfn_range.may_block = range->may_block; | |
536 | ||
537 | /* | |
538 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
539 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
540 | */ | |
541 | gfn_range.start = hva_to_gfn_memslot(hva_start, slot); | |
542 | gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot); | |
543 | gfn_range.slot = slot; | |
544 | ||
8931a454 SC |
545 | if (!locked) { |
546 | locked = true; | |
547 | KVM_MMU_LOCK(kvm); | |
071064f1 PB |
548 | if (!IS_KVM_NULL_FN(range->on_lock)) |
549 | range->on_lock(kvm, range->start, range->end); | |
550 | if (IS_KVM_NULL_FN(range->handler)) | |
551 | break; | |
8931a454 | 552 | } |
3039bcc7 SC |
553 | ret |= range->handler(kvm, &gfn_range); |
554 | } | |
555 | } | |
556 | ||
6bc6db00 | 557 | if (range->flush_on_ret && ret) |
3039bcc7 SC |
558 | kvm_flush_remote_tlbs(kvm); |
559 | ||
8931a454 SC |
560 | if (locked) |
561 | KVM_MMU_UNLOCK(kvm); | |
f922bd9b | 562 | |
3039bcc7 SC |
563 | srcu_read_unlock(&kvm->srcu, idx); |
564 | ||
565 | /* The notifiers are averse to booleans. :-( */ | |
566 | return (int)ret; | |
567 | } | |
568 | ||
569 | static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn, | |
570 | unsigned long start, | |
571 | unsigned long end, | |
572 | pte_t pte, | |
573 | hva_handler_t handler) | |
574 | { | |
575 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
576 | const struct kvm_hva_range range = { | |
577 | .start = start, | |
578 | .end = end, | |
579 | .pte = pte, | |
580 | .handler = handler, | |
f922bd9b | 581 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
582 | .flush_on_ret = true, |
583 | .may_block = false, | |
584 | }; | |
3039bcc7 | 585 | |
f922bd9b | 586 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 SC |
587 | } |
588 | ||
589 | static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn, | |
590 | unsigned long start, | |
591 | unsigned long end, | |
592 | hva_handler_t handler) | |
593 | { | |
594 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
595 | const struct kvm_hva_range range = { | |
596 | .start = start, | |
597 | .end = end, | |
598 | .pte = __pte(0), | |
599 | .handler = handler, | |
f922bd9b | 600 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
601 | .flush_on_ret = false, |
602 | .may_block = false, | |
603 | }; | |
3039bcc7 | 604 | |
f922bd9b | 605 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 | 606 | } |
3da0dd43 IE |
607 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
608 | struct mm_struct *mm, | |
609 | unsigned long address, | |
610 | pte_t pte) | |
611 | { | |
612 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
613 | ||
501b9185 SC |
614 | trace_kvm_set_spte_hva(address); |
615 | ||
c13fda23 | 616 | /* |
52ac8b35 | 617 | * .change_pte() must be surrounded by .invalidate_range_{start,end}(). |
071064f1 PB |
618 | * If mmu_notifier_count is zero, then no in-progress invalidations, |
619 | * including this one, found a relevant memslot at start(); rechecking | |
620 | * memslots here is unnecessary. Note, a false positive (count elevated | |
621 | * by a different invalidation) is sub-optimal but functionally ok. | |
c13fda23 | 622 | */ |
52ac8b35 | 623 | WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count)); |
071064f1 PB |
624 | if (!READ_ONCE(kvm->mmu_notifier_count)) |
625 | return; | |
c13fda23 | 626 | |
3039bcc7 | 627 | kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn); |
3da0dd43 IE |
628 | } |
629 | ||
edb298c6 | 630 | void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 631 | unsigned long end) |
e930bffe | 632 | { |
e930bffe AA |
633 | /* |
634 | * The count increase must become visible at unlock time as no | |
635 | * spte can be established without taking the mmu_lock and | |
636 | * count is also read inside the mmu_lock critical section. | |
637 | */ | |
638 | kvm->mmu_notifier_count++; | |
4a42d848 | 639 | if (likely(kvm->mmu_notifier_count == 1)) { |
f922bd9b SC |
640 | kvm->mmu_notifier_range_start = start; |
641 | kvm->mmu_notifier_range_end = end; | |
4a42d848 DS |
642 | } else { |
643 | /* | |
644 | * Fully tracking multiple concurrent ranges has dimishing | |
645 | * returns. Keep things simple and just find the minimal range | |
646 | * which includes the current and new ranges. As there won't be | |
647 | * enough information to subtract a range after its invalidate | |
648 | * completes, any ranges invalidated concurrently will | |
649 | * accumulate and persist until all outstanding invalidates | |
650 | * complete. | |
651 | */ | |
652 | kvm->mmu_notifier_range_start = | |
f922bd9b | 653 | min(kvm->mmu_notifier_range_start, start); |
4a42d848 | 654 | kvm->mmu_notifier_range_end = |
f922bd9b | 655 | max(kvm->mmu_notifier_range_end, end); |
4a42d848 | 656 | } |
f922bd9b | 657 | } |
3039bcc7 | 658 | |
f922bd9b SC |
659 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
660 | const struct mmu_notifier_range *range) | |
661 | { | |
662 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
663 | const struct kvm_hva_range hva_range = { | |
664 | .start = range->start, | |
665 | .end = range->end, | |
666 | .pte = __pte(0), | |
667 | .handler = kvm_unmap_gfn_range, | |
668 | .on_lock = kvm_inc_notifier_count, | |
669 | .flush_on_ret = true, | |
670 | .may_block = mmu_notifier_range_blockable(range), | |
671 | }; | |
565f3be2 | 672 | |
f922bd9b SC |
673 | trace_kvm_unmap_hva_range(range->start, range->end); |
674 | ||
52ac8b35 PB |
675 | /* |
676 | * Prevent memslot modification between range_start() and range_end() | |
677 | * so that conditionally locking provides the same result in both | |
678 | * functions. Without that guarantee, the mmu_notifier_count | |
679 | * adjustments will be imbalanced. | |
680 | * | |
681 | * Pairs with the decrement in range_end(). | |
682 | */ | |
683 | spin_lock(&kvm->mn_invalidate_lock); | |
684 | kvm->mn_active_invalidate_count++; | |
685 | spin_unlock(&kvm->mn_invalidate_lock); | |
686 | ||
f922bd9b | 687 | __kvm_handle_hva_range(kvm, &hva_range); |
93065ac7 | 688 | |
e649b3f0 | 689 | return 0; |
e930bffe AA |
690 | } |
691 | ||
edb298c6 | 692 | void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 693 | unsigned long end) |
e930bffe | 694 | { |
e930bffe AA |
695 | /* |
696 | * This sequence increase will notify the kvm page fault that | |
697 | * the page that is going to be mapped in the spte could have | |
698 | * been freed. | |
699 | */ | |
700 | kvm->mmu_notifier_seq++; | |
a355aa54 | 701 | smp_wmb(); |
e930bffe AA |
702 | /* |
703 | * The above sequence increase must be visible before the | |
a355aa54 PM |
704 | * below count decrease, which is ensured by the smp_wmb above |
705 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
706 | */ |
707 | kvm->mmu_notifier_count--; | |
f922bd9b SC |
708 | } |
709 | ||
710 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
711 | const struct mmu_notifier_range *range) | |
712 | { | |
713 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
714 | const struct kvm_hva_range hva_range = { | |
715 | .start = range->start, | |
716 | .end = range->end, | |
717 | .pte = __pte(0), | |
718 | .handler = (void *)kvm_null_fn, | |
719 | .on_lock = kvm_dec_notifier_count, | |
720 | .flush_on_ret = false, | |
721 | .may_block = mmu_notifier_range_blockable(range), | |
722 | }; | |
52ac8b35 | 723 | bool wake; |
f922bd9b SC |
724 | |
725 | __kvm_handle_hva_range(kvm, &hva_range); | |
e930bffe | 726 | |
52ac8b35 PB |
727 | /* Pairs with the increment in range_start(). */ |
728 | spin_lock(&kvm->mn_invalidate_lock); | |
729 | wake = (--kvm->mn_active_invalidate_count == 0); | |
730 | spin_unlock(&kvm->mn_invalidate_lock); | |
731 | ||
732 | /* | |
733 | * There can only be one waiter, since the wait happens under | |
734 | * slots_lock. | |
735 | */ | |
736 | if (wake) | |
737 | rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait); | |
738 | ||
e930bffe AA |
739 | BUG_ON(kvm->mmu_notifier_count < 0); |
740 | } | |
741 | ||
742 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
743 | struct mm_struct *mm, | |
57128468 ALC |
744 | unsigned long start, |
745 | unsigned long end) | |
e930bffe | 746 | { |
501b9185 SC |
747 | trace_kvm_age_hva(start, end); |
748 | ||
3039bcc7 | 749 | return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn); |
e930bffe AA |
750 | } |
751 | ||
1d7715c6 VD |
752 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
753 | struct mm_struct *mm, | |
754 | unsigned long start, | |
755 | unsigned long end) | |
756 | { | |
501b9185 SC |
757 | trace_kvm_age_hva(start, end); |
758 | ||
1d7715c6 VD |
759 | /* |
760 | * Even though we do not flush TLB, this will still adversely | |
761 | * affect performance on pre-Haswell Intel EPT, where there is | |
762 | * no EPT Access Bit to clear so that we have to tear down EPT | |
763 | * tables instead. If we find this unacceptable, we can always | |
764 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
765 | * do anything on clear_young. | |
766 | * | |
767 | * Also note that currently we never issue secondary TLB flushes | |
768 | * from clear_young, leaving this job up to the regular system | |
769 | * cadence. If we find this inaccurate, we might come up with a | |
770 | * more sophisticated heuristic later. | |
771 | */ | |
3039bcc7 | 772 | return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn); |
1d7715c6 VD |
773 | } |
774 | ||
8ee53820 AA |
775 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
776 | struct mm_struct *mm, | |
777 | unsigned long address) | |
778 | { | |
501b9185 SC |
779 | trace_kvm_test_age_hva(address); |
780 | ||
3039bcc7 SC |
781 | return kvm_handle_hva_range_no_flush(mn, address, address + 1, |
782 | kvm_test_age_gfn); | |
8ee53820 AA |
783 | } |
784 | ||
85db06e5 MT |
785 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
786 | struct mm_struct *mm) | |
787 | { | |
788 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
789 | int idx; |
790 | ||
791 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 792 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 793 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
794 | } |
795 | ||
e930bffe | 796 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e649b3f0 | 797 | .invalidate_range = kvm_mmu_notifier_invalidate_range, |
e930bffe AA |
798 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
799 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
800 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 801 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 802 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 803 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 804 | .release = kvm_mmu_notifier_release, |
e930bffe | 805 | }; |
4c07b0a4 AK |
806 | |
807 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
808 | { | |
809 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
810 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
811 | } | |
812 | ||
813 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
814 | ||
815 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
816 | { | |
817 | return 0; | |
818 | } | |
819 | ||
e930bffe AA |
820 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
821 | ||
2fdef3a2 SS |
822 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
823 | static int kvm_pm_notifier_call(struct notifier_block *bl, | |
824 | unsigned long state, | |
825 | void *unused) | |
826 | { | |
827 | struct kvm *kvm = container_of(bl, struct kvm, pm_notifier); | |
828 | ||
829 | return kvm_arch_pm_notifier(kvm, state); | |
830 | } | |
831 | ||
832 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
833 | { | |
834 | kvm->pm_notifier.notifier_call = kvm_pm_notifier_call; | |
835 | /* Suspend KVM before we suspend ftrace, RCU, etc. */ | |
836 | kvm->pm_notifier.priority = INT_MAX; | |
837 | register_pm_notifier(&kvm->pm_notifier); | |
838 | } | |
839 | ||
840 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
841 | { | |
842 | unregister_pm_notifier(&kvm->pm_notifier); | |
843 | } | |
844 | #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
845 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
846 | { | |
847 | } | |
848 | ||
849 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
850 | { | |
851 | } | |
852 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
853 | ||
a47d2b07 | 854 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
855 | { |
856 | int i; | |
a47d2b07 | 857 | struct kvm_memslots *slots; |
bf3e05bc | 858 | |
b12ce36a | 859 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
a47d2b07 PB |
860 | if (!slots) |
861 | return NULL; | |
862 | ||
bf3e05bc | 863 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
36947254 | 864 | slots->id_to_index[i] = -1; |
a47d2b07 PB |
865 | |
866 | return slots; | |
867 | } | |
868 | ||
869 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
870 | { | |
871 | if (!memslot->dirty_bitmap) | |
872 | return; | |
873 | ||
874 | kvfree(memslot->dirty_bitmap); | |
875 | memslot->dirty_bitmap = NULL; | |
876 | } | |
877 | ||
e96c81ee | 878 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 879 | { |
e96c81ee | 880 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 881 | |
e96c81ee | 882 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 883 | |
e96c81ee SC |
884 | slot->flags = 0; |
885 | slot->npages = 0; | |
a47d2b07 PB |
886 | } |
887 | ||
888 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
889 | { | |
890 | struct kvm_memory_slot *memslot; | |
891 | ||
892 | if (!slots) | |
893 | return; | |
894 | ||
895 | kvm_for_each_memslot(memslot, slots) | |
e96c81ee | 896 | kvm_free_memslot(kvm, memslot); |
a47d2b07 PB |
897 | |
898 | kvfree(slots); | |
bf3e05bc XG |
899 | } |
900 | ||
bc9e9e67 JZ |
901 | static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc) |
902 | { | |
903 | switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) { | |
904 | case KVM_STATS_TYPE_INSTANT: | |
905 | return 0444; | |
906 | case KVM_STATS_TYPE_CUMULATIVE: | |
907 | case KVM_STATS_TYPE_PEAK: | |
908 | default: | |
909 | return 0644; | |
910 | } | |
911 | } | |
912 | ||
913 | ||
536a6f88 JF |
914 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
915 | { | |
916 | int i; | |
bc9e9e67 JZ |
917 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
918 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
919 | |
920 | if (!kvm->debugfs_dentry) | |
921 | return; | |
922 | ||
923 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
924 | ||
9d5a1dce LC |
925 | if (kvm->debugfs_stat_data) { |
926 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
927 | kfree(kvm->debugfs_stat_data[i]); | |
928 | kfree(kvm->debugfs_stat_data); | |
929 | } | |
536a6f88 JF |
930 | } |
931 | ||
932 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
933 | { | |
85cd39af PB |
934 | static DEFINE_MUTEX(kvm_debugfs_lock); |
935 | struct dentry *dent; | |
536a6f88 JF |
936 | char dir_name[ITOA_MAX_LEN * 2]; |
937 | struct kvm_stat_data *stat_data; | |
bc9e9e67 | 938 | const struct _kvm_stats_desc *pdesc; |
3165af73 | 939 | int i, ret; |
bc9e9e67 JZ |
940 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
941 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
942 | |
943 | if (!debugfs_initialized()) | |
944 | return 0; | |
945 | ||
946 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
85cd39af PB |
947 | mutex_lock(&kvm_debugfs_lock); |
948 | dent = debugfs_lookup(dir_name, kvm_debugfs_dir); | |
949 | if (dent) { | |
950 | pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name); | |
951 | dput(dent); | |
952 | mutex_unlock(&kvm_debugfs_lock); | |
953 | return 0; | |
954 | } | |
955 | dent = debugfs_create_dir(dir_name, kvm_debugfs_dir); | |
956 | mutex_unlock(&kvm_debugfs_lock); | |
957 | if (IS_ERR(dent)) | |
958 | return 0; | |
536a6f88 | 959 | |
85cd39af | 960 | kvm->debugfs_dentry = dent; |
536a6f88 JF |
961 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
962 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 963 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
964 | if (!kvm->debugfs_stat_data) |
965 | return -ENOMEM; | |
966 | ||
bc9e9e67 JZ |
967 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
968 | pdesc = &kvm_vm_stats_desc[i]; | |
b12ce36a | 969 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
970 | if (!stat_data) |
971 | return -ENOMEM; | |
972 | ||
973 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
974 | stat_data->desc = pdesc; |
975 | stat_data->kind = KVM_STAT_VM; | |
976 | kvm->debugfs_stat_data[i] = stat_data; | |
977 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
978 | kvm->debugfs_dentry, stat_data, | |
979 | &stat_fops_per_vm); | |
980 | } | |
981 | ||
982 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
983 | pdesc = &kvm_vcpu_stats_desc[i]; | |
b12ce36a | 984 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
985 | if (!stat_data) |
986 | return -ENOMEM; | |
987 | ||
988 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
989 | stat_data->desc = pdesc; |
990 | stat_data->kind = KVM_STAT_VCPU; | |
004d62eb | 991 | kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data; |
bc9e9e67 | 992 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), |
09cbcef6 MP |
993 | kvm->debugfs_dentry, stat_data, |
994 | &stat_fops_per_vm); | |
536a6f88 | 995 | } |
3165af73 PX |
996 | |
997 | ret = kvm_arch_create_vm_debugfs(kvm); | |
998 | if (ret) { | |
999 | kvm_destroy_vm_debugfs(kvm); | |
1000 | return i; | |
1001 | } | |
1002 | ||
536a6f88 JF |
1003 | return 0; |
1004 | } | |
1005 | ||
1aa9b957 JS |
1006 | /* |
1007 | * Called after the VM is otherwise initialized, but just before adding it to | |
1008 | * the vm_list. | |
1009 | */ | |
1010 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
1011 | { | |
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * Called just after removing the VM from the vm_list, but before doing any | |
1017 | * other destruction. | |
1018 | */ | |
1019 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
1020 | { | |
1021 | } | |
1022 | ||
3165af73 PX |
1023 | /* |
1024 | * Called after per-vm debugfs created. When called kvm->debugfs_dentry should | |
1025 | * be setup already, so we can create arch-specific debugfs entries under it. | |
1026 | * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so | |
1027 | * a per-arch destroy interface is not needed. | |
1028 | */ | |
1029 | int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm) | |
1030 | { | |
1031 | return 0; | |
1032 | } | |
1033 | ||
e08b9637 | 1034 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 1035 | { |
d89f5eff | 1036 | struct kvm *kvm = kvm_arch_alloc_vm(); |
9121923c JM |
1037 | int r = -ENOMEM; |
1038 | int i; | |
6aa8b732 | 1039 | |
d89f5eff JK |
1040 | if (!kvm) |
1041 | return ERR_PTR(-ENOMEM); | |
1042 | ||
531810ca | 1043 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 1044 | mmgrab(current->mm); |
e9ad4ec8 PB |
1045 | kvm->mm = current->mm; |
1046 | kvm_eventfd_init(kvm); | |
1047 | mutex_init(&kvm->lock); | |
1048 | mutex_init(&kvm->irq_lock); | |
1049 | mutex_init(&kvm->slots_lock); | |
b10a038e | 1050 | mutex_init(&kvm->slots_arch_lock); |
52ac8b35 PB |
1051 | spin_lock_init(&kvm->mn_invalidate_lock); |
1052 | rcuwait_init(&kvm->mn_memslots_update_rcuwait); | |
1053 | ||
e9ad4ec8 PB |
1054 | INIT_LIST_HEAD(&kvm->devices); |
1055 | ||
1e702d9a AW |
1056 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
1057 | ||
8a44119a PB |
1058 | if (init_srcu_struct(&kvm->srcu)) |
1059 | goto out_err_no_srcu; | |
1060 | if (init_srcu_struct(&kvm->irq_srcu)) | |
1061 | goto out_err_no_irq_srcu; | |
1062 | ||
e2d3fcaf | 1063 | refcount_set(&kvm->users_count, 1); |
f481b069 | 1064 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4bd518f1 | 1065 | struct kvm_memslots *slots = kvm_alloc_memslots(); |
9121923c | 1066 | |
4bd518f1 | 1067 | if (!slots) |
a97b0e77 | 1068 | goto out_err_no_arch_destroy_vm; |
0e32958e | 1069 | /* Generations must be different for each address space. */ |
164bf7e5 | 1070 | slots->generation = i; |
4bd518f1 | 1071 | rcu_assign_pointer(kvm->memslots[i], slots); |
f481b069 | 1072 | } |
00f034a1 | 1073 | |
e93f8a0f | 1074 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 1075 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 1076 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 1077 | if (!kvm->buses[i]) |
a97b0e77 | 1078 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 1079 | } |
e930bffe | 1080 | |
acd05785 DM |
1081 | kvm->max_halt_poll_ns = halt_poll_ns; |
1082 | ||
e08b9637 | 1083 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 1084 | if (r) |
a97b0e77 | 1085 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
1086 | |
1087 | r = hardware_enable_all(); | |
1088 | if (r) | |
719d93cd | 1089 | goto out_err_no_disable; |
10474ae8 | 1090 | |
c77dcacb | 1091 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 1092 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 1093 | #endif |
6aa8b732 | 1094 | |
74b5c5bf | 1095 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
1096 | if (r) |
1097 | goto out_err_no_mmu_notifier; | |
1098 | ||
1099 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
1100 | if (r) |
1101 | goto out_err; | |
1102 | ||
0d9ce162 | 1103 | mutex_lock(&kvm_lock); |
5e58cfe4 | 1104 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 1105 | mutex_unlock(&kvm_lock); |
d89f5eff | 1106 | |
2ecd9d29 | 1107 | preempt_notifier_inc(); |
2fdef3a2 | 1108 | kvm_init_pm_notifier(kvm); |
2ecd9d29 | 1109 | |
f17abe9a | 1110 | return kvm; |
10474ae8 AG |
1111 | |
1112 | out_err: | |
1aa9b957 JS |
1113 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1114 | if (kvm->mmu_notifier.ops) | |
1115 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
1116 | #endif | |
1117 | out_err_no_mmu_notifier: | |
10474ae8 | 1118 | hardware_disable_all(); |
719d93cd | 1119 | out_err_no_disable: |
a97b0e77 | 1120 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 1121 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 1122 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 1123 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 1124 | kfree(kvm_get_bus(kvm, i)); |
f481b069 | 1125 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 1126 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
8a44119a PB |
1127 | cleanup_srcu_struct(&kvm->irq_srcu); |
1128 | out_err_no_irq_srcu: | |
1129 | cleanup_srcu_struct(&kvm->srcu); | |
1130 | out_err_no_srcu: | |
d89f5eff | 1131 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 1132 | mmdrop(current->mm); |
10474ae8 | 1133 | return ERR_PTR(r); |
f17abe9a AK |
1134 | } |
1135 | ||
07f0a7bd SW |
1136 | static void kvm_destroy_devices(struct kvm *kvm) |
1137 | { | |
e6e3b5a6 | 1138 | struct kvm_device *dev, *tmp; |
07f0a7bd | 1139 | |
a28ebea2 CD |
1140 | /* |
1141 | * We do not need to take the kvm->lock here, because nobody else | |
1142 | * has a reference to the struct kvm at this point and therefore | |
1143 | * cannot access the devices list anyhow. | |
1144 | */ | |
e6e3b5a6 GT |
1145 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
1146 | list_del(&dev->vm_node); | |
07f0a7bd SW |
1147 | dev->ops->destroy(dev); |
1148 | } | |
1149 | } | |
1150 | ||
f17abe9a AK |
1151 | static void kvm_destroy_vm(struct kvm *kvm) |
1152 | { | |
e93f8a0f | 1153 | int i; |
6d4e4c4f AK |
1154 | struct mm_struct *mm = kvm->mm; |
1155 | ||
2fdef3a2 | 1156 | kvm_destroy_pm_notifier(kvm); |
286de8f6 | 1157 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 1158 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 1159 | kvm_arch_sync_events(kvm); |
0d9ce162 | 1160 | mutex_lock(&kvm_lock); |
133de902 | 1161 | list_del(&kvm->vm_list); |
0d9ce162 | 1162 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
1163 | kvm_arch_pre_destroy_vm(kvm); |
1164 | ||
399ec807 | 1165 | kvm_free_irq_routing(kvm); |
df630b8c | 1166 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 1167 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 1168 | |
4a12f951 CB |
1169 | if (bus) |
1170 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
1171 | kvm->buses[i] = NULL; |
1172 | } | |
980da6ce | 1173 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
1174 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1175 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
52ac8b35 PB |
1176 | /* |
1177 | * At this point, pending calls to invalidate_range_start() | |
1178 | * have completed but no more MMU notifiers will run, so | |
1179 | * mn_active_invalidate_count may remain unbalanced. | |
1180 | * No threads can be waiting in install_new_memslots as the | |
1181 | * last reference on KVM has been dropped, but freeing | |
1182 | * memslots would deadlock without this manual intervention. | |
1183 | */ | |
1184 | WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait)); | |
1185 | kvm->mn_active_invalidate_count = 0; | |
f00be0ca | 1186 | #else |
2df72e9b | 1187 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 1188 | #endif |
d19a9cd2 | 1189 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 1190 | kvm_destroy_devices(kvm); |
f481b069 | 1191 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 1192 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
820b3fcd | 1193 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
1194 | cleanup_srcu_struct(&kvm->srcu); |
1195 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 1196 | preempt_notifier_dec(); |
10474ae8 | 1197 | hardware_disable_all(); |
6d4e4c4f | 1198 | mmdrop(mm); |
f17abe9a AK |
1199 | } |
1200 | ||
d39f13b0 IE |
1201 | void kvm_get_kvm(struct kvm *kvm) |
1202 | { | |
e3736c3e | 1203 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
1204 | } |
1205 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
1206 | ||
605c7130 PX |
1207 | /* |
1208 | * Make sure the vm is not during destruction, which is a safe version of | |
1209 | * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise. | |
1210 | */ | |
1211 | bool kvm_get_kvm_safe(struct kvm *kvm) | |
1212 | { | |
1213 | return refcount_inc_not_zero(&kvm->users_count); | |
1214 | } | |
1215 | EXPORT_SYMBOL_GPL(kvm_get_kvm_safe); | |
1216 | ||
d39f13b0 IE |
1217 | void kvm_put_kvm(struct kvm *kvm) |
1218 | { | |
e3736c3e | 1219 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
1220 | kvm_destroy_vm(kvm); |
1221 | } | |
1222 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
1223 | ||
149487bd SC |
1224 | /* |
1225 | * Used to put a reference that was taken on behalf of an object associated | |
1226 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
1227 | * of the new file descriptor fails and the reference cannot be transferred to | |
1228 | * its final owner. In such cases, the caller is still actively using @kvm and | |
1229 | * will fail miserably if the refcount unexpectedly hits zero. | |
1230 | */ | |
1231 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
1232 | { | |
1233 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
1234 | } | |
1235 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 1236 | |
f17abe9a AK |
1237 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
1238 | { | |
1239 | struct kvm *kvm = filp->private_data; | |
1240 | ||
721eecbf GH |
1241 | kvm_irqfd_release(kvm); |
1242 | ||
d39f13b0 | 1243 | kvm_put_kvm(kvm); |
6aa8b732 AK |
1244 | return 0; |
1245 | } | |
1246 | ||
515a0127 TY |
1247 | /* |
1248 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 1249 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 1250 | */ |
3c9bd400 | 1251 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 1252 | { |
515a0127 | 1253 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 1254 | |
b12ce36a | 1255 | memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
1256 | if (!memslot->dirty_bitmap) |
1257 | return -ENOMEM; | |
1258 | ||
a36a57b1 TY |
1259 | return 0; |
1260 | } | |
1261 | ||
bf3e05bc | 1262 | /* |
0577d1ab SC |
1263 | * Delete a memslot by decrementing the number of used slots and shifting all |
1264 | * other entries in the array forward one spot. | |
bf3e05bc | 1265 | */ |
0577d1ab SC |
1266 | static inline void kvm_memslot_delete(struct kvm_memslots *slots, |
1267 | struct kvm_memory_slot *memslot) | |
bf3e05bc | 1268 | { |
063584d4 | 1269 | struct kvm_memory_slot *mslots = slots->memslots; |
0577d1ab | 1270 | int i; |
f85e2cb5 | 1271 | |
0577d1ab SC |
1272 | if (WARN_ON(slots->id_to_index[memslot->id] == -1)) |
1273 | return; | |
0e60b079 | 1274 | |
0577d1ab SC |
1275 | slots->used_slots--; |
1276 | ||
87689270 DM |
1277 | if (atomic_read(&slots->last_used_slot) >= slots->used_slots) |
1278 | atomic_set(&slots->last_used_slot, 0); | |
0774a964 | 1279 | |
0577d1ab | 1280 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) { |
7f379cff IM |
1281 | mslots[i] = mslots[i + 1]; |
1282 | slots->id_to_index[mslots[i].id] = i; | |
7f379cff | 1283 | } |
0577d1ab SC |
1284 | mslots[i] = *memslot; |
1285 | slots->id_to_index[memslot->id] = -1; | |
1286 | } | |
1287 | ||
1288 | /* | |
1289 | * "Insert" a new memslot by incrementing the number of used slots. Returns | |
1290 | * the new slot's initial index into the memslots array. | |
1291 | */ | |
1292 | static inline int kvm_memslot_insert_back(struct kvm_memslots *slots) | |
1293 | { | |
1294 | return slots->used_slots++; | |
1295 | } | |
1296 | ||
1297 | /* | |
1298 | * Move a changed memslot backwards in the array by shifting existing slots | |
1299 | * with a higher GFN toward the front of the array. Note, the changed memslot | |
1300 | * itself is not preserved in the array, i.e. not swapped at this time, only | |
1301 | * its new index into the array is tracked. Returns the changed memslot's | |
1302 | * current index into the memslots array. | |
1303 | */ | |
1304 | static inline int kvm_memslot_move_backward(struct kvm_memslots *slots, | |
1305 | struct kvm_memory_slot *memslot) | |
1306 | { | |
1307 | struct kvm_memory_slot *mslots = slots->memslots; | |
1308 | int i; | |
1309 | ||
1310 | if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) || | |
1311 | WARN_ON_ONCE(!slots->used_slots)) | |
1312 | return -1; | |
efbeec70 PB |
1313 | |
1314 | /* | |
0577d1ab SC |
1315 | * Move the target memslot backward in the array by shifting existing |
1316 | * memslots with a higher GFN (than the target memslot) towards the | |
1317 | * front of the array. | |
efbeec70 | 1318 | */ |
0577d1ab SC |
1319 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) { |
1320 | if (memslot->base_gfn > mslots[i + 1].base_gfn) | |
1321 | break; | |
1322 | ||
1323 | WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn); | |
f85e2cb5 | 1324 | |
0577d1ab SC |
1325 | /* Shift the next memslot forward one and update its index. */ |
1326 | mslots[i] = mslots[i + 1]; | |
1327 | slots->id_to_index[mslots[i].id] = i; | |
1328 | } | |
1329 | return i; | |
1330 | } | |
1331 | ||
1332 | /* | |
1333 | * Move a changed memslot forwards in the array by shifting existing slots with | |
1334 | * a lower GFN toward the back of the array. Note, the changed memslot itself | |
1335 | * is not preserved in the array, i.e. not swapped at this time, only its new | |
1336 | * index into the array is tracked. Returns the changed memslot's final index | |
1337 | * into the memslots array. | |
1338 | */ | |
1339 | static inline int kvm_memslot_move_forward(struct kvm_memslots *slots, | |
1340 | struct kvm_memory_slot *memslot, | |
1341 | int start) | |
1342 | { | |
1343 | struct kvm_memory_slot *mslots = slots->memslots; | |
1344 | int i; | |
1345 | ||
1346 | for (i = start; i > 0; i--) { | |
1347 | if (memslot->base_gfn < mslots[i - 1].base_gfn) | |
1348 | break; | |
1349 | ||
1350 | WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn); | |
1351 | ||
1352 | /* Shift the next memslot back one and update its index. */ | |
1353 | mslots[i] = mslots[i - 1]; | |
1354 | slots->id_to_index[mslots[i].id] = i; | |
1355 | } | |
1356 | return i; | |
1357 | } | |
1358 | ||
1359 | /* | |
1360 | * Re-sort memslots based on their GFN to account for an added, deleted, or | |
1361 | * moved memslot. Sorting memslots by GFN allows using a binary search during | |
1362 | * memslot lookup. | |
1363 | * | |
1364 | * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry | |
1365 | * at memslots[0] has the highest GFN. | |
1366 | * | |
1367 | * The sorting algorithm takes advantage of having initially sorted memslots | |
1368 | * and knowing the position of the changed memslot. Sorting is also optimized | |
1369 | * by not swapping the updated memslot and instead only shifting other memslots | |
1370 | * and tracking the new index for the update memslot. Only once its final | |
1371 | * index is known is the updated memslot copied into its position in the array. | |
1372 | * | |
1373 | * - When deleting a memslot, the deleted memslot simply needs to be moved to | |
1374 | * the end of the array. | |
1375 | * | |
1376 | * - When creating a memslot, the algorithm "inserts" the new memslot at the | |
1377 | * end of the array and then it forward to its correct location. | |
1378 | * | |
1379 | * - When moving a memslot, the algorithm first moves the updated memslot | |
1380 | * backward to handle the scenario where the memslot's GFN was changed to a | |
1381 | * lower value. update_memslots() then falls through and runs the same flow | |
1382 | * as creating a memslot to move the memslot forward to handle the scenario | |
1383 | * where its GFN was changed to a higher value. | |
1384 | * | |
1385 | * Note, slots are sorted from highest->lowest instead of lowest->highest for | |
1386 | * historical reasons. Originally, invalid memslots where denoted by having | |
1387 | * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots | |
1388 | * to the end of the array. The current algorithm uses dedicated logic to | |
1389 | * delete a memslot and thus does not rely on invalid memslots having GFN=0. | |
1390 | * | |
1391 | * The other historical motiviation for highest->lowest was to improve the | |
1392 | * performance of memslot lookup. KVM originally used a linear search starting | |
1393 | * at memslots[0]. On x86, the largest memslot usually has one of the highest, | |
1394 | * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a | |
1395 | * single memslot above the 4gb boundary. As the largest memslot is also the | |
1396 | * most likely to be referenced, sorting it to the front of the array was | |
1397 | * advantageous. The current binary search starts from the middle of the array | |
1398 | * and uses an LRU pointer to improve performance for all memslots and GFNs. | |
1399 | */ | |
1400 | static void update_memslots(struct kvm_memslots *slots, | |
1401 | struct kvm_memory_slot *memslot, | |
1402 | enum kvm_mr_change change) | |
1403 | { | |
1404 | int i; | |
1405 | ||
1406 | if (change == KVM_MR_DELETE) { | |
1407 | kvm_memslot_delete(slots, memslot); | |
1408 | } else { | |
1409 | if (change == KVM_MR_CREATE) | |
1410 | i = kvm_memslot_insert_back(slots); | |
1411 | else | |
1412 | i = kvm_memslot_move_backward(slots, memslot); | |
1413 | i = kvm_memslot_move_forward(slots, memslot, i); | |
1414 | ||
1415 | /* | |
1416 | * Copy the memslot to its new position in memslots and update | |
1417 | * its index accordingly. | |
1418 | */ | |
1419 | slots->memslots[i] = *memslot; | |
1420 | slots->id_to_index[memslot->id] = i; | |
1421 | } | |
bf3e05bc XG |
1422 | } |
1423 | ||
09170a49 | 1424 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 1425 | { |
4d8b81ab XG |
1426 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1427 | ||
0f8a4de3 | 1428 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
1429 | valid_flags |= KVM_MEM_READONLY; |
1430 | #endif | |
1431 | ||
1432 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1433 | return -EINVAL; |
1434 | ||
1435 | return 0; | |
1436 | } | |
1437 | ||
7ec4fb44 | 1438 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 1439 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 1440 | { |
f481b069 | 1441 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
361209e0 | 1442 | u64 gen = old_memslots->generation; |
7ec4fb44 | 1443 | |
361209e0 SC |
1444 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1445 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1446 | |
52ac8b35 PB |
1447 | /* |
1448 | * Do not store the new memslots while there are invalidations in | |
071064f1 PB |
1449 | * progress, otherwise the locking in invalidate_range_start and |
1450 | * invalidate_range_end will be unbalanced. | |
52ac8b35 PB |
1451 | */ |
1452 | spin_lock(&kvm->mn_invalidate_lock); | |
1453 | prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
1454 | while (kvm->mn_active_invalidate_count) { | |
1455 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1456 | spin_unlock(&kvm->mn_invalidate_lock); | |
1457 | schedule(); | |
1458 | spin_lock(&kvm->mn_invalidate_lock); | |
1459 | } | |
1460 | finish_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
f481b069 | 1461 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
52ac8b35 | 1462 | spin_unlock(&kvm->mn_invalidate_lock); |
b10a038e BG |
1463 | |
1464 | /* | |
1465 | * Acquired in kvm_set_memslot. Must be released before synchronize | |
1466 | * SRCU below in order to avoid deadlock with another thread | |
1467 | * acquiring the slots_arch_lock in an srcu critical section. | |
1468 | */ | |
1469 | mutex_unlock(&kvm->slots_arch_lock); | |
1470 | ||
7ec4fb44 | 1471 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1472 | |
ee3d1570 | 1473 | /* |
361209e0 | 1474 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1475 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1476 | * the number of address spaces. This provides a unique and easily |
1477 | * identifiable generation number while the memslots are in flux. | |
1478 | */ | |
1479 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1480 | ||
1481 | /* | |
4bd518f1 PB |
1482 | * Generations must be unique even across address spaces. We do not need |
1483 | * a global counter for that, instead the generation space is evenly split | |
1484 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1485 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1486 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1487 | */ |
164bf7e5 | 1488 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 1489 | |
15248258 | 1490 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1491 | |
15248258 | 1492 | slots->generation = gen; |
e59dbe09 TY |
1493 | |
1494 | return old_memslots; | |
7ec4fb44 GN |
1495 | } |
1496 | ||
ddc12f2a BG |
1497 | static size_t kvm_memslots_size(int slots) |
1498 | { | |
1499 | return sizeof(struct kvm_memslots) + | |
1500 | (sizeof(struct kvm_memory_slot) * slots); | |
1501 | } | |
1502 | ||
1503 | static void kvm_copy_memslots(struct kvm_memslots *to, | |
1504 | struct kvm_memslots *from) | |
1505 | { | |
1506 | memcpy(to, from, kvm_memslots_size(from->used_slots)); | |
1507 | } | |
1508 | ||
36947254 SC |
1509 | /* |
1510 | * Note, at a minimum, the current number of used slots must be allocated, even | |
1511 | * when deleting a memslot, as we need a complete duplicate of the memslots for | |
1512 | * use when invalidating a memslot prior to deleting/moving the memslot. | |
1513 | */ | |
1514 | static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old, | |
1515 | enum kvm_mr_change change) | |
1516 | { | |
1517 | struct kvm_memslots *slots; | |
ddc12f2a | 1518 | size_t new_size; |
36947254 SC |
1519 | |
1520 | if (change == KVM_MR_CREATE) | |
ddc12f2a | 1521 | new_size = kvm_memslots_size(old->used_slots + 1); |
36947254 | 1522 | else |
ddc12f2a | 1523 | new_size = kvm_memslots_size(old->used_slots); |
36947254 SC |
1524 | |
1525 | slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT); | |
1526 | if (likely(slots)) | |
ddc12f2a | 1527 | kvm_copy_memslots(slots, old); |
36947254 SC |
1528 | |
1529 | return slots; | |
1530 | } | |
1531 | ||
cf47f50b SC |
1532 | static int kvm_set_memslot(struct kvm *kvm, |
1533 | const struct kvm_userspace_memory_region *mem, | |
9d4c197c | 1534 | struct kvm_memory_slot *old, |
cf47f50b SC |
1535 | struct kvm_memory_slot *new, int as_id, |
1536 | enum kvm_mr_change change) | |
1537 | { | |
1538 | struct kvm_memory_slot *slot; | |
1539 | struct kvm_memslots *slots; | |
1540 | int r; | |
1541 | ||
b10a038e BG |
1542 | /* |
1543 | * Released in install_new_memslots. | |
1544 | * | |
1545 | * Must be held from before the current memslots are copied until | |
1546 | * after the new memslots are installed with rcu_assign_pointer, | |
1547 | * then released before the synchronize srcu in install_new_memslots. | |
1548 | * | |
1549 | * When modifying memslots outside of the slots_lock, must be held | |
1550 | * before reading the pointer to the current memslots until after all | |
1551 | * changes to those memslots are complete. | |
1552 | * | |
1553 | * These rules ensure that installing new memslots does not lose | |
1554 | * changes made to the previous memslots. | |
1555 | */ | |
1556 | mutex_lock(&kvm->slots_arch_lock); | |
1557 | ||
36947254 | 1558 | slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change); |
b10a038e BG |
1559 | if (!slots) { |
1560 | mutex_unlock(&kvm->slots_arch_lock); | |
cf47f50b | 1561 | return -ENOMEM; |
b10a038e | 1562 | } |
cf47f50b SC |
1563 | |
1564 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { | |
1565 | /* | |
1566 | * Note, the INVALID flag needs to be in the appropriate entry | |
1567 | * in the freshly allocated memslots, not in @old or @new. | |
1568 | */ | |
1569 | slot = id_to_memslot(slots, old->id); | |
1570 | slot->flags |= KVM_MEMSLOT_INVALID; | |
1571 | ||
1572 | /* | |
b10a038e BG |
1573 | * We can re-use the memory from the old memslots. |
1574 | * It will be overwritten with a copy of the new memslots | |
1575 | * after reacquiring the slots_arch_lock below. | |
cf47f50b SC |
1576 | */ |
1577 | slots = install_new_memslots(kvm, as_id, slots); | |
1578 | ||
1579 | /* From this point no new shadow pages pointing to a deleted, | |
1580 | * or moved, memslot will be created. | |
1581 | * | |
1582 | * validation of sp->gfn happens in: | |
1583 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1584 | * - kvm_is_visible_gfn (mmu_check_root) | |
1585 | */ | |
1586 | kvm_arch_flush_shadow_memslot(kvm, slot); | |
b10a038e BG |
1587 | |
1588 | /* Released in install_new_memslots. */ | |
1589 | mutex_lock(&kvm->slots_arch_lock); | |
1590 | ||
1591 | /* | |
1592 | * The arch-specific fields of the memslots could have changed | |
1593 | * between releasing the slots_arch_lock in | |
1594 | * install_new_memslots and here, so get a fresh copy of the | |
1595 | * slots. | |
1596 | */ | |
1597 | kvm_copy_memslots(slots, __kvm_memslots(kvm, as_id)); | |
cf47f50b SC |
1598 | } |
1599 | ||
1600 | r = kvm_arch_prepare_memory_region(kvm, new, mem, change); | |
1601 | if (r) | |
1602 | goto out_slots; | |
1603 | ||
1604 | update_memslots(slots, new, change); | |
1605 | slots = install_new_memslots(kvm, as_id, slots); | |
1606 | ||
1607 | kvm_arch_commit_memory_region(kvm, mem, old, new, change); | |
1608 | ||
1609 | kvfree(slots); | |
1610 | return 0; | |
1611 | ||
1612 | out_slots: | |
b10a038e BG |
1613 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
1614 | slot = id_to_memslot(slots, old->id); | |
1615 | slot->flags &= ~KVM_MEMSLOT_INVALID; | |
cf47f50b | 1616 | slots = install_new_memslots(kvm, as_id, slots); |
b10a038e BG |
1617 | } else { |
1618 | mutex_unlock(&kvm->slots_arch_lock); | |
1619 | } | |
cf47f50b SC |
1620 | kvfree(slots); |
1621 | return r; | |
1622 | } | |
1623 | ||
5c0b4f3d SC |
1624 | static int kvm_delete_memslot(struct kvm *kvm, |
1625 | const struct kvm_userspace_memory_region *mem, | |
1626 | struct kvm_memory_slot *old, int as_id) | |
1627 | { | |
1628 | struct kvm_memory_slot new; | |
1629 | int r; | |
1630 | ||
1631 | if (!old->npages) | |
1632 | return -EINVAL; | |
1633 | ||
1634 | memset(&new, 0, sizeof(new)); | |
1635 | new.id = old->id; | |
9e9eb226 PX |
1636 | /* |
1637 | * This is only for debugging purpose; it should never be referenced | |
1638 | * for a removed memslot. | |
1639 | */ | |
1640 | new.as_id = as_id; | |
5c0b4f3d SC |
1641 | |
1642 | r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE); | |
1643 | if (r) | |
1644 | return r; | |
1645 | ||
e96c81ee | 1646 | kvm_free_memslot(kvm, old); |
5c0b4f3d SC |
1647 | return 0; |
1648 | } | |
1649 | ||
6aa8b732 AK |
1650 | /* |
1651 | * Allocate some memory and give it an address in the guest physical address | |
1652 | * space. | |
1653 | * | |
1654 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1655 | * |
02d5d55b | 1656 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1657 | */ |
f78e0e2e | 1658 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 1659 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 1660 | { |
6aa8b732 | 1661 | struct kvm_memory_slot old, new; |
163da372 | 1662 | struct kvm_memory_slot *tmp; |
f64c0398 | 1663 | enum kvm_mr_change change; |
163da372 SC |
1664 | int as_id, id; |
1665 | int r; | |
6aa8b732 | 1666 | |
a50d64d6 XG |
1667 | r = check_memory_region_flags(mem); |
1668 | if (r) | |
71a4c30b | 1669 | return r; |
a50d64d6 | 1670 | |
f481b069 PB |
1671 | as_id = mem->slot >> 16; |
1672 | id = (u16)mem->slot; | |
1673 | ||
6aa8b732 AK |
1674 | /* General sanity checks */ |
1675 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
71a4c30b | 1676 | return -EINVAL; |
6aa8b732 | 1677 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 1678 | return -EINVAL; |
fa3d315a | 1679 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 1680 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 1681 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 1682 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 1683 | mem->memory_size)) |
71a4c30b | 1684 | return -EINVAL; |
f481b069 | 1685 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 1686 | return -EINVAL; |
6aa8b732 | 1687 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 1688 | return -EINVAL; |
6aa8b732 | 1689 | |
5c0b4f3d SC |
1690 | /* |
1691 | * Make a full copy of the old memslot, the pointer will become stale | |
1692 | * when the memslots are re-sorted by update_memslots(), and the old | |
1693 | * memslot needs to be referenced after calling update_memslots(), e.g. | |
0dff0846 | 1694 | * to free its resources and for arch specific behavior. |
5c0b4f3d | 1695 | */ |
0577d1ab SC |
1696 | tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
1697 | if (tmp) { | |
1698 | old = *tmp; | |
1699 | tmp = NULL; | |
1700 | } else { | |
1701 | memset(&old, 0, sizeof(old)); | |
1702 | old.id = id; | |
1703 | } | |
163da372 | 1704 | |
5c0b4f3d SC |
1705 | if (!mem->memory_size) |
1706 | return kvm_delete_memslot(kvm, mem, &old, as_id); | |
1707 | ||
9e9eb226 | 1708 | new.as_id = as_id; |
f481b069 | 1709 | new.id = id; |
163da372 SC |
1710 | new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
1711 | new.npages = mem->memory_size >> PAGE_SHIFT; | |
6aa8b732 | 1712 | new.flags = mem->flags; |
414de7ab | 1713 | new.userspace_addr = mem->userspace_addr; |
6aa8b732 | 1714 | |
163da372 SC |
1715 | if (new.npages > KVM_MEM_MAX_NR_PAGES) |
1716 | return -EINVAL; | |
1717 | ||
5c0b4f3d SC |
1718 | if (!old.npages) { |
1719 | change = KVM_MR_CREATE; | |
163da372 SC |
1720 | new.dirty_bitmap = NULL; |
1721 | memset(&new.arch, 0, sizeof(new.arch)); | |
5c0b4f3d SC |
1722 | } else { /* Modify an existing slot. */ |
1723 | if ((new.userspace_addr != old.userspace_addr) || | |
163da372 | 1724 | (new.npages != old.npages) || |
5c0b4f3d | 1725 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) |
71a4c30b | 1726 | return -EINVAL; |
09170a49 | 1727 | |
163da372 | 1728 | if (new.base_gfn != old.base_gfn) |
5c0b4f3d SC |
1729 | change = KVM_MR_MOVE; |
1730 | else if (new.flags != old.flags) | |
1731 | change = KVM_MR_FLAGS_ONLY; | |
1732 | else /* Nothing to change. */ | |
1733 | return 0; | |
163da372 SC |
1734 | |
1735 | /* Copy dirty_bitmap and arch from the current memslot. */ | |
1736 | new.dirty_bitmap = old.dirty_bitmap; | |
1737 | memcpy(&new.arch, &old.arch, sizeof(new.arch)); | |
09170a49 | 1738 | } |
6aa8b732 | 1739 | |
f64c0398 | 1740 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee | 1741 | /* Check for overlaps */ |
163da372 SC |
1742 | kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) { |
1743 | if (tmp->id == id) | |
0a706bee | 1744 | continue; |
163da372 SC |
1745 | if (!((new.base_gfn + new.npages <= tmp->base_gfn) || |
1746 | (new.base_gfn >= tmp->base_gfn + tmp->npages))) | |
71a4c30b | 1747 | return -EEXIST; |
0a706bee | 1748 | } |
6aa8b732 | 1749 | } |
6aa8b732 | 1750 | |
414de7ab SC |
1751 | /* Allocate/free page dirty bitmap as needed */ |
1752 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1753 | new.dirty_bitmap = NULL; | |
044c59c4 | 1754 | else if (!new.dirty_bitmap && !kvm->dirty_ring_size) { |
3c9bd400 | 1755 | r = kvm_alloc_dirty_bitmap(&new); |
71a4c30b SC |
1756 | if (r) |
1757 | return r; | |
3c9bd400 JZ |
1758 | |
1759 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1760 | bitmap_set(new.dirty_bitmap, 0, new.npages); | |
6aa8b732 AK |
1761 | } |
1762 | ||
cf47f50b SC |
1763 | r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change); |
1764 | if (r) | |
1765 | goto out_bitmap; | |
82ce2c96 | 1766 | |
5c0b4f3d SC |
1767 | if (old.dirty_bitmap && !new.dirty_bitmap) |
1768 | kvm_destroy_dirty_bitmap(&old); | |
6aa8b732 AK |
1769 | return 0; |
1770 | ||
bd0e96fd SC |
1771 | out_bitmap: |
1772 | if (new.dirty_bitmap && !old.dirty_bitmap) | |
1773 | kvm_destroy_dirty_bitmap(&new); | |
6aa8b732 | 1774 | return r; |
210c7c4d | 1775 | } |
f78e0e2e SY |
1776 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1777 | ||
1778 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1779 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1780 | { |
1781 | int r; | |
1782 | ||
79fac95e | 1783 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1784 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1785 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1786 | return r; |
1787 | } | |
210c7c4d IE |
1788 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1789 | ||
7940876e SH |
1790 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1791 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1792 | { |
f481b069 | 1793 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1794 | return -EINVAL; |
09170a49 | 1795 | |
47ae31e2 | 1796 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1797 | } |
1798 | ||
0dff0846 | 1799 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
1800 | /** |
1801 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
1802 | * @kvm: pointer to kvm instance | |
1803 | * @log: slot id and address to which we copy the log | |
1804 | * @is_dirty: set to '1' if any dirty pages were found | |
1805 | * @memslot: set to the associated memslot, always valid on success | |
1806 | */ | |
1807 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
1808 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 1809 | { |
9f6b8029 | 1810 | struct kvm_memslots *slots; |
843574a3 | 1811 | int i, as_id, id; |
87bf6e7d | 1812 | unsigned long n; |
6aa8b732 AK |
1813 | unsigned long any = 0; |
1814 | ||
b2cc64c4 PX |
1815 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1816 | if (kvm->dirty_ring_size) | |
1817 | return -ENXIO; | |
1818 | ||
2a49f61d SC |
1819 | *memslot = NULL; |
1820 | *is_dirty = 0; | |
1821 | ||
f481b069 PB |
1822 | as_id = log->slot >> 16; |
1823 | id = (u16)log->slot; | |
1824 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 1825 | return -EINVAL; |
6aa8b732 | 1826 | |
f481b069 | 1827 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 1828 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 1829 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 1830 | return -ENOENT; |
6aa8b732 | 1831 | |
2a49f61d SC |
1832 | kvm_arch_sync_dirty_log(kvm, *memslot); |
1833 | ||
1834 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 1835 | |
cd1a4a98 | 1836 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 1837 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 1838 | |
2a49f61d | 1839 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 1840 | return -EFAULT; |
6aa8b732 | 1841 | |
5bb064dc ZX |
1842 | if (any) |
1843 | *is_dirty = 1; | |
843574a3 | 1844 | return 0; |
6aa8b732 | 1845 | } |
2ba9f0d8 | 1846 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1847 | |
0dff0846 | 1848 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 1849 | /** |
b8b00220 | 1850 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 1851 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
1852 | * @kvm: pointer to kvm instance |
1853 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
1854 | * |
1855 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1856 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1857 | * following order: | |
1858 | * | |
1859 | * 1. Take a snapshot of the bit and clear it if needed. | |
1860 | * 2. Write protect the corresponding page. | |
1861 | * 3. Copy the snapshot to the userspace. | |
1862 | * 4. Upon return caller flushes TLB's if needed. | |
1863 | * | |
1864 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1865 | * entry. This is not a problem because the page is reported dirty using | |
1866 | * the snapshot taken before and step 4 ensures that writes done after | |
1867 | * exiting to userspace will be logged for the next call. | |
1868 | * | |
1869 | */ | |
0dff0846 | 1870 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 1871 | { |
9f6b8029 | 1872 | struct kvm_memslots *slots; |
ba0513b5 | 1873 | struct kvm_memory_slot *memslot; |
58d6db34 | 1874 | int i, as_id, id; |
ba0513b5 MS |
1875 | unsigned long n; |
1876 | unsigned long *dirty_bitmap; | |
1877 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 1878 | bool flush; |
ba0513b5 | 1879 | |
b2cc64c4 PX |
1880 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1881 | if (kvm->dirty_ring_size) | |
1882 | return -ENXIO; | |
1883 | ||
f481b069 PB |
1884 | as_id = log->slot >> 16; |
1885 | id = (u16)log->slot; | |
1886 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 1887 | return -EINVAL; |
ba0513b5 | 1888 | |
f481b069 PB |
1889 | slots = __kvm_memslots(kvm, as_id); |
1890 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
1891 | if (!memslot || !memslot->dirty_bitmap) |
1892 | return -ENOENT; | |
ba0513b5 MS |
1893 | |
1894 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 1895 | |
0dff0846 SC |
1896 | kvm_arch_sync_dirty_log(kvm, memslot); |
1897 | ||
ba0513b5 | 1898 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 1899 | flush = false; |
2a31b9db PB |
1900 | if (kvm->manual_dirty_log_protect) { |
1901 | /* | |
1902 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
1903 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
1904 | * is some code duplication between this function and | |
1905 | * kvm_get_dirty_log, but hopefully all architecture | |
1906 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
1907 | * can be eliminated. | |
1908 | */ | |
1909 | dirty_bitmap_buffer = dirty_bitmap; | |
1910 | } else { | |
1911 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
1912 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 1913 | |
531810ca | 1914 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
1915 | for (i = 0; i < n / sizeof(long); i++) { |
1916 | unsigned long mask; | |
1917 | gfn_t offset; | |
ba0513b5 | 1918 | |
2a31b9db PB |
1919 | if (!dirty_bitmap[i]) |
1920 | continue; | |
1921 | ||
0dff0846 | 1922 | flush = true; |
2a31b9db PB |
1923 | mask = xchg(&dirty_bitmap[i], 0); |
1924 | dirty_bitmap_buffer[i] = mask; | |
1925 | ||
a67794ca LT |
1926 | offset = i * BITS_PER_LONG; |
1927 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1928 | offset, mask); | |
2a31b9db | 1929 | } |
531810ca | 1930 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
1931 | } |
1932 | ||
0dff0846 SC |
1933 | if (flush) |
1934 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
1935 | ||
2a31b9db PB |
1936 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
1937 | return -EFAULT; | |
1938 | return 0; | |
1939 | } | |
0dff0846 SC |
1940 | |
1941 | ||
1942 | /** | |
1943 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
1944 | * @kvm: kvm instance | |
1945 | * @log: slot id and address to which we copy the log | |
1946 | * | |
1947 | * Steps 1-4 below provide general overview of dirty page logging. See | |
1948 | * kvm_get_dirty_log_protect() function description for additional details. | |
1949 | * | |
1950 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
1951 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
1952 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
1953 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
1954 | * writes will be marked dirty for next log read. | |
1955 | * | |
1956 | * 1. Take a snapshot of the bit and clear it if needed. | |
1957 | * 2. Write protect the corresponding page. | |
1958 | * 3. Copy the snapshot to the userspace. | |
1959 | * 4. Flush TLB's if needed. | |
1960 | */ | |
1961 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
1962 | struct kvm_dirty_log *log) | |
1963 | { | |
1964 | int r; | |
1965 | ||
1966 | mutex_lock(&kvm->slots_lock); | |
1967 | ||
1968 | r = kvm_get_dirty_log_protect(kvm, log); | |
1969 | ||
1970 | mutex_unlock(&kvm->slots_lock); | |
1971 | return r; | |
1972 | } | |
2a31b9db PB |
1973 | |
1974 | /** | |
1975 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
1976 | * and reenable dirty page tracking for the corresponding pages. | |
1977 | * @kvm: pointer to kvm instance | |
1978 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
1979 | */ | |
0dff0846 SC |
1980 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
1981 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
1982 | { |
1983 | struct kvm_memslots *slots; | |
1984 | struct kvm_memory_slot *memslot; | |
98938aa8 | 1985 | int as_id, id; |
2a31b9db | 1986 | gfn_t offset; |
98938aa8 | 1987 | unsigned long i, n; |
2a31b9db PB |
1988 | unsigned long *dirty_bitmap; |
1989 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 1990 | bool flush; |
2a31b9db | 1991 | |
b2cc64c4 PX |
1992 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1993 | if (kvm->dirty_ring_size) | |
1994 | return -ENXIO; | |
1995 | ||
2a31b9db PB |
1996 | as_id = log->slot >> 16; |
1997 | id = (u16)log->slot; | |
1998 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
1999 | return -EINVAL; | |
2000 | ||
76d58e0f | 2001 | if (log->first_page & 63) |
2a31b9db PB |
2002 | return -EINVAL; |
2003 | ||
2004 | slots = __kvm_memslots(kvm, as_id); | |
2005 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2006 | if (!memslot || !memslot->dirty_bitmap) |
2007 | return -ENOENT; | |
2a31b9db PB |
2008 | |
2009 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 2010 | |
4ddc9204 | 2011 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
2012 | |
2013 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
2014 | log->num_pages > memslot->npages - log->first_page || |
2015 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
2016 | return -EINVAL; | |
98938aa8 | 2017 | |
0dff0846 SC |
2018 | kvm_arch_sync_dirty_log(kvm, memslot); |
2019 | ||
2020 | flush = false; | |
2a31b9db PB |
2021 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
2022 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
2023 | return -EFAULT; | |
ba0513b5 | 2024 | |
531810ca | 2025 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
2026 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
2027 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
2028 | i++, offset += BITS_PER_LONG) { |
2029 | unsigned long mask = *dirty_bitmap_buffer++; | |
2030 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
2031 | if (!mask) | |
ba0513b5 MS |
2032 | continue; |
2033 | ||
2a31b9db | 2034 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 2035 | |
2a31b9db PB |
2036 | /* |
2037 | * mask contains the bits that really have been cleared. This | |
2038 | * never includes any bits beyond the length of the memslot (if | |
2039 | * the length is not aligned to 64 pages), therefore it is not | |
2040 | * a problem if userspace sets them in log->dirty_bitmap. | |
2041 | */ | |
58d2930f | 2042 | if (mask) { |
0dff0846 | 2043 | flush = true; |
58d2930f TY |
2044 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
2045 | offset, mask); | |
2046 | } | |
ba0513b5 | 2047 | } |
531810ca | 2048 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 2049 | |
0dff0846 SC |
2050 | if (flush) |
2051 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2052 | ||
58d6db34 | 2053 | return 0; |
ba0513b5 | 2054 | } |
0dff0846 SC |
2055 | |
2056 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
2057 | struct kvm_clear_dirty_log *log) | |
2058 | { | |
2059 | int r; | |
2060 | ||
2061 | mutex_lock(&kvm->slots_lock); | |
2062 | ||
2063 | r = kvm_clear_dirty_log_protect(kvm, log); | |
2064 | ||
2065 | mutex_unlock(&kvm->slots_lock); | |
2066 | return r; | |
2067 | } | |
2068 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 2069 | |
49c7754c GN |
2070 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
2071 | { | |
2072 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
2073 | } | |
a1f4d395 | 2074 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 2075 | |
8e73485c PB |
2076 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
2077 | { | |
fe22ed82 DM |
2078 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
2079 | struct kvm_memory_slot *slot; | |
2080 | int slot_index; | |
2081 | ||
2082 | slot = try_get_memslot(slots, vcpu->last_used_slot, gfn); | |
2083 | if (slot) | |
2084 | return slot; | |
2085 | ||
2086 | /* | |
2087 | * Fall back to searching all memslots. We purposely use | |
2088 | * search_memslots() instead of __gfn_to_memslot() to avoid | |
2089 | * thrashing the VM-wide last_used_index in kvm_memslots. | |
2090 | */ | |
2091 | slot = search_memslots(slots, gfn, &slot_index); | |
2092 | if (slot) { | |
2093 | vcpu->last_used_slot = slot_index; | |
2094 | return slot; | |
2095 | } | |
2096 | ||
2097 | return NULL; | |
8e73485c | 2098 | } |
e72436bc | 2099 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot); |
8e73485c | 2100 | |
33e94154 | 2101 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 2102 | { |
bf3e05bc | 2103 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 2104 | |
c36b7150 | 2105 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
2106 | } |
2107 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
2108 | ||
995decb6 VK |
2109 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
2110 | { | |
2111 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2112 | ||
2113 | return kvm_is_visible_memslot(memslot); | |
2114 | } | |
2115 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
2116 | ||
f9b84e19 | 2117 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
2118 | { |
2119 | struct vm_area_struct *vma; | |
2120 | unsigned long addr, size; | |
2121 | ||
2122 | size = PAGE_SIZE; | |
2123 | ||
42cde48b | 2124 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
2125 | if (kvm_is_error_hva(addr)) |
2126 | return PAGE_SIZE; | |
2127 | ||
d8ed45c5 | 2128 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
2129 | vma = find_vma(current->mm, addr); |
2130 | if (!vma) | |
2131 | goto out; | |
2132 | ||
2133 | size = vma_kernel_pagesize(vma); | |
2134 | ||
2135 | out: | |
d8ed45c5 | 2136 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
2137 | |
2138 | return size; | |
2139 | } | |
2140 | ||
4d8b81ab XG |
2141 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
2142 | { | |
2143 | return slot->flags & KVM_MEM_READONLY; | |
2144 | } | |
2145 | ||
4d8b81ab XG |
2146 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2147 | gfn_t *nr_pages, bool write) | |
539cb660 | 2148 | { |
bc6678a3 | 2149 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 2150 | return KVM_HVA_ERR_BAD; |
48987781 | 2151 | |
4d8b81ab XG |
2152 | if (memslot_is_readonly(slot) && write) |
2153 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
2154 | |
2155 | if (nr_pages) | |
2156 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
2157 | ||
4d8b81ab | 2158 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 2159 | } |
48987781 | 2160 | |
4d8b81ab XG |
2161 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2162 | gfn_t *nr_pages) | |
2163 | { | |
2164 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 2165 | } |
48987781 | 2166 | |
4d8b81ab | 2167 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 2168 | gfn_t gfn) |
4d8b81ab XG |
2169 | { |
2170 | return gfn_to_hva_many(slot, gfn, NULL); | |
2171 | } | |
2172 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
2173 | ||
48987781 XG |
2174 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
2175 | { | |
49c7754c | 2176 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 2177 | } |
0d150298 | 2178 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 2179 | |
8e73485c PB |
2180 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
2181 | { | |
2182 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
2183 | } | |
2184 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
2185 | ||
86ab8cff | 2186 | /* |
970c0d4b WY |
2187 | * Return the hva of a @gfn and the R/W attribute if possible. |
2188 | * | |
2189 | * @slot: the kvm_memory_slot which contains @gfn | |
2190 | * @gfn: the gfn to be translated | |
2191 | * @writable: used to return the read/write attribute of the @slot if the hva | |
2192 | * is valid and @writable is not NULL | |
86ab8cff | 2193 | */ |
64d83126 CD |
2194 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
2195 | gfn_t gfn, bool *writable) | |
86ab8cff | 2196 | { |
a2ac07fe GN |
2197 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
2198 | ||
2199 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
2200 | *writable = !memslot_is_readonly(slot); |
2201 | ||
a2ac07fe | 2202 | return hva; |
86ab8cff XG |
2203 | } |
2204 | ||
64d83126 CD |
2205 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
2206 | { | |
2207 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2208 | ||
2209 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2210 | } | |
2211 | ||
8e73485c PB |
2212 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
2213 | { | |
2214 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2215 | ||
2216 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2217 | } | |
2218 | ||
fafc3dba HY |
2219 | static inline int check_user_page_hwpoison(unsigned long addr) |
2220 | { | |
0d731759 | 2221 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 2222 | |
0d731759 | 2223 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
2224 | return rc == -EHWPOISON; |
2225 | } | |
2226 | ||
2fc84311 | 2227 | /* |
b9b33da2 PB |
2228 | * The fast path to get the writable pfn which will be stored in @pfn, |
2229 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 2230 | * only part that runs if we can in atomic context. |
2fc84311 | 2231 | */ |
b9b33da2 PB |
2232 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
2233 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 2234 | { |
8d4e1288 | 2235 | struct page *page[1]; |
954bbbc2 | 2236 | |
12ce13fe XG |
2237 | /* |
2238 | * Fast pin a writable pfn only if it is a write fault request | |
2239 | * or the caller allows to map a writable pfn for a read fault | |
2240 | * request. | |
2241 | */ | |
2242 | if (!(write_fault || writable)) | |
2243 | return false; | |
612819c3 | 2244 | |
dadbb612 | 2245 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 2246 | *pfn = page_to_pfn(page[0]); |
612819c3 | 2247 | |
2fc84311 XG |
2248 | if (writable) |
2249 | *writable = true; | |
2250 | return true; | |
2251 | } | |
af585b92 | 2252 | |
2fc84311 XG |
2253 | return false; |
2254 | } | |
612819c3 | 2255 | |
2fc84311 XG |
2256 | /* |
2257 | * The slow path to get the pfn of the specified host virtual address, | |
2258 | * 1 indicates success, -errno is returned if error is detected. | |
2259 | */ | |
2260 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 2261 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 2262 | { |
ce53053c AV |
2263 | unsigned int flags = FOLL_HWPOISON; |
2264 | struct page *page; | |
2fc84311 | 2265 | int npages = 0; |
612819c3 | 2266 | |
2fc84311 XG |
2267 | might_sleep(); |
2268 | ||
2269 | if (writable) | |
2270 | *writable = write_fault; | |
2271 | ||
ce53053c AV |
2272 | if (write_fault) |
2273 | flags |= FOLL_WRITE; | |
2274 | if (async) | |
2275 | flags |= FOLL_NOWAIT; | |
d4944b0e | 2276 | |
ce53053c | 2277 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
2278 | if (npages != 1) |
2279 | return npages; | |
2280 | ||
2281 | /* map read fault as writable if possible */ | |
12ce13fe | 2282 | if (unlikely(!write_fault) && writable) { |
ce53053c | 2283 | struct page *wpage; |
2fc84311 | 2284 | |
dadbb612 | 2285 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 2286 | *writable = true; |
ce53053c AV |
2287 | put_page(page); |
2288 | page = wpage; | |
612819c3 | 2289 | } |
887c08ac | 2290 | } |
ce53053c | 2291 | *pfn = page_to_pfn(page); |
2fc84311 XG |
2292 | return npages; |
2293 | } | |
539cb660 | 2294 | |
4d8b81ab XG |
2295 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
2296 | { | |
2297 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
2298 | return false; | |
2e2e3738 | 2299 | |
4d8b81ab XG |
2300 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
2301 | return false; | |
887c08ac | 2302 | |
4d8b81ab XG |
2303 | return true; |
2304 | } | |
bf998156 | 2305 | |
f8be156b NP |
2306 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
2307 | { | |
2308 | if (kvm_is_reserved_pfn(pfn)) | |
2309 | return 1; | |
2310 | return get_page_unless_zero(pfn_to_page(pfn)); | |
2311 | } | |
2312 | ||
92176a8e PB |
2313 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
2314 | unsigned long addr, bool *async, | |
a340b3e2 KA |
2315 | bool write_fault, bool *writable, |
2316 | kvm_pfn_t *p_pfn) | |
92176a8e | 2317 | { |
a9545779 | 2318 | kvm_pfn_t pfn; |
bd2fae8d PB |
2319 | pte_t *ptep; |
2320 | spinlock_t *ptl; | |
add6a0cd PB |
2321 | int r; |
2322 | ||
9fd6dad1 | 2323 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2324 | if (r) { |
2325 | /* | |
2326 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
2327 | * not call the fault handler, so do it here. | |
2328 | */ | |
2329 | bool unlocked = false; | |
64019a2e | 2330 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
2331 | (write_fault ? FAULT_FLAG_WRITE : 0), |
2332 | &unlocked); | |
a8387d0b PB |
2333 | if (unlocked) |
2334 | return -EAGAIN; | |
add6a0cd PB |
2335 | if (r) |
2336 | return r; | |
2337 | ||
9fd6dad1 | 2338 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2339 | if (r) |
2340 | return r; | |
bd2fae8d | 2341 | } |
add6a0cd | 2342 | |
bd2fae8d PB |
2343 | if (write_fault && !pte_write(*ptep)) { |
2344 | pfn = KVM_PFN_ERR_RO_FAULT; | |
2345 | goto out; | |
add6a0cd PB |
2346 | } |
2347 | ||
a340b3e2 | 2348 | if (writable) |
bd2fae8d PB |
2349 | *writable = pte_write(*ptep); |
2350 | pfn = pte_pfn(*ptep); | |
add6a0cd PB |
2351 | |
2352 | /* | |
2353 | * Get a reference here because callers of *hva_to_pfn* and | |
2354 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
2355 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
36c3ce6c | 2356 | * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will |
add6a0cd PB |
2357 | * simply do nothing for reserved pfns. |
2358 | * | |
2359 | * Whoever called remap_pfn_range is also going to call e.g. | |
2360 | * unmap_mapping_range before the underlying pages are freed, | |
2361 | * causing a call to our MMU notifier. | |
f8be156b NP |
2362 | * |
2363 | * Certain IO or PFNMAP mappings can be backed with valid | |
2364 | * struct pages, but be allocated without refcounting e.g., | |
2365 | * tail pages of non-compound higher order allocations, which | |
2366 | * would then underflow the refcount when the caller does the | |
2367 | * required put_page. Don't allow those pages here. | |
add6a0cd | 2368 | */ |
f8be156b NP |
2369 | if (!kvm_try_get_pfn(pfn)) |
2370 | r = -EFAULT; | |
add6a0cd | 2371 | |
bd2fae8d PB |
2372 | out: |
2373 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 2374 | *p_pfn = pfn; |
f8be156b NP |
2375 | |
2376 | return r; | |
92176a8e PB |
2377 | } |
2378 | ||
12ce13fe XG |
2379 | /* |
2380 | * Pin guest page in memory and return its pfn. | |
2381 | * @addr: host virtual address which maps memory to the guest | |
2382 | * @atomic: whether this function can sleep | |
2383 | * @async: whether this function need to wait IO complete if the | |
2384 | * host page is not in the memory | |
2385 | * @write_fault: whether we should get a writable host page | |
2386 | * @writable: whether it allows to map a writable host page for !@write_fault | |
2387 | * | |
2388 | * The function will map a writable host page for these two cases: | |
2389 | * 1): @write_fault = true | |
2390 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
2391 | * whether the mapping is writable. | |
2392 | */ | |
ba049e93 | 2393 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
2394 | bool write_fault, bool *writable) |
2395 | { | |
2396 | struct vm_area_struct *vma; | |
ba049e93 | 2397 | kvm_pfn_t pfn = 0; |
92176a8e | 2398 | int npages, r; |
2e2e3738 | 2399 | |
2fc84311 XG |
2400 | /* we can do it either atomically or asynchronously, not both */ |
2401 | BUG_ON(atomic && async); | |
8d4e1288 | 2402 | |
b9b33da2 | 2403 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
2404 | return pfn; |
2405 | ||
2406 | if (atomic) | |
2407 | return KVM_PFN_ERR_FAULT; | |
2408 | ||
2409 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
2410 | if (npages == 1) | |
2411 | return pfn; | |
8d4e1288 | 2412 | |
d8ed45c5 | 2413 | mmap_read_lock(current->mm); |
2fc84311 XG |
2414 | if (npages == -EHWPOISON || |
2415 | (!async && check_user_page_hwpoison(addr))) { | |
2416 | pfn = KVM_PFN_ERR_HWPOISON; | |
2417 | goto exit; | |
2418 | } | |
2419 | ||
a8387d0b | 2420 | retry: |
fc98c03b | 2421 | vma = vma_lookup(current->mm, addr); |
2fc84311 XG |
2422 | |
2423 | if (vma == NULL) | |
2424 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 2425 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
a340b3e2 | 2426 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn); |
a8387d0b PB |
2427 | if (r == -EAGAIN) |
2428 | goto retry; | |
92176a8e PB |
2429 | if (r < 0) |
2430 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 2431 | } else { |
4d8b81ab | 2432 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
2433 | *async = true; |
2434 | pfn = KVM_PFN_ERR_FAULT; | |
2435 | } | |
2436 | exit: | |
d8ed45c5 | 2437 | mmap_read_unlock(current->mm); |
2e2e3738 | 2438 | return pfn; |
35149e21 AL |
2439 | } |
2440 | ||
ba049e93 DW |
2441 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
2442 | bool atomic, bool *async, bool write_fault, | |
4a42d848 | 2443 | bool *writable, hva_t *hva) |
887c08ac | 2444 | { |
4d8b81ab XG |
2445 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
2446 | ||
4a42d848 DS |
2447 | if (hva) |
2448 | *hva = addr; | |
2449 | ||
b2740d35 PB |
2450 | if (addr == KVM_HVA_ERR_RO_BAD) { |
2451 | if (writable) | |
2452 | *writable = false; | |
4d8b81ab | 2453 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 2454 | } |
4d8b81ab | 2455 | |
b2740d35 PB |
2456 | if (kvm_is_error_hva(addr)) { |
2457 | if (writable) | |
2458 | *writable = false; | |
81c52c56 | 2459 | return KVM_PFN_NOSLOT; |
b2740d35 | 2460 | } |
4d8b81ab XG |
2461 | |
2462 | /* Do not map writable pfn in the readonly memslot. */ | |
2463 | if (writable && memslot_is_readonly(slot)) { | |
2464 | *writable = false; | |
2465 | writable = NULL; | |
2466 | } | |
2467 | ||
2468 | return hva_to_pfn(addr, atomic, async, write_fault, | |
2469 | writable); | |
887c08ac | 2470 | } |
3520469d | 2471 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 2472 | |
ba049e93 | 2473 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
2474 | bool *writable) |
2475 | { | |
e37afc6e | 2476 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
4a42d848 | 2477 | write_fault, writable, NULL); |
612819c3 MT |
2478 | } |
2479 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
2480 | ||
ba049e93 | 2481 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2482 | { |
4a42d848 | 2483 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL); |
506f0d6f | 2484 | } |
e37afc6e | 2485 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 2486 | |
ba049e93 | 2487 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2488 | { |
4a42d848 | 2489 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL); |
506f0d6f | 2490 | } |
037d92dc | 2491 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 2492 | |
ba049e93 | 2493 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2494 | { |
2495 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2496 | } | |
2497 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
2498 | ||
ba049e93 | 2499 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
2500 | { |
2501 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
2502 | } | |
2503 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
2504 | ||
ba049e93 | 2505 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2506 | { |
2507 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2508 | } | |
2509 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
2510 | ||
d9ef13c2 PB |
2511 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2512 | struct page **pages, int nr_pages) | |
48987781 XG |
2513 | { |
2514 | unsigned long addr; | |
076b925d | 2515 | gfn_t entry = 0; |
48987781 | 2516 | |
d9ef13c2 | 2517 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
2518 | if (kvm_is_error_hva(addr)) |
2519 | return -1; | |
2520 | ||
2521 | if (entry < nr_pages) | |
2522 | return 0; | |
2523 | ||
dadbb612 | 2524 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
2525 | } |
2526 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
2527 | ||
ba049e93 | 2528 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 2529 | { |
81c52c56 | 2530 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 2531 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2532 | |
bf4bea8e | 2533 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 2534 | WARN_ON(1); |
6cede2e6 | 2535 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 2536 | } |
a2766325 XG |
2537 | |
2538 | return pfn_to_page(pfn); | |
2539 | } | |
2540 | ||
35149e21 AL |
2541 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
2542 | { | |
ba049e93 | 2543 | kvm_pfn_t pfn; |
2e2e3738 AL |
2544 | |
2545 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 2546 | |
a2766325 | 2547 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
2548 | } |
2549 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
2550 | ||
91724814 BO |
2551 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache) |
2552 | { | |
2553 | if (pfn == 0) | |
2554 | return; | |
2555 | ||
2556 | if (cache) | |
2557 | cache->pfn = cache->gfn = 0; | |
2558 | ||
2559 | if (dirty) | |
2560 | kvm_release_pfn_dirty(pfn); | |
2561 | else | |
2562 | kvm_release_pfn_clean(pfn); | |
2563 | } | |
2564 | ||
2565 | static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn, | |
2566 | struct gfn_to_pfn_cache *cache, u64 gen) | |
2567 | { | |
2568 | kvm_release_pfn(cache->pfn, cache->dirty, cache); | |
2569 | ||
2570 | cache->pfn = gfn_to_pfn_memslot(slot, gfn); | |
2571 | cache->gfn = gfn; | |
2572 | cache->dirty = false; | |
2573 | cache->generation = gen; | |
2574 | } | |
2575 | ||
1eff70a9 | 2576 | static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn, |
91724814 BO |
2577 | struct kvm_host_map *map, |
2578 | struct gfn_to_pfn_cache *cache, | |
2579 | bool atomic) | |
e45adf66 KA |
2580 | { |
2581 | kvm_pfn_t pfn; | |
2582 | void *hva = NULL; | |
2583 | struct page *page = KVM_UNMAPPED_PAGE; | |
1eff70a9 | 2584 | struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn); |
91724814 | 2585 | u64 gen = slots->generation; |
e45adf66 KA |
2586 | |
2587 | if (!map) | |
2588 | return -EINVAL; | |
2589 | ||
91724814 BO |
2590 | if (cache) { |
2591 | if (!cache->pfn || cache->gfn != gfn || | |
2592 | cache->generation != gen) { | |
2593 | if (atomic) | |
2594 | return -EAGAIN; | |
2595 | kvm_cache_gfn_to_pfn(slot, gfn, cache, gen); | |
2596 | } | |
2597 | pfn = cache->pfn; | |
2598 | } else { | |
2599 | if (atomic) | |
2600 | return -EAGAIN; | |
2601 | pfn = gfn_to_pfn_memslot(slot, gfn); | |
2602 | } | |
e45adf66 KA |
2603 | if (is_error_noslot_pfn(pfn)) |
2604 | return -EINVAL; | |
2605 | ||
2606 | if (pfn_valid(pfn)) { | |
2607 | page = pfn_to_page(pfn); | |
91724814 BO |
2608 | if (atomic) |
2609 | hva = kmap_atomic(page); | |
2610 | else | |
2611 | hva = kmap(page); | |
d30b214d | 2612 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2613 | } else if (!atomic) { |
e45adf66 | 2614 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
91724814 BO |
2615 | } else { |
2616 | return -EINVAL; | |
d30b214d | 2617 | #endif |
e45adf66 KA |
2618 | } |
2619 | ||
2620 | if (!hva) | |
2621 | return -EFAULT; | |
2622 | ||
2623 | map->page = page; | |
2624 | map->hva = hva; | |
2625 | map->pfn = pfn; | |
2626 | map->gfn = gfn; | |
2627 | ||
2628 | return 0; | |
2629 | } | |
2630 | ||
91724814 BO |
2631 | int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map, |
2632 | struct gfn_to_pfn_cache *cache, bool atomic) | |
1eff70a9 | 2633 | { |
91724814 BO |
2634 | return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map, |
2635 | cache, atomic); | |
1eff70a9 BO |
2636 | } |
2637 | EXPORT_SYMBOL_GPL(kvm_map_gfn); | |
2638 | ||
e45adf66 KA |
2639 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
2640 | { | |
91724814 BO |
2641 | return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map, |
2642 | NULL, false); | |
e45adf66 KA |
2643 | } |
2644 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); | |
2645 | ||
28bd726a PX |
2646 | static void __kvm_unmap_gfn(struct kvm *kvm, |
2647 | struct kvm_memory_slot *memslot, | |
91724814 BO |
2648 | struct kvm_host_map *map, |
2649 | struct gfn_to_pfn_cache *cache, | |
2650 | bool dirty, bool atomic) | |
e45adf66 KA |
2651 | { |
2652 | if (!map) | |
2653 | return; | |
2654 | ||
2655 | if (!map->hva) | |
2656 | return; | |
2657 | ||
91724814 BO |
2658 | if (map->page != KVM_UNMAPPED_PAGE) { |
2659 | if (atomic) | |
2660 | kunmap_atomic(map->hva); | |
2661 | else | |
2662 | kunmap(map->page); | |
2663 | } | |
eb1f2f38 | 2664 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2665 | else if (!atomic) |
e45adf66 | 2666 | memunmap(map->hva); |
91724814 BO |
2667 | else |
2668 | WARN_ONCE(1, "Unexpected unmapping in atomic context"); | |
eb1f2f38 | 2669 | #endif |
e45adf66 | 2670 | |
91724814 | 2671 | if (dirty) |
28bd726a | 2672 | mark_page_dirty_in_slot(kvm, memslot, map->gfn); |
91724814 BO |
2673 | |
2674 | if (cache) | |
2675 | cache->dirty |= dirty; | |
2676 | else | |
2677 | kvm_release_pfn(map->pfn, dirty, NULL); | |
e45adf66 KA |
2678 | |
2679 | map->hva = NULL; | |
2680 | map->page = NULL; | |
2681 | } | |
1eff70a9 | 2682 | |
91724814 BO |
2683 | int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map, |
2684 | struct gfn_to_pfn_cache *cache, bool dirty, bool atomic) | |
1eff70a9 | 2685 | { |
28bd726a | 2686 | __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map, |
91724814 | 2687 | cache, dirty, atomic); |
1eff70a9 BO |
2688 | return 0; |
2689 | } | |
2690 | EXPORT_SYMBOL_GPL(kvm_unmap_gfn); | |
2691 | ||
2692 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) | |
2693 | { | |
28bd726a PX |
2694 | __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), |
2695 | map, NULL, dirty, false); | |
1eff70a9 | 2696 | } |
e45adf66 KA |
2697 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); |
2698 | ||
8e73485c PB |
2699 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
2700 | { | |
ba049e93 | 2701 | kvm_pfn_t pfn; |
8e73485c PB |
2702 | |
2703 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
2704 | ||
2705 | return kvm_pfn_to_page(pfn); | |
2706 | } | |
2707 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
2708 | ||
b4231d61 IE |
2709 | void kvm_release_page_clean(struct page *page) |
2710 | { | |
32cad84f XG |
2711 | WARN_ON(is_error_page(page)); |
2712 | ||
35149e21 | 2713 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
2714 | } |
2715 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
2716 | ||
ba049e93 | 2717 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 2718 | { |
bf4bea8e | 2719 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 2720 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
2721 | } |
2722 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
2723 | ||
b4231d61 | 2724 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 2725 | { |
a2766325 XG |
2726 | WARN_ON(is_error_page(page)); |
2727 | ||
35149e21 AL |
2728 | kvm_release_pfn_dirty(page_to_pfn(page)); |
2729 | } | |
2730 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
2731 | ||
f7a6509f | 2732 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
2733 | { |
2734 | kvm_set_pfn_dirty(pfn); | |
2735 | kvm_release_pfn_clean(pfn); | |
2736 | } | |
f7a6509f | 2737 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 2738 | |
ba049e93 | 2739 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2740 | { |
d29c03a5 ML |
2741 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2742 | SetPageDirty(pfn_to_page(pfn)); | |
8a7ae055 | 2743 | } |
35149e21 AL |
2744 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
2745 | ||
ba049e93 | 2746 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 2747 | { |
a78986aa | 2748 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2e2e3738 | 2749 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
2750 | } |
2751 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
2752 | ||
195aefde IE |
2753 | static int next_segment(unsigned long len, int offset) |
2754 | { | |
2755 | if (len > PAGE_SIZE - offset) | |
2756 | return PAGE_SIZE - offset; | |
2757 | else | |
2758 | return len; | |
2759 | } | |
2760 | ||
8e73485c PB |
2761 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
2762 | void *data, int offset, int len) | |
195aefde | 2763 | { |
e0506bcb IE |
2764 | int r; |
2765 | unsigned long addr; | |
195aefde | 2766 | |
8e73485c | 2767 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
2768 | if (kvm_is_error_hva(addr)) |
2769 | return -EFAULT; | |
3180a7fc | 2770 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 2771 | if (r) |
195aefde | 2772 | return -EFAULT; |
195aefde IE |
2773 | return 0; |
2774 | } | |
8e73485c PB |
2775 | |
2776 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
2777 | int len) | |
2778 | { | |
2779 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2780 | ||
2781 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2782 | } | |
195aefde IE |
2783 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
2784 | ||
8e73485c PB |
2785 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
2786 | int offset, int len) | |
2787 | { | |
2788 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2789 | ||
2790 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2791 | } | |
2792 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
2793 | ||
195aefde IE |
2794 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
2795 | { | |
2796 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2797 | int seg; | |
2798 | int offset = offset_in_page(gpa); | |
2799 | int ret; | |
2800 | ||
2801 | while ((seg = next_segment(len, offset)) != 0) { | |
2802 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
2803 | if (ret < 0) | |
2804 | return ret; | |
2805 | offset = 0; | |
2806 | len -= seg; | |
2807 | data += seg; | |
2808 | ++gfn; | |
2809 | } | |
2810 | return 0; | |
2811 | } | |
2812 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
2813 | ||
8e73485c | 2814 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 2815 | { |
7ec54588 | 2816 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 2817 | int seg; |
7ec54588 | 2818 | int offset = offset_in_page(gpa); |
8e73485c PB |
2819 | int ret; |
2820 | ||
2821 | while ((seg = next_segment(len, offset)) != 0) { | |
2822 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
2823 | if (ret < 0) | |
2824 | return ret; | |
2825 | offset = 0; | |
2826 | len -= seg; | |
2827 | data += seg; | |
2828 | ++gfn; | |
2829 | } | |
2830 | return 0; | |
2831 | } | |
2832 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 2833 | |
8e73485c PB |
2834 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2835 | void *data, int offset, unsigned long len) | |
2836 | { | |
2837 | int r; | |
2838 | unsigned long addr; | |
2839 | ||
2840 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
2841 | if (kvm_is_error_hva(addr)) |
2842 | return -EFAULT; | |
0aac03f0 | 2843 | pagefault_disable(); |
3180a7fc | 2844 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 2845 | pagefault_enable(); |
7ec54588 MT |
2846 | if (r) |
2847 | return -EFAULT; | |
2848 | return 0; | |
2849 | } | |
7ec54588 | 2850 | |
8e73485c PB |
2851 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
2852 | void *data, unsigned long len) | |
2853 | { | |
2854 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2855 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2856 | int offset = offset_in_page(gpa); | |
2857 | ||
2858 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2859 | } | |
2860 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
2861 | ||
28bd726a PX |
2862 | static int __kvm_write_guest_page(struct kvm *kvm, |
2863 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 2864 | const void *data, int offset, int len) |
195aefde | 2865 | { |
e0506bcb IE |
2866 | int r; |
2867 | unsigned long addr; | |
195aefde | 2868 | |
251eb841 | 2869 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
2870 | if (kvm_is_error_hva(addr)) |
2871 | return -EFAULT; | |
8b0cedff | 2872 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 2873 | if (r) |
195aefde | 2874 | return -EFAULT; |
28bd726a | 2875 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
2876 | return 0; |
2877 | } | |
8e73485c PB |
2878 | |
2879 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
2880 | const void *data, int offset, int len) | |
2881 | { | |
2882 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2883 | ||
28bd726a | 2884 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 2885 | } |
195aefde IE |
2886 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
2887 | ||
8e73485c PB |
2888 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
2889 | const void *data, int offset, int len) | |
2890 | { | |
2891 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2892 | ||
28bd726a | 2893 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
2894 | } |
2895 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
2896 | ||
195aefde IE |
2897 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
2898 | unsigned long len) | |
2899 | { | |
2900 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2901 | int seg; | |
2902 | int offset = offset_in_page(gpa); | |
2903 | int ret; | |
2904 | ||
2905 | while ((seg = next_segment(len, offset)) != 0) { | |
2906 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
2907 | if (ret < 0) | |
2908 | return ret; | |
2909 | offset = 0; | |
2910 | len -= seg; | |
2911 | data += seg; | |
2912 | ++gfn; | |
2913 | } | |
2914 | return 0; | |
2915 | } | |
ff651cb6 | 2916 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 2917 | |
8e73485c PB |
2918 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
2919 | unsigned long len) | |
2920 | { | |
2921 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2922 | int seg; | |
2923 | int offset = offset_in_page(gpa); | |
2924 | int ret; | |
2925 | ||
2926 | while ((seg = next_segment(len, offset)) != 0) { | |
2927 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
2928 | if (ret < 0) | |
2929 | return ret; | |
2930 | offset = 0; | |
2931 | len -= seg; | |
2932 | data += seg; | |
2933 | ++gfn; | |
2934 | } | |
2935 | return 0; | |
2936 | } | |
2937 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
2938 | ||
5a2d4365 PB |
2939 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
2940 | struct gfn_to_hva_cache *ghc, | |
2941 | gpa_t gpa, unsigned long len) | |
49c7754c | 2942 | { |
49c7754c | 2943 | int offset = offset_in_page(gpa); |
8f964525 AH |
2944 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
2945 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
2946 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
2947 | gfn_t nr_pages_avail; | |
49c7754c | 2948 | |
6ad1e29f | 2949 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 2950 | ghc->generation = slots->generation; |
6ad1e29f SC |
2951 | |
2952 | if (start_gfn > end_gfn) { | |
2953 | ghc->hva = KVM_HVA_ERR_BAD; | |
2954 | return -EINVAL; | |
2955 | } | |
f1b9dd5e JM |
2956 | |
2957 | /* | |
2958 | * If the requested region crosses two memslots, we still | |
2959 | * verify that the entire region is valid here. | |
2960 | */ | |
6ad1e29f | 2961 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
2962 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
2963 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
2964 | &nr_pages_avail); | |
2965 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 2966 | return -EFAULT; |
f1b9dd5e JM |
2967 | } |
2968 | ||
2969 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 2970 | if (nr_pages_needed == 1) |
49c7754c | 2971 | ghc->hva += offset; |
f1b9dd5e | 2972 | else |
8f964525 | 2973 | ghc->memslot = NULL; |
f1b9dd5e | 2974 | |
6ad1e29f SC |
2975 | ghc->gpa = gpa; |
2976 | ghc->len = len; | |
2977 | return 0; | |
49c7754c | 2978 | } |
5a2d4365 | 2979 | |
4e335d9e | 2980 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
2981 | gpa_t gpa, unsigned long len) |
2982 | { | |
4e335d9e | 2983 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
2984 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
2985 | } | |
4e335d9e | 2986 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 2987 | |
4e335d9e | 2988 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
2989 | void *data, unsigned int offset, |
2990 | unsigned long len) | |
49c7754c | 2991 | { |
4e335d9e | 2992 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 2993 | int r; |
4ec6e863 | 2994 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 2995 | |
4ec6e863 | 2996 | BUG_ON(len + offset > ghc->len); |
8f964525 | 2997 | |
dc9ce71e SC |
2998 | if (slots->generation != ghc->generation) { |
2999 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3000 | return -EFAULT; | |
3001 | } | |
8f964525 | 3002 | |
49c7754c GN |
3003 | if (kvm_is_error_hva(ghc->hva)) |
3004 | return -EFAULT; | |
3005 | ||
fcfbc617 SC |
3006 | if (unlikely(!ghc->memslot)) |
3007 | return kvm_write_guest(kvm, gpa, data, len); | |
3008 | ||
4ec6e863 | 3009 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
3010 | if (r) |
3011 | return -EFAULT; | |
28bd726a | 3012 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
3013 | |
3014 | return 0; | |
3015 | } | |
4e335d9e | 3016 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 3017 | |
4e335d9e PB |
3018 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3019 | void *data, unsigned long len) | |
4ec6e863 | 3020 | { |
4e335d9e | 3021 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 3022 | } |
4e335d9e | 3023 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 3024 | |
0958f0ce VK |
3025 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3026 | void *data, unsigned int offset, | |
3027 | unsigned long len) | |
e03b644f | 3028 | { |
4e335d9e | 3029 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 3030 | int r; |
0958f0ce | 3031 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 3032 | |
0958f0ce | 3033 | BUG_ON(len + offset > ghc->len); |
8f964525 | 3034 | |
dc9ce71e SC |
3035 | if (slots->generation != ghc->generation) { |
3036 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3037 | return -EFAULT; | |
3038 | } | |
8f964525 | 3039 | |
e03b644f GN |
3040 | if (kvm_is_error_hva(ghc->hva)) |
3041 | return -EFAULT; | |
3042 | ||
fcfbc617 | 3043 | if (unlikely(!ghc->memslot)) |
0958f0ce | 3044 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 3045 | |
0958f0ce | 3046 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
3047 | if (r) |
3048 | return -EFAULT; | |
3049 | ||
3050 | return 0; | |
3051 | } | |
0958f0ce VK |
3052 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
3053 | ||
3054 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
3055 | void *data, unsigned long len) | |
3056 | { | |
3057 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
3058 | } | |
4e335d9e | 3059 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 3060 | |
195aefde IE |
3061 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
3062 | { | |
2f541442 | 3063 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
3064 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3065 | int seg; | |
3066 | int offset = offset_in_page(gpa); | |
3067 | int ret; | |
3068 | ||
bfda0e84 | 3069 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 3070 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
3071 | if (ret < 0) |
3072 | return ret; | |
3073 | offset = 0; | |
3074 | len -= seg; | |
3075 | ++gfn; | |
3076 | } | |
3077 | return 0; | |
3078 | } | |
3079 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
3080 | ||
28bd726a PX |
3081 | void mark_page_dirty_in_slot(struct kvm *kvm, |
3082 | struct kvm_memory_slot *memslot, | |
3083 | gfn_t gfn) | |
6aa8b732 | 3084 | { |
044c59c4 | 3085 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 3086 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 3087 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 3088 | |
fb04a1ed PX |
3089 | if (kvm->dirty_ring_size) |
3090 | kvm_dirty_ring_push(kvm_dirty_ring_get(kvm), | |
3091 | slot, rel_gfn); | |
3092 | else | |
3093 | set_bit_le(rel_gfn, memslot->dirty_bitmap); | |
6aa8b732 AK |
3094 | } |
3095 | } | |
a6a0b05d | 3096 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 3097 | |
49c7754c GN |
3098 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
3099 | { | |
3100 | struct kvm_memory_slot *memslot; | |
3101 | ||
3102 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 3103 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 3104 | } |
2ba9f0d8 | 3105 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 3106 | |
8e73485c PB |
3107 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
3108 | { | |
3109 | struct kvm_memory_slot *memslot; | |
3110 | ||
3111 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 3112 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
3113 | } |
3114 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
3115 | ||
20b7035c JS |
3116 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
3117 | { | |
3118 | if (!vcpu->sigset_active) | |
3119 | return; | |
3120 | ||
3121 | /* | |
3122 | * This does a lockless modification of ->real_blocked, which is fine | |
3123 | * because, only current can change ->real_blocked and all readers of | |
3124 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
3125 | * of ->blocked. | |
3126 | */ | |
3127 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
3128 | } | |
3129 | ||
3130 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
3131 | { | |
3132 | if (!vcpu->sigset_active) | |
3133 | return; | |
3134 | ||
3135 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
3136 | sigemptyset(¤t->real_blocked); | |
3137 | } | |
3138 | ||
aca6ff29 WL |
3139 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
3140 | { | |
dee339b5 | 3141 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 3142 | |
2cbd7824 | 3143 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 3144 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3145 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
3146 | if (!grow) |
3147 | goto out; | |
3148 | ||
dee339b5 NW |
3149 | val *= grow; |
3150 | if (val < grow_start) | |
3151 | val = grow_start; | |
aca6ff29 | 3152 | |
258785ef DM |
3153 | if (val > vcpu->kvm->max_halt_poll_ns) |
3154 | val = vcpu->kvm->max_halt_poll_ns; | |
313f636d | 3155 | |
aca6ff29 | 3156 | vcpu->halt_poll_ns = val; |
7fa08e71 | 3157 | out: |
2cbd7824 | 3158 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3159 | } |
3160 | ||
3161 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
3162 | { | |
ae232ea4 | 3163 | unsigned int old, val, shrink, grow_start; |
aca6ff29 | 3164 | |
2cbd7824 | 3165 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 3166 | shrink = READ_ONCE(halt_poll_ns_shrink); |
ae232ea4 | 3167 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3168 | if (shrink == 0) |
aca6ff29 WL |
3169 | val = 0; |
3170 | else | |
6b6de68c | 3171 | val /= shrink; |
aca6ff29 | 3172 | |
ae232ea4 SS |
3173 | if (val < grow_start) |
3174 | val = 0; | |
3175 | ||
aca6ff29 | 3176 | vcpu->halt_poll_ns = val; |
2cbd7824 | 3177 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3178 | } |
3179 | ||
f7819512 PB |
3180 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
3181 | { | |
50c28f21 JS |
3182 | int ret = -EINTR; |
3183 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
3184 | ||
f7819512 PB |
3185 | if (kvm_arch_vcpu_runnable(vcpu)) { |
3186 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 3187 | goto out; |
f7819512 PB |
3188 | } |
3189 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 3190 | goto out; |
f7819512 | 3191 | if (signal_pending(current)) |
50c28f21 | 3192 | goto out; |
084071d5 MT |
3193 | if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu)) |
3194 | goto out; | |
f7819512 | 3195 | |
50c28f21 JS |
3196 | ret = 0; |
3197 | out: | |
3198 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
3199 | return ret; | |
f7819512 PB |
3200 | } |
3201 | ||
cb953129 DM |
3202 | static inline void |
3203 | update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited) | |
3204 | { | |
3205 | if (waited) | |
0193cc90 | 3206 | vcpu->stat.generic.halt_poll_fail_ns += poll_ns; |
cb953129 | 3207 | else |
0193cc90 | 3208 | vcpu->stat.generic.halt_poll_success_ns += poll_ns; |
cb953129 DM |
3209 | } |
3210 | ||
b6958ce4 ED |
3211 | /* |
3212 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
3213 | */ | |
8776e519 | 3214 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 3215 | { |
cb953129 | 3216 | ktime_t start, cur, poll_end; |
f7819512 | 3217 | bool waited = false; |
aca6ff29 | 3218 | u64 block_ns; |
f7819512 | 3219 | |
07ab0f8d MZ |
3220 | kvm_arch_vcpu_blocking(vcpu); |
3221 | ||
cb953129 | 3222 | start = cur = poll_end = ktime_get(); |
cdd6ad3a | 3223 | if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) { |
19020f8a | 3224 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); |
f95ef0cd | 3225 | |
0193cc90 | 3226 | ++vcpu->stat.generic.halt_attempted_poll; |
f7819512 PB |
3227 | do { |
3228 | /* | |
3229 | * This sets KVM_REQ_UNHALT if an interrupt | |
3230 | * arrives. | |
3231 | */ | |
3232 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
0193cc90 | 3233 | ++vcpu->stat.generic.halt_successful_poll; |
3491caf2 | 3234 | if (!vcpu_valid_wakeup(vcpu)) |
0193cc90 | 3235 | ++vcpu->stat.generic.halt_poll_invalid; |
8ccba534 JZ |
3236 | |
3237 | KVM_STATS_LOG_HIST_UPDATE( | |
3238 | vcpu->stat.generic.halt_poll_success_hist, | |
3239 | ktime_to_ns(ktime_get()) - | |
3240 | ktime_to_ns(start)); | |
f7819512 PB |
3241 | goto out; |
3242 | } | |
74775654 | 3243 | cpu_relax(); |
cb953129 | 3244 | poll_end = cur = ktime_get(); |
6bd5b743 | 3245 | } while (kvm_vcpu_can_poll(cur, stop)); |
8ccba534 JZ |
3246 | |
3247 | KVM_STATS_LOG_HIST_UPDATE( | |
3248 | vcpu->stat.generic.halt_poll_fail_hist, | |
3249 | ktime_to_ns(ktime_get()) - ktime_to_ns(start)); | |
f7819512 | 3250 | } |
e5c239cf | 3251 | |
8ccba534 | 3252 | |
da4ad88c | 3253 | prepare_to_rcuwait(&vcpu->wait); |
e5c239cf | 3254 | for (;;) { |
da4ad88c | 3255 | set_current_state(TASK_INTERRUPTIBLE); |
e5c239cf | 3256 | |
f7819512 | 3257 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
3258 | break; |
3259 | ||
f7819512 | 3260 | waited = true; |
b6958ce4 | 3261 | schedule(); |
b6958ce4 | 3262 | } |
da4ad88c | 3263 | finish_rcuwait(&vcpu->wait); |
f7819512 | 3264 | cur = ktime_get(); |
87bcc5fa JZ |
3265 | if (waited) { |
3266 | vcpu->stat.generic.halt_wait_ns += | |
3267 | ktime_to_ns(cur) - ktime_to_ns(poll_end); | |
8ccba534 JZ |
3268 | KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist, |
3269 | ktime_to_ns(cur) - ktime_to_ns(poll_end)); | |
87bcc5fa | 3270 | } |
f7819512 | 3271 | out: |
07ab0f8d | 3272 | kvm_arch_vcpu_unblocking(vcpu); |
aca6ff29 WL |
3273 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
3274 | ||
cb953129 DM |
3275 | update_halt_poll_stats( |
3276 | vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited); | |
3277 | ||
44551b2f WL |
3278 | if (!kvm_arch_no_poll(vcpu)) { |
3279 | if (!vcpu_valid_wakeup(vcpu)) { | |
aca6ff29 | 3280 | shrink_halt_poll_ns(vcpu); |
acd05785 | 3281 | } else if (vcpu->kvm->max_halt_poll_ns) { |
44551b2f WL |
3282 | if (block_ns <= vcpu->halt_poll_ns) |
3283 | ; | |
3284 | /* we had a long block, shrink polling */ | |
acd05785 DM |
3285 | else if (vcpu->halt_poll_ns && |
3286 | block_ns > vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
3287 | shrink_halt_poll_ns(vcpu); |
3288 | /* we had a short halt and our poll time is too small */ | |
acd05785 DM |
3289 | else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && |
3290 | block_ns < vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
3291 | grow_halt_poll_ns(vcpu); |
3292 | } else { | |
3293 | vcpu->halt_poll_ns = 0; | |
3294 | } | |
3295 | } | |
aca6ff29 | 3296 | |
3491caf2 CB |
3297 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
3298 | kvm_arch_vcpu_block_finish(vcpu); | |
b6958ce4 | 3299 | } |
2ba9f0d8 | 3300 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 3301 | |
178f02ff | 3302 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 3303 | { |
da4ad88c | 3304 | struct rcuwait *waitp; |
b6d33834 | 3305 | |
da4ad88c DB |
3306 | waitp = kvm_arch_vcpu_get_wait(vcpu); |
3307 | if (rcuwait_wake_up(waitp)) { | |
d73eb57b | 3308 | WRITE_ONCE(vcpu->ready, true); |
0193cc90 | 3309 | ++vcpu->stat.generic.halt_wakeup; |
178f02ff | 3310 | return true; |
b6d33834 CD |
3311 | } |
3312 | ||
178f02ff | 3313 | return false; |
dd1a4cc1 RK |
3314 | } |
3315 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
3316 | ||
0266c894 | 3317 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
3318 | /* |
3319 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
3320 | */ | |
3321 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
3322 | { | |
85b64045 | 3323 | int me, cpu; |
dd1a4cc1 | 3324 | |
178f02ff RK |
3325 | if (kvm_vcpu_wake_up(vcpu)) |
3326 | return; | |
3327 | ||
85b64045 SC |
3328 | /* |
3329 | * Note, the vCPU could get migrated to a different pCPU at any point | |
3330 | * after kvm_arch_vcpu_should_kick(), which could result in sending an | |
3331 | * IPI to the previous pCPU. But, that's ok because the purpose of the | |
3332 | * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the | |
3333 | * vCPU also requires it to leave IN_GUEST_MODE. | |
3334 | */ | |
b6d33834 | 3335 | me = get_cpu(); |
85b64045 SC |
3336 | if (kvm_arch_vcpu_should_kick(vcpu)) { |
3337 | cpu = READ_ONCE(vcpu->cpu); | |
3338 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
b6d33834 | 3339 | smp_send_reschedule(cpu); |
85b64045 | 3340 | } |
b6d33834 CD |
3341 | put_cpu(); |
3342 | } | |
a20ed54d | 3343 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 3344 | #endif /* !CONFIG_S390 */ |
b6d33834 | 3345 | |
fa93384f | 3346 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
3347 | { |
3348 | struct pid *pid; | |
3349 | struct task_struct *task = NULL; | |
fa93384f | 3350 | int ret = 0; |
41628d33 KW |
3351 | |
3352 | rcu_read_lock(); | |
3353 | pid = rcu_dereference(target->pid); | |
3354 | if (pid) | |
27fbe64b | 3355 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
3356 | rcu_read_unlock(); |
3357 | if (!task) | |
c45c528e | 3358 | return ret; |
c45c528e | 3359 | ret = yield_to(task, 1); |
41628d33 | 3360 | put_task_struct(task); |
c45c528e R |
3361 | |
3362 | return ret; | |
41628d33 KW |
3363 | } |
3364 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
3365 | ||
06e48c51 R |
3366 | /* |
3367 | * Helper that checks whether a VCPU is eligible for directed yield. | |
3368 | * Most eligible candidate to yield is decided by following heuristics: | |
3369 | * | |
3370 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
3371 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 3372 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
3373 | * |
3374 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
3375 | * chance last time (mostly it has become eligible now since we have probably | |
3376 | * yielded to lockholder in last iteration. This is done by toggling | |
3377 | * @dy_eligible each time a VCPU checked for eligibility.) | |
3378 | * | |
3379 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
3380 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
3381 | * burning. Giving priority for a potential lock-holder increases lock | |
3382 | * progress. | |
3383 | * | |
3384 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
3385 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
3386 | * and continue with next VCPU and so on. | |
3387 | */ | |
7940876e | 3388 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 3389 | { |
4a55dd72 | 3390 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
3391 | bool eligible; |
3392 | ||
3393 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 3394 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
3395 | |
3396 | if (vcpu->spin_loop.in_spin_loop) | |
3397 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
3398 | ||
3399 | return eligible; | |
4a55dd72 SW |
3400 | #else |
3401 | return true; | |
06e48c51 | 3402 | #endif |
4a55dd72 | 3403 | } |
c45c528e | 3404 | |
17e433b5 WL |
3405 | /* |
3406 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
3407 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
3408 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
3409 | */ | |
3410 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
3411 | { | |
3412 | return kvm_arch_vcpu_runnable(vcpu); | |
3413 | } | |
3414 | ||
3415 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
3416 | { | |
3417 | if (kvm_arch_dy_runnable(vcpu)) | |
3418 | return true; | |
3419 | ||
3420 | #ifdef CONFIG_KVM_ASYNC_PF | |
3421 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
3422 | return true; | |
3423 | #endif | |
3424 | ||
3425 | return false; | |
3426 | } | |
3427 | ||
52acd22f WL |
3428 | bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
3429 | { | |
3430 | return false; | |
3431 | } | |
3432 | ||
199b5763 | 3433 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 3434 | { |
217ece61 RR |
3435 | struct kvm *kvm = me->kvm; |
3436 | struct kvm_vcpu *vcpu; | |
3437 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
3438 | int yielded = 0; | |
c45c528e | 3439 | int try = 3; |
217ece61 RR |
3440 | int pass; |
3441 | int i; | |
d255f4f2 | 3442 | |
4c088493 | 3443 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
3444 | /* |
3445 | * We boost the priority of a VCPU that is runnable but not | |
3446 | * currently running, because it got preempted by something | |
3447 | * else and called schedule in __vcpu_run. Hopefully that | |
3448 | * VCPU is holding the lock that we need and will release it. | |
3449 | * We approximate round-robin by starting at the last boosted VCPU. | |
3450 | */ | |
c45c528e | 3451 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 3452 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 3453 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
3454 | i = last_boosted_vcpu; |
3455 | continue; | |
3456 | } else if (pass && i > last_boosted_vcpu) | |
3457 | break; | |
d73eb57b | 3458 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 3459 | continue; |
217ece61 RR |
3460 | if (vcpu == me) |
3461 | continue; | |
da4ad88c DB |
3462 | if (rcuwait_active(&vcpu->wait) && |
3463 | !vcpu_dy_runnable(vcpu)) | |
217ece61 | 3464 | continue; |
046ddeed | 3465 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
52acd22f WL |
3466 | !kvm_arch_dy_has_pending_interrupt(vcpu) && |
3467 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 3468 | continue; |
06e48c51 R |
3469 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
3470 | continue; | |
c45c528e R |
3471 | |
3472 | yielded = kvm_vcpu_yield_to(vcpu); | |
3473 | if (yielded > 0) { | |
217ece61 | 3474 | kvm->last_boosted_vcpu = i; |
217ece61 | 3475 | break; |
c45c528e R |
3476 | } else if (yielded < 0) { |
3477 | try--; | |
3478 | if (!try) | |
3479 | break; | |
217ece61 | 3480 | } |
217ece61 RR |
3481 | } |
3482 | } | |
4c088493 | 3483 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
3484 | |
3485 | /* Ensure vcpu is not eligible during next spinloop */ | |
3486 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
3487 | } |
3488 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
3489 | ||
fb04a1ed PX |
3490 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
3491 | { | |
3492 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
3493 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && | |
3494 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
3495 | kvm->dirty_ring_size / PAGE_SIZE); | |
3496 | #else | |
3497 | return false; | |
3498 | #endif | |
3499 | } | |
3500 | ||
1499fa80 | 3501 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 3502 | { |
11bac800 | 3503 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
3504 | struct page *page; |
3505 | ||
e4a533a4 | 3506 | if (vmf->pgoff == 0) |
039576c0 | 3507 | page = virt_to_page(vcpu->run); |
09566765 | 3508 | #ifdef CONFIG_X86 |
e4a533a4 | 3509 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 3510 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 3511 | #endif |
4b4357e0 | 3512 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3513 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
3514 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 3515 | #endif |
fb04a1ed PX |
3516 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
3517 | page = kvm_dirty_ring_get_page( | |
3518 | &vcpu->dirty_ring, | |
3519 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 3520 | else |
5b1c1493 | 3521 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 3522 | get_page(page); |
e4a533a4 | 3523 | vmf->page = page; |
3524 | return 0; | |
9a2bb7f4 AK |
3525 | } |
3526 | ||
f0f37e2f | 3527 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 3528 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
3529 | }; |
3530 | ||
3531 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
3532 | { | |
fb04a1ed | 3533 | struct kvm_vcpu *vcpu = file->private_data; |
11476d27 | 3534 | unsigned long pages = vma_pages(vma); |
fb04a1ed PX |
3535 | |
3536 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
3537 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
3538 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
3539 | return -EINVAL; | |
3540 | ||
9a2bb7f4 AK |
3541 | vma->vm_ops = &kvm_vcpu_vm_ops; |
3542 | return 0; | |
3543 | } | |
3544 | ||
bccf2150 AK |
3545 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
3546 | { | |
3547 | struct kvm_vcpu *vcpu = filp->private_data; | |
3548 | ||
66c0b394 | 3549 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
3550 | return 0; |
3551 | } | |
3552 | ||
3d3aab1b | 3553 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
3554 | .release = kvm_vcpu_release, |
3555 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 3556 | .mmap = kvm_vcpu_mmap, |
6038f373 | 3557 | .llseek = noop_llseek, |
7ddfd3e0 | 3558 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
3559 | }; |
3560 | ||
3561 | /* | |
3562 | * Allocates an inode for the vcpu. | |
3563 | */ | |
3564 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
3565 | { | |
e46b4692 MY |
3566 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
3567 | ||
3568 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
3569 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
3570 | } |
3571 | ||
3e7093d0 | 3572 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 3573 | { |
741cbbae | 3574 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
d56f5136 | 3575 | struct dentry *debugfs_dentry; |
45b5939e | 3576 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 3577 | |
45b5939e | 3578 | if (!debugfs_initialized()) |
3e7093d0 | 3579 | return; |
45b5939e LC |
3580 | |
3581 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
3582 | debugfs_dentry = debugfs_create_dir(dir_name, |
3583 | vcpu->kvm->debugfs_dentry); | |
45b5939e | 3584 | |
d56f5136 | 3585 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
741cbbae | 3586 | #endif |
45b5939e LC |
3587 | } |
3588 | ||
c5ea7660 AK |
3589 | /* |
3590 | * Creates some virtual cpus. Good luck creating more than one. | |
3591 | */ | |
73880c80 | 3592 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
3593 | { |
3594 | int r; | |
e09fefde | 3595 | struct kvm_vcpu *vcpu; |
8bd826d6 | 3596 | struct page *page; |
c5ea7660 | 3597 | |
a1c42dde | 3598 | if (id >= KVM_MAX_VCPU_IDS) |
338c7dba AH |
3599 | return -EINVAL; |
3600 | ||
6c7caebc PB |
3601 | mutex_lock(&kvm->lock); |
3602 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
3603 | mutex_unlock(&kvm->lock); | |
3604 | return -EINVAL; | |
3605 | } | |
3606 | ||
3607 | kvm->created_vcpus++; | |
3608 | mutex_unlock(&kvm->lock); | |
3609 | ||
897cc38e SC |
3610 | r = kvm_arch_vcpu_precreate(kvm, id); |
3611 | if (r) | |
3612 | goto vcpu_decrement; | |
3613 | ||
85f47930 | 3614 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
e529ef66 SC |
3615 | if (!vcpu) { |
3616 | r = -ENOMEM; | |
6c7caebc PB |
3617 | goto vcpu_decrement; |
3618 | } | |
c5ea7660 | 3619 | |
fcd97ad5 | 3620 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 3621 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
3622 | if (!page) { |
3623 | r = -ENOMEM; | |
e529ef66 | 3624 | goto vcpu_free; |
8bd826d6 SC |
3625 | } |
3626 | vcpu->run = page_address(page); | |
3627 | ||
3628 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
3629 | |
3630 | r = kvm_arch_vcpu_create(vcpu); | |
3631 | if (r) | |
8bd826d6 | 3632 | goto vcpu_free_run_page; |
e529ef66 | 3633 | |
fb04a1ed PX |
3634 | if (kvm->dirty_ring_size) { |
3635 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
3636 | id, kvm->dirty_ring_size); | |
3637 | if (r) | |
3638 | goto arch_vcpu_destroy; | |
3639 | } | |
3640 | ||
11ec2804 | 3641 | mutex_lock(&kvm->lock); |
e09fefde DH |
3642 | if (kvm_get_vcpu_by_id(kvm, id)) { |
3643 | r = -EEXIST; | |
3644 | goto unlock_vcpu_destroy; | |
3645 | } | |
73880c80 | 3646 | |
8750e72a RK |
3647 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
3648 | BUG_ON(kvm->vcpus[vcpu->vcpu_idx]); | |
c5ea7660 | 3649 | |
ce55c049 JZ |
3650 | /* Fill the stats id string for the vcpu */ |
3651 | snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d", | |
3652 | task_pid_nr(current), id); | |
3653 | ||
fb3f0f51 | 3654 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 3655 | kvm_get_kvm(kvm); |
bccf2150 | 3656 | r = create_vcpu_fd(vcpu); |
73880c80 | 3657 | if (r < 0) { |
149487bd | 3658 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 3659 | goto unlock_vcpu_destroy; |
73880c80 GN |
3660 | } |
3661 | ||
8750e72a | 3662 | kvm->vcpus[vcpu->vcpu_idx] = vcpu; |
dd489240 PB |
3663 | |
3664 | /* | |
3665 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
3666 | * before kvm->online_vcpu's incremented value. | |
3667 | */ | |
73880c80 GN |
3668 | smp_wmb(); |
3669 | atomic_inc(&kvm->online_vcpus); | |
3670 | ||
73880c80 | 3671 | mutex_unlock(&kvm->lock); |
42897d86 | 3672 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 3673 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 3674 | return r; |
39c3b86e | 3675 | |
d780592b | 3676 | unlock_vcpu_destroy: |
7d8fece6 | 3677 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
3678 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
3679 | arch_vcpu_destroy: | |
d40ccc62 | 3680 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
3681 | vcpu_free_run_page: |
3682 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
3683 | vcpu_free: |
3684 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
3685 | vcpu_decrement: |
3686 | mutex_lock(&kvm->lock); | |
3687 | kvm->created_vcpus--; | |
3688 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
3689 | return r; |
3690 | } | |
3691 | ||
1961d276 AK |
3692 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
3693 | { | |
3694 | if (sigset) { | |
3695 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
3696 | vcpu->sigset_active = 1; | |
3697 | vcpu->sigset = *sigset; | |
3698 | } else | |
3699 | vcpu->sigset_active = 0; | |
3700 | return 0; | |
3701 | } | |
3702 | ||
ce55c049 JZ |
3703 | static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer, |
3704 | size_t size, loff_t *offset) | |
3705 | { | |
3706 | struct kvm_vcpu *vcpu = file->private_data; | |
3707 | ||
3708 | return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header, | |
3709 | &kvm_vcpu_stats_desc[0], &vcpu->stat, | |
3710 | sizeof(vcpu->stat), user_buffer, size, offset); | |
3711 | } | |
3712 | ||
3713 | static const struct file_operations kvm_vcpu_stats_fops = { | |
3714 | .read = kvm_vcpu_stats_read, | |
3715 | .llseek = noop_llseek, | |
3716 | }; | |
3717 | ||
3718 | static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu) | |
3719 | { | |
3720 | int fd; | |
3721 | struct file *file; | |
3722 | char name[15 + ITOA_MAX_LEN + 1]; | |
3723 | ||
3724 | snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id); | |
3725 | ||
3726 | fd = get_unused_fd_flags(O_CLOEXEC); | |
3727 | if (fd < 0) | |
3728 | return fd; | |
3729 | ||
3730 | file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY); | |
3731 | if (IS_ERR(file)) { | |
3732 | put_unused_fd(fd); | |
3733 | return PTR_ERR(file); | |
3734 | } | |
3735 | file->f_mode |= FMODE_PREAD; | |
3736 | fd_install(fd, file); | |
3737 | ||
3738 | return fd; | |
3739 | } | |
3740 | ||
bccf2150 AK |
3741 | static long kvm_vcpu_ioctl(struct file *filp, |
3742 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 3743 | { |
bccf2150 | 3744 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 3745 | void __user *argp = (void __user *)arg; |
313a3dc7 | 3746 | int r; |
fa3795a7 DH |
3747 | struct kvm_fpu *fpu = NULL; |
3748 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 3749 | |
0b8f1173 | 3750 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged) |
6d4e4c4f | 3751 | return -EIO; |
2122ff5e | 3752 | |
2ea75be3 DM |
3753 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
3754 | return -EINVAL; | |
3755 | ||
2122ff5e | 3756 | /* |
5cb0944c PB |
3757 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
3758 | * execution; mutex_lock() would break them. | |
2122ff5e | 3759 | */ |
5cb0944c PB |
3760 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
3761 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 3762 | return r; |
2122ff5e | 3763 | |
ec7660cc CD |
3764 | if (mutex_lock_killable(&vcpu->mutex)) |
3765 | return -EINTR; | |
6aa8b732 | 3766 | switch (ioctl) { |
0e4524a5 CB |
3767 | case KVM_RUN: { |
3768 | struct pid *oldpid; | |
f0fe5108 AK |
3769 | r = -EINVAL; |
3770 | if (arg) | |
3771 | goto out; | |
0e4524a5 | 3772 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 3773 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 3774 | /* The thread running this VCPU changed. */ |
bd2a6394 | 3775 | struct pid *newpid; |
f95ef0cd | 3776 | |
bd2a6394 CD |
3777 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
3778 | if (r) | |
3779 | break; | |
3780 | ||
3781 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
3782 | rcu_assign_pointer(vcpu->pid, newpid); |
3783 | if (oldpid) | |
3784 | synchronize_rcu(); | |
3785 | put_pid(oldpid); | |
3786 | } | |
1b94f6f8 | 3787 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 3788 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 3789 | break; |
0e4524a5 | 3790 | } |
6aa8b732 | 3791 | case KVM_GET_REGS: { |
3e4bb3ac | 3792 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3793 | |
3e4bb3ac | 3794 | r = -ENOMEM; |
b12ce36a | 3795 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 3796 | if (!kvm_regs) |
6aa8b732 | 3797 | goto out; |
3e4bb3ac XZ |
3798 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
3799 | if (r) | |
3800 | goto out_free1; | |
6aa8b732 | 3801 | r = -EFAULT; |
3e4bb3ac XZ |
3802 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
3803 | goto out_free1; | |
6aa8b732 | 3804 | r = 0; |
3e4bb3ac XZ |
3805 | out_free1: |
3806 | kfree(kvm_regs); | |
6aa8b732 AK |
3807 | break; |
3808 | } | |
3809 | case KVM_SET_REGS: { | |
3e4bb3ac | 3810 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3811 | |
ff5c2c03 SL |
3812 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
3813 | if (IS_ERR(kvm_regs)) { | |
3814 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 3815 | goto out; |
ff5c2c03 | 3816 | } |
3e4bb3ac | 3817 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 3818 | kfree(kvm_regs); |
6aa8b732 AK |
3819 | break; |
3820 | } | |
3821 | case KVM_GET_SREGS: { | |
b12ce36a BG |
3822 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
3823 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
3824 | r = -ENOMEM; |
3825 | if (!kvm_sregs) | |
3826 | goto out; | |
3827 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
3828 | if (r) |
3829 | goto out; | |
3830 | r = -EFAULT; | |
fa3795a7 | 3831 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
3832 | goto out; |
3833 | r = 0; | |
3834 | break; | |
3835 | } | |
3836 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
3837 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
3838 | if (IS_ERR(kvm_sregs)) { | |
3839 | r = PTR_ERR(kvm_sregs); | |
18595411 | 3840 | kvm_sregs = NULL; |
6aa8b732 | 3841 | goto out; |
ff5c2c03 | 3842 | } |
fa3795a7 | 3843 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
3844 | break; |
3845 | } | |
62d9f0db MT |
3846 | case KVM_GET_MP_STATE: { |
3847 | struct kvm_mp_state mp_state; | |
3848 | ||
3849 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
3850 | if (r) | |
3851 | goto out; | |
3852 | r = -EFAULT; | |
893bdbf1 | 3853 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
3854 | goto out; |
3855 | r = 0; | |
3856 | break; | |
3857 | } | |
3858 | case KVM_SET_MP_STATE: { | |
3859 | struct kvm_mp_state mp_state; | |
3860 | ||
3861 | r = -EFAULT; | |
893bdbf1 | 3862 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
3863 | goto out; |
3864 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
3865 | break; |
3866 | } | |
6aa8b732 AK |
3867 | case KVM_TRANSLATE: { |
3868 | struct kvm_translation tr; | |
3869 | ||
3870 | r = -EFAULT; | |
893bdbf1 | 3871 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 3872 | goto out; |
8b006791 | 3873 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
3874 | if (r) |
3875 | goto out; | |
3876 | r = -EFAULT; | |
893bdbf1 | 3877 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
3878 | goto out; |
3879 | r = 0; | |
3880 | break; | |
3881 | } | |
d0bfb940 JK |
3882 | case KVM_SET_GUEST_DEBUG: { |
3883 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
3884 | |
3885 | r = -EFAULT; | |
893bdbf1 | 3886 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 3887 | goto out; |
d0bfb940 | 3888 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
3889 | break; |
3890 | } | |
1961d276 AK |
3891 | case KVM_SET_SIGNAL_MASK: { |
3892 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3893 | struct kvm_signal_mask kvm_sigmask; | |
3894 | sigset_t sigset, *p; | |
3895 | ||
3896 | p = NULL; | |
3897 | if (argp) { | |
3898 | r = -EFAULT; | |
3899 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3900 | sizeof(kvm_sigmask))) |
1961d276 AK |
3901 | goto out; |
3902 | r = -EINVAL; | |
893bdbf1 | 3903 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
3904 | goto out; |
3905 | r = -EFAULT; | |
3906 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 3907 | sizeof(sigset))) |
1961d276 AK |
3908 | goto out; |
3909 | p = &sigset; | |
3910 | } | |
376d41ff | 3911 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
3912 | break; |
3913 | } | |
b8836737 | 3914 | case KVM_GET_FPU: { |
b12ce36a | 3915 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
3916 | r = -ENOMEM; |
3917 | if (!fpu) | |
3918 | goto out; | |
3919 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
3920 | if (r) |
3921 | goto out; | |
3922 | r = -EFAULT; | |
fa3795a7 | 3923 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
3924 | goto out; |
3925 | r = 0; | |
3926 | break; | |
3927 | } | |
3928 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
3929 | fpu = memdup_user(argp, sizeof(*fpu)); |
3930 | if (IS_ERR(fpu)) { | |
3931 | r = PTR_ERR(fpu); | |
18595411 | 3932 | fpu = NULL; |
b8836737 | 3933 | goto out; |
ff5c2c03 | 3934 | } |
fa3795a7 | 3935 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
3936 | break; |
3937 | } | |
ce55c049 JZ |
3938 | case KVM_GET_STATS_FD: { |
3939 | r = kvm_vcpu_ioctl_get_stats_fd(vcpu); | |
3940 | break; | |
3941 | } | |
bccf2150 | 3942 | default: |
313a3dc7 | 3943 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
3944 | } |
3945 | out: | |
ec7660cc | 3946 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
3947 | kfree(fpu); |
3948 | kfree(kvm_sregs); | |
bccf2150 AK |
3949 | return r; |
3950 | } | |
3951 | ||
de8e5d74 | 3952 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
3953 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
3954 | unsigned int ioctl, unsigned long arg) | |
3955 | { | |
3956 | struct kvm_vcpu *vcpu = filp->private_data; | |
3957 | void __user *argp = compat_ptr(arg); | |
3958 | int r; | |
3959 | ||
0b8f1173 | 3960 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged) |
1dda606c AG |
3961 | return -EIO; |
3962 | ||
3963 | switch (ioctl) { | |
3964 | case KVM_SET_SIGNAL_MASK: { | |
3965 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3966 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
3967 | sigset_t sigset; |
3968 | ||
3969 | if (argp) { | |
3970 | r = -EFAULT; | |
3971 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3972 | sizeof(kvm_sigmask))) |
1dda606c AG |
3973 | goto out; |
3974 | r = -EINVAL; | |
3968cf62 | 3975 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
3976 | goto out; |
3977 | r = -EFAULT; | |
1393b4aa PB |
3978 | if (get_compat_sigset(&sigset, |
3979 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 3980 | goto out; |
760a9a30 AC |
3981 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
3982 | } else | |
3983 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
3984 | break; |
3985 | } | |
3986 | default: | |
3987 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
3988 | } | |
3989 | ||
3990 | out: | |
3991 | return r; | |
3992 | } | |
3993 | #endif | |
3994 | ||
a1cd3f08 CLG |
3995 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
3996 | { | |
3997 | struct kvm_device *dev = filp->private_data; | |
3998 | ||
3999 | if (dev->ops->mmap) | |
4000 | return dev->ops->mmap(dev, vma); | |
4001 | ||
4002 | return -ENODEV; | |
4003 | } | |
4004 | ||
852b6d57 SW |
4005 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
4006 | int (*accessor)(struct kvm_device *dev, | |
4007 | struct kvm_device_attr *attr), | |
4008 | unsigned long arg) | |
4009 | { | |
4010 | struct kvm_device_attr attr; | |
4011 | ||
4012 | if (!accessor) | |
4013 | return -EPERM; | |
4014 | ||
4015 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
4016 | return -EFAULT; | |
4017 | ||
4018 | return accessor(dev, &attr); | |
4019 | } | |
4020 | ||
4021 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
4022 | unsigned long arg) | |
4023 | { | |
4024 | struct kvm_device *dev = filp->private_data; | |
4025 | ||
0b8f1173 | 4026 | if (dev->kvm->mm != current->mm || dev->kvm->vm_bugged) |
ddba9180 SC |
4027 | return -EIO; |
4028 | ||
852b6d57 SW |
4029 | switch (ioctl) { |
4030 | case KVM_SET_DEVICE_ATTR: | |
4031 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
4032 | case KVM_GET_DEVICE_ATTR: | |
4033 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
4034 | case KVM_HAS_DEVICE_ATTR: | |
4035 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
4036 | default: | |
4037 | if (dev->ops->ioctl) | |
4038 | return dev->ops->ioctl(dev, ioctl, arg); | |
4039 | ||
4040 | return -ENOTTY; | |
4041 | } | |
4042 | } | |
4043 | ||
852b6d57 SW |
4044 | static int kvm_device_release(struct inode *inode, struct file *filp) |
4045 | { | |
4046 | struct kvm_device *dev = filp->private_data; | |
4047 | struct kvm *kvm = dev->kvm; | |
4048 | ||
2bde9b3e CLG |
4049 | if (dev->ops->release) { |
4050 | mutex_lock(&kvm->lock); | |
4051 | list_del(&dev->vm_node); | |
4052 | dev->ops->release(dev); | |
4053 | mutex_unlock(&kvm->lock); | |
4054 | } | |
4055 | ||
852b6d57 SW |
4056 | kvm_put_kvm(kvm); |
4057 | return 0; | |
4058 | } | |
4059 | ||
4060 | static const struct file_operations kvm_device_fops = { | |
4061 | .unlocked_ioctl = kvm_device_ioctl, | |
4062 | .release = kvm_device_release, | |
7ddfd3e0 | 4063 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 4064 | .mmap = kvm_device_mmap, |
852b6d57 SW |
4065 | }; |
4066 | ||
4067 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
4068 | { | |
4069 | if (filp->f_op != &kvm_device_fops) | |
4070 | return NULL; | |
4071 | ||
4072 | return filp->private_data; | |
4073 | } | |
4074 | ||
8538cb22 | 4075 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 4076 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
4077 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
4078 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 4079 | #endif |
d60eacb0 WD |
4080 | }; |
4081 | ||
8538cb22 | 4082 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
4083 | { |
4084 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
4085 | return -ENOSPC; | |
4086 | ||
4087 | if (kvm_device_ops_table[type] != NULL) | |
4088 | return -EEXIST; | |
4089 | ||
4090 | kvm_device_ops_table[type] = ops; | |
4091 | return 0; | |
4092 | } | |
4093 | ||
571ee1b6 WL |
4094 | void kvm_unregister_device_ops(u32 type) |
4095 | { | |
4096 | if (kvm_device_ops_table[type] != NULL) | |
4097 | kvm_device_ops_table[type] = NULL; | |
4098 | } | |
4099 | ||
852b6d57 SW |
4100 | static int kvm_ioctl_create_device(struct kvm *kvm, |
4101 | struct kvm_create_device *cd) | |
4102 | { | |
8538cb22 | 4103 | const struct kvm_device_ops *ops = NULL; |
852b6d57 SW |
4104 | struct kvm_device *dev; |
4105 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 4106 | int type; |
852b6d57 SW |
4107 | int ret; |
4108 | ||
d60eacb0 WD |
4109 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
4110 | return -ENODEV; | |
4111 | ||
1d487e9b PB |
4112 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
4113 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 4114 | if (ops == NULL) |
852b6d57 | 4115 | return -ENODEV; |
852b6d57 SW |
4116 | |
4117 | if (test) | |
4118 | return 0; | |
4119 | ||
b12ce36a | 4120 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
4121 | if (!dev) |
4122 | return -ENOMEM; | |
4123 | ||
4124 | dev->ops = ops; | |
4125 | dev->kvm = kvm; | |
852b6d57 | 4126 | |
a28ebea2 | 4127 | mutex_lock(&kvm->lock); |
1d487e9b | 4128 | ret = ops->create(dev, type); |
852b6d57 | 4129 | if (ret < 0) { |
a28ebea2 | 4130 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
4131 | kfree(dev); |
4132 | return ret; | |
4133 | } | |
a28ebea2 CD |
4134 | list_add(&dev->vm_node, &kvm->devices); |
4135 | mutex_unlock(&kvm->lock); | |
852b6d57 | 4136 | |
023e9fdd CD |
4137 | if (ops->init) |
4138 | ops->init(dev); | |
4139 | ||
cfa39381 | 4140 | kvm_get_kvm(kvm); |
24009b05 | 4141 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 4142 | if (ret < 0) { |
149487bd | 4143 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
4144 | mutex_lock(&kvm->lock); |
4145 | list_del(&dev->vm_node); | |
4146 | mutex_unlock(&kvm->lock); | |
a0f1d21c | 4147 | ops->destroy(dev); |
852b6d57 SW |
4148 | return ret; |
4149 | } | |
4150 | ||
852b6d57 SW |
4151 | cd->fd = ret; |
4152 | return 0; | |
4153 | } | |
4154 | ||
92b591a4 AG |
4155 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
4156 | { | |
4157 | switch (arg) { | |
4158 | case KVM_CAP_USER_MEMORY: | |
4159 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
4160 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
4161 | case KVM_CAP_INTERNAL_ERROR_DATA: |
4162 | #ifdef CONFIG_HAVE_KVM_MSI | |
4163 | case KVM_CAP_SIGNAL_MSI: | |
4164 | #endif | |
297e2105 | 4165 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 4166 | case KVM_CAP_IRQFD: |
92b591a4 AG |
4167 | case KVM_CAP_IRQFD_RESAMPLE: |
4168 | #endif | |
e9ea5069 | 4169 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 4170 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 4171 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 4172 | case KVM_CAP_HALT_POLL: |
92b591a4 | 4173 | return 1; |
4b4357e0 | 4174 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
4175 | case KVM_CAP_COALESCED_MMIO: |
4176 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
4177 | case KVM_CAP_COALESCED_PIO: |
4178 | return 1; | |
30422558 | 4179 | #endif |
3c9bd400 JZ |
4180 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4181 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
4182 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
4183 | #endif | |
92b591a4 AG |
4184 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4185 | case KVM_CAP_IRQ_ROUTING: | |
4186 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
4187 | #endif |
4188 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
4189 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
4190 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 4191 | #endif |
c110ae57 PB |
4192 | case KVM_CAP_NR_MEMSLOTS: |
4193 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed PX |
4194 | case KVM_CAP_DIRTY_LOG_RING: |
4195 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
4196 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); | |
4197 | #else | |
4198 | return 0; | |
4199 | #endif | |
ce55c049 JZ |
4200 | case KVM_CAP_BINARY_STATS_FD: |
4201 | return 1; | |
92b591a4 AG |
4202 | default: |
4203 | break; | |
4204 | } | |
4205 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
4206 | } | |
4207 | ||
fb04a1ed PX |
4208 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
4209 | { | |
4210 | int r; | |
4211 | ||
4212 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
4213 | return -EINVAL; | |
4214 | ||
4215 | /* the size should be power of 2 */ | |
4216 | if (!size || (size & (size - 1))) | |
4217 | return -EINVAL; | |
4218 | ||
4219 | /* Should be bigger to keep the reserved entries, or a page */ | |
4220 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
4221 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
4222 | return -EINVAL; | |
4223 | ||
4224 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
4225 | sizeof(struct kvm_dirty_gfn)) | |
4226 | return -E2BIG; | |
4227 | ||
4228 | /* We only allow it to set once */ | |
4229 | if (kvm->dirty_ring_size) | |
4230 | return -EINVAL; | |
4231 | ||
4232 | mutex_lock(&kvm->lock); | |
4233 | ||
4234 | if (kvm->created_vcpus) { | |
4235 | /* We don't allow to change this value after vcpu created */ | |
4236 | r = -EINVAL; | |
4237 | } else { | |
4238 | kvm->dirty_ring_size = size; | |
4239 | r = 0; | |
4240 | } | |
4241 | ||
4242 | mutex_unlock(&kvm->lock); | |
4243 | return r; | |
4244 | } | |
4245 | ||
4246 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
4247 | { | |
4248 | int i; | |
4249 | struct kvm_vcpu *vcpu; | |
4250 | int cleared = 0; | |
4251 | ||
4252 | if (!kvm->dirty_ring_size) | |
4253 | return -EINVAL; | |
4254 | ||
4255 | mutex_lock(&kvm->slots_lock); | |
4256 | ||
4257 | kvm_for_each_vcpu(i, vcpu, kvm) | |
4258 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
4259 | ||
4260 | mutex_unlock(&kvm->slots_lock); | |
4261 | ||
4262 | if (cleared) | |
4263 | kvm_flush_remote_tlbs(kvm); | |
4264 | ||
4265 | return cleared; | |
4266 | } | |
4267 | ||
e5d83c74 PB |
4268 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4269 | struct kvm_enable_cap *cap) | |
4270 | { | |
4271 | return -EINVAL; | |
4272 | } | |
4273 | ||
4274 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
4275 | struct kvm_enable_cap *cap) | |
4276 | { | |
4277 | switch (cap->cap) { | |
2a31b9db | 4278 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
4279 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
4280 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
4281 | ||
4282 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
4283 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
4284 | ||
4285 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
4286 | return -EINVAL; |
4287 | kvm->manual_dirty_log_protect = cap->args[0]; | |
4288 | return 0; | |
3c9bd400 | 4289 | } |
2a31b9db | 4290 | #endif |
acd05785 DM |
4291 | case KVM_CAP_HALT_POLL: { |
4292 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
4293 | return -EINVAL; | |
4294 | ||
4295 | kvm->max_halt_poll_ns = cap->args[0]; | |
4296 | return 0; | |
4297 | } | |
fb04a1ed PX |
4298 | case KVM_CAP_DIRTY_LOG_RING: |
4299 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); | |
e5d83c74 PB |
4300 | default: |
4301 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
4302 | } | |
4303 | } | |
4304 | ||
fcfe1bae JZ |
4305 | static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer, |
4306 | size_t size, loff_t *offset) | |
4307 | { | |
4308 | struct kvm *kvm = file->private_data; | |
4309 | ||
4310 | return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header, | |
4311 | &kvm_vm_stats_desc[0], &kvm->stat, | |
4312 | sizeof(kvm->stat), user_buffer, size, offset); | |
4313 | } | |
4314 | ||
4315 | static const struct file_operations kvm_vm_stats_fops = { | |
4316 | .read = kvm_vm_stats_read, | |
4317 | .llseek = noop_llseek, | |
4318 | }; | |
4319 | ||
4320 | static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm) | |
4321 | { | |
4322 | int fd; | |
4323 | struct file *file; | |
4324 | ||
4325 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4326 | if (fd < 0) | |
4327 | return fd; | |
4328 | ||
4329 | file = anon_inode_getfile("kvm-vm-stats", | |
4330 | &kvm_vm_stats_fops, kvm, O_RDONLY); | |
4331 | if (IS_ERR(file)) { | |
4332 | put_unused_fd(fd); | |
4333 | return PTR_ERR(file); | |
4334 | } | |
4335 | file->f_mode |= FMODE_PREAD; | |
4336 | fd_install(fd, file); | |
4337 | ||
4338 | return fd; | |
4339 | } | |
4340 | ||
bccf2150 AK |
4341 | static long kvm_vm_ioctl(struct file *filp, |
4342 | unsigned int ioctl, unsigned long arg) | |
4343 | { | |
4344 | struct kvm *kvm = filp->private_data; | |
4345 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 4346 | int r; |
bccf2150 | 4347 | |
0b8f1173 | 4348 | if (kvm->mm != current->mm || kvm->vm_bugged) |
6d4e4c4f | 4349 | return -EIO; |
bccf2150 AK |
4350 | switch (ioctl) { |
4351 | case KVM_CREATE_VCPU: | |
4352 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 4353 | break; |
e5d83c74 PB |
4354 | case KVM_ENABLE_CAP: { |
4355 | struct kvm_enable_cap cap; | |
4356 | ||
4357 | r = -EFAULT; | |
4358 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4359 | goto out; | |
4360 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
4361 | break; | |
4362 | } | |
6fc138d2 IE |
4363 | case KVM_SET_USER_MEMORY_REGION: { |
4364 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
4365 | ||
4366 | r = -EFAULT; | |
4367 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 4368 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
4369 | goto out; |
4370 | ||
47ae31e2 | 4371 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
4372 | break; |
4373 | } | |
4374 | case KVM_GET_DIRTY_LOG: { | |
4375 | struct kvm_dirty_log log; | |
4376 | ||
4377 | r = -EFAULT; | |
893bdbf1 | 4378 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 4379 | goto out; |
2c6f5df9 | 4380 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
4381 | break; |
4382 | } | |
2a31b9db PB |
4383 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4384 | case KVM_CLEAR_DIRTY_LOG: { | |
4385 | struct kvm_clear_dirty_log log; | |
4386 | ||
4387 | r = -EFAULT; | |
4388 | if (copy_from_user(&log, argp, sizeof(log))) | |
4389 | goto out; | |
4390 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4391 | break; | |
4392 | } | |
4393 | #endif | |
4b4357e0 | 4394 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
4395 | case KVM_REGISTER_COALESCED_MMIO: { |
4396 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4397 | |
5f94c174 | 4398 | r = -EFAULT; |
893bdbf1 | 4399 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4400 | goto out; |
5f94c174 | 4401 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4402 | break; |
4403 | } | |
4404 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
4405 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4406 | |
5f94c174 | 4407 | r = -EFAULT; |
893bdbf1 | 4408 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4409 | goto out; |
5f94c174 | 4410 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4411 | break; |
4412 | } | |
4413 | #endif | |
721eecbf GH |
4414 | case KVM_IRQFD: { |
4415 | struct kvm_irqfd data; | |
4416 | ||
4417 | r = -EFAULT; | |
893bdbf1 | 4418 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 4419 | goto out; |
d4db2935 | 4420 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
4421 | break; |
4422 | } | |
d34e6b17 GH |
4423 | case KVM_IOEVENTFD: { |
4424 | struct kvm_ioeventfd data; | |
4425 | ||
4426 | r = -EFAULT; | |
893bdbf1 | 4427 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
4428 | goto out; |
4429 | r = kvm_ioeventfd(kvm, &data); | |
4430 | break; | |
4431 | } | |
07975ad3 JK |
4432 | #ifdef CONFIG_HAVE_KVM_MSI |
4433 | case KVM_SIGNAL_MSI: { | |
4434 | struct kvm_msi msi; | |
4435 | ||
4436 | r = -EFAULT; | |
893bdbf1 | 4437 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
4438 | goto out; |
4439 | r = kvm_send_userspace_msi(kvm, &msi); | |
4440 | break; | |
4441 | } | |
23d43cf9 CD |
4442 | #endif |
4443 | #ifdef __KVM_HAVE_IRQ_LINE | |
4444 | case KVM_IRQ_LINE_STATUS: | |
4445 | case KVM_IRQ_LINE: { | |
4446 | struct kvm_irq_level irq_event; | |
4447 | ||
4448 | r = -EFAULT; | |
893bdbf1 | 4449 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
4450 | goto out; |
4451 | ||
aa2fbe6d YZ |
4452 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
4453 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
4454 | if (r) |
4455 | goto out; | |
4456 | ||
4457 | r = -EFAULT; | |
4458 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 4459 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
4460 | goto out; |
4461 | } | |
4462 | ||
4463 | r = 0; | |
4464 | break; | |
4465 | } | |
73880c80 | 4466 | #endif |
aa8d5944 AG |
4467 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4468 | case KVM_SET_GSI_ROUTING: { | |
4469 | struct kvm_irq_routing routing; | |
4470 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 4471 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
4472 | |
4473 | r = -EFAULT; | |
4474 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
4475 | goto out; | |
4476 | r = -EINVAL; | |
5c0aea0e DH |
4477 | if (!kvm_arch_can_set_irq_routing(kvm)) |
4478 | goto out; | |
caf1ff26 | 4479 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
4480 | goto out; |
4481 | if (routing.flags) | |
4482 | goto out; | |
f8c1b85b | 4483 | if (routing.nr) { |
f8c1b85b | 4484 | urouting = argp; |
7ec28e26 DE |
4485 | entries = vmemdup_user(urouting->entries, |
4486 | array_size(sizeof(*entries), | |
4487 | routing.nr)); | |
4488 | if (IS_ERR(entries)) { | |
4489 | r = PTR_ERR(entries); | |
4490 | goto out; | |
4491 | } | |
f8c1b85b | 4492 | } |
aa8d5944 AG |
4493 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
4494 | routing.flags); | |
7ec28e26 | 4495 | kvfree(entries); |
aa8d5944 AG |
4496 | break; |
4497 | } | |
4498 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
4499 | case KVM_CREATE_DEVICE: { |
4500 | struct kvm_create_device cd; | |
4501 | ||
4502 | r = -EFAULT; | |
4503 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
4504 | goto out; | |
4505 | ||
4506 | r = kvm_ioctl_create_device(kvm, &cd); | |
4507 | if (r) | |
4508 | goto out; | |
4509 | ||
4510 | r = -EFAULT; | |
4511 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
4512 | goto out; | |
4513 | ||
4514 | r = 0; | |
4515 | break; | |
4516 | } | |
92b591a4 AG |
4517 | case KVM_CHECK_EXTENSION: |
4518 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
4519 | break; | |
fb04a1ed PX |
4520 | case KVM_RESET_DIRTY_RINGS: |
4521 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
4522 | break; | |
fcfe1bae JZ |
4523 | case KVM_GET_STATS_FD: |
4524 | r = kvm_vm_ioctl_get_stats_fd(kvm); | |
4525 | break; | |
f17abe9a | 4526 | default: |
1fe779f8 | 4527 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
4528 | } |
4529 | out: | |
4530 | return r; | |
4531 | } | |
4532 | ||
de8e5d74 | 4533 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
4534 | struct compat_kvm_dirty_log { |
4535 | __u32 slot; | |
4536 | __u32 padding1; | |
4537 | union { | |
4538 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4539 | __u64 padding2; | |
4540 | }; | |
4541 | }; | |
4542 | ||
8750f9bb PB |
4543 | struct compat_kvm_clear_dirty_log { |
4544 | __u32 slot; | |
4545 | __u32 num_pages; | |
4546 | __u64 first_page; | |
4547 | union { | |
4548 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4549 | __u64 padding2; | |
4550 | }; | |
4551 | }; | |
4552 | ||
6ff5894c AB |
4553 | static long kvm_vm_compat_ioctl(struct file *filp, |
4554 | unsigned int ioctl, unsigned long arg) | |
4555 | { | |
4556 | struct kvm *kvm = filp->private_data; | |
4557 | int r; | |
4558 | ||
0b8f1173 | 4559 | if (kvm->mm != current->mm || kvm->vm_bugged) |
6ff5894c AB |
4560 | return -EIO; |
4561 | switch (ioctl) { | |
8750f9bb PB |
4562 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4563 | case KVM_CLEAR_DIRTY_LOG: { | |
4564 | struct compat_kvm_clear_dirty_log compat_log; | |
4565 | struct kvm_clear_dirty_log log; | |
4566 | ||
4567 | if (copy_from_user(&compat_log, (void __user *)arg, | |
4568 | sizeof(compat_log))) | |
4569 | return -EFAULT; | |
4570 | log.slot = compat_log.slot; | |
4571 | log.num_pages = compat_log.num_pages; | |
4572 | log.first_page = compat_log.first_page; | |
4573 | log.padding2 = compat_log.padding2; | |
4574 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4575 | ||
4576 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4577 | break; | |
4578 | } | |
4579 | #endif | |
6ff5894c AB |
4580 | case KVM_GET_DIRTY_LOG: { |
4581 | struct compat_kvm_dirty_log compat_log; | |
4582 | struct kvm_dirty_log log; | |
4583 | ||
6ff5894c AB |
4584 | if (copy_from_user(&compat_log, (void __user *)arg, |
4585 | sizeof(compat_log))) | |
f6a3b168 | 4586 | return -EFAULT; |
6ff5894c AB |
4587 | log.slot = compat_log.slot; |
4588 | log.padding1 = compat_log.padding1; | |
4589 | log.padding2 = compat_log.padding2; | |
4590 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4591 | ||
4592 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
4593 | break; |
4594 | } | |
4595 | default: | |
4596 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
4597 | } | |
6ff5894c AB |
4598 | return r; |
4599 | } | |
4600 | #endif | |
4601 | ||
3d3aab1b | 4602 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
4603 | .release = kvm_vm_release, |
4604 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 4605 | .llseek = noop_llseek, |
7ddfd3e0 | 4606 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
4607 | }; |
4608 | ||
54526d1f NT |
4609 | bool file_is_kvm(struct file *file) |
4610 | { | |
4611 | return file && file->f_op == &kvm_vm_fops; | |
4612 | } | |
4613 | EXPORT_SYMBOL_GPL(file_is_kvm); | |
4614 | ||
e08b9637 | 4615 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 4616 | { |
aac87636 | 4617 | int r; |
f17abe9a | 4618 | struct kvm *kvm; |
506cfba9 | 4619 | struct file *file; |
f17abe9a | 4620 | |
e08b9637 | 4621 | kvm = kvm_create_vm(type); |
d6d28168 AK |
4622 | if (IS_ERR(kvm)) |
4623 | return PTR_ERR(kvm); | |
4b4357e0 | 4624 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 4625 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
4626 | if (r < 0) |
4627 | goto put_kvm; | |
6ce5a090 | 4628 | #endif |
506cfba9 | 4629 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
4630 | if (r < 0) |
4631 | goto put_kvm; | |
4632 | ||
fcfe1bae JZ |
4633 | snprintf(kvm->stats_id, sizeof(kvm->stats_id), |
4634 | "kvm-%d", task_pid_nr(current)); | |
4635 | ||
506cfba9 AV |
4636 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
4637 | if (IS_ERR(file)) { | |
4638 | put_unused_fd(r); | |
78588335 ME |
4639 | r = PTR_ERR(file); |
4640 | goto put_kvm; | |
506cfba9 | 4641 | } |
536a6f88 | 4642 | |
525df861 PB |
4643 | /* |
4644 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
4645 | * already set, with ->release() being kvm_vm_release(). In error | |
4646 | * cases it will be called by the final fput(file) and will take | |
4647 | * care of doing kvm_put_kvm(kvm). | |
4648 | */ | |
536a6f88 | 4649 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
4650 | put_unused_fd(r); |
4651 | fput(file); | |
536a6f88 JF |
4652 | return -ENOMEM; |
4653 | } | |
286de8f6 | 4654 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 4655 | |
506cfba9 | 4656 | fd_install(r, file); |
aac87636 | 4657 | return r; |
78588335 ME |
4658 | |
4659 | put_kvm: | |
4660 | kvm_put_kvm(kvm); | |
4661 | return r; | |
f17abe9a AK |
4662 | } |
4663 | ||
4664 | static long kvm_dev_ioctl(struct file *filp, | |
4665 | unsigned int ioctl, unsigned long arg) | |
4666 | { | |
07c45a36 | 4667 | long r = -EINVAL; |
f17abe9a AK |
4668 | |
4669 | switch (ioctl) { | |
4670 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
4671 | if (arg) |
4672 | goto out; | |
f17abe9a AK |
4673 | r = KVM_API_VERSION; |
4674 | break; | |
4675 | case KVM_CREATE_VM: | |
e08b9637 | 4676 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 4677 | break; |
018d00d2 | 4678 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 4679 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 4680 | break; |
07c45a36 | 4681 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
4682 | if (arg) |
4683 | goto out; | |
adb1ff46 AK |
4684 | r = PAGE_SIZE; /* struct kvm_run */ |
4685 | #ifdef CONFIG_X86 | |
4686 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 4687 | #endif |
4b4357e0 | 4688 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 4689 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 4690 | #endif |
07c45a36 | 4691 | break; |
d4c9ff2d FEL |
4692 | case KVM_TRACE_ENABLE: |
4693 | case KVM_TRACE_PAUSE: | |
4694 | case KVM_TRACE_DISABLE: | |
2023a29c | 4695 | r = -EOPNOTSUPP; |
d4c9ff2d | 4696 | break; |
6aa8b732 | 4697 | default: |
043405e1 | 4698 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
4699 | } |
4700 | out: | |
4701 | return r; | |
4702 | } | |
4703 | ||
6aa8b732 | 4704 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 4705 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 4706 | .llseek = noop_llseek, |
7ddfd3e0 | 4707 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
4708 | }; |
4709 | ||
4710 | static struct miscdevice kvm_dev = { | |
bbe4432e | 4711 | KVM_MINOR, |
6aa8b732 AK |
4712 | "kvm", |
4713 | &kvm_chardev_ops, | |
4714 | }; | |
4715 | ||
75b7127c | 4716 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
4717 | { |
4718 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 4719 | int r; |
1b6c0168 | 4720 | |
7f59f492 | 4721 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4722 | return; |
10474ae8 | 4723 | |
7f59f492 | 4724 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 4725 | |
13a34e06 | 4726 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
4727 | |
4728 | if (r) { | |
4729 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
4730 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 4731 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 4732 | } |
1b6c0168 AK |
4733 | } |
4734 | ||
8c18b2d2 | 4735 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 4736 | { |
4a937f96 | 4737 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4738 | if (kvm_usage_count) |
4739 | hardware_enable_nolock(NULL); | |
4a937f96 | 4740 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4741 | return 0; |
75b7127c TY |
4742 | } |
4743 | ||
4744 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
4745 | { |
4746 | int cpu = raw_smp_processor_id(); | |
4747 | ||
7f59f492 | 4748 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4749 | return; |
7f59f492 | 4750 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 4751 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
4752 | } |
4753 | ||
8c18b2d2 | 4754 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 4755 | { |
4a937f96 | 4756 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4757 | if (kvm_usage_count) |
4758 | hardware_disable_nolock(NULL); | |
4a937f96 | 4759 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4760 | return 0; |
75b7127c TY |
4761 | } |
4762 | ||
10474ae8 AG |
4763 | static void hardware_disable_all_nolock(void) |
4764 | { | |
4765 | BUG_ON(!kvm_usage_count); | |
4766 | ||
4767 | kvm_usage_count--; | |
4768 | if (!kvm_usage_count) | |
75b7127c | 4769 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
4770 | } |
4771 | ||
4772 | static void hardware_disable_all(void) | |
4773 | { | |
4a937f96 | 4774 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 4775 | hardware_disable_all_nolock(); |
4a937f96 | 4776 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4777 | } |
4778 | ||
4779 | static int hardware_enable_all(void) | |
4780 | { | |
4781 | int r = 0; | |
4782 | ||
4a937f96 | 4783 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
4784 | |
4785 | kvm_usage_count++; | |
4786 | if (kvm_usage_count == 1) { | |
4787 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 4788 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
4789 | |
4790 | if (atomic_read(&hardware_enable_failed)) { | |
4791 | hardware_disable_all_nolock(); | |
4792 | r = -EBUSY; | |
4793 | } | |
4794 | } | |
4795 | ||
4a937f96 | 4796 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4797 | |
4798 | return r; | |
4799 | } | |
4800 | ||
9a2b85c6 | 4801 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 4802 | void *v) |
9a2b85c6 | 4803 | { |
8e1c1815 SY |
4804 | /* |
4805 | * Some (well, at least mine) BIOSes hang on reboot if | |
4806 | * in vmx root mode. | |
4807 | * | |
4808 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
4809 | */ | |
1170adc6 | 4810 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 4811 | kvm_rebooting = true; |
75b7127c | 4812 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
4813 | return NOTIFY_OK; |
4814 | } | |
4815 | ||
4816 | static struct notifier_block kvm_reboot_notifier = { | |
4817 | .notifier_call = kvm_reboot, | |
4818 | .priority = 0, | |
4819 | }; | |
4820 | ||
e93f8a0f | 4821 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
4822 | { |
4823 | int i; | |
4824 | ||
4825 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 4826 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
4827 | |
4828 | kvm_iodevice_destructor(pos); | |
4829 | } | |
e93f8a0f | 4830 | kfree(bus); |
2eeb2e94 GH |
4831 | } |
4832 | ||
c21fbff1 | 4833 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 4834 | const struct kvm_io_range *r2) |
743eeb0b | 4835 | { |
8f4216c7 JW |
4836 | gpa_t addr1 = r1->addr; |
4837 | gpa_t addr2 = r2->addr; | |
4838 | ||
4839 | if (addr1 < addr2) | |
743eeb0b | 4840 | return -1; |
8f4216c7 JW |
4841 | |
4842 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
4843 | * accept any overlapping write. Any order is acceptable for | |
4844 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
4845 | * we process all of them. | |
4846 | */ | |
4847 | if (r2->len) { | |
4848 | addr1 += r1->len; | |
4849 | addr2 += r2->len; | |
4850 | } | |
4851 | ||
4852 | if (addr1 > addr2) | |
743eeb0b | 4853 | return 1; |
8f4216c7 | 4854 | |
743eeb0b SL |
4855 | return 0; |
4856 | } | |
4857 | ||
a343c9b7 PB |
4858 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
4859 | { | |
c21fbff1 | 4860 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
4861 | } |
4862 | ||
39369f7a | 4863 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
4864 | gpa_t addr, int len) |
4865 | { | |
4866 | struct kvm_io_range *range, key; | |
4867 | int off; | |
4868 | ||
4869 | key = (struct kvm_io_range) { | |
4870 | .addr = addr, | |
4871 | .len = len, | |
4872 | }; | |
4873 | ||
4874 | range = bsearch(&key, bus->range, bus->dev_count, | |
4875 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
4876 | if (range == NULL) | |
4877 | return -ENOENT; | |
4878 | ||
4879 | off = range - bus->range; | |
4880 | ||
c21fbff1 | 4881 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
4882 | off--; |
4883 | ||
4884 | return off; | |
4885 | } | |
4886 | ||
e32edf4f | 4887 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
4888 | struct kvm_io_range *range, const void *val) |
4889 | { | |
4890 | int idx; | |
4891 | ||
4892 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
4893 | if (idx < 0) | |
4894 | return -EOPNOTSUPP; | |
4895 | ||
4896 | while (idx < bus->dev_count && | |
c21fbff1 | 4897 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4898 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4899 | range->len, val)) |
4900 | return idx; | |
4901 | idx++; | |
4902 | } | |
4903 | ||
4904 | return -EOPNOTSUPP; | |
4905 | } | |
4906 | ||
bda9020e | 4907 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 4908 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 4909 | int len, const void *val) |
2eeb2e94 | 4910 | { |
90d83dc3 | 4911 | struct kvm_io_bus *bus; |
743eeb0b | 4912 | struct kvm_io_range range; |
126a5af5 | 4913 | int r; |
743eeb0b SL |
4914 | |
4915 | range = (struct kvm_io_range) { | |
4916 | .addr = addr, | |
4917 | .len = len, | |
4918 | }; | |
90d83dc3 | 4919 | |
e32edf4f | 4920 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4921 | if (!bus) |
4922 | return -ENOMEM; | |
e32edf4f | 4923 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4924 | return r < 0 ? r : 0; |
4925 | } | |
a2420107 | 4926 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
4927 | |
4928 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
4929 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
4930 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
4931 | { |
4932 | struct kvm_io_bus *bus; | |
4933 | struct kvm_io_range range; | |
4934 | ||
4935 | range = (struct kvm_io_range) { | |
4936 | .addr = addr, | |
4937 | .len = len, | |
4938 | }; | |
4939 | ||
e32edf4f | 4940 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4941 | if (!bus) |
4942 | return -ENOMEM; | |
126a5af5 CH |
4943 | |
4944 | /* First try the device referenced by cookie. */ | |
4945 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 4946 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 4947 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
4948 | val)) |
4949 | return cookie; | |
4950 | ||
4951 | /* | |
4952 | * cookie contained garbage; fall back to search and return the | |
4953 | * correct cookie value. | |
4954 | */ | |
e32edf4f | 4955 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4956 | } |
4957 | ||
e32edf4f NN |
4958 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
4959 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
4960 | { |
4961 | int idx; | |
4962 | ||
4963 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
4964 | if (idx < 0) |
4965 | return -EOPNOTSUPP; | |
4966 | ||
4967 | while (idx < bus->dev_count && | |
c21fbff1 | 4968 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4969 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4970 | range->len, val)) |
4971 | return idx; | |
743eeb0b SL |
4972 | idx++; |
4973 | } | |
4974 | ||
bda9020e MT |
4975 | return -EOPNOTSUPP; |
4976 | } | |
2eeb2e94 | 4977 | |
bda9020e | 4978 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 4979 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 4980 | int len, void *val) |
bda9020e | 4981 | { |
90d83dc3 | 4982 | struct kvm_io_bus *bus; |
743eeb0b | 4983 | struct kvm_io_range range; |
126a5af5 | 4984 | int r; |
743eeb0b SL |
4985 | |
4986 | range = (struct kvm_io_range) { | |
4987 | .addr = addr, | |
4988 | .len = len, | |
4989 | }; | |
e93f8a0f | 4990 | |
e32edf4f | 4991 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4992 | if (!bus) |
4993 | return -ENOMEM; | |
e32edf4f | 4994 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
4995 | return r < 0 ? r : 0; |
4996 | } | |
743eeb0b | 4997 | |
79fac95e | 4998 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
4999 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
5000 | int len, struct kvm_io_device *dev) | |
6c474694 | 5001 | { |
d4c67a7a | 5002 | int i; |
e93f8a0f | 5003 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 5004 | struct kvm_io_range range; |
090b7aff | 5005 | |
4a12f951 | 5006 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
5007 | if (!bus) |
5008 | return -ENOMEM; | |
5009 | ||
6ea34c9b AK |
5010 | /* exclude ioeventfd which is limited by maximum fd */ |
5011 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 5012 | return -ENOSPC; |
2eeb2e94 | 5013 | |
90952cd3 | 5014 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 5015 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
5016 | if (!new_bus) |
5017 | return -ENOMEM; | |
d4c67a7a GH |
5018 | |
5019 | range = (struct kvm_io_range) { | |
5020 | .addr = addr, | |
5021 | .len = len, | |
5022 | .dev = dev, | |
5023 | }; | |
5024 | ||
5025 | for (i = 0; i < bus->dev_count; i++) | |
5026 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
5027 | break; | |
5028 | ||
5029 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
5030 | new_bus->dev_count++; | |
5031 | new_bus->range[i] = range; | |
5032 | memcpy(new_bus->range + i + 1, bus->range + i, | |
5033 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
5034 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
5035 | synchronize_srcu_expedited(&kvm->srcu); | |
5036 | kfree(bus); | |
090b7aff GH |
5037 | |
5038 | return 0; | |
5039 | } | |
5040 | ||
5d3c4c79 SC |
5041 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5042 | struct kvm_io_device *dev) | |
090b7aff | 5043 | { |
f6588660 | 5044 | int i, j; |
e93f8a0f | 5045 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 5046 | |
7c896d37 SC |
5047 | lockdep_assert_held(&kvm->slots_lock); |
5048 | ||
4a12f951 | 5049 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 5050 | if (!bus) |
5d3c4c79 | 5051 | return 0; |
df630b8c | 5052 | |
7c896d37 | 5053 | for (i = 0; i < bus->dev_count; i++) { |
a1300716 | 5054 | if (bus->range[i].dev == dev) { |
090b7aff GH |
5055 | break; |
5056 | } | |
7c896d37 | 5057 | } |
e93f8a0f | 5058 | |
90db1043 | 5059 | if (i == bus->dev_count) |
5d3c4c79 | 5060 | return 0; |
a1300716 | 5061 | |
90952cd3 | 5062 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 5063 | GFP_KERNEL_ACCOUNT); |
f6588660 | 5064 | if (new_bus) { |
871c433b | 5065 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
5066 | new_bus->dev_count--; |
5067 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 5068 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
2ee37574 SC |
5069 | } |
5070 | ||
5071 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
5072 | synchronize_srcu_expedited(&kvm->srcu); | |
5073 | ||
5074 | /* Destroy the old bus _after_ installing the (null) bus. */ | |
5075 | if (!new_bus) { | |
90db1043 | 5076 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
f6588660 RK |
5077 | for (j = 0; j < bus->dev_count; j++) { |
5078 | if (j == i) | |
5079 | continue; | |
5080 | kvm_iodevice_destructor(bus->range[j].dev); | |
5081 | } | |
90db1043 | 5082 | } |
a1300716 | 5083 | |
e93f8a0f | 5084 | kfree(bus); |
5d3c4c79 | 5085 | return new_bus ? 0 : -ENOMEM; |
2eeb2e94 GH |
5086 | } |
5087 | ||
8a39d006 AP |
5088 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5089 | gpa_t addr) | |
5090 | { | |
5091 | struct kvm_io_bus *bus; | |
5092 | int dev_idx, srcu_idx; | |
5093 | struct kvm_io_device *iodev = NULL; | |
5094 | ||
5095 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
5096 | ||
5097 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
5098 | if (!bus) |
5099 | goto out_unlock; | |
8a39d006 AP |
5100 | |
5101 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
5102 | if (dev_idx < 0) | |
5103 | goto out_unlock; | |
5104 | ||
5105 | iodev = bus->range[dev_idx].dev; | |
5106 | ||
5107 | out_unlock: | |
5108 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
5109 | ||
5110 | return iodev; | |
5111 | } | |
5112 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
5113 | ||
536a6f88 JF |
5114 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
5115 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
5116 | const char *fmt) | |
5117 | { | |
5118 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5119 | inode->i_private; | |
5120 | ||
605c7130 PX |
5121 | /* |
5122 | * The debugfs files are a reference to the kvm struct which | |
5123 | * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe | |
5124 | * avoids the race between open and the removal of the debugfs directory. | |
536a6f88 | 5125 | */ |
605c7130 | 5126 | if (!kvm_get_kvm_safe(stat_data->kvm)) |
536a6f88 JF |
5127 | return -ENOENT; |
5128 | ||
833b45de | 5129 | if (simple_attr_open(inode, file, get, |
bc9e9e67 | 5130 | kvm_stats_debugfs_mode(stat_data->desc) & 0222 |
09cbcef6 MP |
5131 | ? set : NULL, |
5132 | fmt)) { | |
536a6f88 JF |
5133 | kvm_put_kvm(stat_data->kvm); |
5134 | return -ENOMEM; | |
5135 | } | |
5136 | ||
5137 | return 0; | |
5138 | } | |
5139 | ||
5140 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
5141 | { | |
5142 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5143 | inode->i_private; | |
5144 | ||
5145 | simple_attr_release(inode, file); | |
5146 | kvm_put_kvm(stat_data->kvm); | |
5147 | ||
5148 | return 0; | |
5149 | } | |
5150 | ||
09cbcef6 | 5151 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 5152 | { |
bc9e9e67 | 5153 | *val = *(u64 *)((void *)(&kvm->stat) + offset); |
536a6f88 | 5154 | |
09cbcef6 MP |
5155 | return 0; |
5156 | } | |
5157 | ||
5158 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
5159 | { | |
bc9e9e67 | 5160 | *(u64 *)((void *)(&kvm->stat) + offset) = 0; |
536a6f88 JF |
5161 | |
5162 | return 0; | |
5163 | } | |
5164 | ||
09cbcef6 | 5165 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 5166 | { |
09cbcef6 MP |
5167 | int i; |
5168 | struct kvm_vcpu *vcpu; | |
ce35ef27 | 5169 | |
09cbcef6 | 5170 | *val = 0; |
ce35ef27 | 5171 | |
09cbcef6 | 5172 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5173 | *val += *(u64 *)((void *)(&vcpu->stat) + offset); |
ce35ef27 SJS |
5174 | |
5175 | return 0; | |
5176 | } | |
5177 | ||
09cbcef6 | 5178 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 5179 | { |
09cbcef6 MP |
5180 | int i; |
5181 | struct kvm_vcpu *vcpu; | |
536a6f88 | 5182 | |
09cbcef6 | 5183 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5184 | *(u64 *)((void *)(&vcpu->stat) + offset) = 0; |
09cbcef6 MP |
5185 | |
5186 | return 0; | |
5187 | } | |
536a6f88 | 5188 | |
09cbcef6 | 5189 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 5190 | { |
09cbcef6 | 5191 | int r = -EFAULT; |
536a6f88 | 5192 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
536a6f88 | 5193 | |
bc9e9e67 | 5194 | switch (stat_data->kind) { |
09cbcef6 MP |
5195 | case KVM_STAT_VM: |
5196 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5197 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5198 | break; |
5199 | case KVM_STAT_VCPU: | |
5200 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5201 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5202 | break; |
5203 | } | |
536a6f88 | 5204 | |
09cbcef6 | 5205 | return r; |
536a6f88 JF |
5206 | } |
5207 | ||
09cbcef6 | 5208 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 5209 | { |
09cbcef6 | 5210 | int r = -EFAULT; |
ce35ef27 | 5211 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
ce35ef27 SJS |
5212 | |
5213 | if (val) | |
5214 | return -EINVAL; | |
5215 | ||
bc9e9e67 | 5216 | switch (stat_data->kind) { |
09cbcef6 MP |
5217 | case KVM_STAT_VM: |
5218 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5219 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5220 | break; |
5221 | case KVM_STAT_VCPU: | |
5222 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5223 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5224 | break; |
5225 | } | |
ce35ef27 | 5226 | |
09cbcef6 | 5227 | return r; |
ce35ef27 SJS |
5228 | } |
5229 | ||
09cbcef6 | 5230 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
5231 | { |
5232 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
5233 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
5234 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
5235 | } |
5236 | ||
09cbcef6 MP |
5237 | static const struct file_operations stat_fops_per_vm = { |
5238 | .owner = THIS_MODULE, | |
5239 | .open = kvm_stat_data_open, | |
536a6f88 | 5240 | .release = kvm_debugfs_release, |
09cbcef6 MP |
5241 | .read = simple_attr_read, |
5242 | .write = simple_attr_write, | |
5243 | .llseek = no_llseek, | |
536a6f88 JF |
5244 | }; |
5245 | ||
8b88b099 | 5246 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
5247 | { |
5248 | unsigned offset = (long)_offset; | |
ba1389b7 | 5249 | struct kvm *kvm; |
536a6f88 | 5250 | u64 tmp_val; |
ba1389b7 | 5251 | |
8b88b099 | 5252 | *val = 0; |
0d9ce162 | 5253 | mutex_lock(&kvm_lock); |
536a6f88 | 5254 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5255 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
5256 | *val += tmp_val; |
5257 | } | |
0d9ce162 | 5258 | mutex_unlock(&kvm_lock); |
8b88b099 | 5259 | return 0; |
ba1389b7 AK |
5260 | } |
5261 | ||
ce35ef27 SJS |
5262 | static int vm_stat_clear(void *_offset, u64 val) |
5263 | { | |
5264 | unsigned offset = (long)_offset; | |
5265 | struct kvm *kvm; | |
ce35ef27 SJS |
5266 | |
5267 | if (val) | |
5268 | return -EINVAL; | |
5269 | ||
0d9ce162 | 5270 | mutex_lock(&kvm_lock); |
ce35ef27 | 5271 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5272 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 5273 | } |
0d9ce162 | 5274 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5275 | |
5276 | return 0; | |
5277 | } | |
5278 | ||
5279 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
bc9e9e67 | 5280 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n"); |
ba1389b7 | 5281 | |
8b88b099 | 5282 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
5283 | { |
5284 | unsigned offset = (long)_offset; | |
1165f5fe | 5285 | struct kvm *kvm; |
536a6f88 | 5286 | u64 tmp_val; |
1165f5fe | 5287 | |
8b88b099 | 5288 | *val = 0; |
0d9ce162 | 5289 | mutex_lock(&kvm_lock); |
536a6f88 | 5290 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5291 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
5292 | *val += tmp_val; |
5293 | } | |
0d9ce162 | 5294 | mutex_unlock(&kvm_lock); |
8b88b099 | 5295 | return 0; |
1165f5fe AK |
5296 | } |
5297 | ||
ce35ef27 SJS |
5298 | static int vcpu_stat_clear(void *_offset, u64 val) |
5299 | { | |
5300 | unsigned offset = (long)_offset; | |
5301 | struct kvm *kvm; | |
ce35ef27 SJS |
5302 | |
5303 | if (val) | |
5304 | return -EINVAL; | |
5305 | ||
0d9ce162 | 5306 | mutex_lock(&kvm_lock); |
ce35ef27 | 5307 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5308 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 5309 | } |
0d9ce162 | 5310 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5311 | |
5312 | return 0; | |
5313 | } | |
5314 | ||
5315 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
5316 | "%llu\n"); | |
bc9e9e67 | 5317 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n"); |
1165f5fe | 5318 | |
286de8f6 CI |
5319 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
5320 | { | |
5321 | struct kobj_uevent_env *env; | |
286de8f6 CI |
5322 | unsigned long long created, active; |
5323 | ||
5324 | if (!kvm_dev.this_device || !kvm) | |
5325 | return; | |
5326 | ||
0d9ce162 | 5327 | mutex_lock(&kvm_lock); |
286de8f6 CI |
5328 | if (type == KVM_EVENT_CREATE_VM) { |
5329 | kvm_createvm_count++; | |
5330 | kvm_active_vms++; | |
5331 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
5332 | kvm_active_vms--; | |
5333 | } | |
5334 | created = kvm_createvm_count; | |
5335 | active = kvm_active_vms; | |
0d9ce162 | 5336 | mutex_unlock(&kvm_lock); |
286de8f6 | 5337 | |
b12ce36a | 5338 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
5339 | if (!env) |
5340 | return; | |
5341 | ||
5342 | add_uevent_var(env, "CREATED=%llu", created); | |
5343 | add_uevent_var(env, "COUNT=%llu", active); | |
5344 | ||
fdeaf7e3 | 5345 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 5346 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
5347 | kvm->userspace_pid = task_pid_nr(current); |
5348 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 5349 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
5350 | } |
5351 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 5352 | |
85cd39af | 5353 | if (kvm->debugfs_dentry) { |
b12ce36a | 5354 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
5355 | |
5356 | if (p) { | |
5357 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
5358 | if (!IS_ERR(tmp)) | |
5359 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
5360 | kfree(p); | |
286de8f6 CI |
5361 | } |
5362 | } | |
5363 | /* no need for checks, since we are adding at most only 5 keys */ | |
5364 | env->envp[env->envp_idx++] = NULL; | |
5365 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
5366 | kfree(env); | |
286de8f6 CI |
5367 | } |
5368 | ||
929f45e3 | 5369 | static void kvm_init_debug(void) |
6aa8b732 | 5370 | { |
bc9e9e67 JZ |
5371 | const struct file_operations *fops; |
5372 | const struct _kvm_stats_desc *pdesc; | |
5373 | int i; | |
6aa8b732 | 5374 | |
76f7c879 | 5375 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 5376 | |
bc9e9e67 JZ |
5377 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
5378 | pdesc = &kvm_vm_stats_desc[i]; | |
5379 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5380 | fops = &vm_stat_fops; | |
5381 | else | |
5382 | fops = &vm_stat_readonly_fops; | |
5383 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5384 | kvm_debugfs_dir, | |
5385 | (void *)(long)pdesc->desc.offset, fops); | |
5386 | } | |
5387 | ||
5388 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
5389 | pdesc = &kvm_vcpu_stats_desc[i]; | |
5390 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5391 | fops = &vcpu_stat_fops; | |
5392 | else | |
5393 | fops = &vcpu_stat_readonly_fops; | |
5394 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5395 | kvm_debugfs_dir, | |
5396 | (void *)(long)pdesc->desc.offset, fops); | |
4f69b680 | 5397 | } |
6aa8b732 AK |
5398 | } |
5399 | ||
fb3600cc | 5400 | static int kvm_suspend(void) |
59ae6c6b | 5401 | { |
10474ae8 | 5402 | if (kvm_usage_count) |
75b7127c | 5403 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
5404 | return 0; |
5405 | } | |
5406 | ||
fb3600cc | 5407 | static void kvm_resume(void) |
59ae6c6b | 5408 | { |
ca84d1a2 | 5409 | if (kvm_usage_count) { |
2eb06c30 WL |
5410 | #ifdef CONFIG_LOCKDEP |
5411 | WARN_ON(lockdep_is_held(&kvm_count_lock)); | |
5412 | #endif | |
75b7127c | 5413 | hardware_enable_nolock(NULL); |
ca84d1a2 | 5414 | } |
59ae6c6b AK |
5415 | } |
5416 | ||
fb3600cc | 5417 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
5418 | .suspend = kvm_suspend, |
5419 | .resume = kvm_resume, | |
5420 | }; | |
5421 | ||
15ad7146 AK |
5422 | static inline |
5423 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
5424 | { | |
5425 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
5426 | } | |
5427 | ||
5428 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
5429 | { | |
5430 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 5431 | |
046ddeed | 5432 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 5433 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 5434 | |
7495e22b | 5435 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 5436 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 5437 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
5438 | } |
5439 | ||
5440 | static void kvm_sched_out(struct preempt_notifier *pn, | |
5441 | struct task_struct *next) | |
5442 | { | |
5443 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
5444 | ||
3ba9f93b | 5445 | if (current->on_rq) { |
046ddeed | 5446 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
5447 | WRITE_ONCE(vcpu->ready, true); |
5448 | } | |
e9b11c17 | 5449 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
5450 | __this_cpu_write(kvm_running_vcpu, NULL); |
5451 | } | |
5452 | ||
5453 | /** | |
5454 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
5455 | * |
5456 | * We can disable preemption locally around accessing the per-CPU variable, | |
5457 | * and use the resolved vcpu pointer after enabling preemption again, | |
5458 | * because even if the current thread is migrated to another CPU, reading | |
5459 | * the per-CPU value later will give us the same value as we update the | |
5460 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
5461 | */ |
5462 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
5463 | { | |
1f03b2bc MZ |
5464 | struct kvm_vcpu *vcpu; |
5465 | ||
5466 | preempt_disable(); | |
5467 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
5468 | preempt_enable(); | |
5469 | ||
5470 | return vcpu; | |
7495e22b | 5471 | } |
379a3c8e | 5472 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
5473 | |
5474 | /** | |
5475 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
5476 | */ | |
5477 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
5478 | { | |
5479 | return &kvm_running_vcpu; | |
15ad7146 AK |
5480 | } |
5481 | ||
b9904085 SC |
5482 | struct kvm_cpu_compat_check { |
5483 | void *opaque; | |
5484 | int *ret; | |
5485 | }; | |
5486 | ||
5487 | static void check_processor_compat(void *data) | |
f257d6dc | 5488 | { |
b9904085 SC |
5489 | struct kvm_cpu_compat_check *c = data; |
5490 | ||
5491 | *c->ret = kvm_arch_check_processor_compat(c->opaque); | |
f257d6dc SC |
5492 | } |
5493 | ||
0ee75bea | 5494 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 5495 | struct module *module) |
6aa8b732 | 5496 | { |
b9904085 | 5497 | struct kvm_cpu_compat_check c; |
6aa8b732 | 5498 | int r; |
002c7f7c | 5499 | int cpu; |
6aa8b732 | 5500 | |
f8c16bba ZX |
5501 | r = kvm_arch_init(opaque); |
5502 | if (r) | |
d2308784 | 5503 | goto out_fail; |
cb498ea2 | 5504 | |
7dac16c3 AH |
5505 | /* |
5506 | * kvm_arch_init makes sure there's at most one caller | |
5507 | * for architectures that support multiple implementations, | |
5508 | * like intel and amd on x86. | |
36343f6e PB |
5509 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
5510 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 5511 | */ |
36343f6e PB |
5512 | r = kvm_irqfd_init(); |
5513 | if (r) | |
5514 | goto out_irqfd; | |
7dac16c3 | 5515 | |
8437a617 | 5516 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
5517 | r = -ENOMEM; |
5518 | goto out_free_0; | |
5519 | } | |
5520 | ||
b9904085 | 5521 | r = kvm_arch_hardware_setup(opaque); |
6aa8b732 | 5522 | if (r < 0) |
faf0be22 | 5523 | goto out_free_1; |
6aa8b732 | 5524 | |
b9904085 SC |
5525 | c.ret = &r; |
5526 | c.opaque = opaque; | |
002c7f7c | 5527 | for_each_online_cpu(cpu) { |
b9904085 | 5528 | smp_call_function_single(cpu, check_processor_compat, &c, 1); |
002c7f7c | 5529 | if (r < 0) |
faf0be22 | 5530 | goto out_free_2; |
002c7f7c YS |
5531 | } |
5532 | ||
73c1b41e | 5533 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 5534 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 5535 | if (r) |
d2308784 | 5536 | goto out_free_2; |
6aa8b732 AK |
5537 | register_reboot_notifier(&kvm_reboot_notifier); |
5538 | ||
c16f862d | 5539 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
5540 | if (!vcpu_align) |
5541 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
5542 | kvm_vcpu_cache = |
5543 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
5544 | SLAB_ACCOUNT, | |
5545 | offsetof(struct kvm_vcpu, arch), | |
ce55c049 JZ |
5546 | offsetofend(struct kvm_vcpu, stats_id) |
5547 | - offsetof(struct kvm_vcpu, arch), | |
46515736 | 5548 | NULL); |
c16f862d RR |
5549 | if (!kvm_vcpu_cache) { |
5550 | r = -ENOMEM; | |
fb3600cc | 5551 | goto out_free_3; |
c16f862d RR |
5552 | } |
5553 | ||
baff59cc VK |
5554 | for_each_possible_cpu(cpu) { |
5555 | if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu), | |
5556 | GFP_KERNEL, cpu_to_node(cpu))) { | |
5557 | r = -ENOMEM; | |
5558 | goto out_free_4; | |
5559 | } | |
5560 | } | |
5561 | ||
af585b92 GN |
5562 | r = kvm_async_pf_init(); |
5563 | if (r) | |
baff59cc | 5564 | goto out_free_5; |
af585b92 | 5565 | |
6aa8b732 | 5566 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
5567 | kvm_vm_fops.owner = module; |
5568 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
5569 | |
5570 | r = misc_register(&kvm_dev); | |
5571 | if (r) { | |
1170adc6 | 5572 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 5573 | goto out_unreg; |
6aa8b732 AK |
5574 | } |
5575 | ||
fb3600cc RW |
5576 | register_syscore_ops(&kvm_syscore_ops); |
5577 | ||
15ad7146 AK |
5578 | kvm_preempt_ops.sched_in = kvm_sched_in; |
5579 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
5580 | ||
929f45e3 | 5581 | kvm_init_debug(); |
0ea4ed8e | 5582 | |
3c3c29fd PB |
5583 | r = kvm_vfio_ops_init(); |
5584 | WARN_ON(r); | |
5585 | ||
c7addb90 | 5586 | return 0; |
6aa8b732 | 5587 | |
af585b92 GN |
5588 | out_unreg: |
5589 | kvm_async_pf_deinit(); | |
baff59cc VK |
5590 | out_free_5: |
5591 | for_each_possible_cpu(cpu) | |
5592 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
5593 | out_free_4: | |
c16f862d | 5594 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 5595 | out_free_3: |
6aa8b732 | 5596 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5597 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 5598 | out_free_2: |
e9b11c17 | 5599 | kvm_arch_hardware_unsetup(); |
faf0be22 | 5600 | out_free_1: |
7f59f492 | 5601 | free_cpumask_var(cpus_hardware_enabled); |
d2308784 | 5602 | out_free_0: |
a0f155e9 | 5603 | kvm_irqfd_exit(); |
36343f6e | 5604 | out_irqfd: |
7dac16c3 AH |
5605 | kvm_arch_exit(); |
5606 | out_fail: | |
6aa8b732 AK |
5607 | return r; |
5608 | } | |
cb498ea2 | 5609 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 5610 | |
cb498ea2 | 5611 | void kvm_exit(void) |
6aa8b732 | 5612 | { |
baff59cc VK |
5613 | int cpu; |
5614 | ||
4bd33b56 | 5615 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 5616 | misc_deregister(&kvm_dev); |
baff59cc VK |
5617 | for_each_possible_cpu(cpu) |
5618 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
c16f862d | 5619 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 5620 | kvm_async_pf_deinit(); |
fb3600cc | 5621 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 5622 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5623 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 5624 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 5625 | kvm_arch_hardware_unsetup(); |
f8c16bba | 5626 | kvm_arch_exit(); |
a0f155e9 | 5627 | kvm_irqfd_exit(); |
7f59f492 | 5628 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 5629 | kvm_vfio_ops_exit(); |
6aa8b732 | 5630 | } |
cb498ea2 | 5631 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
5632 | |
5633 | struct kvm_vm_worker_thread_context { | |
5634 | struct kvm *kvm; | |
5635 | struct task_struct *parent; | |
5636 | struct completion init_done; | |
5637 | kvm_vm_thread_fn_t thread_fn; | |
5638 | uintptr_t data; | |
5639 | int err; | |
5640 | }; | |
5641 | ||
5642 | static int kvm_vm_worker_thread(void *context) | |
5643 | { | |
5644 | /* | |
5645 | * The init_context is allocated on the stack of the parent thread, so | |
5646 | * we have to locally copy anything that is needed beyond initialization | |
5647 | */ | |
5648 | struct kvm_vm_worker_thread_context *init_context = context; | |
5649 | struct kvm *kvm = init_context->kvm; | |
5650 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
5651 | uintptr_t data = init_context->data; | |
5652 | int err; | |
5653 | ||
5654 | err = kthread_park(current); | |
5655 | /* kthread_park(current) is never supposed to return an error */ | |
5656 | WARN_ON(err != 0); | |
5657 | if (err) | |
5658 | goto init_complete; | |
5659 | ||
5660 | err = cgroup_attach_task_all(init_context->parent, current); | |
5661 | if (err) { | |
5662 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
5663 | __func__, err); | |
5664 | goto init_complete; | |
5665 | } | |
5666 | ||
5667 | set_user_nice(current, task_nice(init_context->parent)); | |
5668 | ||
5669 | init_complete: | |
5670 | init_context->err = err; | |
5671 | complete(&init_context->init_done); | |
5672 | init_context = NULL; | |
5673 | ||
5674 | if (err) | |
5675 | return err; | |
5676 | ||
5677 | /* Wait to be woken up by the spawner before proceeding. */ | |
5678 | kthread_parkme(); | |
5679 | ||
5680 | if (!kthread_should_stop()) | |
5681 | err = thread_fn(kvm, data); | |
5682 | ||
5683 | return err; | |
5684 | } | |
5685 | ||
5686 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
5687 | uintptr_t data, const char *name, | |
5688 | struct task_struct **thread_ptr) | |
5689 | { | |
5690 | struct kvm_vm_worker_thread_context init_context = {}; | |
5691 | struct task_struct *thread; | |
5692 | ||
5693 | *thread_ptr = NULL; | |
5694 | init_context.kvm = kvm; | |
5695 | init_context.parent = current; | |
5696 | init_context.thread_fn = thread_fn; | |
5697 | init_context.data = data; | |
5698 | init_completion(&init_context.init_done); | |
5699 | ||
5700 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
5701 | "%s-%d", name, task_pid_nr(current)); | |
5702 | if (IS_ERR(thread)) | |
5703 | return PTR_ERR(thread); | |
5704 | ||
5705 | /* kthread_run is never supposed to return NULL */ | |
5706 | WARN_ON(thread == NULL); | |
5707 | ||
5708 | wait_for_completion(&init_context.init_done); | |
5709 | ||
5710 | if (!init_context.err) | |
5711 | *thread_ptr = thread; | |
5712 | ||
5713 | return init_context.err; | |
5714 | } |