Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
6aa8b732 AK |
2 | /* |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * This module enables machines with Intel VT-x extensions to run virtual | |
6 | * machines without emulation or binary translation. | |
7 | * | |
8 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 9 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
10 | * |
11 | * Authors: | |
12 | * Avi Kivity <avi@qumranet.com> | |
13 | * Yaniv Kamay <yaniv@qumranet.com> | |
6aa8b732 AK |
14 | */ |
15 | ||
af669ac6 | 16 | #include <kvm/iodev.h> |
6aa8b732 | 17 | |
edf88417 | 18 | #include <linux/kvm_host.h> |
6aa8b732 AK |
19 | #include <linux/kvm.h> |
20 | #include <linux/module.h> | |
21 | #include <linux/errno.h> | |
6aa8b732 | 22 | #include <linux/percpu.h> |
6aa8b732 AK |
23 | #include <linux/mm.h> |
24 | #include <linux/miscdevice.h> | |
25 | #include <linux/vmalloc.h> | |
6aa8b732 | 26 | #include <linux/reboot.h> |
6aa8b732 AK |
27 | #include <linux/debugfs.h> |
28 | #include <linux/highmem.h> | |
29 | #include <linux/file.h> | |
fb3600cc | 30 | #include <linux/syscore_ops.h> |
774c47f1 | 31 | #include <linux/cpu.h> |
174cd4b1 | 32 | #include <linux/sched/signal.h> |
6e84f315 | 33 | #include <linux/sched/mm.h> |
03441a34 | 34 | #include <linux/sched/stat.h> |
d9e368d6 AK |
35 | #include <linux/cpumask.h> |
36 | #include <linux/smp.h> | |
d6d28168 | 37 | #include <linux/anon_inodes.h> |
04d2cc77 | 38 | #include <linux/profile.h> |
7aa81cc0 | 39 | #include <linux/kvm_para.h> |
6fc138d2 | 40 | #include <linux/pagemap.h> |
8d4e1288 | 41 | #include <linux/mman.h> |
35149e21 | 42 | #include <linux/swap.h> |
e56d532f | 43 | #include <linux/bitops.h> |
547de29e | 44 | #include <linux/spinlock.h> |
6ff5894c | 45 | #include <linux/compat.h> |
bc6678a3 | 46 | #include <linux/srcu.h> |
8f0b1ab6 | 47 | #include <linux/hugetlb.h> |
5a0e3ad6 | 48 | #include <linux/slab.h> |
743eeb0b SL |
49 | #include <linux/sort.h> |
50 | #include <linux/bsearch.h> | |
c011d23b | 51 | #include <linux/io.h> |
2eb06c30 | 52 | #include <linux/lockdep.h> |
c57c8046 | 53 | #include <linux/kthread.h> |
2fdef3a2 | 54 | #include <linux/suspend.h> |
6aa8b732 | 55 | |
e495606d | 56 | #include <asm/processor.h> |
2ea75be3 | 57 | #include <asm/ioctl.h> |
7c0f6ba6 | 58 | #include <linux/uaccess.h> |
6aa8b732 | 59 | |
5f94c174 | 60 | #include "coalesced_mmio.h" |
af585b92 | 61 | #include "async_pf.h" |
982ed0de | 62 | #include "kvm_mm.h" |
3c3c29fd | 63 | #include "vfio.h" |
5f94c174 | 64 | |
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 | |
5f6de5cb DM |
120 | static struct file_operations kvm_chardev_ops; |
121 | ||
bccf2150 AK |
122 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
123 | unsigned long arg); | |
de8e5d74 | 124 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
125 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
126 | unsigned long arg); | |
7ddfd3e0 MZ |
127 | #define KVM_COMPAT(c) .compat_ioctl = (c) |
128 | #else | |
9cb09e7c MZ |
129 | /* |
130 | * For architectures that don't implement a compat infrastructure, | |
131 | * adopt a double line of defense: | |
132 | * - Prevent a compat task from opening /dev/kvm | |
133 | * - If the open has been done by a 64bit task, and the KVM fd | |
134 | * passed to a compat task, let the ioctls fail. | |
135 | */ | |
7ddfd3e0 MZ |
136 | static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl, |
137 | unsigned long arg) { return -EINVAL; } | |
b9876e6d MZ |
138 | |
139 | static int kvm_no_compat_open(struct inode *inode, struct file *file) | |
140 | { | |
141 | return is_compat_task() ? -ENODEV : 0; | |
142 | } | |
143 | #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \ | |
144 | .open = kvm_no_compat_open | |
1dda606c | 145 | #endif |
10474ae8 AG |
146 | static int hardware_enable_all(void); |
147 | static void hardware_disable_all(void); | |
bccf2150 | 148 | |
e93f8a0f | 149 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 150 | |
52480137 | 151 | __visible bool kvm_rebooting; |
b7c4145b | 152 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 153 | |
286de8f6 CI |
154 | #define KVM_EVENT_CREATE_VM 0 |
155 | #define KVM_EVENT_DESTROY_VM 1 | |
156 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); | |
157 | static unsigned long long kvm_createvm_count; | |
158 | static unsigned long long kvm_active_vms; | |
159 | ||
baff59cc VK |
160 | static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask); |
161 | ||
e649b3f0 ET |
162 | __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
163 | unsigned long start, unsigned long end) | |
b1394e74 RK |
164 | { |
165 | } | |
166 | ||
683412cc MZ |
167 | __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm) |
168 | { | |
169 | } | |
170 | ||
a78986aa SC |
171 | bool kvm_is_zone_device_pfn(kvm_pfn_t pfn) |
172 | { | |
173 | /* | |
174 | * The metadata used by is_zone_device_page() to determine whether or | |
175 | * not a page is ZONE_DEVICE is guaranteed to be valid if and only if | |
176 | * the device has been pinned, e.g. by get_user_pages(). WARN if the | |
177 | * page_count() is zero to help detect bad usage of this helper. | |
178 | */ | |
179 | if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn)))) | |
180 | return false; | |
181 | ||
182 | return is_zone_device_page(pfn_to_page(pfn)); | |
183 | } | |
184 | ||
ba049e93 | 185 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
cbff90a7 | 186 | { |
a78986aa SC |
187 | /* |
188 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting | |
189 | * perspective they are "normal" pages, albeit with slightly different | |
190 | * usage rules. | |
191 | */ | |
11feeb49 | 192 | if (pfn_valid(pfn)) |
a78986aa | 193 | return PageReserved(pfn_to_page(pfn)) && |
7df003c8 | 194 | !is_zero_pfn(pfn) && |
a78986aa | 195 | !kvm_is_zone_device_pfn(pfn); |
cbff90a7 BAY |
196 | |
197 | return true; | |
198 | } | |
199 | ||
bccf2150 AK |
200 | /* |
201 | * Switches to specified vcpu, until a matching vcpu_put() | |
202 | */ | |
ec7660cc | 203 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 204 | { |
ec7660cc | 205 | int cpu = get_cpu(); |
7495e22b PB |
206 | |
207 | __this_cpu_write(kvm_running_vcpu, vcpu); | |
15ad7146 | 208 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 209 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 210 | put_cpu(); |
6aa8b732 | 211 | } |
2f1fe811 | 212 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 213 | |
313a3dc7 | 214 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 215 | { |
15ad7146 | 216 | preempt_disable(); |
313a3dc7 | 217 | kvm_arch_vcpu_put(vcpu); |
15ad7146 | 218 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
7495e22b | 219 | __this_cpu_write(kvm_running_vcpu, NULL); |
15ad7146 | 220 | preempt_enable(); |
6aa8b732 | 221 | } |
2f1fe811 | 222 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 223 | |
7a97cec2 PB |
224 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
225 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
226 | { | |
227 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
228 | ||
229 | /* | |
230 | * We need to wait for the VCPU to reenable interrupts and get out of | |
231 | * READING_SHADOW_PAGE_TABLES mode. | |
232 | */ | |
233 | if (req & KVM_REQUEST_WAIT) | |
234 | return mode != OUTSIDE_GUEST_MODE; | |
235 | ||
236 | /* | |
237 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
238 | */ | |
239 | return mode == IN_GUEST_MODE; | |
240 | } | |
241 | ||
f24b44e4 | 242 | static void ack_kick(void *_completed) |
d9e368d6 | 243 | { |
d9e368d6 AK |
244 | } |
245 | ||
620b2438 | 246 | static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait) |
b49defe8 | 247 | { |
b49defe8 PB |
248 | if (cpumask_empty(cpus)) |
249 | return false; | |
250 | ||
f24b44e4 | 251 | smp_call_function_many(cpus, ack_kick, NULL, wait); |
b49defe8 PB |
252 | return true; |
253 | } | |
254 | ||
b56bd8e0 JL |
255 | static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req, |
256 | struct cpumask *tmp, int current_cpu) | |
ae0946cd VK |
257 | { |
258 | int cpu; | |
259 | ||
df06dae3 SC |
260 | if (likely(!(req & KVM_REQUEST_NO_ACTION))) |
261 | __kvm_make_request(req, vcpu); | |
ae0946cd VK |
262 | |
263 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) | |
264 | return; | |
265 | ||
ae0946cd VK |
266 | /* |
267 | * Note, the vCPU could get migrated to a different pCPU at any point | |
268 | * after kvm_request_needs_ipi(), which could result in sending an IPI | |
269 | * to the previous pCPU. But, that's OK because the purpose of the IPI | |
270 | * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is | |
271 | * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES | |
272 | * after this point is also OK, as the requirement is only that KVM wait | |
273 | * for vCPUs that were reading SPTEs _before_ any changes were | |
274 | * finalized. See kvm_vcpu_kick() for more details on handling requests. | |
275 | */ | |
276 | if (kvm_request_needs_ipi(vcpu, req)) { | |
277 | cpu = READ_ONCE(vcpu->cpu); | |
278 | if (cpu != -1 && cpu != current_cpu) | |
279 | __cpumask_set_cpu(cpu, tmp); | |
280 | } | |
281 | } | |
282 | ||
7053df4e | 283 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
620b2438 | 284 | unsigned long *vcpu_bitmap) |
d9e368d6 | 285 | { |
d9e368d6 | 286 | struct kvm_vcpu *vcpu; |
620b2438 | 287 | struct cpumask *cpus; |
ae0946cd | 288 | int i, me; |
7053df4e | 289 | bool called; |
6ef7a1bc | 290 | |
3cba4130 | 291 | me = get_cpu(); |
7053df4e | 292 | |
620b2438 VK |
293 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
294 | cpumask_clear(cpus); | |
295 | ||
ae0946cd VK |
296 | for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) { |
297 | vcpu = kvm_get_vcpu(kvm, i); | |
381cecc5 | 298 | if (!vcpu) |
7053df4e | 299 | continue; |
b56bd8e0 | 300 | kvm_make_vcpu_request(vcpu, req, cpus, me); |
49846896 | 301 | } |
7053df4e | 302 | |
620b2438 | 303 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); |
3cba4130 | 304 | put_cpu(); |
7053df4e VK |
305 | |
306 | return called; | |
307 | } | |
308 | ||
54163a34 SS |
309 | bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req, |
310 | struct kvm_vcpu *except) | |
7053df4e | 311 | { |
ae0946cd | 312 | struct kvm_vcpu *vcpu; |
baff59cc | 313 | struct cpumask *cpus; |
46808a4c | 314 | unsigned long i; |
7053df4e | 315 | bool called; |
46808a4c | 316 | int me; |
7053df4e | 317 | |
ae0946cd VK |
318 | me = get_cpu(); |
319 | ||
baff59cc VK |
320 | cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask); |
321 | cpumask_clear(cpus); | |
322 | ||
ae0946cd VK |
323 | kvm_for_each_vcpu(i, vcpu, kvm) { |
324 | if (vcpu == except) | |
325 | continue; | |
b56bd8e0 | 326 | kvm_make_vcpu_request(vcpu, req, cpus, me); |
ae0946cd VK |
327 | } |
328 | ||
329 | called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT)); | |
330 | put_cpu(); | |
7053df4e | 331 | |
49846896 | 332 | return called; |
d9e368d6 AK |
333 | } |
334 | ||
54163a34 SS |
335 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
336 | { | |
337 | return kvm_make_all_cpus_request_except(kvm, req, NULL); | |
338 | } | |
a2486020 | 339 | EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request); |
54163a34 | 340 | |
a6d51016 | 341 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 342 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 343 | { |
3cc4e148 | 344 | ++kvm->stat.generic.remote_tlb_flush_requests; |
6bc6db00 | 345 | |
4ae3cb3a LT |
346 | /* |
347 | * We want to publish modifications to the page tables before reading | |
348 | * mode. Pairs with a memory barrier in arch-specific code. | |
349 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
350 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
351 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
352 | * | |
353 | * There is already an smp_mb__after_atomic() before | |
354 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
355 | * barrier here. | |
356 | */ | |
b08660e5 TL |
357 | if (!kvm_arch_flush_remote_tlb(kvm) |
358 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
0193cc90 | 359 | ++kvm->stat.generic.remote_tlb_flush; |
2e53d63a | 360 | } |
2ba9f0d8 | 361 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 362 | #endif |
2e53d63a | 363 | |
683412cc MZ |
364 | static void kvm_flush_shadow_all(struct kvm *kvm) |
365 | { | |
366 | kvm_arch_flush_shadow_all(kvm); | |
367 | kvm_arch_guest_memory_reclaimed(kvm); | |
368 | } | |
369 | ||
6926f95a SC |
370 | #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE |
371 | static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc, | |
372 | gfp_t gfp_flags) | |
373 | { | |
374 | gfp_flags |= mc->gfp_zero; | |
375 | ||
376 | if (mc->kmem_cache) | |
377 | return kmem_cache_alloc(mc->kmem_cache, gfp_flags); | |
378 | else | |
379 | return (void *)__get_free_page(gfp_flags); | |
380 | } | |
381 | ||
382 | int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min) | |
383 | { | |
384 | void *obj; | |
385 | ||
386 | if (mc->nobjs >= min) | |
387 | return 0; | |
388 | while (mc->nobjs < ARRAY_SIZE(mc->objects)) { | |
389 | obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT); | |
390 | if (!obj) | |
391 | return mc->nobjs >= min ? 0 : -ENOMEM; | |
392 | mc->objects[mc->nobjs++] = obj; | |
393 | } | |
394 | return 0; | |
395 | } | |
396 | ||
397 | int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc) | |
398 | { | |
399 | return mc->nobjs; | |
400 | } | |
401 | ||
402 | void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
403 | { | |
404 | while (mc->nobjs) { | |
405 | if (mc->kmem_cache) | |
406 | kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]); | |
407 | else | |
408 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
409 | } | |
410 | } | |
411 | ||
412 | void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
413 | { | |
414 | void *p; | |
415 | ||
416 | if (WARN_ON(!mc->nobjs)) | |
417 | p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT); | |
418 | else | |
419 | p = mc->objects[--mc->nobjs]; | |
420 | BUG_ON(!p); | |
421 | return p; | |
422 | } | |
423 | #endif | |
424 | ||
8bd826d6 | 425 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 426 | { |
fb3f0f51 RR |
427 | mutex_init(&vcpu->mutex); |
428 | vcpu->cpu = -1; | |
fb3f0f51 RR |
429 | vcpu->kvm = kvm; |
430 | vcpu->vcpu_id = id; | |
34bb10b7 | 431 | vcpu->pid = NULL; |
510958e9 | 432 | #ifndef __KVM_HAVE_ARCH_WQP |
da4ad88c | 433 | rcuwait_init(&vcpu->wait); |
510958e9 | 434 | #endif |
af585b92 | 435 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 436 | |
4c088493 R |
437 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
438 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 439 | vcpu->preempted = false; |
d73eb57b | 440 | vcpu->ready = false; |
d5c48deb | 441 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
a54d8066 | 442 | vcpu->last_used_slot = NULL; |
fb3f0f51 | 443 | } |
fb3f0f51 | 444 | |
27592ae8 | 445 | static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
4543bdc0 SC |
446 | { |
447 | kvm_arch_vcpu_destroy(vcpu); | |
5593473a | 448 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
e529ef66 | 449 | |
9941d224 SC |
450 | /* |
451 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
452 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
453 | * are already gone. | |
454 | */ | |
455 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
456 | ||
8bd826d6 | 457 | free_page((unsigned long)vcpu->run); |
e529ef66 | 458 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 | 459 | } |
27592ae8 MZ |
460 | |
461 | void kvm_destroy_vcpus(struct kvm *kvm) | |
462 | { | |
46808a4c | 463 | unsigned long i; |
27592ae8 MZ |
464 | struct kvm_vcpu *vcpu; |
465 | ||
466 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
467 | kvm_vcpu_destroy(vcpu); | |
c5b07754 | 468 | xa_erase(&kvm->vcpu_array, i); |
27592ae8 MZ |
469 | } |
470 | ||
471 | atomic_set(&kvm->online_vcpus, 0); | |
472 | } | |
473 | EXPORT_SYMBOL_GPL(kvm_destroy_vcpus); | |
4543bdc0 | 474 | |
e930bffe AA |
475 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
476 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
477 | { | |
478 | return container_of(mn, struct kvm, mmu_notifier); | |
479 | } | |
480 | ||
e649b3f0 ET |
481 | static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn, |
482 | struct mm_struct *mm, | |
483 | unsigned long start, unsigned long end) | |
484 | { | |
485 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
486 | int idx; | |
487 | ||
488 | idx = srcu_read_lock(&kvm->srcu); | |
489 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
490 | srcu_read_unlock(&kvm->srcu, idx); | |
491 | } | |
492 | ||
3039bcc7 SC |
493 | typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range); |
494 | ||
f922bd9b SC |
495 | typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start, |
496 | unsigned long end); | |
497 | ||
683412cc MZ |
498 | typedef void (*on_unlock_fn_t)(struct kvm *kvm); |
499 | ||
3039bcc7 SC |
500 | struct kvm_hva_range { |
501 | unsigned long start; | |
502 | unsigned long end; | |
503 | pte_t pte; | |
504 | hva_handler_t handler; | |
f922bd9b | 505 | on_lock_fn_t on_lock; |
683412cc | 506 | on_unlock_fn_t on_unlock; |
3039bcc7 SC |
507 | bool flush_on_ret; |
508 | bool may_block; | |
509 | }; | |
510 | ||
f922bd9b SC |
511 | /* |
512 | * Use a dedicated stub instead of NULL to indicate that there is no callback | |
513 | * function/handler. The compiler technically can't guarantee that a real | |
514 | * function will have a non-zero address, and so it will generate code to | |
515 | * check for !NULL, whereas comparing against a stub will be elided at compile | |
516 | * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9). | |
517 | */ | |
518 | static void kvm_null_fn(void) | |
519 | { | |
520 | ||
521 | } | |
522 | #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn) | |
523 | ||
ed922739 MS |
524 | /* Iterate over each memslot intersecting [start, last] (inclusive) range */ |
525 | #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \ | |
526 | for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \ | |
527 | node; \ | |
528 | node = interval_tree_iter_next(node, start, last)) \ | |
529 | ||
3039bcc7 SC |
530 | static __always_inline int __kvm_handle_hva_range(struct kvm *kvm, |
531 | const struct kvm_hva_range *range) | |
532 | { | |
8931a454 | 533 | bool ret = false, locked = false; |
f922bd9b | 534 | struct kvm_gfn_range gfn_range; |
3039bcc7 SC |
535 | struct kvm_memory_slot *slot; |
536 | struct kvm_memslots *slots; | |
3039bcc7 SC |
537 | int i, idx; |
538 | ||
ed922739 MS |
539 | if (WARN_ON_ONCE(range->end <= range->start)) |
540 | return 0; | |
541 | ||
f922bd9b SC |
542 | /* A null handler is allowed if and only if on_lock() is provided. */ |
543 | if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) && | |
544 | IS_KVM_NULL_FN(range->handler))) | |
545 | return 0; | |
546 | ||
3039bcc7 SC |
547 | idx = srcu_read_lock(&kvm->srcu); |
548 | ||
549 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
ed922739 MS |
550 | struct interval_tree_node *node; |
551 | ||
3039bcc7 | 552 | slots = __kvm_memslots(kvm, i); |
ed922739 MS |
553 | kvm_for_each_memslot_in_hva_range(node, slots, |
554 | range->start, range->end - 1) { | |
3039bcc7 SC |
555 | unsigned long hva_start, hva_end; |
556 | ||
a54d8066 | 557 | slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]); |
3039bcc7 SC |
558 | hva_start = max(range->start, slot->userspace_addr); |
559 | hva_end = min(range->end, slot->userspace_addr + | |
560 | (slot->npages << PAGE_SHIFT)); | |
3039bcc7 SC |
561 | |
562 | /* | |
563 | * To optimize for the likely case where the address | |
564 | * range is covered by zero or one memslots, don't | |
565 | * bother making these conditional (to avoid writes on | |
566 | * the second or later invocation of the handler). | |
567 | */ | |
568 | gfn_range.pte = range->pte; | |
569 | gfn_range.may_block = range->may_block; | |
570 | ||
571 | /* | |
572 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
573 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
574 | */ | |
575 | gfn_range.start = hva_to_gfn_memslot(hva_start, slot); | |
576 | gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot); | |
577 | gfn_range.slot = slot; | |
578 | ||
8931a454 SC |
579 | if (!locked) { |
580 | locked = true; | |
581 | KVM_MMU_LOCK(kvm); | |
071064f1 PB |
582 | if (!IS_KVM_NULL_FN(range->on_lock)) |
583 | range->on_lock(kvm, range->start, range->end); | |
584 | if (IS_KVM_NULL_FN(range->handler)) | |
585 | break; | |
8931a454 | 586 | } |
3039bcc7 SC |
587 | ret |= range->handler(kvm, &gfn_range); |
588 | } | |
589 | } | |
590 | ||
6bc6db00 | 591 | if (range->flush_on_ret && ret) |
3039bcc7 SC |
592 | kvm_flush_remote_tlbs(kvm); |
593 | ||
683412cc | 594 | if (locked) { |
8931a454 | 595 | KVM_MMU_UNLOCK(kvm); |
683412cc MZ |
596 | if (!IS_KVM_NULL_FN(range->on_unlock)) |
597 | range->on_unlock(kvm); | |
598 | } | |
f922bd9b | 599 | |
3039bcc7 SC |
600 | srcu_read_unlock(&kvm->srcu, idx); |
601 | ||
602 | /* The notifiers are averse to booleans. :-( */ | |
603 | return (int)ret; | |
604 | } | |
605 | ||
606 | static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn, | |
607 | unsigned long start, | |
608 | unsigned long end, | |
609 | pte_t pte, | |
610 | hva_handler_t handler) | |
611 | { | |
612 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
613 | const struct kvm_hva_range range = { | |
614 | .start = start, | |
615 | .end = end, | |
616 | .pte = pte, | |
617 | .handler = handler, | |
f922bd9b | 618 | .on_lock = (void *)kvm_null_fn, |
683412cc | 619 | .on_unlock = (void *)kvm_null_fn, |
3039bcc7 SC |
620 | .flush_on_ret = true, |
621 | .may_block = false, | |
622 | }; | |
3039bcc7 | 623 | |
f922bd9b | 624 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 SC |
625 | } |
626 | ||
627 | static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn, | |
628 | unsigned long start, | |
629 | unsigned long end, | |
630 | hva_handler_t handler) | |
631 | { | |
632 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
633 | const struct kvm_hva_range range = { | |
634 | .start = start, | |
635 | .end = end, | |
636 | .pte = __pte(0), | |
637 | .handler = handler, | |
f922bd9b | 638 | .on_lock = (void *)kvm_null_fn, |
683412cc | 639 | .on_unlock = (void *)kvm_null_fn, |
3039bcc7 SC |
640 | .flush_on_ret = false, |
641 | .may_block = false, | |
642 | }; | |
3039bcc7 | 643 | |
f922bd9b | 644 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 | 645 | } |
3da0dd43 IE |
646 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
647 | struct mm_struct *mm, | |
648 | unsigned long address, | |
649 | pte_t pte) | |
650 | { | |
651 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
652 | ||
501b9185 SC |
653 | trace_kvm_set_spte_hva(address); |
654 | ||
c13fda23 | 655 | /* |
52ac8b35 | 656 | * .change_pte() must be surrounded by .invalidate_range_{start,end}(). |
071064f1 PB |
657 | * If mmu_notifier_count is zero, then no in-progress invalidations, |
658 | * including this one, found a relevant memslot at start(); rechecking | |
659 | * memslots here is unnecessary. Note, a false positive (count elevated | |
660 | * by a different invalidation) is sub-optimal but functionally ok. | |
c13fda23 | 661 | */ |
52ac8b35 | 662 | WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count)); |
071064f1 PB |
663 | if (!READ_ONCE(kvm->mmu_notifier_count)) |
664 | return; | |
c13fda23 | 665 | |
3039bcc7 | 666 | kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn); |
3da0dd43 IE |
667 | } |
668 | ||
edb298c6 | 669 | void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 670 | unsigned long end) |
e930bffe | 671 | { |
e930bffe AA |
672 | /* |
673 | * The count increase must become visible at unlock time as no | |
674 | * spte can be established without taking the mmu_lock and | |
675 | * count is also read inside the mmu_lock critical section. | |
676 | */ | |
677 | kvm->mmu_notifier_count++; | |
4a42d848 | 678 | if (likely(kvm->mmu_notifier_count == 1)) { |
f922bd9b SC |
679 | kvm->mmu_notifier_range_start = start; |
680 | kvm->mmu_notifier_range_end = end; | |
4a42d848 DS |
681 | } else { |
682 | /* | |
a413a625 | 683 | * Fully tracking multiple concurrent ranges has diminishing |
4a42d848 DS |
684 | * returns. Keep things simple and just find the minimal range |
685 | * which includes the current and new ranges. As there won't be | |
686 | * enough information to subtract a range after its invalidate | |
687 | * completes, any ranges invalidated concurrently will | |
688 | * accumulate and persist until all outstanding invalidates | |
689 | * complete. | |
690 | */ | |
691 | kvm->mmu_notifier_range_start = | |
f922bd9b | 692 | min(kvm->mmu_notifier_range_start, start); |
4a42d848 | 693 | kvm->mmu_notifier_range_end = |
f922bd9b | 694 | max(kvm->mmu_notifier_range_end, end); |
4a42d848 | 695 | } |
f922bd9b | 696 | } |
3039bcc7 | 697 | |
f922bd9b SC |
698 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
699 | const struct mmu_notifier_range *range) | |
700 | { | |
701 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
702 | const struct kvm_hva_range hva_range = { | |
703 | .start = range->start, | |
704 | .end = range->end, | |
705 | .pte = __pte(0), | |
706 | .handler = kvm_unmap_gfn_range, | |
707 | .on_lock = kvm_inc_notifier_count, | |
683412cc | 708 | .on_unlock = kvm_arch_guest_memory_reclaimed, |
f922bd9b SC |
709 | .flush_on_ret = true, |
710 | .may_block = mmu_notifier_range_blockable(range), | |
711 | }; | |
565f3be2 | 712 | |
f922bd9b SC |
713 | trace_kvm_unmap_hva_range(range->start, range->end); |
714 | ||
52ac8b35 PB |
715 | /* |
716 | * Prevent memslot modification between range_start() and range_end() | |
717 | * so that conditionally locking provides the same result in both | |
718 | * functions. Without that guarantee, the mmu_notifier_count | |
719 | * adjustments will be imbalanced. | |
720 | * | |
721 | * Pairs with the decrement in range_end(). | |
722 | */ | |
723 | spin_lock(&kvm->mn_invalidate_lock); | |
724 | kvm->mn_active_invalidate_count++; | |
725 | spin_unlock(&kvm->mn_invalidate_lock); | |
726 | ||
58cd407c SC |
727 | /* |
728 | * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e. | |
729 | * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring | |
730 | * each cache's lock. There are relatively few caches in existence at | |
731 | * any given time, and the caches themselves can check for hva overlap, | |
732 | * i.e. don't need to rely on memslot overlap checks for performance. | |
733 | * Because this runs without holding mmu_lock, the pfn caches must use | |
734 | * mn_active_invalidate_count (see above) instead of mmu_notifier_count. | |
735 | */ | |
982ed0de DW |
736 | gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end, |
737 | hva_range.may_block); | |
738 | ||
f922bd9b | 739 | __kvm_handle_hva_range(kvm, &hva_range); |
93065ac7 | 740 | |
e649b3f0 | 741 | return 0; |
e930bffe AA |
742 | } |
743 | ||
edb298c6 | 744 | void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 745 | unsigned long end) |
e930bffe | 746 | { |
e930bffe AA |
747 | /* |
748 | * This sequence increase will notify the kvm page fault that | |
749 | * the page that is going to be mapped in the spte could have | |
750 | * been freed. | |
751 | */ | |
752 | kvm->mmu_notifier_seq++; | |
a355aa54 | 753 | smp_wmb(); |
e930bffe AA |
754 | /* |
755 | * The above sequence increase must be visible before the | |
a355aa54 PM |
756 | * below count decrease, which is ensured by the smp_wmb above |
757 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
758 | */ |
759 | kvm->mmu_notifier_count--; | |
f922bd9b SC |
760 | } |
761 | ||
762 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
763 | const struct mmu_notifier_range *range) | |
764 | { | |
765 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
766 | const struct kvm_hva_range hva_range = { | |
767 | .start = range->start, | |
768 | .end = range->end, | |
769 | .pte = __pte(0), | |
770 | .handler = (void *)kvm_null_fn, | |
771 | .on_lock = kvm_dec_notifier_count, | |
683412cc | 772 | .on_unlock = (void *)kvm_null_fn, |
f922bd9b SC |
773 | .flush_on_ret = false, |
774 | .may_block = mmu_notifier_range_blockable(range), | |
775 | }; | |
52ac8b35 | 776 | bool wake; |
f922bd9b SC |
777 | |
778 | __kvm_handle_hva_range(kvm, &hva_range); | |
e930bffe | 779 | |
52ac8b35 PB |
780 | /* Pairs with the increment in range_start(). */ |
781 | spin_lock(&kvm->mn_invalidate_lock); | |
782 | wake = (--kvm->mn_active_invalidate_count == 0); | |
783 | spin_unlock(&kvm->mn_invalidate_lock); | |
784 | ||
785 | /* | |
786 | * There can only be one waiter, since the wait happens under | |
787 | * slots_lock. | |
788 | */ | |
789 | if (wake) | |
790 | rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait); | |
791 | ||
e930bffe AA |
792 | BUG_ON(kvm->mmu_notifier_count < 0); |
793 | } | |
794 | ||
795 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
796 | struct mm_struct *mm, | |
57128468 ALC |
797 | unsigned long start, |
798 | unsigned long end) | |
e930bffe | 799 | { |
501b9185 SC |
800 | trace_kvm_age_hva(start, end); |
801 | ||
3039bcc7 | 802 | return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn); |
e930bffe AA |
803 | } |
804 | ||
1d7715c6 VD |
805 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
806 | struct mm_struct *mm, | |
807 | unsigned long start, | |
808 | unsigned long end) | |
809 | { | |
501b9185 SC |
810 | trace_kvm_age_hva(start, end); |
811 | ||
1d7715c6 VD |
812 | /* |
813 | * Even though we do not flush TLB, this will still adversely | |
814 | * affect performance on pre-Haswell Intel EPT, where there is | |
815 | * no EPT Access Bit to clear so that we have to tear down EPT | |
816 | * tables instead. If we find this unacceptable, we can always | |
817 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
818 | * do anything on clear_young. | |
819 | * | |
820 | * Also note that currently we never issue secondary TLB flushes | |
821 | * from clear_young, leaving this job up to the regular system | |
822 | * cadence. If we find this inaccurate, we might come up with a | |
823 | * more sophisticated heuristic later. | |
824 | */ | |
3039bcc7 | 825 | return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn); |
1d7715c6 VD |
826 | } |
827 | ||
8ee53820 AA |
828 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
829 | struct mm_struct *mm, | |
830 | unsigned long address) | |
831 | { | |
501b9185 SC |
832 | trace_kvm_test_age_hva(address); |
833 | ||
3039bcc7 SC |
834 | return kvm_handle_hva_range_no_flush(mn, address, address + 1, |
835 | kvm_test_age_gfn); | |
8ee53820 AA |
836 | } |
837 | ||
85db06e5 MT |
838 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
839 | struct mm_struct *mm) | |
840 | { | |
841 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
842 | int idx; |
843 | ||
844 | idx = srcu_read_lock(&kvm->srcu); | |
683412cc | 845 | kvm_flush_shadow_all(kvm); |
eda2beda | 846 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
847 | } |
848 | ||
e930bffe | 849 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e649b3f0 | 850 | .invalidate_range = kvm_mmu_notifier_invalidate_range, |
e930bffe AA |
851 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
852 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
853 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 854 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 855 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 856 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 857 | .release = kvm_mmu_notifier_release, |
e930bffe | 858 | }; |
4c07b0a4 AK |
859 | |
860 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
861 | { | |
862 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
863 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
864 | } | |
865 | ||
866 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
867 | ||
868 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
869 | { | |
870 | return 0; | |
871 | } | |
872 | ||
e930bffe AA |
873 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
874 | ||
2fdef3a2 SS |
875 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
876 | static int kvm_pm_notifier_call(struct notifier_block *bl, | |
877 | unsigned long state, | |
878 | void *unused) | |
879 | { | |
880 | struct kvm *kvm = container_of(bl, struct kvm, pm_notifier); | |
881 | ||
882 | return kvm_arch_pm_notifier(kvm, state); | |
883 | } | |
884 | ||
885 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
886 | { | |
887 | kvm->pm_notifier.notifier_call = kvm_pm_notifier_call; | |
888 | /* Suspend KVM before we suspend ftrace, RCU, etc. */ | |
889 | kvm->pm_notifier.priority = INT_MAX; | |
890 | register_pm_notifier(&kvm->pm_notifier); | |
891 | } | |
892 | ||
893 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
894 | { | |
895 | unregister_pm_notifier(&kvm->pm_notifier); | |
896 | } | |
897 | #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
898 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
899 | { | |
900 | } | |
901 | ||
902 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
903 | { | |
904 | } | |
905 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
906 | ||
a47d2b07 PB |
907 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
908 | { | |
909 | if (!memslot->dirty_bitmap) | |
910 | return; | |
911 | ||
912 | kvfree(memslot->dirty_bitmap); | |
913 | memslot->dirty_bitmap = NULL; | |
914 | } | |
915 | ||
a54d8066 | 916 | /* This does not remove the slot from struct kvm_memslots data structures */ |
e96c81ee | 917 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 918 | { |
e96c81ee | 919 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 920 | |
e96c81ee | 921 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 922 | |
a54d8066 | 923 | kfree(slot); |
a47d2b07 PB |
924 | } |
925 | ||
926 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
927 | { | |
a54d8066 | 928 | struct hlist_node *idnode; |
a47d2b07 | 929 | struct kvm_memory_slot *memslot; |
a54d8066 | 930 | int bkt; |
a47d2b07 | 931 | |
a54d8066 MS |
932 | /* |
933 | * The same memslot objects live in both active and inactive sets, | |
934 | * arbitrarily free using index '1' so the second invocation of this | |
935 | * function isn't operating over a structure with dangling pointers | |
936 | * (even though this function isn't actually touching them). | |
937 | */ | |
938 | if (!slots->node_idx) | |
a47d2b07 PB |
939 | return; |
940 | ||
a54d8066 | 941 | hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1]) |
e96c81ee | 942 | kvm_free_memslot(kvm, memslot); |
bf3e05bc XG |
943 | } |
944 | ||
bc9e9e67 JZ |
945 | static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc) |
946 | { | |
947 | switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) { | |
948 | case KVM_STATS_TYPE_INSTANT: | |
949 | return 0444; | |
950 | case KVM_STATS_TYPE_CUMULATIVE: | |
951 | case KVM_STATS_TYPE_PEAK: | |
952 | default: | |
953 | return 0644; | |
954 | } | |
955 | } | |
956 | ||
957 | ||
536a6f88 JF |
958 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
959 | { | |
960 | int i; | |
bc9e9e67 JZ |
961 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
962 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 | 963 | |
a44a4cc1 | 964 | if (IS_ERR(kvm->debugfs_dentry)) |
536a6f88 JF |
965 | return; |
966 | ||
967 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
968 | ||
9d5a1dce LC |
969 | if (kvm->debugfs_stat_data) { |
970 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
971 | kfree(kvm->debugfs_stat_data[i]); | |
972 | kfree(kvm->debugfs_stat_data); | |
973 | } | |
536a6f88 JF |
974 | } |
975 | ||
976 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
977 | { | |
85cd39af PB |
978 | static DEFINE_MUTEX(kvm_debugfs_lock); |
979 | struct dentry *dent; | |
536a6f88 JF |
980 | char dir_name[ITOA_MAX_LEN * 2]; |
981 | struct kvm_stat_data *stat_data; | |
bc9e9e67 | 982 | const struct _kvm_stats_desc *pdesc; |
3165af73 | 983 | int i, ret; |
bc9e9e67 JZ |
984 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
985 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
986 | |
987 | if (!debugfs_initialized()) | |
988 | return 0; | |
989 | ||
990 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
85cd39af PB |
991 | mutex_lock(&kvm_debugfs_lock); |
992 | dent = debugfs_lookup(dir_name, kvm_debugfs_dir); | |
993 | if (dent) { | |
994 | pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name); | |
995 | dput(dent); | |
996 | mutex_unlock(&kvm_debugfs_lock); | |
997 | return 0; | |
998 | } | |
999 | dent = debugfs_create_dir(dir_name, kvm_debugfs_dir); | |
1000 | mutex_unlock(&kvm_debugfs_lock); | |
1001 | if (IS_ERR(dent)) | |
1002 | return 0; | |
536a6f88 | 1003 | |
85cd39af | 1004 | kvm->debugfs_dentry = dent; |
536a6f88 JF |
1005 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
1006 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 1007 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
1008 | if (!kvm->debugfs_stat_data) |
1009 | return -ENOMEM; | |
1010 | ||
bc9e9e67 JZ |
1011 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
1012 | pdesc = &kvm_vm_stats_desc[i]; | |
b12ce36a | 1013 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
1014 | if (!stat_data) |
1015 | return -ENOMEM; | |
1016 | ||
1017 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1018 | stat_data->desc = pdesc; |
1019 | stat_data->kind = KVM_STAT_VM; | |
1020 | kvm->debugfs_stat_data[i] = stat_data; | |
1021 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
1022 | kvm->debugfs_dentry, stat_data, | |
1023 | &stat_fops_per_vm); | |
1024 | } | |
1025 | ||
1026 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
1027 | pdesc = &kvm_vcpu_stats_desc[i]; | |
b12ce36a | 1028 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
1029 | if (!stat_data) |
1030 | return -ENOMEM; | |
1031 | ||
1032 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
1033 | stat_data->desc = pdesc; |
1034 | stat_data->kind = KVM_STAT_VCPU; | |
004d62eb | 1035 | kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data; |
bc9e9e67 | 1036 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), |
09cbcef6 MP |
1037 | kvm->debugfs_dentry, stat_data, |
1038 | &stat_fops_per_vm); | |
536a6f88 | 1039 | } |
3165af73 PX |
1040 | |
1041 | ret = kvm_arch_create_vm_debugfs(kvm); | |
1042 | if (ret) { | |
1043 | kvm_destroy_vm_debugfs(kvm); | |
1044 | return i; | |
1045 | } | |
1046 | ||
536a6f88 JF |
1047 | return 0; |
1048 | } | |
1049 | ||
1aa9b957 JS |
1050 | /* |
1051 | * Called after the VM is otherwise initialized, but just before adding it to | |
1052 | * the vm_list. | |
1053 | */ | |
1054 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
1055 | { | |
1056 | return 0; | |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * Called just after removing the VM from the vm_list, but before doing any | |
1061 | * other destruction. | |
1062 | */ | |
1063 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
1064 | { | |
1065 | } | |
1066 | ||
3165af73 PX |
1067 | /* |
1068 | * Called after per-vm debugfs created. When called kvm->debugfs_dentry should | |
1069 | * be setup already, so we can create arch-specific debugfs entries under it. | |
1070 | * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so | |
1071 | * a per-arch destroy interface is not needed. | |
1072 | */ | |
1073 | int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm) | |
1074 | { | |
1075 | return 0; | |
1076 | } | |
1077 | ||
e08b9637 | 1078 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 1079 | { |
d89f5eff | 1080 | struct kvm *kvm = kvm_arch_alloc_vm(); |
a54d8066 | 1081 | struct kvm_memslots *slots; |
9121923c | 1082 | int r = -ENOMEM; |
a54d8066 | 1083 | int i, j; |
6aa8b732 | 1084 | |
d89f5eff JK |
1085 | if (!kvm) |
1086 | return ERR_PTR(-ENOMEM); | |
1087 | ||
531810ca | 1088 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 1089 | mmgrab(current->mm); |
e9ad4ec8 PB |
1090 | kvm->mm = current->mm; |
1091 | kvm_eventfd_init(kvm); | |
1092 | mutex_init(&kvm->lock); | |
1093 | mutex_init(&kvm->irq_lock); | |
1094 | mutex_init(&kvm->slots_lock); | |
b10a038e | 1095 | mutex_init(&kvm->slots_arch_lock); |
52ac8b35 PB |
1096 | spin_lock_init(&kvm->mn_invalidate_lock); |
1097 | rcuwait_init(&kvm->mn_memslots_update_rcuwait); | |
c5b07754 | 1098 | xa_init(&kvm->vcpu_array); |
52ac8b35 | 1099 | |
982ed0de DW |
1100 | INIT_LIST_HEAD(&kvm->gpc_list); |
1101 | spin_lock_init(&kvm->gpc_lock); | |
52ac8b35 | 1102 | |
e9ad4ec8 | 1103 | INIT_LIST_HEAD(&kvm->devices); |
f502cc56 | 1104 | kvm->max_vcpus = KVM_MAX_VCPUS; |
e9ad4ec8 | 1105 | |
1e702d9a AW |
1106 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
1107 | ||
5c697c36 SC |
1108 | /* |
1109 | * Force subsequent debugfs file creations to fail if the VM directory | |
1110 | * is not created (by kvm_create_vm_debugfs()). | |
1111 | */ | |
1112 | kvm->debugfs_dentry = ERR_PTR(-ENOENT); | |
1113 | ||
8a44119a PB |
1114 | if (init_srcu_struct(&kvm->srcu)) |
1115 | goto out_err_no_srcu; | |
1116 | if (init_srcu_struct(&kvm->irq_srcu)) | |
1117 | goto out_err_no_irq_srcu; | |
1118 | ||
e2d3fcaf | 1119 | refcount_set(&kvm->users_count, 1); |
f481b069 | 1120 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
a54d8066 MS |
1121 | for (j = 0; j < 2; j++) { |
1122 | slots = &kvm->__memslots[i][j]; | |
9121923c | 1123 | |
a54d8066 MS |
1124 | atomic_long_set(&slots->last_used_slot, (unsigned long)NULL); |
1125 | slots->hva_tree = RB_ROOT_CACHED; | |
1126 | slots->gfn_tree = RB_ROOT; | |
1127 | hash_init(slots->id_hash); | |
1128 | slots->node_idx = j; | |
1129 | ||
1130 | /* Generations must be different for each address space. */ | |
1131 | slots->generation = i; | |
1132 | } | |
1133 | ||
1134 | rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]); | |
f481b069 | 1135 | } |
00f034a1 | 1136 | |
e93f8a0f | 1137 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 1138 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 1139 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 1140 | if (!kvm->buses[i]) |
a97b0e77 | 1141 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 1142 | } |
e930bffe | 1143 | |
acd05785 DM |
1144 | kvm->max_halt_poll_ns = halt_poll_ns; |
1145 | ||
e08b9637 | 1146 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 1147 | if (r) |
a97b0e77 | 1148 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
1149 | |
1150 | r = hardware_enable_all(); | |
1151 | if (r) | |
719d93cd | 1152 | goto out_err_no_disable; |
10474ae8 | 1153 | |
c77dcacb | 1154 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 1155 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 1156 | #endif |
6aa8b732 | 1157 | |
74b5c5bf | 1158 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
1159 | if (r) |
1160 | goto out_err_no_mmu_notifier; | |
1161 | ||
1162 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
1163 | if (r) |
1164 | goto out_err; | |
1165 | ||
0d9ce162 | 1166 | mutex_lock(&kvm_lock); |
5e58cfe4 | 1167 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 1168 | mutex_unlock(&kvm_lock); |
d89f5eff | 1169 | |
2ecd9d29 | 1170 | preempt_notifier_inc(); |
2fdef3a2 | 1171 | kvm_init_pm_notifier(kvm); |
2ecd9d29 | 1172 | |
5f6de5cb DM |
1173 | /* |
1174 | * When the fd passed to this ioctl() is opened it pins the module, | |
1175 | * but try_module_get() also prevents getting a reference if the module | |
1176 | * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait"). | |
1177 | */ | |
1178 | if (!try_module_get(kvm_chardev_ops.owner)) { | |
1179 | r = -ENODEV; | |
1180 | goto out_err; | |
1181 | } | |
1182 | ||
f17abe9a | 1183 | return kvm; |
10474ae8 AG |
1184 | |
1185 | out_err: | |
1aa9b957 JS |
1186 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1187 | if (kvm->mmu_notifier.ops) | |
1188 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
1189 | #endif | |
1190 | out_err_no_mmu_notifier: | |
10474ae8 | 1191 | hardware_disable_all(); |
719d93cd | 1192 | out_err_no_disable: |
a97b0e77 | 1193 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 1194 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 1195 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 1196 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 1197 | kfree(kvm_get_bus(kvm, i)); |
8a44119a PB |
1198 | cleanup_srcu_struct(&kvm->irq_srcu); |
1199 | out_err_no_irq_srcu: | |
1200 | cleanup_srcu_struct(&kvm->srcu); | |
1201 | out_err_no_srcu: | |
d89f5eff | 1202 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 1203 | mmdrop(current->mm); |
10474ae8 | 1204 | return ERR_PTR(r); |
f17abe9a AK |
1205 | } |
1206 | ||
07f0a7bd SW |
1207 | static void kvm_destroy_devices(struct kvm *kvm) |
1208 | { | |
e6e3b5a6 | 1209 | struct kvm_device *dev, *tmp; |
07f0a7bd | 1210 | |
a28ebea2 CD |
1211 | /* |
1212 | * We do not need to take the kvm->lock here, because nobody else | |
1213 | * has a reference to the struct kvm at this point and therefore | |
1214 | * cannot access the devices list anyhow. | |
1215 | */ | |
e6e3b5a6 GT |
1216 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
1217 | list_del(&dev->vm_node); | |
07f0a7bd SW |
1218 | dev->ops->destroy(dev); |
1219 | } | |
1220 | } | |
1221 | ||
f17abe9a AK |
1222 | static void kvm_destroy_vm(struct kvm *kvm) |
1223 | { | |
e93f8a0f | 1224 | int i; |
6d4e4c4f AK |
1225 | struct mm_struct *mm = kvm->mm; |
1226 | ||
2fdef3a2 | 1227 | kvm_destroy_pm_notifier(kvm); |
286de8f6 | 1228 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 1229 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 1230 | kvm_arch_sync_events(kvm); |
0d9ce162 | 1231 | mutex_lock(&kvm_lock); |
133de902 | 1232 | list_del(&kvm->vm_list); |
0d9ce162 | 1233 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
1234 | kvm_arch_pre_destroy_vm(kvm); |
1235 | ||
399ec807 | 1236 | kvm_free_irq_routing(kvm); |
df630b8c | 1237 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 1238 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 1239 | |
4a12f951 CB |
1240 | if (bus) |
1241 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
1242 | kvm->buses[i] = NULL; |
1243 | } | |
980da6ce | 1244 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
1245 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1246 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
52ac8b35 PB |
1247 | /* |
1248 | * At this point, pending calls to invalidate_range_start() | |
1249 | * have completed but no more MMU notifiers will run, so | |
1250 | * mn_active_invalidate_count may remain unbalanced. | |
1251 | * No threads can be waiting in install_new_memslots as the | |
1252 | * last reference on KVM has been dropped, but freeing | |
1253 | * memslots would deadlock without this manual intervention. | |
1254 | */ | |
1255 | WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait)); | |
1256 | kvm->mn_active_invalidate_count = 0; | |
f00be0ca | 1257 | #else |
683412cc | 1258 | kvm_flush_shadow_all(kvm); |
5f94c174 | 1259 | #endif |
d19a9cd2 | 1260 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 1261 | kvm_destroy_devices(kvm); |
a54d8066 MS |
1262 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
1263 | kvm_free_memslots(kvm, &kvm->__memslots[i][0]); | |
1264 | kvm_free_memslots(kvm, &kvm->__memslots[i][1]); | |
1265 | } | |
820b3fcd | 1266 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
1267 | cleanup_srcu_struct(&kvm->srcu); |
1268 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 1269 | preempt_notifier_dec(); |
10474ae8 | 1270 | hardware_disable_all(); |
6d4e4c4f | 1271 | mmdrop(mm); |
5f6de5cb | 1272 | module_put(kvm_chardev_ops.owner); |
f17abe9a AK |
1273 | } |
1274 | ||
d39f13b0 IE |
1275 | void kvm_get_kvm(struct kvm *kvm) |
1276 | { | |
e3736c3e | 1277 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
1278 | } |
1279 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
1280 | ||
605c7130 PX |
1281 | /* |
1282 | * Make sure the vm is not during destruction, which is a safe version of | |
1283 | * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise. | |
1284 | */ | |
1285 | bool kvm_get_kvm_safe(struct kvm *kvm) | |
1286 | { | |
1287 | return refcount_inc_not_zero(&kvm->users_count); | |
1288 | } | |
1289 | EXPORT_SYMBOL_GPL(kvm_get_kvm_safe); | |
1290 | ||
d39f13b0 IE |
1291 | void kvm_put_kvm(struct kvm *kvm) |
1292 | { | |
e3736c3e | 1293 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
1294 | kvm_destroy_vm(kvm); |
1295 | } | |
1296 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
1297 | ||
149487bd SC |
1298 | /* |
1299 | * Used to put a reference that was taken on behalf of an object associated | |
1300 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
1301 | * of the new file descriptor fails and the reference cannot be transferred to | |
1302 | * its final owner. In such cases, the caller is still actively using @kvm and | |
1303 | * will fail miserably if the refcount unexpectedly hits zero. | |
1304 | */ | |
1305 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
1306 | { | |
1307 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
1308 | } | |
1309 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 1310 | |
f17abe9a AK |
1311 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
1312 | { | |
1313 | struct kvm *kvm = filp->private_data; | |
1314 | ||
721eecbf GH |
1315 | kvm_irqfd_release(kvm); |
1316 | ||
d39f13b0 | 1317 | kvm_put_kvm(kvm); |
6aa8b732 AK |
1318 | return 0; |
1319 | } | |
1320 | ||
515a0127 TY |
1321 | /* |
1322 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 1323 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 1324 | */ |
3c9bd400 | 1325 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 1326 | { |
37b2a651 | 1327 | unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 1328 | |
37b2a651 | 1329 | memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
1330 | if (!memslot->dirty_bitmap) |
1331 | return -ENOMEM; | |
1332 | ||
a36a57b1 TY |
1333 | return 0; |
1334 | } | |
1335 | ||
a54d8066 | 1336 | static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id) |
bf3e05bc | 1337 | { |
a54d8066 MS |
1338 | struct kvm_memslots *active = __kvm_memslots(kvm, as_id); |
1339 | int node_idx_inactive = active->node_idx ^ 1; | |
0e60b079 | 1340 | |
a54d8066 | 1341 | return &kvm->__memslots[as_id][node_idx_inactive]; |
0577d1ab SC |
1342 | } |
1343 | ||
1344 | /* | |
a54d8066 MS |
1345 | * Helper to get the address space ID when one of memslot pointers may be NULL. |
1346 | * This also serves as a sanity that at least one of the pointers is non-NULL, | |
1347 | * and that their address space IDs don't diverge. | |
0577d1ab | 1348 | */ |
a54d8066 MS |
1349 | static int kvm_memslots_get_as_id(struct kvm_memory_slot *a, |
1350 | struct kvm_memory_slot *b) | |
0577d1ab | 1351 | { |
a54d8066 MS |
1352 | if (WARN_ON_ONCE(!a && !b)) |
1353 | return 0; | |
0577d1ab | 1354 | |
a54d8066 MS |
1355 | if (!a) |
1356 | return b->as_id; | |
1357 | if (!b) | |
1358 | return a->as_id; | |
0577d1ab | 1359 | |
a54d8066 MS |
1360 | WARN_ON_ONCE(a->as_id != b->as_id); |
1361 | return a->as_id; | |
0577d1ab | 1362 | } |
efbeec70 | 1363 | |
a54d8066 MS |
1364 | static void kvm_insert_gfn_node(struct kvm_memslots *slots, |
1365 | struct kvm_memory_slot *slot) | |
0577d1ab | 1366 | { |
a54d8066 MS |
1367 | struct rb_root *gfn_tree = &slots->gfn_tree; |
1368 | struct rb_node **node, *parent; | |
1369 | int idx = slots->node_idx; | |
0577d1ab | 1370 | |
a54d8066 MS |
1371 | parent = NULL; |
1372 | for (node = &gfn_tree->rb_node; *node; ) { | |
1373 | struct kvm_memory_slot *tmp; | |
f85e2cb5 | 1374 | |
a54d8066 MS |
1375 | tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]); |
1376 | parent = *node; | |
1377 | if (slot->base_gfn < tmp->base_gfn) | |
1378 | node = &(*node)->rb_left; | |
1379 | else if (slot->base_gfn > tmp->base_gfn) | |
1380 | node = &(*node)->rb_right; | |
1381 | else | |
1382 | BUG(); | |
0577d1ab | 1383 | } |
a54d8066 MS |
1384 | |
1385 | rb_link_node(&slot->gfn_node[idx], parent, node); | |
1386 | rb_insert_color(&slot->gfn_node[idx], gfn_tree); | |
0577d1ab SC |
1387 | } |
1388 | ||
a54d8066 MS |
1389 | static void kvm_erase_gfn_node(struct kvm_memslots *slots, |
1390 | struct kvm_memory_slot *slot) | |
0577d1ab | 1391 | { |
a54d8066 MS |
1392 | rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree); |
1393 | } | |
0577d1ab | 1394 | |
a54d8066 MS |
1395 | static void kvm_replace_gfn_node(struct kvm_memslots *slots, |
1396 | struct kvm_memory_slot *old, | |
1397 | struct kvm_memory_slot *new) | |
1398 | { | |
1399 | int idx = slots->node_idx; | |
0577d1ab | 1400 | |
a54d8066 | 1401 | WARN_ON_ONCE(old->base_gfn != new->base_gfn); |
0577d1ab | 1402 | |
a54d8066 MS |
1403 | rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx], |
1404 | &slots->gfn_tree); | |
0577d1ab SC |
1405 | } |
1406 | ||
1407 | /* | |
a54d8066 | 1408 | * Replace @old with @new in the inactive memslots. |
0577d1ab | 1409 | * |
a54d8066 MS |
1410 | * With NULL @old this simply adds @new. |
1411 | * With NULL @new this simply removes @old. | |
0577d1ab | 1412 | * |
a54d8066 MS |
1413 | * If @new is non-NULL its hva_node[slots_idx] range has to be set |
1414 | * appropriately. | |
0577d1ab | 1415 | */ |
a54d8066 MS |
1416 | static void kvm_replace_memslot(struct kvm *kvm, |
1417 | struct kvm_memory_slot *old, | |
1418 | struct kvm_memory_slot *new) | |
0577d1ab | 1419 | { |
a54d8066 MS |
1420 | int as_id = kvm_memslots_get_as_id(old, new); |
1421 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); | |
1422 | int idx = slots->node_idx; | |
0577d1ab | 1423 | |
a54d8066 MS |
1424 | if (old) { |
1425 | hash_del(&old->id_node[idx]); | |
1426 | interval_tree_remove(&old->hva_node[idx], &slots->hva_tree); | |
0577d1ab | 1427 | |
a54d8066 MS |
1428 | if ((long)old == atomic_long_read(&slots->last_used_slot)) |
1429 | atomic_long_set(&slots->last_used_slot, (long)new); | |
0577d1ab | 1430 | |
a54d8066 MS |
1431 | if (!new) { |
1432 | kvm_erase_gfn_node(slots, old); | |
1e8617d3 | 1433 | return; |
a54d8066 MS |
1434 | } |
1435 | } | |
1e8617d3 | 1436 | |
a54d8066 MS |
1437 | /* |
1438 | * Initialize @new's hva range. Do this even when replacing an @old | |
1439 | * slot, kvm_copy_memslot() deliberately does not touch node data. | |
1440 | */ | |
1441 | new->hva_node[idx].start = new->userspace_addr; | |
1442 | new->hva_node[idx].last = new->userspace_addr + | |
1443 | (new->npages << PAGE_SHIFT) - 1; | |
1444 | ||
1445 | /* | |
1446 | * (Re)Add the new memslot. There is no O(1) interval_tree_replace(), | |
1447 | * hva_node needs to be swapped with remove+insert even though hva can't | |
1448 | * change when replacing an existing slot. | |
1449 | */ | |
1450 | hash_add(slots->id_hash, &new->id_node[idx], new->id); | |
1451 | interval_tree_insert(&new->hva_node[idx], &slots->hva_tree); | |
1452 | ||
1453 | /* | |
1454 | * If the memslot gfn is unchanged, rb_replace_node() can be used to | |
1455 | * switch the node in the gfn tree instead of removing the old and | |
1456 | * inserting the new as two separate operations. Replacement is a | |
1457 | * single O(1) operation versus two O(log(n)) operations for | |
1458 | * remove+insert. | |
1459 | */ | |
1460 | if (old && old->base_gfn == new->base_gfn) { | |
1461 | kvm_replace_gfn_node(slots, old, new); | |
1462 | } else { | |
1463 | if (old) | |
1464 | kvm_erase_gfn_node(slots, old); | |
1465 | kvm_insert_gfn_node(slots, new); | |
0577d1ab | 1466 | } |
bf3e05bc XG |
1467 | } |
1468 | ||
09170a49 | 1469 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 1470 | { |
4d8b81ab XG |
1471 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1472 | ||
0f8a4de3 | 1473 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
1474 | valid_flags |= KVM_MEM_READONLY; |
1475 | #endif | |
1476 | ||
1477 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1478 | return -EINVAL; |
1479 | ||
1480 | return 0; | |
1481 | } | |
1482 | ||
a54d8066 | 1483 | static void kvm_swap_active_memslots(struct kvm *kvm, int as_id) |
7ec4fb44 | 1484 | { |
a54d8066 MS |
1485 | struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id); |
1486 | ||
1487 | /* Grab the generation from the activate memslots. */ | |
1488 | u64 gen = __kvm_memslots(kvm, as_id)->generation; | |
7ec4fb44 | 1489 | |
361209e0 SC |
1490 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1491 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1492 | |
52ac8b35 PB |
1493 | /* |
1494 | * Do not store the new memslots while there are invalidations in | |
071064f1 PB |
1495 | * progress, otherwise the locking in invalidate_range_start and |
1496 | * invalidate_range_end will be unbalanced. | |
52ac8b35 PB |
1497 | */ |
1498 | spin_lock(&kvm->mn_invalidate_lock); | |
1499 | prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
1500 | while (kvm->mn_active_invalidate_count) { | |
1501 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1502 | spin_unlock(&kvm->mn_invalidate_lock); | |
1503 | schedule(); | |
1504 | spin_lock(&kvm->mn_invalidate_lock); | |
1505 | } | |
1506 | finish_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
f481b069 | 1507 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
52ac8b35 | 1508 | spin_unlock(&kvm->mn_invalidate_lock); |
b10a038e BG |
1509 | |
1510 | /* | |
1511 | * Acquired in kvm_set_memslot. Must be released before synchronize | |
1512 | * SRCU below in order to avoid deadlock with another thread | |
1513 | * acquiring the slots_arch_lock in an srcu critical section. | |
1514 | */ | |
1515 | mutex_unlock(&kvm->slots_arch_lock); | |
1516 | ||
7ec4fb44 | 1517 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1518 | |
ee3d1570 | 1519 | /* |
361209e0 | 1520 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1521 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1522 | * the number of address spaces. This provides a unique and easily |
1523 | * identifiable generation number while the memslots are in flux. | |
1524 | */ | |
1525 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1526 | ||
1527 | /* | |
4bd518f1 PB |
1528 | * Generations must be unique even across address spaces. We do not need |
1529 | * a global counter for that, instead the generation space is evenly split | |
1530 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1531 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1532 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1533 | */ |
164bf7e5 | 1534 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 1535 | |
15248258 | 1536 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1537 | |
15248258 | 1538 | slots->generation = gen; |
7ec4fb44 GN |
1539 | } |
1540 | ||
07921665 SC |
1541 | static int kvm_prepare_memory_region(struct kvm *kvm, |
1542 | const struct kvm_memory_slot *old, | |
1543 | struct kvm_memory_slot *new, | |
1544 | enum kvm_mr_change change) | |
ddc12f2a | 1545 | { |
07921665 SC |
1546 | int r; |
1547 | ||
1548 | /* | |
1549 | * If dirty logging is disabled, nullify the bitmap; the old bitmap | |
1550 | * will be freed on "commit". If logging is enabled in both old and | |
1551 | * new, reuse the existing bitmap. If logging is enabled only in the | |
1552 | * new and KVM isn't using a ring buffer, allocate and initialize a | |
1553 | * new bitmap. | |
1554 | */ | |
244893fa SC |
1555 | if (change != KVM_MR_DELETE) { |
1556 | if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1557 | new->dirty_bitmap = NULL; | |
1558 | else if (old && old->dirty_bitmap) | |
1559 | new->dirty_bitmap = old->dirty_bitmap; | |
1560 | else if (!kvm->dirty_ring_size) { | |
1561 | r = kvm_alloc_dirty_bitmap(new); | |
1562 | if (r) | |
1563 | return r; | |
1564 | ||
1565 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1566 | bitmap_set(new->dirty_bitmap, 0, new->npages); | |
1567 | } | |
07921665 SC |
1568 | } |
1569 | ||
1570 | r = kvm_arch_prepare_memory_region(kvm, old, new, change); | |
1571 | ||
1572 | /* Free the bitmap on failure if it was allocated above. */ | |
c87661f8 | 1573 | if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap)) |
07921665 SC |
1574 | kvm_destroy_dirty_bitmap(new); |
1575 | ||
1576 | return r; | |
ddc12f2a BG |
1577 | } |
1578 | ||
07921665 SC |
1579 | static void kvm_commit_memory_region(struct kvm *kvm, |
1580 | struct kvm_memory_slot *old, | |
1581 | const struct kvm_memory_slot *new, | |
1582 | enum kvm_mr_change change) | |
ddc12f2a | 1583 | { |
07921665 SC |
1584 | /* |
1585 | * Update the total number of memslot pages before calling the arch | |
1586 | * hook so that architectures can consume the result directly. | |
1587 | */ | |
1588 | if (change == KVM_MR_DELETE) | |
1589 | kvm->nr_memslot_pages -= old->npages; | |
1590 | else if (change == KVM_MR_CREATE) | |
1591 | kvm->nr_memslot_pages += new->npages; | |
1592 | ||
1593 | kvm_arch_commit_memory_region(kvm, old, new, change); | |
1594 | ||
a54d8066 MS |
1595 | switch (change) { |
1596 | case KVM_MR_CREATE: | |
1597 | /* Nothing more to do. */ | |
1598 | break; | |
1599 | case KVM_MR_DELETE: | |
1600 | /* Free the old memslot and all its metadata. */ | |
1601 | kvm_free_memslot(kvm, old); | |
1602 | break; | |
1603 | case KVM_MR_MOVE: | |
1604 | case KVM_MR_FLAGS_ONLY: | |
1605 | /* | |
1606 | * Free the dirty bitmap as needed; the below check encompasses | |
1607 | * both the flags and whether a ring buffer is being used) | |
1608 | */ | |
1609 | if (old->dirty_bitmap && !new->dirty_bitmap) | |
1610 | kvm_destroy_dirty_bitmap(old); | |
1611 | ||
1612 | /* | |
1613 | * The final quirk. Free the detached, old slot, but only its | |
1614 | * memory, not any metadata. Metadata, including arch specific | |
1615 | * data, may be reused by @new. | |
1616 | */ | |
1617 | kfree(old); | |
1618 | break; | |
1619 | default: | |
1620 | BUG(); | |
1621 | } | |
ddc12f2a BG |
1622 | } |
1623 | ||
36947254 | 1624 | /* |
a54d8066 MS |
1625 | * Activate @new, which must be installed in the inactive slots by the caller, |
1626 | * by swapping the active slots and then propagating @new to @old once @old is | |
1627 | * unreachable and can be safely modified. | |
1628 | * | |
1629 | * With NULL @old this simply adds @new to @active (while swapping the sets). | |
1630 | * With NULL @new this simply removes @old from @active and frees it | |
1631 | * (while also swapping the sets). | |
36947254 | 1632 | */ |
a54d8066 MS |
1633 | static void kvm_activate_memslot(struct kvm *kvm, |
1634 | struct kvm_memory_slot *old, | |
1635 | struct kvm_memory_slot *new) | |
36947254 | 1636 | { |
a54d8066 | 1637 | int as_id = kvm_memslots_get_as_id(old, new); |
36947254 | 1638 | |
a54d8066 MS |
1639 | kvm_swap_active_memslots(kvm, as_id); |
1640 | ||
1641 | /* Propagate the new memslot to the now inactive memslots. */ | |
1642 | kvm_replace_memslot(kvm, old, new); | |
1643 | } | |
1644 | ||
1645 | static void kvm_copy_memslot(struct kvm_memory_slot *dest, | |
1646 | const struct kvm_memory_slot *src) | |
1647 | { | |
1648 | dest->base_gfn = src->base_gfn; | |
1649 | dest->npages = src->npages; | |
1650 | dest->dirty_bitmap = src->dirty_bitmap; | |
1651 | dest->arch = src->arch; | |
1652 | dest->userspace_addr = src->userspace_addr; | |
1653 | dest->flags = src->flags; | |
1654 | dest->id = src->id; | |
1655 | dest->as_id = src->as_id; | |
1656 | } | |
1657 | ||
1658 | static void kvm_invalidate_memslot(struct kvm *kvm, | |
1659 | struct kvm_memory_slot *old, | |
244893fa | 1660 | struct kvm_memory_slot *invalid_slot) |
a54d8066 | 1661 | { |
07921665 | 1662 | /* |
a54d8066 MS |
1663 | * Mark the current slot INVALID. As with all memslot modifications, |
1664 | * this must be done on an unreachable slot to avoid modifying the | |
1665 | * current slot in the active tree. | |
07921665 | 1666 | */ |
244893fa SC |
1667 | kvm_copy_memslot(invalid_slot, old); |
1668 | invalid_slot->flags |= KVM_MEMSLOT_INVALID; | |
1669 | kvm_replace_memslot(kvm, old, invalid_slot); | |
a54d8066 MS |
1670 | |
1671 | /* | |
1672 | * Activate the slot that is now marked INVALID, but don't propagate | |
1673 | * the slot to the now inactive slots. The slot is either going to be | |
1674 | * deleted or recreated as a new slot. | |
1675 | */ | |
1676 | kvm_swap_active_memslots(kvm, old->as_id); | |
1677 | ||
1678 | /* | |
1679 | * From this point no new shadow pages pointing to a deleted, or moved, | |
1680 | * memslot will be created. Validation of sp->gfn happens in: | |
1681 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1682 | * - kvm_is_visible_gfn (mmu_check_root) | |
1683 | */ | |
bcb63dcd | 1684 | kvm_arch_flush_shadow_memslot(kvm, old); |
683412cc | 1685 | kvm_arch_guest_memory_reclaimed(kvm); |
a54d8066 MS |
1686 | |
1687 | /* Was released by kvm_swap_active_memslots, reacquire. */ | |
1688 | mutex_lock(&kvm->slots_arch_lock); | |
1689 | ||
1690 | /* | |
1691 | * Copy the arch-specific field of the newly-installed slot back to the | |
1692 | * old slot as the arch data could have changed between releasing | |
1693 | * slots_arch_lock in install_new_memslots() and re-acquiring the lock | |
1694 | * above. Writers are required to retrieve memslots *after* acquiring | |
1695 | * slots_arch_lock, thus the active slot's data is guaranteed to be fresh. | |
1696 | */ | |
244893fa | 1697 | old->arch = invalid_slot->arch; |
a54d8066 MS |
1698 | } |
1699 | ||
1700 | static void kvm_create_memslot(struct kvm *kvm, | |
244893fa | 1701 | struct kvm_memory_slot *new) |
a54d8066 | 1702 | { |
244893fa SC |
1703 | /* Add the new memslot to the inactive set and activate. */ |
1704 | kvm_replace_memslot(kvm, NULL, new); | |
1705 | kvm_activate_memslot(kvm, NULL, new); | |
a54d8066 MS |
1706 | } |
1707 | ||
1708 | static void kvm_delete_memslot(struct kvm *kvm, | |
1709 | struct kvm_memory_slot *old, | |
1710 | struct kvm_memory_slot *invalid_slot) | |
1711 | { | |
1712 | /* | |
1713 | * Remove the old memslot (in the inactive memslots) by passing NULL as | |
244893fa | 1714 | * the "new" slot, and for the invalid version in the active slots. |
a54d8066 MS |
1715 | */ |
1716 | kvm_replace_memslot(kvm, old, NULL); | |
a54d8066 | 1717 | kvm_activate_memslot(kvm, invalid_slot, NULL); |
a54d8066 | 1718 | } |
36947254 | 1719 | |
244893fa SC |
1720 | static void kvm_move_memslot(struct kvm *kvm, |
1721 | struct kvm_memory_slot *old, | |
1722 | struct kvm_memory_slot *new, | |
1723 | struct kvm_memory_slot *invalid_slot) | |
a54d8066 | 1724 | { |
a54d8066 | 1725 | /* |
244893fa SC |
1726 | * Replace the old memslot in the inactive slots, and then swap slots |
1727 | * and replace the current INVALID with the new as well. | |
a54d8066 | 1728 | */ |
244893fa SC |
1729 | kvm_replace_memslot(kvm, old, new); |
1730 | kvm_activate_memslot(kvm, invalid_slot, new); | |
a54d8066 | 1731 | } |
36947254 | 1732 | |
a54d8066 MS |
1733 | static void kvm_update_flags_memslot(struct kvm *kvm, |
1734 | struct kvm_memory_slot *old, | |
244893fa | 1735 | struct kvm_memory_slot *new) |
a54d8066 MS |
1736 | { |
1737 | /* | |
1738 | * Similar to the MOVE case, but the slot doesn't need to be zapped as | |
1739 | * an intermediate step. Instead, the old memslot is simply replaced | |
1740 | * with a new, updated copy in both memslot sets. | |
1741 | */ | |
244893fa SC |
1742 | kvm_replace_memslot(kvm, old, new); |
1743 | kvm_activate_memslot(kvm, old, new); | |
36947254 SC |
1744 | } |
1745 | ||
cf47f50b | 1746 | static int kvm_set_memslot(struct kvm *kvm, |
a54d8066 | 1747 | struct kvm_memory_slot *old, |
ce5f0215 | 1748 | struct kvm_memory_slot *new, |
cf47f50b SC |
1749 | enum kvm_mr_change change) |
1750 | { | |
244893fa | 1751 | struct kvm_memory_slot *invalid_slot; |
cf47f50b SC |
1752 | int r; |
1753 | ||
b10a038e | 1754 | /* |
a54d8066 | 1755 | * Released in kvm_swap_active_memslots. |
b10a038e BG |
1756 | * |
1757 | * Must be held from before the current memslots are copied until | |
1758 | * after the new memslots are installed with rcu_assign_pointer, | |
a54d8066 | 1759 | * then released before the synchronize srcu in kvm_swap_active_memslots. |
b10a038e BG |
1760 | * |
1761 | * When modifying memslots outside of the slots_lock, must be held | |
1762 | * before reading the pointer to the current memslots until after all | |
1763 | * changes to those memslots are complete. | |
1764 | * | |
1765 | * These rules ensure that installing new memslots does not lose | |
1766 | * changes made to the previous memslots. | |
1767 | */ | |
1768 | mutex_lock(&kvm->slots_arch_lock); | |
1769 | ||
a54d8066 MS |
1770 | /* |
1771 | * Invalidate the old slot if it's being deleted or moved. This is | |
1772 | * done prior to actually deleting/moving the memslot to allow vCPUs to | |
1773 | * continue running by ensuring there are no mappings or shadow pages | |
1774 | * for the memslot when it is deleted/moved. Without pre-invalidation | |
1775 | * (and without a lock), a window would exist between effecting the | |
1776 | * delete/move and committing the changes in arch code where KVM or a | |
1777 | * guest could access a non-existent memslot. | |
244893fa SC |
1778 | * |
1779 | * Modifications are done on a temporary, unreachable slot. The old | |
1780 | * slot needs to be preserved in case a later step fails and the | |
1781 | * invalidation needs to be reverted. | |
a54d8066 | 1782 | */ |
cf47f50b | 1783 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
244893fa SC |
1784 | invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT); |
1785 | if (!invalid_slot) { | |
1786 | mutex_unlock(&kvm->slots_arch_lock); | |
1787 | return -ENOMEM; | |
1788 | } | |
1789 | kvm_invalidate_memslot(kvm, old, invalid_slot); | |
1790 | } | |
b10a038e | 1791 | |
a54d8066 MS |
1792 | r = kvm_prepare_memory_region(kvm, old, new, change); |
1793 | if (r) { | |
b10a038e | 1794 | /* |
a54d8066 MS |
1795 | * For DELETE/MOVE, revert the above INVALID change. No |
1796 | * modifications required since the original slot was preserved | |
1797 | * in the inactive slots. Changing the active memslots also | |
1798 | * release slots_arch_lock. | |
b10a038e | 1799 | */ |
244893fa SC |
1800 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
1801 | kvm_activate_memslot(kvm, invalid_slot, old); | |
1802 | kfree(invalid_slot); | |
1803 | } else { | |
a54d8066 | 1804 | mutex_unlock(&kvm->slots_arch_lock); |
244893fa | 1805 | } |
a54d8066 | 1806 | return r; |
cf47f50b SC |
1807 | } |
1808 | ||
bda44d84 | 1809 | /* |
a54d8066 MS |
1810 | * For DELETE and MOVE, the working slot is now active as the INVALID |
1811 | * version of the old slot. MOVE is particularly special as it reuses | |
1812 | * the old slot and returns a copy of the old slot (in working_slot). | |
1813 | * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the | |
1814 | * old slot is detached but otherwise preserved. | |
bda44d84 | 1815 | */ |
a54d8066 | 1816 | if (change == KVM_MR_CREATE) |
244893fa | 1817 | kvm_create_memslot(kvm, new); |
a54d8066 | 1818 | else if (change == KVM_MR_DELETE) |
244893fa | 1819 | kvm_delete_memslot(kvm, old, invalid_slot); |
a54d8066 | 1820 | else if (change == KVM_MR_MOVE) |
244893fa | 1821 | kvm_move_memslot(kvm, old, new, invalid_slot); |
a54d8066 | 1822 | else if (change == KVM_MR_FLAGS_ONLY) |
244893fa | 1823 | kvm_update_flags_memslot(kvm, old, new); |
a54d8066 MS |
1824 | else |
1825 | BUG(); | |
cf47f50b | 1826 | |
244893fa SC |
1827 | /* Free the temporary INVALID slot used for DELETE and MOVE. */ |
1828 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) | |
1829 | kfree(invalid_slot); | |
bda44d84 | 1830 | |
a54d8066 MS |
1831 | /* |
1832 | * No need to refresh new->arch, changes after dropping slots_arch_lock | |
a413a625 | 1833 | * will directly hit the final, active memslot. Architectures are |
a54d8066 MS |
1834 | * responsible for knowing that new->arch may be stale. |
1835 | */ | |
1836 | kvm_commit_memory_region(kvm, old, new, change); | |
cf47f50b | 1837 | |
cf47f50b | 1838 | return 0; |
cf47f50b SC |
1839 | } |
1840 | ||
44401a20 MS |
1841 | static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id, |
1842 | gfn_t start, gfn_t end) | |
5c0b4f3d | 1843 | { |
44401a20 | 1844 | struct kvm_memslot_iter iter; |
5c0b4f3d | 1845 | |
44401a20 MS |
1846 | kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) { |
1847 | if (iter.slot->id != id) | |
1848 | return true; | |
1849 | } | |
5c0b4f3d | 1850 | |
44401a20 | 1851 | return false; |
5c0b4f3d SC |
1852 | } |
1853 | ||
6aa8b732 AK |
1854 | /* |
1855 | * Allocate some memory and give it an address in the guest physical address | |
1856 | * space. | |
1857 | * | |
1858 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1859 | * |
02d5d55b | 1860 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1861 | */ |
f78e0e2e | 1862 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 1863 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 1864 | { |
244893fa | 1865 | struct kvm_memory_slot *old, *new; |
44401a20 | 1866 | struct kvm_memslots *slots; |
f64c0398 | 1867 | enum kvm_mr_change change; |
0f9bdef3 SC |
1868 | unsigned long npages; |
1869 | gfn_t base_gfn; | |
163da372 SC |
1870 | int as_id, id; |
1871 | int r; | |
6aa8b732 | 1872 | |
a50d64d6 XG |
1873 | r = check_memory_region_flags(mem); |
1874 | if (r) | |
71a4c30b | 1875 | return r; |
a50d64d6 | 1876 | |
f481b069 PB |
1877 | as_id = mem->slot >> 16; |
1878 | id = (u16)mem->slot; | |
1879 | ||
6aa8b732 | 1880 | /* General sanity checks */ |
6b285a55 SC |
1881 | if ((mem->memory_size & (PAGE_SIZE - 1)) || |
1882 | (mem->memory_size != (unsigned long)mem->memory_size)) | |
71a4c30b | 1883 | return -EINVAL; |
6aa8b732 | 1884 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 1885 | return -EINVAL; |
fa3d315a | 1886 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 1887 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 1888 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 1889 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 1890 | mem->memory_size)) |
71a4c30b | 1891 | return -EINVAL; |
f481b069 | 1892 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 1893 | return -EINVAL; |
6aa8b732 | 1894 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 1895 | return -EINVAL; |
0f9bdef3 SC |
1896 | if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES) |
1897 | return -EINVAL; | |
6aa8b732 | 1898 | |
44401a20 | 1899 | slots = __kvm_memslots(kvm, as_id); |
6aa8b732 | 1900 | |
5c0b4f3d | 1901 | /* |
7cd08553 SC |
1902 | * Note, the old memslot (and the pointer itself!) may be invalidated |
1903 | * and/or destroyed by kvm_set_memslot(). | |
5c0b4f3d | 1904 | */ |
44401a20 | 1905 | old = id_to_memslot(slots, id); |
163da372 | 1906 | |
47ea7d90 | 1907 | if (!mem->memory_size) { |
7cd08553 | 1908 | if (!old || !old->npages) |
47ea7d90 | 1909 | return -EINVAL; |
5c0b4f3d | 1910 | |
7cd08553 | 1911 | if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages)) |
47ea7d90 | 1912 | return -EIO; |
6aa8b732 | 1913 | |
244893fa | 1914 | return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE); |
47ea7d90 | 1915 | } |
5c0b4f3d | 1916 | |
0f9bdef3 SC |
1917 | base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT); |
1918 | npages = (mem->memory_size >> PAGE_SHIFT); | |
163da372 | 1919 | |
7cd08553 | 1920 | if (!old || !old->npages) { |
5c0b4f3d | 1921 | change = KVM_MR_CREATE; |
afa319a5 SC |
1922 | |
1923 | /* | |
1924 | * To simplify KVM internals, the total number of pages across | |
1925 | * all memslots must fit in an unsigned long. | |
1926 | */ | |
0f9bdef3 | 1927 | if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages) |
afa319a5 | 1928 | return -EINVAL; |
5c0b4f3d | 1929 | } else { /* Modify an existing slot. */ |
0f9bdef3 SC |
1930 | if ((mem->userspace_addr != old->userspace_addr) || |
1931 | (npages != old->npages) || | |
1932 | ((mem->flags ^ old->flags) & KVM_MEM_READONLY)) | |
71a4c30b | 1933 | return -EINVAL; |
09170a49 | 1934 | |
0f9bdef3 | 1935 | if (base_gfn != old->base_gfn) |
5c0b4f3d | 1936 | change = KVM_MR_MOVE; |
0f9bdef3 | 1937 | else if (mem->flags != old->flags) |
5c0b4f3d SC |
1938 | change = KVM_MR_FLAGS_ONLY; |
1939 | else /* Nothing to change. */ | |
1940 | return 0; | |
09170a49 | 1941 | } |
6aa8b732 | 1942 | |
44401a20 | 1943 | if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) && |
0f9bdef3 | 1944 | kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages)) |
44401a20 | 1945 | return -EEXIST; |
6aa8b732 | 1946 | |
244893fa SC |
1947 | /* Allocate a slot that will persist in the memslot. */ |
1948 | new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT); | |
1949 | if (!new) | |
1950 | return -ENOMEM; | |
3c9bd400 | 1951 | |
244893fa SC |
1952 | new->as_id = as_id; |
1953 | new->id = id; | |
1954 | new->base_gfn = base_gfn; | |
1955 | new->npages = npages; | |
1956 | new->flags = mem->flags; | |
1957 | new->userspace_addr = mem->userspace_addr; | |
6aa8b732 | 1958 | |
244893fa | 1959 | r = kvm_set_memslot(kvm, old, new, change); |
cf47f50b | 1960 | if (r) |
244893fa | 1961 | kfree(new); |
6aa8b732 | 1962 | return r; |
210c7c4d | 1963 | } |
f78e0e2e SY |
1964 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1965 | ||
1966 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1967 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1968 | { |
1969 | int r; | |
1970 | ||
79fac95e | 1971 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1972 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1973 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1974 | return r; |
1975 | } | |
210c7c4d IE |
1976 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1977 | ||
7940876e SH |
1978 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1979 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1980 | { |
f481b069 | 1981 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1982 | return -EINVAL; |
09170a49 | 1983 | |
47ae31e2 | 1984 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1985 | } |
1986 | ||
0dff0846 | 1987 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
1988 | /** |
1989 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
1990 | * @kvm: pointer to kvm instance | |
1991 | * @log: slot id and address to which we copy the log | |
1992 | * @is_dirty: set to '1' if any dirty pages were found | |
1993 | * @memslot: set to the associated memslot, always valid on success | |
1994 | */ | |
1995 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
1996 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 1997 | { |
9f6b8029 | 1998 | struct kvm_memslots *slots; |
843574a3 | 1999 | int i, as_id, id; |
87bf6e7d | 2000 | unsigned long n; |
6aa8b732 AK |
2001 | unsigned long any = 0; |
2002 | ||
b2cc64c4 PX |
2003 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
2004 | if (kvm->dirty_ring_size) | |
2005 | return -ENXIO; | |
2006 | ||
2a49f61d SC |
2007 | *memslot = NULL; |
2008 | *is_dirty = 0; | |
2009 | ||
f481b069 PB |
2010 | as_id = log->slot >> 16; |
2011 | id = (u16)log->slot; | |
2012 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 2013 | return -EINVAL; |
6aa8b732 | 2014 | |
f481b069 | 2015 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 2016 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 2017 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 2018 | return -ENOENT; |
6aa8b732 | 2019 | |
2a49f61d SC |
2020 | kvm_arch_sync_dirty_log(kvm, *memslot); |
2021 | ||
2022 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 2023 | |
cd1a4a98 | 2024 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 2025 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 2026 | |
2a49f61d | 2027 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 2028 | return -EFAULT; |
6aa8b732 | 2029 | |
5bb064dc ZX |
2030 | if (any) |
2031 | *is_dirty = 1; | |
843574a3 | 2032 | return 0; |
6aa8b732 | 2033 | } |
2ba9f0d8 | 2034 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 2035 | |
0dff0846 | 2036 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 2037 | /** |
b8b00220 | 2038 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 2039 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
2040 | * @kvm: pointer to kvm instance |
2041 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
2042 | * |
2043 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
2044 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
2045 | * following order: | |
2046 | * | |
2047 | * 1. Take a snapshot of the bit and clear it if needed. | |
2048 | * 2. Write protect the corresponding page. | |
2049 | * 3. Copy the snapshot to the userspace. | |
2050 | * 4. Upon return caller flushes TLB's if needed. | |
2051 | * | |
2052 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
2053 | * entry. This is not a problem because the page is reported dirty using | |
2054 | * the snapshot taken before and step 4 ensures that writes done after | |
2055 | * exiting to userspace will be logged for the next call. | |
2056 | * | |
2057 | */ | |
0dff0846 | 2058 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 2059 | { |
9f6b8029 | 2060 | struct kvm_memslots *slots; |
ba0513b5 | 2061 | struct kvm_memory_slot *memslot; |
58d6db34 | 2062 | int i, as_id, id; |
ba0513b5 MS |
2063 | unsigned long n; |
2064 | unsigned long *dirty_bitmap; | |
2065 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2066 | bool flush; |
ba0513b5 | 2067 | |
b2cc64c4 PX |
2068 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
2069 | if (kvm->dirty_ring_size) | |
2070 | return -ENXIO; | |
2071 | ||
f481b069 PB |
2072 | as_id = log->slot >> 16; |
2073 | id = (u16)log->slot; | |
2074 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 2075 | return -EINVAL; |
ba0513b5 | 2076 | |
f481b069 PB |
2077 | slots = __kvm_memslots(kvm, as_id); |
2078 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2079 | if (!memslot || !memslot->dirty_bitmap) |
2080 | return -ENOENT; | |
ba0513b5 MS |
2081 | |
2082 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 2083 | |
0dff0846 SC |
2084 | kvm_arch_sync_dirty_log(kvm, memslot); |
2085 | ||
ba0513b5 | 2086 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 2087 | flush = false; |
2a31b9db PB |
2088 | if (kvm->manual_dirty_log_protect) { |
2089 | /* | |
2090 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
2091 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
2092 | * is some code duplication between this function and | |
2093 | * kvm_get_dirty_log, but hopefully all architecture | |
2094 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
2095 | * can be eliminated. | |
2096 | */ | |
2097 | dirty_bitmap_buffer = dirty_bitmap; | |
2098 | } else { | |
2099 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
2100 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 2101 | |
531810ca | 2102 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
2103 | for (i = 0; i < n / sizeof(long); i++) { |
2104 | unsigned long mask; | |
2105 | gfn_t offset; | |
ba0513b5 | 2106 | |
2a31b9db PB |
2107 | if (!dirty_bitmap[i]) |
2108 | continue; | |
2109 | ||
0dff0846 | 2110 | flush = true; |
2a31b9db PB |
2111 | mask = xchg(&dirty_bitmap[i], 0); |
2112 | dirty_bitmap_buffer[i] = mask; | |
2113 | ||
a67794ca LT |
2114 | offset = i * BITS_PER_LONG; |
2115 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
2116 | offset, mask); | |
2a31b9db | 2117 | } |
531810ca | 2118 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
2119 | } |
2120 | ||
0dff0846 SC |
2121 | if (flush) |
2122 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2123 | ||
2a31b9db PB |
2124 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
2125 | return -EFAULT; | |
2126 | return 0; | |
2127 | } | |
0dff0846 SC |
2128 | |
2129 | ||
2130 | /** | |
2131 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
2132 | * @kvm: kvm instance | |
2133 | * @log: slot id and address to which we copy the log | |
2134 | * | |
2135 | * Steps 1-4 below provide general overview of dirty page logging. See | |
2136 | * kvm_get_dirty_log_protect() function description for additional details. | |
2137 | * | |
2138 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
2139 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
2140 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
2141 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
2142 | * writes will be marked dirty for next log read. | |
2143 | * | |
2144 | * 1. Take a snapshot of the bit and clear it if needed. | |
2145 | * 2. Write protect the corresponding page. | |
2146 | * 3. Copy the snapshot to the userspace. | |
2147 | * 4. Flush TLB's if needed. | |
2148 | */ | |
2149 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
2150 | struct kvm_dirty_log *log) | |
2151 | { | |
2152 | int r; | |
2153 | ||
2154 | mutex_lock(&kvm->slots_lock); | |
2155 | ||
2156 | r = kvm_get_dirty_log_protect(kvm, log); | |
2157 | ||
2158 | mutex_unlock(&kvm->slots_lock); | |
2159 | return r; | |
2160 | } | |
2a31b9db PB |
2161 | |
2162 | /** | |
2163 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
2164 | * and reenable dirty page tracking for the corresponding pages. | |
2165 | * @kvm: pointer to kvm instance | |
2166 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
2167 | */ | |
0dff0846 SC |
2168 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
2169 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
2170 | { |
2171 | struct kvm_memslots *slots; | |
2172 | struct kvm_memory_slot *memslot; | |
98938aa8 | 2173 | int as_id, id; |
2a31b9db | 2174 | gfn_t offset; |
98938aa8 | 2175 | unsigned long i, n; |
2a31b9db PB |
2176 | unsigned long *dirty_bitmap; |
2177 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 2178 | bool flush; |
2a31b9db | 2179 | |
b2cc64c4 PX |
2180 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
2181 | if (kvm->dirty_ring_size) | |
2182 | return -ENXIO; | |
2183 | ||
2a31b9db PB |
2184 | as_id = log->slot >> 16; |
2185 | id = (u16)log->slot; | |
2186 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
2187 | return -EINVAL; | |
2188 | ||
76d58e0f | 2189 | if (log->first_page & 63) |
2a31b9db PB |
2190 | return -EINVAL; |
2191 | ||
2192 | slots = __kvm_memslots(kvm, as_id); | |
2193 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2194 | if (!memslot || !memslot->dirty_bitmap) |
2195 | return -ENOENT; | |
2a31b9db PB |
2196 | |
2197 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 2198 | |
4ddc9204 | 2199 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
2200 | |
2201 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
2202 | log->num_pages > memslot->npages - log->first_page || |
2203 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
2204 | return -EINVAL; | |
98938aa8 | 2205 | |
0dff0846 SC |
2206 | kvm_arch_sync_dirty_log(kvm, memslot); |
2207 | ||
2208 | flush = false; | |
2a31b9db PB |
2209 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
2210 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
2211 | return -EFAULT; | |
ba0513b5 | 2212 | |
531810ca | 2213 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
2214 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
2215 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
2216 | i++, offset += BITS_PER_LONG) { |
2217 | unsigned long mask = *dirty_bitmap_buffer++; | |
2218 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
2219 | if (!mask) | |
ba0513b5 MS |
2220 | continue; |
2221 | ||
2a31b9db | 2222 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 2223 | |
2a31b9db PB |
2224 | /* |
2225 | * mask contains the bits that really have been cleared. This | |
2226 | * never includes any bits beyond the length of the memslot (if | |
2227 | * the length is not aligned to 64 pages), therefore it is not | |
2228 | * a problem if userspace sets them in log->dirty_bitmap. | |
2229 | */ | |
58d2930f | 2230 | if (mask) { |
0dff0846 | 2231 | flush = true; |
58d2930f TY |
2232 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
2233 | offset, mask); | |
2234 | } | |
ba0513b5 | 2235 | } |
531810ca | 2236 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 2237 | |
0dff0846 SC |
2238 | if (flush) |
2239 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2240 | ||
58d6db34 | 2241 | return 0; |
ba0513b5 | 2242 | } |
0dff0846 SC |
2243 | |
2244 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
2245 | struct kvm_clear_dirty_log *log) | |
2246 | { | |
2247 | int r; | |
2248 | ||
2249 | mutex_lock(&kvm->slots_lock); | |
2250 | ||
2251 | r = kvm_clear_dirty_log_protect(kvm, log); | |
2252 | ||
2253 | mutex_unlock(&kvm->slots_lock); | |
2254 | return r; | |
2255 | } | |
2256 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 2257 | |
49c7754c GN |
2258 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
2259 | { | |
2260 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
2261 | } | |
a1f4d395 | 2262 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 2263 | |
8e73485c PB |
2264 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
2265 | { | |
fe22ed82 | 2266 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
a54d8066 | 2267 | u64 gen = slots->generation; |
fe22ed82 | 2268 | struct kvm_memory_slot *slot; |
fe22ed82 | 2269 | |
a54d8066 MS |
2270 | /* |
2271 | * This also protects against using a memslot from a different address space, | |
2272 | * since different address spaces have different generation numbers. | |
2273 | */ | |
2274 | if (unlikely(gen != vcpu->last_used_slot_gen)) { | |
2275 | vcpu->last_used_slot = NULL; | |
2276 | vcpu->last_used_slot_gen = gen; | |
2277 | } | |
2278 | ||
2279 | slot = try_get_memslot(vcpu->last_used_slot, gfn); | |
fe22ed82 DM |
2280 | if (slot) |
2281 | return slot; | |
2282 | ||
2283 | /* | |
2284 | * Fall back to searching all memslots. We purposely use | |
2285 | * search_memslots() instead of __gfn_to_memslot() to avoid | |
a54d8066 | 2286 | * thrashing the VM-wide last_used_slot in kvm_memslots. |
fe22ed82 | 2287 | */ |
a54d8066 | 2288 | slot = search_memslots(slots, gfn, false); |
fe22ed82 | 2289 | if (slot) { |
a54d8066 | 2290 | vcpu->last_used_slot = slot; |
fe22ed82 DM |
2291 | return slot; |
2292 | } | |
2293 | ||
2294 | return NULL; | |
8e73485c PB |
2295 | } |
2296 | ||
33e94154 | 2297 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 2298 | { |
bf3e05bc | 2299 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 2300 | |
c36b7150 | 2301 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
2302 | } |
2303 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
2304 | ||
995decb6 VK |
2305 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
2306 | { | |
2307 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2308 | ||
2309 | return kvm_is_visible_memslot(memslot); | |
2310 | } | |
2311 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
2312 | ||
f9b84e19 | 2313 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
2314 | { |
2315 | struct vm_area_struct *vma; | |
2316 | unsigned long addr, size; | |
2317 | ||
2318 | size = PAGE_SIZE; | |
2319 | ||
42cde48b | 2320 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
2321 | if (kvm_is_error_hva(addr)) |
2322 | return PAGE_SIZE; | |
2323 | ||
d8ed45c5 | 2324 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
2325 | vma = find_vma(current->mm, addr); |
2326 | if (!vma) | |
2327 | goto out; | |
2328 | ||
2329 | size = vma_kernel_pagesize(vma); | |
2330 | ||
2331 | out: | |
d8ed45c5 | 2332 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
2333 | |
2334 | return size; | |
2335 | } | |
2336 | ||
8283e36a | 2337 | static bool memslot_is_readonly(const struct kvm_memory_slot *slot) |
4d8b81ab XG |
2338 | { |
2339 | return slot->flags & KVM_MEM_READONLY; | |
2340 | } | |
2341 | ||
8283e36a | 2342 | static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn, |
4d8b81ab | 2343 | gfn_t *nr_pages, bool write) |
539cb660 | 2344 | { |
bc6678a3 | 2345 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 2346 | return KVM_HVA_ERR_BAD; |
48987781 | 2347 | |
4d8b81ab XG |
2348 | if (memslot_is_readonly(slot) && write) |
2349 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
2350 | |
2351 | if (nr_pages) | |
2352 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
2353 | ||
4d8b81ab | 2354 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 2355 | } |
48987781 | 2356 | |
4d8b81ab XG |
2357 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2358 | gfn_t *nr_pages) | |
2359 | { | |
2360 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 2361 | } |
48987781 | 2362 | |
4d8b81ab | 2363 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 2364 | gfn_t gfn) |
4d8b81ab XG |
2365 | { |
2366 | return gfn_to_hva_many(slot, gfn, NULL); | |
2367 | } | |
2368 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
2369 | ||
48987781 XG |
2370 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
2371 | { | |
49c7754c | 2372 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 2373 | } |
0d150298 | 2374 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 2375 | |
8e73485c PB |
2376 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
2377 | { | |
2378 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
2379 | } | |
2380 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
2381 | ||
86ab8cff | 2382 | /* |
970c0d4b WY |
2383 | * Return the hva of a @gfn and the R/W attribute if possible. |
2384 | * | |
2385 | * @slot: the kvm_memory_slot which contains @gfn | |
2386 | * @gfn: the gfn to be translated | |
2387 | * @writable: used to return the read/write attribute of the @slot if the hva | |
2388 | * is valid and @writable is not NULL | |
86ab8cff | 2389 | */ |
64d83126 CD |
2390 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
2391 | gfn_t gfn, bool *writable) | |
86ab8cff | 2392 | { |
a2ac07fe GN |
2393 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
2394 | ||
2395 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
2396 | *writable = !memslot_is_readonly(slot); |
2397 | ||
a2ac07fe | 2398 | return hva; |
86ab8cff XG |
2399 | } |
2400 | ||
64d83126 CD |
2401 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
2402 | { | |
2403 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2404 | ||
2405 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2406 | } | |
2407 | ||
8e73485c PB |
2408 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
2409 | { | |
2410 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2411 | ||
2412 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2413 | } | |
2414 | ||
fafc3dba HY |
2415 | static inline int check_user_page_hwpoison(unsigned long addr) |
2416 | { | |
0d731759 | 2417 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 2418 | |
0d731759 | 2419 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
2420 | return rc == -EHWPOISON; |
2421 | } | |
2422 | ||
2fc84311 | 2423 | /* |
b9b33da2 PB |
2424 | * The fast path to get the writable pfn which will be stored in @pfn, |
2425 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 2426 | * only part that runs if we can in atomic context. |
2fc84311 | 2427 | */ |
b9b33da2 PB |
2428 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
2429 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 2430 | { |
8d4e1288 | 2431 | struct page *page[1]; |
954bbbc2 | 2432 | |
12ce13fe XG |
2433 | /* |
2434 | * Fast pin a writable pfn only if it is a write fault request | |
2435 | * or the caller allows to map a writable pfn for a read fault | |
2436 | * request. | |
2437 | */ | |
2438 | if (!(write_fault || writable)) | |
2439 | return false; | |
612819c3 | 2440 | |
dadbb612 | 2441 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 2442 | *pfn = page_to_pfn(page[0]); |
612819c3 | 2443 | |
2fc84311 XG |
2444 | if (writable) |
2445 | *writable = true; | |
2446 | return true; | |
2447 | } | |
af585b92 | 2448 | |
2fc84311 XG |
2449 | return false; |
2450 | } | |
612819c3 | 2451 | |
2fc84311 XG |
2452 | /* |
2453 | * The slow path to get the pfn of the specified host virtual address, | |
2454 | * 1 indicates success, -errno is returned if error is detected. | |
2455 | */ | |
2456 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 2457 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 2458 | { |
ce53053c AV |
2459 | unsigned int flags = FOLL_HWPOISON; |
2460 | struct page *page; | |
2fc84311 | 2461 | int npages = 0; |
612819c3 | 2462 | |
2fc84311 XG |
2463 | might_sleep(); |
2464 | ||
2465 | if (writable) | |
2466 | *writable = write_fault; | |
2467 | ||
ce53053c AV |
2468 | if (write_fault) |
2469 | flags |= FOLL_WRITE; | |
2470 | if (async) | |
2471 | flags |= FOLL_NOWAIT; | |
d4944b0e | 2472 | |
ce53053c | 2473 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
2474 | if (npages != 1) |
2475 | return npages; | |
2476 | ||
2477 | /* map read fault as writable if possible */ | |
12ce13fe | 2478 | if (unlikely(!write_fault) && writable) { |
ce53053c | 2479 | struct page *wpage; |
2fc84311 | 2480 | |
dadbb612 | 2481 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 2482 | *writable = true; |
ce53053c AV |
2483 | put_page(page); |
2484 | page = wpage; | |
612819c3 | 2485 | } |
887c08ac | 2486 | } |
ce53053c | 2487 | *pfn = page_to_pfn(page); |
2fc84311 XG |
2488 | return npages; |
2489 | } | |
539cb660 | 2490 | |
4d8b81ab XG |
2491 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
2492 | { | |
2493 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
2494 | return false; | |
2e2e3738 | 2495 | |
4d8b81ab XG |
2496 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
2497 | return false; | |
887c08ac | 2498 | |
4d8b81ab XG |
2499 | return true; |
2500 | } | |
bf998156 | 2501 | |
f8be156b NP |
2502 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
2503 | { | |
2504 | if (kvm_is_reserved_pfn(pfn)) | |
2505 | return 1; | |
2506 | return get_page_unless_zero(pfn_to_page(pfn)); | |
2507 | } | |
2508 | ||
92176a8e | 2509 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
1625566e XT |
2510 | unsigned long addr, bool write_fault, |
2511 | bool *writable, kvm_pfn_t *p_pfn) | |
92176a8e | 2512 | { |
a9545779 | 2513 | kvm_pfn_t pfn; |
bd2fae8d PB |
2514 | pte_t *ptep; |
2515 | spinlock_t *ptl; | |
add6a0cd PB |
2516 | int r; |
2517 | ||
9fd6dad1 | 2518 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2519 | if (r) { |
2520 | /* | |
2521 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
2522 | * not call the fault handler, so do it here. | |
2523 | */ | |
2524 | bool unlocked = false; | |
64019a2e | 2525 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
2526 | (write_fault ? FAULT_FLAG_WRITE : 0), |
2527 | &unlocked); | |
a8387d0b PB |
2528 | if (unlocked) |
2529 | return -EAGAIN; | |
add6a0cd PB |
2530 | if (r) |
2531 | return r; | |
2532 | ||
9fd6dad1 | 2533 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2534 | if (r) |
2535 | return r; | |
bd2fae8d | 2536 | } |
add6a0cd | 2537 | |
bd2fae8d PB |
2538 | if (write_fault && !pte_write(*ptep)) { |
2539 | pfn = KVM_PFN_ERR_RO_FAULT; | |
2540 | goto out; | |
add6a0cd PB |
2541 | } |
2542 | ||
a340b3e2 | 2543 | if (writable) |
bd2fae8d PB |
2544 | *writable = pte_write(*ptep); |
2545 | pfn = pte_pfn(*ptep); | |
add6a0cd PB |
2546 | |
2547 | /* | |
2548 | * Get a reference here because callers of *hva_to_pfn* and | |
2549 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
2550 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
36c3ce6c | 2551 | * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will |
add6a0cd PB |
2552 | * simply do nothing for reserved pfns. |
2553 | * | |
2554 | * Whoever called remap_pfn_range is also going to call e.g. | |
2555 | * unmap_mapping_range before the underlying pages are freed, | |
2556 | * causing a call to our MMU notifier. | |
f8be156b NP |
2557 | * |
2558 | * Certain IO or PFNMAP mappings can be backed with valid | |
2559 | * struct pages, but be allocated without refcounting e.g., | |
2560 | * tail pages of non-compound higher order allocations, which | |
2561 | * would then underflow the refcount when the caller does the | |
2562 | * required put_page. Don't allow those pages here. | |
add6a0cd | 2563 | */ |
f8be156b NP |
2564 | if (!kvm_try_get_pfn(pfn)) |
2565 | r = -EFAULT; | |
add6a0cd | 2566 | |
bd2fae8d PB |
2567 | out: |
2568 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 2569 | *p_pfn = pfn; |
f8be156b NP |
2570 | |
2571 | return r; | |
92176a8e PB |
2572 | } |
2573 | ||
12ce13fe XG |
2574 | /* |
2575 | * Pin guest page in memory and return its pfn. | |
2576 | * @addr: host virtual address which maps memory to the guest | |
2577 | * @atomic: whether this function can sleep | |
2578 | * @async: whether this function need to wait IO complete if the | |
2579 | * host page is not in the memory | |
2580 | * @write_fault: whether we should get a writable host page | |
2581 | * @writable: whether it allows to map a writable host page for !@write_fault | |
2582 | * | |
2583 | * The function will map a writable host page for these two cases: | |
2584 | * 1): @write_fault = true | |
2585 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
2586 | * whether the mapping is writable. | |
2587 | */ | |
982ed0de DW |
2588 | kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2589 | bool write_fault, bool *writable) | |
2fc84311 XG |
2590 | { |
2591 | struct vm_area_struct *vma; | |
ba049e93 | 2592 | kvm_pfn_t pfn = 0; |
92176a8e | 2593 | int npages, r; |
2e2e3738 | 2594 | |
2fc84311 XG |
2595 | /* we can do it either atomically or asynchronously, not both */ |
2596 | BUG_ON(atomic && async); | |
8d4e1288 | 2597 | |
b9b33da2 | 2598 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
2599 | return pfn; |
2600 | ||
2601 | if (atomic) | |
2602 | return KVM_PFN_ERR_FAULT; | |
2603 | ||
2604 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
2605 | if (npages == 1) | |
2606 | return pfn; | |
8d4e1288 | 2607 | |
d8ed45c5 | 2608 | mmap_read_lock(current->mm); |
2fc84311 XG |
2609 | if (npages == -EHWPOISON || |
2610 | (!async && check_user_page_hwpoison(addr))) { | |
2611 | pfn = KVM_PFN_ERR_HWPOISON; | |
2612 | goto exit; | |
2613 | } | |
2614 | ||
a8387d0b | 2615 | retry: |
fc98c03b | 2616 | vma = vma_lookup(current->mm, addr); |
2fc84311 XG |
2617 | |
2618 | if (vma == NULL) | |
2619 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 2620 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
1625566e | 2621 | r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn); |
a8387d0b PB |
2622 | if (r == -EAGAIN) |
2623 | goto retry; | |
92176a8e PB |
2624 | if (r < 0) |
2625 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 2626 | } else { |
4d8b81ab | 2627 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
2628 | *async = true; |
2629 | pfn = KVM_PFN_ERR_FAULT; | |
2630 | } | |
2631 | exit: | |
d8ed45c5 | 2632 | mmap_read_unlock(current->mm); |
2e2e3738 | 2633 | return pfn; |
35149e21 AL |
2634 | } |
2635 | ||
8283e36a | 2636 | kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn, |
ba049e93 | 2637 | bool atomic, bool *async, bool write_fault, |
4a42d848 | 2638 | bool *writable, hva_t *hva) |
887c08ac | 2639 | { |
4d8b81ab XG |
2640 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
2641 | ||
4a42d848 DS |
2642 | if (hva) |
2643 | *hva = addr; | |
2644 | ||
b2740d35 PB |
2645 | if (addr == KVM_HVA_ERR_RO_BAD) { |
2646 | if (writable) | |
2647 | *writable = false; | |
4d8b81ab | 2648 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 2649 | } |
4d8b81ab | 2650 | |
b2740d35 PB |
2651 | if (kvm_is_error_hva(addr)) { |
2652 | if (writable) | |
2653 | *writable = false; | |
81c52c56 | 2654 | return KVM_PFN_NOSLOT; |
b2740d35 | 2655 | } |
4d8b81ab XG |
2656 | |
2657 | /* Do not map writable pfn in the readonly memslot. */ | |
2658 | if (writable && memslot_is_readonly(slot)) { | |
2659 | *writable = false; | |
2660 | writable = NULL; | |
2661 | } | |
2662 | ||
2663 | return hva_to_pfn(addr, atomic, async, write_fault, | |
2664 | writable); | |
887c08ac | 2665 | } |
3520469d | 2666 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 2667 | |
ba049e93 | 2668 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
2669 | bool *writable) |
2670 | { | |
e37afc6e | 2671 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
4a42d848 | 2672 | write_fault, writable, NULL); |
612819c3 MT |
2673 | } |
2674 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
2675 | ||
8283e36a | 2676 | kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2677 | { |
4a42d848 | 2678 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL); |
506f0d6f | 2679 | } |
e37afc6e | 2680 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 2681 | |
8283e36a | 2682 | kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2683 | { |
4a42d848 | 2684 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL); |
506f0d6f | 2685 | } |
037d92dc | 2686 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 2687 | |
ba049e93 | 2688 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2689 | { |
2690 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2691 | } | |
2692 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
2693 | ||
ba049e93 | 2694 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
2695 | { |
2696 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
2697 | } | |
2698 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
2699 | ||
ba049e93 | 2700 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2701 | { |
2702 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2703 | } | |
2704 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
2705 | ||
d9ef13c2 PB |
2706 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2707 | struct page **pages, int nr_pages) | |
48987781 XG |
2708 | { |
2709 | unsigned long addr; | |
076b925d | 2710 | gfn_t entry = 0; |
48987781 | 2711 | |
d9ef13c2 | 2712 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
2713 | if (kvm_is_error_hva(addr)) |
2714 | return -1; | |
2715 | ||
2716 | if (entry < nr_pages) | |
2717 | return 0; | |
2718 | ||
dadbb612 | 2719 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
2720 | } |
2721 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
2722 | ||
b1624f99 SC |
2723 | /* |
2724 | * Do not use this helper unless you are absolutely certain the gfn _must_ be | |
2725 | * backed by 'struct page'. A valid example is if the backing memslot is | |
2726 | * controlled by KVM. Note, if the returned page is valid, it's refcount has | |
2727 | * been elevated by gfn_to_pfn(). | |
2728 | */ | |
2729 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) | |
a2766325 | 2730 | { |
b1624f99 SC |
2731 | kvm_pfn_t pfn; |
2732 | ||
2733 | pfn = gfn_to_pfn(kvm, gfn); | |
2734 | ||
81c52c56 | 2735 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 2736 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2737 | |
6573a691 | 2738 | if (kvm_is_reserved_pfn(pfn)) |
6cede2e6 | 2739 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 XG |
2740 | |
2741 | return pfn_to_page(pfn); | |
2742 | } | |
954bbbc2 AK |
2743 | EXPORT_SYMBOL_GPL(gfn_to_page); |
2744 | ||
357a18ad | 2745 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty) |
91724814 | 2746 | { |
91724814 BO |
2747 | if (dirty) |
2748 | kvm_release_pfn_dirty(pfn); | |
2749 | else | |
2750 | kvm_release_pfn_clean(pfn); | |
2751 | } | |
2752 | ||
357a18ad | 2753 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
e45adf66 KA |
2754 | { |
2755 | kvm_pfn_t pfn; | |
2756 | void *hva = NULL; | |
2757 | struct page *page = KVM_UNMAPPED_PAGE; | |
2758 | ||
2759 | if (!map) | |
2760 | return -EINVAL; | |
2761 | ||
357a18ad | 2762 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
e45adf66 KA |
2763 | if (is_error_noslot_pfn(pfn)) |
2764 | return -EINVAL; | |
2765 | ||
2766 | if (pfn_valid(pfn)) { | |
2767 | page = pfn_to_page(pfn); | |
357a18ad | 2768 | hva = kmap(page); |
d30b214d | 2769 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2770 | } else { |
357a18ad | 2771 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
d30b214d | 2772 | #endif |
e45adf66 KA |
2773 | } |
2774 | ||
2775 | if (!hva) | |
2776 | return -EFAULT; | |
2777 | ||
2778 | map->page = page; | |
2779 | map->hva = hva; | |
2780 | map->pfn = pfn; | |
2781 | map->gfn = gfn; | |
2782 | ||
2783 | return 0; | |
2784 | } | |
e45adf66 KA |
2785 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); |
2786 | ||
357a18ad | 2787 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) |
e45adf66 KA |
2788 | { |
2789 | if (!map) | |
2790 | return; | |
2791 | ||
2792 | if (!map->hva) | |
2793 | return; | |
2794 | ||
357a18ad DW |
2795 | if (map->page != KVM_UNMAPPED_PAGE) |
2796 | kunmap(map->page); | |
eb1f2f38 | 2797 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2798 | else |
357a18ad | 2799 | memunmap(map->hva); |
eb1f2f38 | 2800 | #endif |
e45adf66 | 2801 | |
91724814 | 2802 | if (dirty) |
357a18ad | 2803 | kvm_vcpu_mark_page_dirty(vcpu, map->gfn); |
91724814 | 2804 | |
357a18ad | 2805 | kvm_release_pfn(map->pfn, dirty); |
e45adf66 KA |
2806 | |
2807 | map->hva = NULL; | |
2808 | map->page = NULL; | |
2809 | } | |
2810 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); | |
2811 | ||
8e1c6914 SC |
2812 | static bool kvm_is_ad_tracked_page(struct page *page) |
2813 | { | |
2814 | /* | |
2815 | * Per page-flags.h, pages tagged PG_reserved "should in general not be | |
2816 | * touched (e.g. set dirty) except by its owner". | |
2817 | */ | |
2818 | return !PageReserved(page); | |
2819 | } | |
2820 | ||
2821 | static void kvm_set_page_dirty(struct page *page) | |
2822 | { | |
2823 | if (kvm_is_ad_tracked_page(page)) | |
2824 | SetPageDirty(page); | |
2825 | } | |
2826 | ||
2827 | static void kvm_set_page_accessed(struct page *page) | |
2828 | { | |
2829 | if (kvm_is_ad_tracked_page(page)) | |
2830 | mark_page_accessed(page); | |
2831 | } | |
2832 | ||
b4231d61 IE |
2833 | void kvm_release_page_clean(struct page *page) |
2834 | { | |
32cad84f XG |
2835 | WARN_ON(is_error_page(page)); |
2836 | ||
8e1c6914 SC |
2837 | kvm_set_page_accessed(page); |
2838 | put_page(page); | |
b4231d61 IE |
2839 | } |
2840 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
2841 | ||
ba049e93 | 2842 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 2843 | { |
bf4bea8e | 2844 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
8e1c6914 | 2845 | kvm_release_page_clean(pfn_to_page(pfn)); |
35149e21 AL |
2846 | } |
2847 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
2848 | ||
b4231d61 | 2849 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 2850 | { |
a2766325 XG |
2851 | WARN_ON(is_error_page(page)); |
2852 | ||
8e1c6914 SC |
2853 | kvm_set_page_dirty(page); |
2854 | kvm_release_page_clean(page); | |
35149e21 AL |
2855 | } |
2856 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
2857 | ||
f7a6509f | 2858 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2859 | { |
8e1c6914 SC |
2860 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2861 | kvm_release_page_dirty(pfn_to_page(pfn)); | |
35149e21 | 2862 | } |
f7a6509f | 2863 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 2864 | |
8e1c6914 SC |
2865 | /* |
2866 | * Note, checking for an error/noslot pfn is the caller's responsibility when | |
2867 | * directly marking a page dirty/accessed. Unlike the "release" helpers, the | |
2868 | * "set" helpers are not to be used when the pfn might point at garbage. | |
2869 | */ | |
ba049e93 | 2870 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2871 | { |
8e1c6914 SC |
2872 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
2873 | return; | |
2874 | ||
2875 | if (pfn_valid(pfn)) | |
2876 | kvm_set_page_dirty(pfn_to_page(pfn)); | |
8a7ae055 | 2877 | } |
35149e21 AL |
2878 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
2879 | ||
ba049e93 | 2880 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 2881 | { |
8e1c6914 SC |
2882 | if (WARN_ON(is_error_noslot_pfn(pfn))) |
2883 | return; | |
2884 | ||
2885 | if (pfn_valid(pfn)) | |
2886 | kvm_set_page_accessed(pfn_to_page(pfn)); | |
35149e21 AL |
2887 | } |
2888 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
2889 | ||
195aefde IE |
2890 | static int next_segment(unsigned long len, int offset) |
2891 | { | |
2892 | if (len > PAGE_SIZE - offset) | |
2893 | return PAGE_SIZE - offset; | |
2894 | else | |
2895 | return len; | |
2896 | } | |
2897 | ||
8e73485c PB |
2898 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
2899 | void *data, int offset, int len) | |
195aefde | 2900 | { |
e0506bcb IE |
2901 | int r; |
2902 | unsigned long addr; | |
195aefde | 2903 | |
8e73485c | 2904 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
2905 | if (kvm_is_error_hva(addr)) |
2906 | return -EFAULT; | |
3180a7fc | 2907 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 2908 | if (r) |
195aefde | 2909 | return -EFAULT; |
195aefde IE |
2910 | return 0; |
2911 | } | |
8e73485c PB |
2912 | |
2913 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
2914 | int len) | |
2915 | { | |
2916 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2917 | ||
2918 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2919 | } | |
195aefde IE |
2920 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
2921 | ||
8e73485c PB |
2922 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
2923 | int offset, int len) | |
2924 | { | |
2925 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2926 | ||
2927 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2928 | } | |
2929 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
2930 | ||
195aefde IE |
2931 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
2932 | { | |
2933 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2934 | int seg; | |
2935 | int offset = offset_in_page(gpa); | |
2936 | int ret; | |
2937 | ||
2938 | while ((seg = next_segment(len, offset)) != 0) { | |
2939 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
2940 | if (ret < 0) | |
2941 | return ret; | |
2942 | offset = 0; | |
2943 | len -= seg; | |
2944 | data += seg; | |
2945 | ++gfn; | |
2946 | } | |
2947 | return 0; | |
2948 | } | |
2949 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
2950 | ||
8e73485c | 2951 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 2952 | { |
7ec54588 | 2953 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 2954 | int seg; |
7ec54588 | 2955 | int offset = offset_in_page(gpa); |
8e73485c PB |
2956 | int ret; |
2957 | ||
2958 | while ((seg = next_segment(len, offset)) != 0) { | |
2959 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
2960 | if (ret < 0) | |
2961 | return ret; | |
2962 | offset = 0; | |
2963 | len -= seg; | |
2964 | data += seg; | |
2965 | ++gfn; | |
2966 | } | |
2967 | return 0; | |
2968 | } | |
2969 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 2970 | |
8e73485c PB |
2971 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2972 | void *data, int offset, unsigned long len) | |
2973 | { | |
2974 | int r; | |
2975 | unsigned long addr; | |
2976 | ||
2977 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
2978 | if (kvm_is_error_hva(addr)) |
2979 | return -EFAULT; | |
0aac03f0 | 2980 | pagefault_disable(); |
3180a7fc | 2981 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 2982 | pagefault_enable(); |
7ec54588 MT |
2983 | if (r) |
2984 | return -EFAULT; | |
2985 | return 0; | |
2986 | } | |
7ec54588 | 2987 | |
8e73485c PB |
2988 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
2989 | void *data, unsigned long len) | |
2990 | { | |
2991 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2992 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2993 | int offset = offset_in_page(gpa); | |
2994 | ||
2995 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2996 | } | |
2997 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
2998 | ||
28bd726a PX |
2999 | static int __kvm_write_guest_page(struct kvm *kvm, |
3000 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 3001 | const void *data, int offset, int len) |
195aefde | 3002 | { |
e0506bcb IE |
3003 | int r; |
3004 | unsigned long addr; | |
195aefde | 3005 | |
251eb841 | 3006 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
3007 | if (kvm_is_error_hva(addr)) |
3008 | return -EFAULT; | |
8b0cedff | 3009 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 3010 | if (r) |
195aefde | 3011 | return -EFAULT; |
28bd726a | 3012 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
3013 | return 0; |
3014 | } | |
8e73485c PB |
3015 | |
3016 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
3017 | const void *data, int offset, int len) | |
3018 | { | |
3019 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
3020 | ||
28bd726a | 3021 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 3022 | } |
195aefde IE |
3023 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
3024 | ||
8e73485c PB |
3025 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3026 | const void *data, int offset, int len) | |
3027 | { | |
3028 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3029 | ||
28bd726a | 3030 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
3031 | } |
3032 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
3033 | ||
195aefde IE |
3034 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
3035 | unsigned long len) | |
3036 | { | |
3037 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3038 | int seg; | |
3039 | int offset = offset_in_page(gpa); | |
3040 | int ret; | |
3041 | ||
3042 | while ((seg = next_segment(len, offset)) != 0) { | |
3043 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
3044 | if (ret < 0) | |
3045 | return ret; | |
3046 | offset = 0; | |
3047 | len -= seg; | |
3048 | data += seg; | |
3049 | ++gfn; | |
3050 | } | |
3051 | return 0; | |
3052 | } | |
ff651cb6 | 3053 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 3054 | |
8e73485c PB |
3055 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
3056 | unsigned long len) | |
3057 | { | |
3058 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3059 | int seg; | |
3060 | int offset = offset_in_page(gpa); | |
3061 | int ret; | |
3062 | ||
3063 | while ((seg = next_segment(len, offset)) != 0) { | |
3064 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
3065 | if (ret < 0) | |
3066 | return ret; | |
3067 | offset = 0; | |
3068 | len -= seg; | |
3069 | data += seg; | |
3070 | ++gfn; | |
3071 | } | |
3072 | return 0; | |
3073 | } | |
3074 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
3075 | ||
5a2d4365 PB |
3076 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
3077 | struct gfn_to_hva_cache *ghc, | |
3078 | gpa_t gpa, unsigned long len) | |
49c7754c | 3079 | { |
49c7754c | 3080 | int offset = offset_in_page(gpa); |
8f964525 AH |
3081 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
3082 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
3083 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
3084 | gfn_t nr_pages_avail; | |
49c7754c | 3085 | |
6ad1e29f | 3086 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 3087 | ghc->generation = slots->generation; |
6ad1e29f SC |
3088 | |
3089 | if (start_gfn > end_gfn) { | |
3090 | ghc->hva = KVM_HVA_ERR_BAD; | |
3091 | return -EINVAL; | |
3092 | } | |
f1b9dd5e JM |
3093 | |
3094 | /* | |
3095 | * If the requested region crosses two memslots, we still | |
3096 | * verify that the entire region is valid here. | |
3097 | */ | |
6ad1e29f | 3098 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
3099 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
3100 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
3101 | &nr_pages_avail); | |
3102 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 3103 | return -EFAULT; |
f1b9dd5e JM |
3104 | } |
3105 | ||
3106 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 3107 | if (nr_pages_needed == 1) |
49c7754c | 3108 | ghc->hva += offset; |
f1b9dd5e | 3109 | else |
8f964525 | 3110 | ghc->memslot = NULL; |
f1b9dd5e | 3111 | |
6ad1e29f SC |
3112 | ghc->gpa = gpa; |
3113 | ghc->len = len; | |
3114 | return 0; | |
49c7754c | 3115 | } |
5a2d4365 | 3116 | |
4e335d9e | 3117 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
3118 | gpa_t gpa, unsigned long len) |
3119 | { | |
4e335d9e | 3120 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
3121 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
3122 | } | |
4e335d9e | 3123 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 3124 | |
4e335d9e | 3125 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
3126 | void *data, unsigned int offset, |
3127 | unsigned long len) | |
49c7754c | 3128 | { |
4e335d9e | 3129 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 3130 | int r; |
4ec6e863 | 3131 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 3132 | |
5f25e71e PB |
3133 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3134 | return -EINVAL; | |
8f964525 | 3135 | |
dc9ce71e SC |
3136 | if (slots->generation != ghc->generation) { |
3137 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3138 | return -EFAULT; | |
3139 | } | |
8f964525 | 3140 | |
49c7754c GN |
3141 | if (kvm_is_error_hva(ghc->hva)) |
3142 | return -EFAULT; | |
3143 | ||
fcfbc617 SC |
3144 | if (unlikely(!ghc->memslot)) |
3145 | return kvm_write_guest(kvm, gpa, data, len); | |
3146 | ||
4ec6e863 | 3147 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
3148 | if (r) |
3149 | return -EFAULT; | |
28bd726a | 3150 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
3151 | |
3152 | return 0; | |
3153 | } | |
4e335d9e | 3154 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 3155 | |
4e335d9e PB |
3156 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3157 | void *data, unsigned long len) | |
4ec6e863 | 3158 | { |
4e335d9e | 3159 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 3160 | } |
4e335d9e | 3161 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 3162 | |
0958f0ce VK |
3163 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3164 | void *data, unsigned int offset, | |
3165 | unsigned long len) | |
e03b644f | 3166 | { |
4e335d9e | 3167 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 3168 | int r; |
0958f0ce | 3169 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 3170 | |
5f25e71e PB |
3171 | if (WARN_ON_ONCE(len + offset > ghc->len)) |
3172 | return -EINVAL; | |
8f964525 | 3173 | |
dc9ce71e SC |
3174 | if (slots->generation != ghc->generation) { |
3175 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3176 | return -EFAULT; | |
3177 | } | |
8f964525 | 3178 | |
e03b644f GN |
3179 | if (kvm_is_error_hva(ghc->hva)) |
3180 | return -EFAULT; | |
3181 | ||
fcfbc617 | 3182 | if (unlikely(!ghc->memslot)) |
0958f0ce | 3183 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 3184 | |
0958f0ce | 3185 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
3186 | if (r) |
3187 | return -EFAULT; | |
3188 | ||
3189 | return 0; | |
3190 | } | |
0958f0ce VK |
3191 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
3192 | ||
3193 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
3194 | void *data, unsigned long len) | |
3195 | { | |
3196 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
3197 | } | |
4e335d9e | 3198 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 3199 | |
195aefde IE |
3200 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
3201 | { | |
2f541442 | 3202 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
3203 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3204 | int seg; | |
3205 | int offset = offset_in_page(gpa); | |
3206 | int ret; | |
3207 | ||
bfda0e84 | 3208 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 3209 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
3210 | if (ret < 0) |
3211 | return ret; | |
3212 | offset = 0; | |
3213 | len -= seg; | |
3214 | ++gfn; | |
3215 | } | |
3216 | return 0; | |
3217 | } | |
3218 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
3219 | ||
28bd726a | 3220 | void mark_page_dirty_in_slot(struct kvm *kvm, |
8283e36a | 3221 | const struct kvm_memory_slot *memslot, |
28bd726a | 3222 | gfn_t gfn) |
6aa8b732 | 3223 | { |
2efd61a6 DW |
3224 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); |
3225 | ||
e09fccb5 | 3226 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
2efd61a6 DW |
3227 | if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm)) |
3228 | return; | |
e09fccb5 | 3229 | #endif |
2efd61a6 | 3230 | |
044c59c4 | 3231 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 3232 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 3233 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 3234 | |
fb04a1ed | 3235 | if (kvm->dirty_ring_size) |
2efd61a6 | 3236 | kvm_dirty_ring_push(&vcpu->dirty_ring, |
fb04a1ed PX |
3237 | slot, rel_gfn); |
3238 | else | |
3239 | set_bit_le(rel_gfn, memslot->dirty_bitmap); | |
6aa8b732 AK |
3240 | } |
3241 | } | |
a6a0b05d | 3242 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 3243 | |
49c7754c GN |
3244 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
3245 | { | |
3246 | struct kvm_memory_slot *memslot; | |
3247 | ||
3248 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 3249 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 3250 | } |
2ba9f0d8 | 3251 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 3252 | |
8e73485c PB |
3253 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
3254 | { | |
3255 | struct kvm_memory_slot *memslot; | |
3256 | ||
3257 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 3258 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
3259 | } |
3260 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
3261 | ||
20b7035c JS |
3262 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
3263 | { | |
3264 | if (!vcpu->sigset_active) | |
3265 | return; | |
3266 | ||
3267 | /* | |
3268 | * This does a lockless modification of ->real_blocked, which is fine | |
3269 | * because, only current can change ->real_blocked and all readers of | |
3270 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
3271 | * of ->blocked. | |
3272 | */ | |
3273 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
3274 | } | |
3275 | ||
3276 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
3277 | { | |
3278 | if (!vcpu->sigset_active) | |
3279 | return; | |
3280 | ||
3281 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
3282 | sigemptyset(¤t->real_blocked); | |
3283 | } | |
3284 | ||
aca6ff29 WL |
3285 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
3286 | { | |
dee339b5 | 3287 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 3288 | |
2cbd7824 | 3289 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 3290 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3291 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
3292 | if (!grow) |
3293 | goto out; | |
3294 | ||
dee339b5 NW |
3295 | val *= grow; |
3296 | if (val < grow_start) | |
3297 | val = grow_start; | |
aca6ff29 | 3298 | |
258785ef DM |
3299 | if (val > vcpu->kvm->max_halt_poll_ns) |
3300 | val = vcpu->kvm->max_halt_poll_ns; | |
313f636d | 3301 | |
aca6ff29 | 3302 | vcpu->halt_poll_ns = val; |
7fa08e71 | 3303 | out: |
2cbd7824 | 3304 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3305 | } |
3306 | ||
3307 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
3308 | { | |
ae232ea4 | 3309 | unsigned int old, val, shrink, grow_start; |
aca6ff29 | 3310 | |
2cbd7824 | 3311 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 3312 | shrink = READ_ONCE(halt_poll_ns_shrink); |
ae232ea4 | 3313 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3314 | if (shrink == 0) |
aca6ff29 WL |
3315 | val = 0; |
3316 | else | |
6b6de68c | 3317 | val /= shrink; |
aca6ff29 | 3318 | |
ae232ea4 SS |
3319 | if (val < grow_start) |
3320 | val = 0; | |
3321 | ||
aca6ff29 | 3322 | vcpu->halt_poll_ns = val; |
2cbd7824 | 3323 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3324 | } |
3325 | ||
f7819512 PB |
3326 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
3327 | { | |
50c28f21 JS |
3328 | int ret = -EINTR; |
3329 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
3330 | ||
f7819512 PB |
3331 | if (kvm_arch_vcpu_runnable(vcpu)) { |
3332 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 3333 | goto out; |
f7819512 PB |
3334 | } |
3335 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 3336 | goto out; |
f7819512 | 3337 | if (signal_pending(current)) |
50c28f21 | 3338 | goto out; |
084071d5 MT |
3339 | if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu)) |
3340 | goto out; | |
f7819512 | 3341 | |
50c28f21 JS |
3342 | ret = 0; |
3343 | out: | |
3344 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
3345 | return ret; | |
f7819512 PB |
3346 | } |
3347 | ||
fac42688 SC |
3348 | /* |
3349 | * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is | |
3350 | * pending. This is mostly used when halting a vCPU, but may also be used | |
3351 | * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI. | |
3352 | */ | |
3353 | bool kvm_vcpu_block(struct kvm_vcpu *vcpu) | |
cb953129 | 3354 | { |
fac42688 SC |
3355 | struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); |
3356 | bool waited = false; | |
3357 | ||
c3858335 JZ |
3358 | vcpu->stat.generic.blocking = 1; |
3359 | ||
18869f26 | 3360 | preempt_disable(); |
fac42688 | 3361 | kvm_arch_vcpu_blocking(vcpu); |
fac42688 | 3362 | prepare_to_rcuwait(wait); |
18869f26 ML |
3363 | preempt_enable(); |
3364 | ||
fac42688 SC |
3365 | for (;;) { |
3366 | set_current_state(TASK_INTERRUPTIBLE); | |
3367 | ||
3368 | if (kvm_vcpu_check_block(vcpu) < 0) | |
3369 | break; | |
3370 | ||
3371 | waited = true; | |
3372 | schedule(); | |
3373 | } | |
fac42688 | 3374 | |
18869f26 ML |
3375 | preempt_disable(); |
3376 | finish_rcuwait(wait); | |
fac42688 | 3377 | kvm_arch_vcpu_unblocking(vcpu); |
18869f26 | 3378 | preempt_enable(); |
fac42688 | 3379 | |
c3858335 JZ |
3380 | vcpu->stat.generic.blocking = 0; |
3381 | ||
fac42688 SC |
3382 | return waited; |
3383 | } | |
3384 | ||
29e72893 SC |
3385 | static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start, |
3386 | ktime_t end, bool success) | |
cb953129 | 3387 | { |
30c94347 | 3388 | struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic; |
29e72893 SC |
3389 | u64 poll_ns = ktime_to_ns(ktime_sub(end, start)); |
3390 | ||
30c94347 SC |
3391 | ++vcpu->stat.generic.halt_attempted_poll; |
3392 | ||
3393 | if (success) { | |
3394 | ++vcpu->stat.generic.halt_successful_poll; | |
3395 | ||
3396 | if (!vcpu_valid_wakeup(vcpu)) | |
3397 | ++vcpu->stat.generic.halt_poll_invalid; | |
3398 | ||
3399 | stats->halt_poll_success_ns += poll_ns; | |
3400 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns); | |
3401 | } else { | |
3402 | stats->halt_poll_fail_ns += poll_ns; | |
3403 | KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns); | |
3404 | } | |
cb953129 DM |
3405 | } |
3406 | ||
b6958ce4 | 3407 | /* |
fac42688 SC |
3408 | * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt |
3409 | * polling is enabled, busy wait for a short time before blocking to avoid the | |
3410 | * expensive block+unblock sequence if a wake event arrives soon after the vCPU | |
3411 | * is halted. | |
b6958ce4 | 3412 | */ |
91b99ea7 | 3413 | void kvm_vcpu_halt(struct kvm_vcpu *vcpu) |
d3bef15f | 3414 | { |
6f390916 | 3415 | bool halt_poll_allowed = !kvm_arch_no_poll(vcpu); |
8df6a61c | 3416 | bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns; |
cb953129 | 3417 | ktime_t start, cur, poll_end; |
f7819512 | 3418 | bool waited = false; |
91b99ea7 | 3419 | u64 halt_ns; |
07ab0f8d | 3420 | |
cb953129 | 3421 | start = cur = poll_end = ktime_get(); |
8df6a61c | 3422 | if (do_halt_poll) { |
109a9826 | 3423 | ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns); |
f95ef0cd | 3424 | |
f7819512 PB |
3425 | do { |
3426 | /* | |
3427 | * This sets KVM_REQ_UNHALT if an interrupt | |
3428 | * arrives. | |
3429 | */ | |
30c94347 | 3430 | if (kvm_vcpu_check_block(vcpu) < 0) |
f7819512 | 3431 | goto out; |
74775654 | 3432 | cpu_relax(); |
cb953129 | 3433 | poll_end = cur = ktime_get(); |
6bd5b743 | 3434 | } while (kvm_vcpu_can_poll(cur, stop)); |
f7819512 | 3435 | } |
e5c239cf | 3436 | |
fac42688 | 3437 | waited = kvm_vcpu_block(vcpu); |
8ccba534 | 3438 | |
f7819512 | 3439 | cur = ktime_get(); |
87bcc5fa JZ |
3440 | if (waited) { |
3441 | vcpu->stat.generic.halt_wait_ns += | |
3442 | ktime_to_ns(cur) - ktime_to_ns(poll_end); | |
8ccba534 JZ |
3443 | KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist, |
3444 | ktime_to_ns(cur) - ktime_to_ns(poll_end)); | |
87bcc5fa | 3445 | } |
f7819512 | 3446 | out: |
91b99ea7 SC |
3447 | /* The total time the vCPU was "halted", including polling time. */ |
3448 | halt_ns = ktime_to_ns(cur) - ktime_to_ns(start); | |
aca6ff29 | 3449 | |
29e72893 SC |
3450 | /* |
3451 | * Note, halt-polling is considered successful so long as the vCPU was | |
3452 | * never actually scheduled out, i.e. even if the wake event arrived | |
3453 | * after of the halt-polling loop itself, but before the full wait. | |
3454 | */ | |
8df6a61c | 3455 | if (do_halt_poll) |
29e72893 | 3456 | update_halt_poll_stats(vcpu, start, poll_end, !waited); |
cb953129 | 3457 | |
6f390916 | 3458 | if (halt_poll_allowed) { |
44551b2f | 3459 | if (!vcpu_valid_wakeup(vcpu)) { |
aca6ff29 | 3460 | shrink_halt_poll_ns(vcpu); |
acd05785 | 3461 | } else if (vcpu->kvm->max_halt_poll_ns) { |
91b99ea7 | 3462 | if (halt_ns <= vcpu->halt_poll_ns) |
44551b2f WL |
3463 | ; |
3464 | /* we had a long block, shrink polling */ | |
acd05785 | 3465 | else if (vcpu->halt_poll_ns && |
91b99ea7 | 3466 | halt_ns > vcpu->kvm->max_halt_poll_ns) |
44551b2f WL |
3467 | shrink_halt_poll_ns(vcpu); |
3468 | /* we had a short halt and our poll time is too small */ | |
acd05785 | 3469 | else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && |
91b99ea7 | 3470 | halt_ns < vcpu->kvm->max_halt_poll_ns) |
44551b2f WL |
3471 | grow_halt_poll_ns(vcpu); |
3472 | } else { | |
3473 | vcpu->halt_poll_ns = 0; | |
3474 | } | |
3475 | } | |
aca6ff29 | 3476 | |
91b99ea7 | 3477 | trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu)); |
b6958ce4 | 3478 | } |
91b99ea7 | 3479 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
b6958ce4 | 3480 | |
178f02ff | 3481 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 3482 | { |
d92a5d1c | 3483 | if (__kvm_vcpu_wake_up(vcpu)) { |
d73eb57b | 3484 | WRITE_ONCE(vcpu->ready, true); |
0193cc90 | 3485 | ++vcpu->stat.generic.halt_wakeup; |
178f02ff | 3486 | return true; |
b6d33834 CD |
3487 | } |
3488 | ||
178f02ff | 3489 | return false; |
dd1a4cc1 RK |
3490 | } |
3491 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
3492 | ||
0266c894 | 3493 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
3494 | /* |
3495 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
3496 | */ | |
3497 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
3498 | { | |
85b64045 | 3499 | int me, cpu; |
dd1a4cc1 | 3500 | |
178f02ff RK |
3501 | if (kvm_vcpu_wake_up(vcpu)) |
3502 | return; | |
3503 | ||
aefdc2ed PB |
3504 | me = get_cpu(); |
3505 | /* | |
3506 | * The only state change done outside the vcpu mutex is IN_GUEST_MODE | |
3507 | * to EXITING_GUEST_MODE. Therefore the moderately expensive "should | |
3508 | * kick" check does not need atomic operations if kvm_vcpu_kick is used | |
3509 | * within the vCPU thread itself. | |
3510 | */ | |
3511 | if (vcpu == __this_cpu_read(kvm_running_vcpu)) { | |
3512 | if (vcpu->mode == IN_GUEST_MODE) | |
3513 | WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE); | |
3514 | goto out; | |
3515 | } | |
3516 | ||
85b64045 SC |
3517 | /* |
3518 | * Note, the vCPU could get migrated to a different pCPU at any point | |
3519 | * after kvm_arch_vcpu_should_kick(), which could result in sending an | |
3520 | * IPI to the previous pCPU. But, that's ok because the purpose of the | |
3521 | * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the | |
3522 | * vCPU also requires it to leave IN_GUEST_MODE. | |
3523 | */ | |
85b64045 SC |
3524 | if (kvm_arch_vcpu_should_kick(vcpu)) { |
3525 | cpu = READ_ONCE(vcpu->cpu); | |
3526 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
b6d33834 | 3527 | smp_send_reschedule(cpu); |
85b64045 | 3528 | } |
aefdc2ed | 3529 | out: |
b6d33834 CD |
3530 | put_cpu(); |
3531 | } | |
a20ed54d | 3532 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 3533 | #endif /* !CONFIG_S390 */ |
b6d33834 | 3534 | |
fa93384f | 3535 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
3536 | { |
3537 | struct pid *pid; | |
3538 | struct task_struct *task = NULL; | |
fa93384f | 3539 | int ret = 0; |
41628d33 KW |
3540 | |
3541 | rcu_read_lock(); | |
3542 | pid = rcu_dereference(target->pid); | |
3543 | if (pid) | |
27fbe64b | 3544 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
3545 | rcu_read_unlock(); |
3546 | if (!task) | |
c45c528e | 3547 | return ret; |
c45c528e | 3548 | ret = yield_to(task, 1); |
41628d33 | 3549 | put_task_struct(task); |
c45c528e R |
3550 | |
3551 | return ret; | |
41628d33 KW |
3552 | } |
3553 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
3554 | ||
06e48c51 R |
3555 | /* |
3556 | * Helper that checks whether a VCPU is eligible for directed yield. | |
3557 | * Most eligible candidate to yield is decided by following heuristics: | |
3558 | * | |
3559 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
3560 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 3561 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
3562 | * |
3563 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
3564 | * chance last time (mostly it has become eligible now since we have probably | |
3565 | * yielded to lockholder in last iteration. This is done by toggling | |
3566 | * @dy_eligible each time a VCPU checked for eligibility.) | |
3567 | * | |
3568 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
3569 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
3570 | * burning. Giving priority for a potential lock-holder increases lock | |
3571 | * progress. | |
3572 | * | |
3573 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
3574 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
3575 | * and continue with next VCPU and so on. | |
3576 | */ | |
7940876e | 3577 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 3578 | { |
4a55dd72 | 3579 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
3580 | bool eligible; |
3581 | ||
3582 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 3583 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
3584 | |
3585 | if (vcpu->spin_loop.in_spin_loop) | |
3586 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
3587 | ||
3588 | return eligible; | |
4a55dd72 SW |
3589 | #else |
3590 | return true; | |
06e48c51 | 3591 | #endif |
4a55dd72 | 3592 | } |
c45c528e | 3593 | |
17e433b5 WL |
3594 | /* |
3595 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
3596 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
3597 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
3598 | */ | |
3599 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
3600 | { | |
3601 | return kvm_arch_vcpu_runnable(vcpu); | |
3602 | } | |
3603 | ||
3604 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
3605 | { | |
3606 | if (kvm_arch_dy_runnable(vcpu)) | |
3607 | return true; | |
3608 | ||
3609 | #ifdef CONFIG_KVM_ASYNC_PF | |
3610 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
3611 | return true; | |
3612 | #endif | |
3613 | ||
3614 | return false; | |
3615 | } | |
3616 | ||
52acd22f WL |
3617 | bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
3618 | { | |
3619 | return false; | |
3620 | } | |
3621 | ||
199b5763 | 3622 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 3623 | { |
217ece61 RR |
3624 | struct kvm *kvm = me->kvm; |
3625 | struct kvm_vcpu *vcpu; | |
3626 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
46808a4c | 3627 | unsigned long i; |
217ece61 | 3628 | int yielded = 0; |
c45c528e | 3629 | int try = 3; |
217ece61 | 3630 | int pass; |
d255f4f2 | 3631 | |
4c088493 | 3632 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
3633 | /* |
3634 | * We boost the priority of a VCPU that is runnable but not | |
3635 | * currently running, because it got preempted by something | |
3636 | * else and called schedule in __vcpu_run. Hopefully that | |
3637 | * VCPU is holding the lock that we need and will release it. | |
3638 | * We approximate round-robin by starting at the last boosted VCPU. | |
3639 | */ | |
c45c528e | 3640 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 3641 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 3642 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
3643 | i = last_boosted_vcpu; |
3644 | continue; | |
3645 | } else if (pass && i > last_boosted_vcpu) | |
3646 | break; | |
d73eb57b | 3647 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 3648 | continue; |
217ece61 RR |
3649 | if (vcpu == me) |
3650 | continue; | |
d92a5d1c | 3651 | if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu)) |
217ece61 | 3652 | continue; |
046ddeed | 3653 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
52acd22f WL |
3654 | !kvm_arch_dy_has_pending_interrupt(vcpu) && |
3655 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 3656 | continue; |
06e48c51 R |
3657 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
3658 | continue; | |
c45c528e R |
3659 | |
3660 | yielded = kvm_vcpu_yield_to(vcpu); | |
3661 | if (yielded > 0) { | |
217ece61 | 3662 | kvm->last_boosted_vcpu = i; |
217ece61 | 3663 | break; |
c45c528e R |
3664 | } else if (yielded < 0) { |
3665 | try--; | |
3666 | if (!try) | |
3667 | break; | |
217ece61 | 3668 | } |
217ece61 RR |
3669 | } |
3670 | } | |
4c088493 | 3671 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
3672 | |
3673 | /* Ensure vcpu is not eligible during next spinloop */ | |
3674 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
3675 | } |
3676 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
3677 | ||
fb04a1ed PX |
3678 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
3679 | { | |
dc70ec21 | 3680 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
fb04a1ed PX |
3681 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && |
3682 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
3683 | kvm->dirty_ring_size / PAGE_SIZE); | |
3684 | #else | |
3685 | return false; | |
3686 | #endif | |
3687 | } | |
3688 | ||
1499fa80 | 3689 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 3690 | { |
11bac800 | 3691 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
3692 | struct page *page; |
3693 | ||
e4a533a4 | 3694 | if (vmf->pgoff == 0) |
039576c0 | 3695 | page = virt_to_page(vcpu->run); |
09566765 | 3696 | #ifdef CONFIG_X86 |
e4a533a4 | 3697 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 3698 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 3699 | #endif |
4b4357e0 | 3700 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3701 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
3702 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 3703 | #endif |
fb04a1ed PX |
3704 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
3705 | page = kvm_dirty_ring_get_page( | |
3706 | &vcpu->dirty_ring, | |
3707 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 3708 | else |
5b1c1493 | 3709 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 3710 | get_page(page); |
e4a533a4 | 3711 | vmf->page = page; |
3712 | return 0; | |
9a2bb7f4 AK |
3713 | } |
3714 | ||
f0f37e2f | 3715 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 3716 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
3717 | }; |
3718 | ||
3719 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
3720 | { | |
fb04a1ed | 3721 | struct kvm_vcpu *vcpu = file->private_data; |
11476d27 | 3722 | unsigned long pages = vma_pages(vma); |
fb04a1ed PX |
3723 | |
3724 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
3725 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
3726 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
3727 | return -EINVAL; | |
3728 | ||
9a2bb7f4 AK |
3729 | vma->vm_ops = &kvm_vcpu_vm_ops; |
3730 | return 0; | |
3731 | } | |
3732 | ||
bccf2150 AK |
3733 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
3734 | { | |
3735 | struct kvm_vcpu *vcpu = filp->private_data; | |
3736 | ||
66c0b394 | 3737 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
3738 | return 0; |
3739 | } | |
3740 | ||
70375c2d | 3741 | static const struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
3742 | .release = kvm_vcpu_release, |
3743 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 3744 | .mmap = kvm_vcpu_mmap, |
6038f373 | 3745 | .llseek = noop_llseek, |
7ddfd3e0 | 3746 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
3747 | }; |
3748 | ||
3749 | /* | |
3750 | * Allocates an inode for the vcpu. | |
3751 | */ | |
3752 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
3753 | { | |
e46b4692 MY |
3754 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
3755 | ||
3756 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
3757 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
3758 | } |
3759 | ||
3e7093d0 | 3760 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 3761 | { |
741cbbae | 3762 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
d56f5136 | 3763 | struct dentry *debugfs_dentry; |
45b5939e | 3764 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 3765 | |
45b5939e | 3766 | if (!debugfs_initialized()) |
3e7093d0 | 3767 | return; |
45b5939e LC |
3768 | |
3769 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
3770 | debugfs_dentry = debugfs_create_dir(dir_name, |
3771 | vcpu->kvm->debugfs_dentry); | |
45b5939e | 3772 | |
d56f5136 | 3773 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
741cbbae | 3774 | #endif |
45b5939e LC |
3775 | } |
3776 | ||
c5ea7660 AK |
3777 | /* |
3778 | * Creates some virtual cpus. Good luck creating more than one. | |
3779 | */ | |
73880c80 | 3780 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
3781 | { |
3782 | int r; | |
e09fefde | 3783 | struct kvm_vcpu *vcpu; |
8bd826d6 | 3784 | struct page *page; |
c5ea7660 | 3785 | |
a1c42dde | 3786 | if (id >= KVM_MAX_VCPU_IDS) |
338c7dba AH |
3787 | return -EINVAL; |
3788 | ||
6c7caebc | 3789 | mutex_lock(&kvm->lock); |
f502cc56 | 3790 | if (kvm->created_vcpus >= kvm->max_vcpus) { |
6c7caebc PB |
3791 | mutex_unlock(&kvm->lock); |
3792 | return -EINVAL; | |
3793 | } | |
3794 | ||
1d5e740d ZG |
3795 | r = kvm_arch_vcpu_precreate(kvm, id); |
3796 | if (r) { | |
3797 | mutex_unlock(&kvm->lock); | |
3798 | return r; | |
3799 | } | |
3800 | ||
6c7caebc PB |
3801 | kvm->created_vcpus++; |
3802 | mutex_unlock(&kvm->lock); | |
3803 | ||
85f47930 | 3804 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
e529ef66 SC |
3805 | if (!vcpu) { |
3806 | r = -ENOMEM; | |
6c7caebc PB |
3807 | goto vcpu_decrement; |
3808 | } | |
c5ea7660 | 3809 | |
fcd97ad5 | 3810 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 3811 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
3812 | if (!page) { |
3813 | r = -ENOMEM; | |
e529ef66 | 3814 | goto vcpu_free; |
8bd826d6 SC |
3815 | } |
3816 | vcpu->run = page_address(page); | |
3817 | ||
3818 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
3819 | |
3820 | r = kvm_arch_vcpu_create(vcpu); | |
3821 | if (r) | |
8bd826d6 | 3822 | goto vcpu_free_run_page; |
e529ef66 | 3823 | |
fb04a1ed PX |
3824 | if (kvm->dirty_ring_size) { |
3825 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
3826 | id, kvm->dirty_ring_size); | |
3827 | if (r) | |
3828 | goto arch_vcpu_destroy; | |
3829 | } | |
3830 | ||
11ec2804 | 3831 | mutex_lock(&kvm->lock); |
e09fefde DH |
3832 | if (kvm_get_vcpu_by_id(kvm, id)) { |
3833 | r = -EEXIST; | |
3834 | goto unlock_vcpu_destroy; | |
3835 | } | |
73880c80 | 3836 | |
8750e72a | 3837 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
c5b07754 MZ |
3838 | r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT); |
3839 | BUG_ON(r == -EBUSY); | |
3840 | if (r) | |
3841 | goto unlock_vcpu_destroy; | |
c5ea7660 | 3842 | |
ce55c049 JZ |
3843 | /* Fill the stats id string for the vcpu */ |
3844 | snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d", | |
3845 | task_pid_nr(current), id); | |
3846 | ||
fb3f0f51 | 3847 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 3848 | kvm_get_kvm(kvm); |
bccf2150 | 3849 | r = create_vcpu_fd(vcpu); |
73880c80 | 3850 | if (r < 0) { |
c5b07754 | 3851 | xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx); |
149487bd | 3852 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 3853 | goto unlock_vcpu_destroy; |
73880c80 GN |
3854 | } |
3855 | ||
dd489240 | 3856 | /* |
c5b07754 MZ |
3857 | * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu |
3858 | * pointer before kvm->online_vcpu's incremented value. | |
dd489240 | 3859 | */ |
73880c80 GN |
3860 | smp_wmb(); |
3861 | atomic_inc(&kvm->online_vcpus); | |
3862 | ||
73880c80 | 3863 | mutex_unlock(&kvm->lock); |
42897d86 | 3864 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 3865 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 3866 | return r; |
39c3b86e | 3867 | |
d780592b | 3868 | unlock_vcpu_destroy: |
7d8fece6 | 3869 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
3870 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
3871 | arch_vcpu_destroy: | |
d40ccc62 | 3872 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
3873 | vcpu_free_run_page: |
3874 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
3875 | vcpu_free: |
3876 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
3877 | vcpu_decrement: |
3878 | mutex_lock(&kvm->lock); | |
3879 | kvm->created_vcpus--; | |
3880 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
3881 | return r; |
3882 | } | |
3883 | ||
1961d276 AK |
3884 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
3885 | { | |
3886 | if (sigset) { | |
3887 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
3888 | vcpu->sigset_active = 1; | |
3889 | vcpu->sigset = *sigset; | |
3890 | } else | |
3891 | vcpu->sigset_active = 0; | |
3892 | return 0; | |
3893 | } | |
3894 | ||
ce55c049 JZ |
3895 | static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer, |
3896 | size_t size, loff_t *offset) | |
3897 | { | |
3898 | struct kvm_vcpu *vcpu = file->private_data; | |
3899 | ||
3900 | return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header, | |
3901 | &kvm_vcpu_stats_desc[0], &vcpu->stat, | |
3902 | sizeof(vcpu->stat), user_buffer, size, offset); | |
3903 | } | |
3904 | ||
3905 | static const struct file_operations kvm_vcpu_stats_fops = { | |
3906 | .read = kvm_vcpu_stats_read, | |
3907 | .llseek = noop_llseek, | |
3908 | }; | |
3909 | ||
3910 | static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu) | |
3911 | { | |
3912 | int fd; | |
3913 | struct file *file; | |
3914 | char name[15 + ITOA_MAX_LEN + 1]; | |
3915 | ||
3916 | snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id); | |
3917 | ||
3918 | fd = get_unused_fd_flags(O_CLOEXEC); | |
3919 | if (fd < 0) | |
3920 | return fd; | |
3921 | ||
3922 | file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY); | |
3923 | if (IS_ERR(file)) { | |
3924 | put_unused_fd(fd); | |
3925 | return PTR_ERR(file); | |
3926 | } | |
3927 | file->f_mode |= FMODE_PREAD; | |
3928 | fd_install(fd, file); | |
3929 | ||
3930 | return fd; | |
3931 | } | |
3932 | ||
bccf2150 AK |
3933 | static long kvm_vcpu_ioctl(struct file *filp, |
3934 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 3935 | { |
bccf2150 | 3936 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 3937 | void __user *argp = (void __user *)arg; |
313a3dc7 | 3938 | int r; |
fa3795a7 DH |
3939 | struct kvm_fpu *fpu = NULL; |
3940 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 3941 | |
f4d31653 | 3942 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
6d4e4c4f | 3943 | return -EIO; |
2122ff5e | 3944 | |
2ea75be3 DM |
3945 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
3946 | return -EINVAL; | |
3947 | ||
2122ff5e | 3948 | /* |
5cb0944c PB |
3949 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
3950 | * execution; mutex_lock() would break them. | |
2122ff5e | 3951 | */ |
5cb0944c PB |
3952 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
3953 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 3954 | return r; |
2122ff5e | 3955 | |
ec7660cc CD |
3956 | if (mutex_lock_killable(&vcpu->mutex)) |
3957 | return -EINTR; | |
6aa8b732 | 3958 | switch (ioctl) { |
0e4524a5 CB |
3959 | case KVM_RUN: { |
3960 | struct pid *oldpid; | |
f0fe5108 AK |
3961 | r = -EINVAL; |
3962 | if (arg) | |
3963 | goto out; | |
0e4524a5 | 3964 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 3965 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 3966 | /* The thread running this VCPU changed. */ |
bd2a6394 | 3967 | struct pid *newpid; |
f95ef0cd | 3968 | |
bd2a6394 CD |
3969 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
3970 | if (r) | |
3971 | break; | |
3972 | ||
3973 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
3974 | rcu_assign_pointer(vcpu->pid, newpid); |
3975 | if (oldpid) | |
3976 | synchronize_rcu(); | |
3977 | put_pid(oldpid); | |
3978 | } | |
1b94f6f8 | 3979 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 3980 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 3981 | break; |
0e4524a5 | 3982 | } |
6aa8b732 | 3983 | case KVM_GET_REGS: { |
3e4bb3ac | 3984 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3985 | |
3e4bb3ac | 3986 | r = -ENOMEM; |
b12ce36a | 3987 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 3988 | if (!kvm_regs) |
6aa8b732 | 3989 | goto out; |
3e4bb3ac XZ |
3990 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
3991 | if (r) | |
3992 | goto out_free1; | |
6aa8b732 | 3993 | r = -EFAULT; |
3e4bb3ac XZ |
3994 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
3995 | goto out_free1; | |
6aa8b732 | 3996 | r = 0; |
3e4bb3ac XZ |
3997 | out_free1: |
3998 | kfree(kvm_regs); | |
6aa8b732 AK |
3999 | break; |
4000 | } | |
4001 | case KVM_SET_REGS: { | |
3e4bb3ac | 4002 | struct kvm_regs *kvm_regs; |
6aa8b732 | 4003 | |
ff5c2c03 SL |
4004 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
4005 | if (IS_ERR(kvm_regs)) { | |
4006 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 4007 | goto out; |
ff5c2c03 | 4008 | } |
3e4bb3ac | 4009 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 4010 | kfree(kvm_regs); |
6aa8b732 AK |
4011 | break; |
4012 | } | |
4013 | case KVM_GET_SREGS: { | |
b12ce36a BG |
4014 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
4015 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
4016 | r = -ENOMEM; |
4017 | if (!kvm_sregs) | |
4018 | goto out; | |
4019 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
4020 | if (r) |
4021 | goto out; | |
4022 | r = -EFAULT; | |
fa3795a7 | 4023 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
4024 | goto out; |
4025 | r = 0; | |
4026 | break; | |
4027 | } | |
4028 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
4029 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
4030 | if (IS_ERR(kvm_sregs)) { | |
4031 | r = PTR_ERR(kvm_sregs); | |
18595411 | 4032 | kvm_sregs = NULL; |
6aa8b732 | 4033 | goto out; |
ff5c2c03 | 4034 | } |
fa3795a7 | 4035 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
4036 | break; |
4037 | } | |
62d9f0db MT |
4038 | case KVM_GET_MP_STATE: { |
4039 | struct kvm_mp_state mp_state; | |
4040 | ||
4041 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
4042 | if (r) | |
4043 | goto out; | |
4044 | r = -EFAULT; | |
893bdbf1 | 4045 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
4046 | goto out; |
4047 | r = 0; | |
4048 | break; | |
4049 | } | |
4050 | case KVM_SET_MP_STATE: { | |
4051 | struct kvm_mp_state mp_state; | |
4052 | ||
4053 | r = -EFAULT; | |
893bdbf1 | 4054 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
4055 | goto out; |
4056 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
4057 | break; |
4058 | } | |
6aa8b732 AK |
4059 | case KVM_TRANSLATE: { |
4060 | struct kvm_translation tr; | |
4061 | ||
4062 | r = -EFAULT; | |
893bdbf1 | 4063 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 4064 | goto out; |
8b006791 | 4065 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
4066 | if (r) |
4067 | goto out; | |
4068 | r = -EFAULT; | |
893bdbf1 | 4069 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
4070 | goto out; |
4071 | r = 0; | |
4072 | break; | |
4073 | } | |
d0bfb940 JK |
4074 | case KVM_SET_GUEST_DEBUG: { |
4075 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
4076 | |
4077 | r = -EFAULT; | |
893bdbf1 | 4078 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 4079 | goto out; |
d0bfb940 | 4080 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
4081 | break; |
4082 | } | |
1961d276 AK |
4083 | case KVM_SET_SIGNAL_MASK: { |
4084 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4085 | struct kvm_signal_mask kvm_sigmask; | |
4086 | sigset_t sigset, *p; | |
4087 | ||
4088 | p = NULL; | |
4089 | if (argp) { | |
4090 | r = -EFAULT; | |
4091 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4092 | sizeof(kvm_sigmask))) |
1961d276 AK |
4093 | goto out; |
4094 | r = -EINVAL; | |
893bdbf1 | 4095 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
4096 | goto out; |
4097 | r = -EFAULT; | |
4098 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 4099 | sizeof(sigset))) |
1961d276 AK |
4100 | goto out; |
4101 | p = &sigset; | |
4102 | } | |
376d41ff | 4103 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
4104 | break; |
4105 | } | |
b8836737 | 4106 | case KVM_GET_FPU: { |
b12ce36a | 4107 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
4108 | r = -ENOMEM; |
4109 | if (!fpu) | |
4110 | goto out; | |
4111 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
4112 | if (r) |
4113 | goto out; | |
4114 | r = -EFAULT; | |
fa3795a7 | 4115 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
4116 | goto out; |
4117 | r = 0; | |
4118 | break; | |
4119 | } | |
4120 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
4121 | fpu = memdup_user(argp, sizeof(*fpu)); |
4122 | if (IS_ERR(fpu)) { | |
4123 | r = PTR_ERR(fpu); | |
18595411 | 4124 | fpu = NULL; |
b8836737 | 4125 | goto out; |
ff5c2c03 | 4126 | } |
fa3795a7 | 4127 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
4128 | break; |
4129 | } | |
ce55c049 JZ |
4130 | case KVM_GET_STATS_FD: { |
4131 | r = kvm_vcpu_ioctl_get_stats_fd(vcpu); | |
4132 | break; | |
4133 | } | |
bccf2150 | 4134 | default: |
313a3dc7 | 4135 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
4136 | } |
4137 | out: | |
ec7660cc | 4138 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
4139 | kfree(fpu); |
4140 | kfree(kvm_sregs); | |
bccf2150 AK |
4141 | return r; |
4142 | } | |
4143 | ||
de8e5d74 | 4144 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
4145 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
4146 | unsigned int ioctl, unsigned long arg) | |
4147 | { | |
4148 | struct kvm_vcpu *vcpu = filp->private_data; | |
4149 | void __user *argp = compat_ptr(arg); | |
4150 | int r; | |
4151 | ||
f4d31653 | 4152 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead) |
1dda606c AG |
4153 | return -EIO; |
4154 | ||
4155 | switch (ioctl) { | |
4156 | case KVM_SET_SIGNAL_MASK: { | |
4157 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
4158 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
4159 | sigset_t sigset; |
4160 | ||
4161 | if (argp) { | |
4162 | r = -EFAULT; | |
4163 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 4164 | sizeof(kvm_sigmask))) |
1dda606c AG |
4165 | goto out; |
4166 | r = -EINVAL; | |
3968cf62 | 4167 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
4168 | goto out; |
4169 | r = -EFAULT; | |
1393b4aa PB |
4170 | if (get_compat_sigset(&sigset, |
4171 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 4172 | goto out; |
760a9a30 AC |
4173 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
4174 | } else | |
4175 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
4176 | break; |
4177 | } | |
4178 | default: | |
4179 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
4180 | } | |
4181 | ||
4182 | out: | |
4183 | return r; | |
4184 | } | |
4185 | #endif | |
4186 | ||
a1cd3f08 CLG |
4187 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
4188 | { | |
4189 | struct kvm_device *dev = filp->private_data; | |
4190 | ||
4191 | if (dev->ops->mmap) | |
4192 | return dev->ops->mmap(dev, vma); | |
4193 | ||
4194 | return -ENODEV; | |
4195 | } | |
4196 | ||
852b6d57 SW |
4197 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
4198 | int (*accessor)(struct kvm_device *dev, | |
4199 | struct kvm_device_attr *attr), | |
4200 | unsigned long arg) | |
4201 | { | |
4202 | struct kvm_device_attr attr; | |
4203 | ||
4204 | if (!accessor) | |
4205 | return -EPERM; | |
4206 | ||
4207 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
4208 | return -EFAULT; | |
4209 | ||
4210 | return accessor(dev, &attr); | |
4211 | } | |
4212 | ||
4213 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
4214 | unsigned long arg) | |
4215 | { | |
4216 | struct kvm_device *dev = filp->private_data; | |
4217 | ||
f4d31653 | 4218 | if (dev->kvm->mm != current->mm || dev->kvm->vm_dead) |
ddba9180 SC |
4219 | return -EIO; |
4220 | ||
852b6d57 SW |
4221 | switch (ioctl) { |
4222 | case KVM_SET_DEVICE_ATTR: | |
4223 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
4224 | case KVM_GET_DEVICE_ATTR: | |
4225 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
4226 | case KVM_HAS_DEVICE_ATTR: | |
4227 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
4228 | default: | |
4229 | if (dev->ops->ioctl) | |
4230 | return dev->ops->ioctl(dev, ioctl, arg); | |
4231 | ||
4232 | return -ENOTTY; | |
4233 | } | |
4234 | } | |
4235 | ||
852b6d57 SW |
4236 | static int kvm_device_release(struct inode *inode, struct file *filp) |
4237 | { | |
4238 | struct kvm_device *dev = filp->private_data; | |
4239 | struct kvm *kvm = dev->kvm; | |
4240 | ||
2bde9b3e CLG |
4241 | if (dev->ops->release) { |
4242 | mutex_lock(&kvm->lock); | |
4243 | list_del(&dev->vm_node); | |
4244 | dev->ops->release(dev); | |
4245 | mutex_unlock(&kvm->lock); | |
4246 | } | |
4247 | ||
852b6d57 SW |
4248 | kvm_put_kvm(kvm); |
4249 | return 0; | |
4250 | } | |
4251 | ||
4252 | static const struct file_operations kvm_device_fops = { | |
4253 | .unlocked_ioctl = kvm_device_ioctl, | |
4254 | .release = kvm_device_release, | |
7ddfd3e0 | 4255 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 4256 | .mmap = kvm_device_mmap, |
852b6d57 SW |
4257 | }; |
4258 | ||
4259 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
4260 | { | |
4261 | if (filp->f_op != &kvm_device_fops) | |
4262 | return NULL; | |
4263 | ||
4264 | return filp->private_data; | |
4265 | } | |
4266 | ||
8538cb22 | 4267 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 4268 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
4269 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
4270 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 4271 | #endif |
d60eacb0 WD |
4272 | }; |
4273 | ||
8538cb22 | 4274 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
4275 | { |
4276 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
4277 | return -ENOSPC; | |
4278 | ||
4279 | if (kvm_device_ops_table[type] != NULL) | |
4280 | return -EEXIST; | |
4281 | ||
4282 | kvm_device_ops_table[type] = ops; | |
4283 | return 0; | |
4284 | } | |
4285 | ||
571ee1b6 WL |
4286 | void kvm_unregister_device_ops(u32 type) |
4287 | { | |
4288 | if (kvm_device_ops_table[type] != NULL) | |
4289 | kvm_device_ops_table[type] = NULL; | |
4290 | } | |
4291 | ||
852b6d57 SW |
4292 | static int kvm_ioctl_create_device(struct kvm *kvm, |
4293 | struct kvm_create_device *cd) | |
4294 | { | |
8538cb22 | 4295 | const struct kvm_device_ops *ops = NULL; |
852b6d57 SW |
4296 | struct kvm_device *dev; |
4297 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 4298 | int type; |
852b6d57 SW |
4299 | int ret; |
4300 | ||
d60eacb0 WD |
4301 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
4302 | return -ENODEV; | |
4303 | ||
1d487e9b PB |
4304 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
4305 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 4306 | if (ops == NULL) |
852b6d57 | 4307 | return -ENODEV; |
852b6d57 SW |
4308 | |
4309 | if (test) | |
4310 | return 0; | |
4311 | ||
b12ce36a | 4312 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
4313 | if (!dev) |
4314 | return -ENOMEM; | |
4315 | ||
4316 | dev->ops = ops; | |
4317 | dev->kvm = kvm; | |
852b6d57 | 4318 | |
a28ebea2 | 4319 | mutex_lock(&kvm->lock); |
1d487e9b | 4320 | ret = ops->create(dev, type); |
852b6d57 | 4321 | if (ret < 0) { |
a28ebea2 | 4322 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
4323 | kfree(dev); |
4324 | return ret; | |
4325 | } | |
a28ebea2 CD |
4326 | list_add(&dev->vm_node, &kvm->devices); |
4327 | mutex_unlock(&kvm->lock); | |
852b6d57 | 4328 | |
023e9fdd CD |
4329 | if (ops->init) |
4330 | ops->init(dev); | |
4331 | ||
cfa39381 | 4332 | kvm_get_kvm(kvm); |
24009b05 | 4333 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 4334 | if (ret < 0) { |
149487bd | 4335 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
4336 | mutex_lock(&kvm->lock); |
4337 | list_del(&dev->vm_node); | |
e8bc2427 AK |
4338 | if (ops->release) |
4339 | ops->release(dev); | |
a28ebea2 | 4340 | mutex_unlock(&kvm->lock); |
e8bc2427 AK |
4341 | if (ops->destroy) |
4342 | ops->destroy(dev); | |
852b6d57 SW |
4343 | return ret; |
4344 | } | |
4345 | ||
852b6d57 SW |
4346 | cd->fd = ret; |
4347 | return 0; | |
4348 | } | |
4349 | ||
92b591a4 AG |
4350 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
4351 | { | |
4352 | switch (arg) { | |
4353 | case KVM_CAP_USER_MEMORY: | |
4354 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
4355 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
4356 | case KVM_CAP_INTERNAL_ERROR_DATA: |
4357 | #ifdef CONFIG_HAVE_KVM_MSI | |
4358 | case KVM_CAP_SIGNAL_MSI: | |
4359 | #endif | |
297e2105 | 4360 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 4361 | case KVM_CAP_IRQFD: |
92b591a4 AG |
4362 | case KVM_CAP_IRQFD_RESAMPLE: |
4363 | #endif | |
e9ea5069 | 4364 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 4365 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 4366 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 4367 | case KVM_CAP_HALT_POLL: |
92b591a4 | 4368 | return 1; |
4b4357e0 | 4369 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
4370 | case KVM_CAP_COALESCED_MMIO: |
4371 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
4372 | case KVM_CAP_COALESCED_PIO: |
4373 | return 1; | |
30422558 | 4374 | #endif |
3c9bd400 JZ |
4375 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4376 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
4377 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
4378 | #endif | |
92b591a4 AG |
4379 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4380 | case KVM_CAP_IRQ_ROUTING: | |
4381 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
4382 | #endif |
4383 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
4384 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
4385 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 4386 | #endif |
c110ae57 PB |
4387 | case KVM_CAP_NR_MEMSLOTS: |
4388 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed | 4389 | case KVM_CAP_DIRTY_LOG_RING: |
dc70ec21 | 4390 | #ifdef CONFIG_HAVE_KVM_DIRTY_RING |
fb04a1ed PX |
4391 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); |
4392 | #else | |
4393 | return 0; | |
4394 | #endif | |
ce55c049 | 4395 | case KVM_CAP_BINARY_STATS_FD: |
d495f942 | 4396 | case KVM_CAP_SYSTEM_EVENT_DATA: |
ce55c049 | 4397 | return 1; |
92b591a4 AG |
4398 | default: |
4399 | break; | |
4400 | } | |
4401 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
4402 | } | |
4403 | ||
fb04a1ed PX |
4404 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
4405 | { | |
4406 | int r; | |
4407 | ||
4408 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
4409 | return -EINVAL; | |
4410 | ||
4411 | /* the size should be power of 2 */ | |
4412 | if (!size || (size & (size - 1))) | |
4413 | return -EINVAL; | |
4414 | ||
4415 | /* Should be bigger to keep the reserved entries, or a page */ | |
4416 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
4417 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
4418 | return -EINVAL; | |
4419 | ||
4420 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
4421 | sizeof(struct kvm_dirty_gfn)) | |
4422 | return -E2BIG; | |
4423 | ||
4424 | /* We only allow it to set once */ | |
4425 | if (kvm->dirty_ring_size) | |
4426 | return -EINVAL; | |
4427 | ||
4428 | mutex_lock(&kvm->lock); | |
4429 | ||
4430 | if (kvm->created_vcpus) { | |
4431 | /* We don't allow to change this value after vcpu created */ | |
4432 | r = -EINVAL; | |
4433 | } else { | |
4434 | kvm->dirty_ring_size = size; | |
4435 | r = 0; | |
4436 | } | |
4437 | ||
4438 | mutex_unlock(&kvm->lock); | |
4439 | return r; | |
4440 | } | |
4441 | ||
4442 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
4443 | { | |
46808a4c | 4444 | unsigned long i; |
fb04a1ed PX |
4445 | struct kvm_vcpu *vcpu; |
4446 | int cleared = 0; | |
4447 | ||
4448 | if (!kvm->dirty_ring_size) | |
4449 | return -EINVAL; | |
4450 | ||
4451 | mutex_lock(&kvm->slots_lock); | |
4452 | ||
4453 | kvm_for_each_vcpu(i, vcpu, kvm) | |
4454 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
4455 | ||
4456 | mutex_unlock(&kvm->slots_lock); | |
4457 | ||
4458 | if (cleared) | |
4459 | kvm_flush_remote_tlbs(kvm); | |
4460 | ||
4461 | return cleared; | |
4462 | } | |
4463 | ||
e5d83c74 PB |
4464 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4465 | struct kvm_enable_cap *cap) | |
4466 | { | |
4467 | return -EINVAL; | |
4468 | } | |
4469 | ||
4470 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
4471 | struct kvm_enable_cap *cap) | |
4472 | { | |
4473 | switch (cap->cap) { | |
2a31b9db | 4474 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
4475 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
4476 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
4477 | ||
4478 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
4479 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
4480 | ||
4481 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
4482 | return -EINVAL; |
4483 | kvm->manual_dirty_log_protect = cap->args[0]; | |
4484 | return 0; | |
3c9bd400 | 4485 | } |
2a31b9db | 4486 | #endif |
acd05785 DM |
4487 | case KVM_CAP_HALT_POLL: { |
4488 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
4489 | return -EINVAL; | |
4490 | ||
4491 | kvm->max_halt_poll_ns = cap->args[0]; | |
4492 | return 0; | |
4493 | } | |
fb04a1ed PX |
4494 | case KVM_CAP_DIRTY_LOG_RING: |
4495 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); | |
e5d83c74 PB |
4496 | default: |
4497 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
4498 | } | |
4499 | } | |
4500 | ||
fcfe1bae JZ |
4501 | static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer, |
4502 | size_t size, loff_t *offset) | |
4503 | { | |
4504 | struct kvm *kvm = file->private_data; | |
4505 | ||
4506 | return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header, | |
4507 | &kvm_vm_stats_desc[0], &kvm->stat, | |
4508 | sizeof(kvm->stat), user_buffer, size, offset); | |
4509 | } | |
4510 | ||
4511 | static const struct file_operations kvm_vm_stats_fops = { | |
4512 | .read = kvm_vm_stats_read, | |
4513 | .llseek = noop_llseek, | |
4514 | }; | |
4515 | ||
4516 | static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm) | |
4517 | { | |
4518 | int fd; | |
4519 | struct file *file; | |
4520 | ||
4521 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4522 | if (fd < 0) | |
4523 | return fd; | |
4524 | ||
4525 | file = anon_inode_getfile("kvm-vm-stats", | |
4526 | &kvm_vm_stats_fops, kvm, O_RDONLY); | |
4527 | if (IS_ERR(file)) { | |
4528 | put_unused_fd(fd); | |
4529 | return PTR_ERR(file); | |
4530 | } | |
4531 | file->f_mode |= FMODE_PREAD; | |
4532 | fd_install(fd, file); | |
4533 | ||
4534 | return fd; | |
4535 | } | |
4536 | ||
bccf2150 AK |
4537 | static long kvm_vm_ioctl(struct file *filp, |
4538 | unsigned int ioctl, unsigned long arg) | |
4539 | { | |
4540 | struct kvm *kvm = filp->private_data; | |
4541 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 4542 | int r; |
bccf2150 | 4543 | |
f4d31653 | 4544 | if (kvm->mm != current->mm || kvm->vm_dead) |
6d4e4c4f | 4545 | return -EIO; |
bccf2150 AK |
4546 | switch (ioctl) { |
4547 | case KVM_CREATE_VCPU: | |
4548 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 4549 | break; |
e5d83c74 PB |
4550 | case KVM_ENABLE_CAP: { |
4551 | struct kvm_enable_cap cap; | |
4552 | ||
4553 | r = -EFAULT; | |
4554 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4555 | goto out; | |
4556 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
4557 | break; | |
4558 | } | |
6fc138d2 IE |
4559 | case KVM_SET_USER_MEMORY_REGION: { |
4560 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
4561 | ||
4562 | r = -EFAULT; | |
4563 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 4564 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
4565 | goto out; |
4566 | ||
47ae31e2 | 4567 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
4568 | break; |
4569 | } | |
4570 | case KVM_GET_DIRTY_LOG: { | |
4571 | struct kvm_dirty_log log; | |
4572 | ||
4573 | r = -EFAULT; | |
893bdbf1 | 4574 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 4575 | goto out; |
2c6f5df9 | 4576 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
4577 | break; |
4578 | } | |
2a31b9db PB |
4579 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4580 | case KVM_CLEAR_DIRTY_LOG: { | |
4581 | struct kvm_clear_dirty_log log; | |
4582 | ||
4583 | r = -EFAULT; | |
4584 | if (copy_from_user(&log, argp, sizeof(log))) | |
4585 | goto out; | |
4586 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4587 | break; | |
4588 | } | |
4589 | #endif | |
4b4357e0 | 4590 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
4591 | case KVM_REGISTER_COALESCED_MMIO: { |
4592 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4593 | |
5f94c174 | 4594 | r = -EFAULT; |
893bdbf1 | 4595 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4596 | goto out; |
5f94c174 | 4597 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4598 | break; |
4599 | } | |
4600 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
4601 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4602 | |
5f94c174 | 4603 | r = -EFAULT; |
893bdbf1 | 4604 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4605 | goto out; |
5f94c174 | 4606 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4607 | break; |
4608 | } | |
4609 | #endif | |
721eecbf GH |
4610 | case KVM_IRQFD: { |
4611 | struct kvm_irqfd data; | |
4612 | ||
4613 | r = -EFAULT; | |
893bdbf1 | 4614 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 4615 | goto out; |
d4db2935 | 4616 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
4617 | break; |
4618 | } | |
d34e6b17 GH |
4619 | case KVM_IOEVENTFD: { |
4620 | struct kvm_ioeventfd data; | |
4621 | ||
4622 | r = -EFAULT; | |
893bdbf1 | 4623 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
4624 | goto out; |
4625 | r = kvm_ioeventfd(kvm, &data); | |
4626 | break; | |
4627 | } | |
07975ad3 JK |
4628 | #ifdef CONFIG_HAVE_KVM_MSI |
4629 | case KVM_SIGNAL_MSI: { | |
4630 | struct kvm_msi msi; | |
4631 | ||
4632 | r = -EFAULT; | |
893bdbf1 | 4633 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
4634 | goto out; |
4635 | r = kvm_send_userspace_msi(kvm, &msi); | |
4636 | break; | |
4637 | } | |
23d43cf9 CD |
4638 | #endif |
4639 | #ifdef __KVM_HAVE_IRQ_LINE | |
4640 | case KVM_IRQ_LINE_STATUS: | |
4641 | case KVM_IRQ_LINE: { | |
4642 | struct kvm_irq_level irq_event; | |
4643 | ||
4644 | r = -EFAULT; | |
893bdbf1 | 4645 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
4646 | goto out; |
4647 | ||
aa2fbe6d YZ |
4648 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
4649 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
4650 | if (r) |
4651 | goto out; | |
4652 | ||
4653 | r = -EFAULT; | |
4654 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 4655 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
4656 | goto out; |
4657 | } | |
4658 | ||
4659 | r = 0; | |
4660 | break; | |
4661 | } | |
73880c80 | 4662 | #endif |
aa8d5944 AG |
4663 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4664 | case KVM_SET_GSI_ROUTING: { | |
4665 | struct kvm_irq_routing routing; | |
4666 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 4667 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
4668 | |
4669 | r = -EFAULT; | |
4670 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
4671 | goto out; | |
4672 | r = -EINVAL; | |
5c0aea0e DH |
4673 | if (!kvm_arch_can_set_irq_routing(kvm)) |
4674 | goto out; | |
caf1ff26 | 4675 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
4676 | goto out; |
4677 | if (routing.flags) | |
4678 | goto out; | |
f8c1b85b | 4679 | if (routing.nr) { |
f8c1b85b | 4680 | urouting = argp; |
7ec28e26 DE |
4681 | entries = vmemdup_user(urouting->entries, |
4682 | array_size(sizeof(*entries), | |
4683 | routing.nr)); | |
4684 | if (IS_ERR(entries)) { | |
4685 | r = PTR_ERR(entries); | |
4686 | goto out; | |
4687 | } | |
f8c1b85b | 4688 | } |
aa8d5944 AG |
4689 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
4690 | routing.flags); | |
7ec28e26 | 4691 | kvfree(entries); |
aa8d5944 AG |
4692 | break; |
4693 | } | |
4694 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
4695 | case KVM_CREATE_DEVICE: { |
4696 | struct kvm_create_device cd; | |
4697 | ||
4698 | r = -EFAULT; | |
4699 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
4700 | goto out; | |
4701 | ||
4702 | r = kvm_ioctl_create_device(kvm, &cd); | |
4703 | if (r) | |
4704 | goto out; | |
4705 | ||
4706 | r = -EFAULT; | |
4707 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
4708 | goto out; | |
4709 | ||
4710 | r = 0; | |
4711 | break; | |
4712 | } | |
92b591a4 AG |
4713 | case KVM_CHECK_EXTENSION: |
4714 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
4715 | break; | |
fb04a1ed PX |
4716 | case KVM_RESET_DIRTY_RINGS: |
4717 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
4718 | break; | |
fcfe1bae JZ |
4719 | case KVM_GET_STATS_FD: |
4720 | r = kvm_vm_ioctl_get_stats_fd(kvm); | |
4721 | break; | |
f17abe9a | 4722 | default: |
1fe779f8 | 4723 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
4724 | } |
4725 | out: | |
4726 | return r; | |
4727 | } | |
4728 | ||
de8e5d74 | 4729 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
4730 | struct compat_kvm_dirty_log { |
4731 | __u32 slot; | |
4732 | __u32 padding1; | |
4733 | union { | |
4734 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4735 | __u64 padding2; | |
4736 | }; | |
4737 | }; | |
4738 | ||
8750f9bb PB |
4739 | struct compat_kvm_clear_dirty_log { |
4740 | __u32 slot; | |
4741 | __u32 num_pages; | |
4742 | __u64 first_page; | |
4743 | union { | |
4744 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4745 | __u64 padding2; | |
4746 | }; | |
4747 | }; | |
4748 | ||
6ff5894c AB |
4749 | static long kvm_vm_compat_ioctl(struct file *filp, |
4750 | unsigned int ioctl, unsigned long arg) | |
4751 | { | |
4752 | struct kvm *kvm = filp->private_data; | |
4753 | int r; | |
4754 | ||
f4d31653 | 4755 | if (kvm->mm != current->mm || kvm->vm_dead) |
6ff5894c AB |
4756 | return -EIO; |
4757 | switch (ioctl) { | |
8750f9bb PB |
4758 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4759 | case KVM_CLEAR_DIRTY_LOG: { | |
4760 | struct compat_kvm_clear_dirty_log compat_log; | |
4761 | struct kvm_clear_dirty_log log; | |
4762 | ||
4763 | if (copy_from_user(&compat_log, (void __user *)arg, | |
4764 | sizeof(compat_log))) | |
4765 | return -EFAULT; | |
4766 | log.slot = compat_log.slot; | |
4767 | log.num_pages = compat_log.num_pages; | |
4768 | log.first_page = compat_log.first_page; | |
4769 | log.padding2 = compat_log.padding2; | |
4770 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4771 | ||
4772 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4773 | break; | |
4774 | } | |
4775 | #endif | |
6ff5894c AB |
4776 | case KVM_GET_DIRTY_LOG: { |
4777 | struct compat_kvm_dirty_log compat_log; | |
4778 | struct kvm_dirty_log log; | |
4779 | ||
6ff5894c AB |
4780 | if (copy_from_user(&compat_log, (void __user *)arg, |
4781 | sizeof(compat_log))) | |
f6a3b168 | 4782 | return -EFAULT; |
6ff5894c AB |
4783 | log.slot = compat_log.slot; |
4784 | log.padding1 = compat_log.padding1; | |
4785 | log.padding2 = compat_log.padding2; | |
4786 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4787 | ||
4788 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
4789 | break; |
4790 | } | |
4791 | default: | |
4792 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
4793 | } | |
6ff5894c AB |
4794 | return r; |
4795 | } | |
4796 | #endif | |
4797 | ||
70375c2d | 4798 | static const struct file_operations kvm_vm_fops = { |
f17abe9a AK |
4799 | .release = kvm_vm_release, |
4800 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 4801 | .llseek = noop_llseek, |
7ddfd3e0 | 4802 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
4803 | }; |
4804 | ||
54526d1f NT |
4805 | bool file_is_kvm(struct file *file) |
4806 | { | |
4807 | return file && file->f_op == &kvm_vm_fops; | |
4808 | } | |
4809 | EXPORT_SYMBOL_GPL(file_is_kvm); | |
4810 | ||
e08b9637 | 4811 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 4812 | { |
aac87636 | 4813 | int r; |
f17abe9a | 4814 | struct kvm *kvm; |
506cfba9 | 4815 | struct file *file; |
f17abe9a | 4816 | |
e08b9637 | 4817 | kvm = kvm_create_vm(type); |
d6d28168 AK |
4818 | if (IS_ERR(kvm)) |
4819 | return PTR_ERR(kvm); | |
4b4357e0 | 4820 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 4821 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
4822 | if (r < 0) |
4823 | goto put_kvm; | |
6ce5a090 | 4824 | #endif |
506cfba9 | 4825 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
4826 | if (r < 0) |
4827 | goto put_kvm; | |
4828 | ||
fcfe1bae JZ |
4829 | snprintf(kvm->stats_id, sizeof(kvm->stats_id), |
4830 | "kvm-%d", task_pid_nr(current)); | |
4831 | ||
506cfba9 AV |
4832 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
4833 | if (IS_ERR(file)) { | |
4834 | put_unused_fd(r); | |
78588335 ME |
4835 | r = PTR_ERR(file); |
4836 | goto put_kvm; | |
506cfba9 | 4837 | } |
536a6f88 | 4838 | |
525df861 PB |
4839 | /* |
4840 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
4841 | * already set, with ->release() being kvm_vm_release(). In error | |
4842 | * cases it will be called by the final fput(file) and will take | |
4843 | * care of doing kvm_put_kvm(kvm). | |
4844 | */ | |
536a6f88 | 4845 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
4846 | put_unused_fd(r); |
4847 | fput(file); | |
536a6f88 JF |
4848 | return -ENOMEM; |
4849 | } | |
286de8f6 | 4850 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 4851 | |
506cfba9 | 4852 | fd_install(r, file); |
aac87636 | 4853 | return r; |
78588335 ME |
4854 | |
4855 | put_kvm: | |
4856 | kvm_put_kvm(kvm); | |
4857 | return r; | |
f17abe9a AK |
4858 | } |
4859 | ||
4860 | static long kvm_dev_ioctl(struct file *filp, | |
4861 | unsigned int ioctl, unsigned long arg) | |
4862 | { | |
07c45a36 | 4863 | long r = -EINVAL; |
f17abe9a AK |
4864 | |
4865 | switch (ioctl) { | |
4866 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
4867 | if (arg) |
4868 | goto out; | |
f17abe9a AK |
4869 | r = KVM_API_VERSION; |
4870 | break; | |
4871 | case KVM_CREATE_VM: | |
e08b9637 | 4872 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 4873 | break; |
018d00d2 | 4874 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 4875 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 4876 | break; |
07c45a36 | 4877 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
4878 | if (arg) |
4879 | goto out; | |
adb1ff46 AK |
4880 | r = PAGE_SIZE; /* struct kvm_run */ |
4881 | #ifdef CONFIG_X86 | |
4882 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 4883 | #endif |
4b4357e0 | 4884 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 4885 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 4886 | #endif |
07c45a36 | 4887 | break; |
d4c9ff2d FEL |
4888 | case KVM_TRACE_ENABLE: |
4889 | case KVM_TRACE_PAUSE: | |
4890 | case KVM_TRACE_DISABLE: | |
2023a29c | 4891 | r = -EOPNOTSUPP; |
d4c9ff2d | 4892 | break; |
6aa8b732 | 4893 | default: |
043405e1 | 4894 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
4895 | } |
4896 | out: | |
4897 | return r; | |
4898 | } | |
4899 | ||
6aa8b732 | 4900 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 4901 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 4902 | .llseek = noop_llseek, |
7ddfd3e0 | 4903 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
4904 | }; |
4905 | ||
4906 | static struct miscdevice kvm_dev = { | |
bbe4432e | 4907 | KVM_MINOR, |
6aa8b732 AK |
4908 | "kvm", |
4909 | &kvm_chardev_ops, | |
4910 | }; | |
4911 | ||
75b7127c | 4912 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
4913 | { |
4914 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 4915 | int r; |
1b6c0168 | 4916 | |
7f59f492 | 4917 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4918 | return; |
10474ae8 | 4919 | |
7f59f492 | 4920 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 4921 | |
13a34e06 | 4922 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
4923 | |
4924 | if (r) { | |
4925 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
4926 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 4927 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 4928 | } |
1b6c0168 AK |
4929 | } |
4930 | ||
8c18b2d2 | 4931 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 4932 | { |
4a937f96 | 4933 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4934 | if (kvm_usage_count) |
4935 | hardware_enable_nolock(NULL); | |
4a937f96 | 4936 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4937 | return 0; |
75b7127c TY |
4938 | } |
4939 | ||
4940 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
4941 | { |
4942 | int cpu = raw_smp_processor_id(); | |
4943 | ||
7f59f492 | 4944 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4945 | return; |
7f59f492 | 4946 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 4947 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
4948 | } |
4949 | ||
8c18b2d2 | 4950 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 4951 | { |
4a937f96 | 4952 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4953 | if (kvm_usage_count) |
4954 | hardware_disable_nolock(NULL); | |
4a937f96 | 4955 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4956 | return 0; |
75b7127c TY |
4957 | } |
4958 | ||
10474ae8 AG |
4959 | static void hardware_disable_all_nolock(void) |
4960 | { | |
4961 | BUG_ON(!kvm_usage_count); | |
4962 | ||
4963 | kvm_usage_count--; | |
4964 | if (!kvm_usage_count) | |
75b7127c | 4965 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
4966 | } |
4967 | ||
4968 | static void hardware_disable_all(void) | |
4969 | { | |
4a937f96 | 4970 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 4971 | hardware_disable_all_nolock(); |
4a937f96 | 4972 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4973 | } |
4974 | ||
4975 | static int hardware_enable_all(void) | |
4976 | { | |
4977 | int r = 0; | |
4978 | ||
4a937f96 | 4979 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
4980 | |
4981 | kvm_usage_count++; | |
4982 | if (kvm_usage_count == 1) { | |
4983 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 4984 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
4985 | |
4986 | if (atomic_read(&hardware_enable_failed)) { | |
4987 | hardware_disable_all_nolock(); | |
4988 | r = -EBUSY; | |
4989 | } | |
4990 | } | |
4991 | ||
4a937f96 | 4992 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4993 | |
4994 | return r; | |
4995 | } | |
4996 | ||
9a2b85c6 | 4997 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 4998 | void *v) |
9a2b85c6 | 4999 | { |
8e1c1815 SY |
5000 | /* |
5001 | * Some (well, at least mine) BIOSes hang on reboot if | |
5002 | * in vmx root mode. | |
5003 | * | |
5004 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
5005 | */ | |
1170adc6 | 5006 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 5007 | kvm_rebooting = true; |
75b7127c | 5008 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
5009 | return NOTIFY_OK; |
5010 | } | |
5011 | ||
5012 | static struct notifier_block kvm_reboot_notifier = { | |
5013 | .notifier_call = kvm_reboot, | |
5014 | .priority = 0, | |
5015 | }; | |
5016 | ||
e93f8a0f | 5017 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
5018 | { |
5019 | int i; | |
5020 | ||
5021 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 5022 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
5023 | |
5024 | kvm_iodevice_destructor(pos); | |
5025 | } | |
e93f8a0f | 5026 | kfree(bus); |
2eeb2e94 GH |
5027 | } |
5028 | ||
c21fbff1 | 5029 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 5030 | const struct kvm_io_range *r2) |
743eeb0b | 5031 | { |
8f4216c7 JW |
5032 | gpa_t addr1 = r1->addr; |
5033 | gpa_t addr2 = r2->addr; | |
5034 | ||
5035 | if (addr1 < addr2) | |
743eeb0b | 5036 | return -1; |
8f4216c7 JW |
5037 | |
5038 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
5039 | * accept any overlapping write. Any order is acceptable for | |
5040 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
5041 | * we process all of them. | |
5042 | */ | |
5043 | if (r2->len) { | |
5044 | addr1 += r1->len; | |
5045 | addr2 += r2->len; | |
5046 | } | |
5047 | ||
5048 | if (addr1 > addr2) | |
743eeb0b | 5049 | return 1; |
8f4216c7 | 5050 | |
743eeb0b SL |
5051 | return 0; |
5052 | } | |
5053 | ||
a343c9b7 PB |
5054 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
5055 | { | |
c21fbff1 | 5056 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
5057 | } |
5058 | ||
39369f7a | 5059 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
5060 | gpa_t addr, int len) |
5061 | { | |
5062 | struct kvm_io_range *range, key; | |
5063 | int off; | |
5064 | ||
5065 | key = (struct kvm_io_range) { | |
5066 | .addr = addr, | |
5067 | .len = len, | |
5068 | }; | |
5069 | ||
5070 | range = bsearch(&key, bus->range, bus->dev_count, | |
5071 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
5072 | if (range == NULL) | |
5073 | return -ENOENT; | |
5074 | ||
5075 | off = range - bus->range; | |
5076 | ||
c21fbff1 | 5077 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
5078 | off--; |
5079 | ||
5080 | return off; | |
5081 | } | |
5082 | ||
e32edf4f | 5083 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
5084 | struct kvm_io_range *range, const void *val) |
5085 | { | |
5086 | int idx; | |
5087 | ||
5088 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
5089 | if (idx < 0) | |
5090 | return -EOPNOTSUPP; | |
5091 | ||
5092 | while (idx < bus->dev_count && | |
c21fbff1 | 5093 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5094 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5095 | range->len, val)) |
5096 | return idx; | |
5097 | idx++; | |
5098 | } | |
5099 | ||
5100 | return -EOPNOTSUPP; | |
5101 | } | |
5102 | ||
bda9020e | 5103 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 5104 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 5105 | int len, const void *val) |
2eeb2e94 | 5106 | { |
90d83dc3 | 5107 | struct kvm_io_bus *bus; |
743eeb0b | 5108 | struct kvm_io_range range; |
126a5af5 | 5109 | int r; |
743eeb0b SL |
5110 | |
5111 | range = (struct kvm_io_range) { | |
5112 | .addr = addr, | |
5113 | .len = len, | |
5114 | }; | |
90d83dc3 | 5115 | |
e32edf4f | 5116 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5117 | if (!bus) |
5118 | return -ENOMEM; | |
e32edf4f | 5119 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5120 | return r < 0 ? r : 0; |
5121 | } | |
a2420107 | 5122 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
5123 | |
5124 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
5125 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
5126 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
5127 | { |
5128 | struct kvm_io_bus *bus; | |
5129 | struct kvm_io_range range; | |
5130 | ||
5131 | range = (struct kvm_io_range) { | |
5132 | .addr = addr, | |
5133 | .len = len, | |
5134 | }; | |
5135 | ||
e32edf4f | 5136 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5137 | if (!bus) |
5138 | return -ENOMEM; | |
126a5af5 CH |
5139 | |
5140 | /* First try the device referenced by cookie. */ | |
5141 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 5142 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 5143 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
5144 | val)) |
5145 | return cookie; | |
5146 | ||
5147 | /* | |
5148 | * cookie contained garbage; fall back to search and return the | |
5149 | * correct cookie value. | |
5150 | */ | |
e32edf4f | 5151 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
5152 | } |
5153 | ||
e32edf4f NN |
5154 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
5155 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
5156 | { |
5157 | int idx; | |
5158 | ||
5159 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
5160 | if (idx < 0) |
5161 | return -EOPNOTSUPP; | |
5162 | ||
5163 | while (idx < bus->dev_count && | |
c21fbff1 | 5164 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 5165 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
5166 | range->len, val)) |
5167 | return idx; | |
743eeb0b SL |
5168 | idx++; |
5169 | } | |
5170 | ||
bda9020e MT |
5171 | return -EOPNOTSUPP; |
5172 | } | |
2eeb2e94 | 5173 | |
bda9020e | 5174 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 5175 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 5176 | int len, void *val) |
bda9020e | 5177 | { |
90d83dc3 | 5178 | struct kvm_io_bus *bus; |
743eeb0b | 5179 | struct kvm_io_range range; |
126a5af5 | 5180 | int r; |
743eeb0b SL |
5181 | |
5182 | range = (struct kvm_io_range) { | |
5183 | .addr = addr, | |
5184 | .len = len, | |
5185 | }; | |
e93f8a0f | 5186 | |
e32edf4f | 5187 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
5188 | if (!bus) |
5189 | return -ENOMEM; | |
e32edf4f | 5190 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
5191 | return r < 0 ? r : 0; |
5192 | } | |
743eeb0b | 5193 | |
79fac95e | 5194 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
5195 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
5196 | int len, struct kvm_io_device *dev) | |
6c474694 | 5197 | { |
d4c67a7a | 5198 | int i; |
e93f8a0f | 5199 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 5200 | struct kvm_io_range range; |
090b7aff | 5201 | |
4a12f951 | 5202 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
5203 | if (!bus) |
5204 | return -ENOMEM; | |
5205 | ||
6ea34c9b AK |
5206 | /* exclude ioeventfd which is limited by maximum fd */ |
5207 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 5208 | return -ENOSPC; |
2eeb2e94 | 5209 | |
90952cd3 | 5210 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 5211 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
5212 | if (!new_bus) |
5213 | return -ENOMEM; | |
d4c67a7a GH |
5214 | |
5215 | range = (struct kvm_io_range) { | |
5216 | .addr = addr, | |
5217 | .len = len, | |
5218 | .dev = dev, | |
5219 | }; | |
5220 | ||
5221 | for (i = 0; i < bus->dev_count; i++) | |
5222 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
5223 | break; | |
5224 | ||
5225 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
5226 | new_bus->dev_count++; | |
5227 | new_bus->range[i] = range; | |
5228 | memcpy(new_bus->range + i + 1, bus->range + i, | |
5229 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
5230 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
5231 | synchronize_srcu_expedited(&kvm->srcu); | |
5232 | kfree(bus); | |
090b7aff GH |
5233 | |
5234 | return 0; | |
5235 | } | |
5236 | ||
5d3c4c79 SC |
5237 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5238 | struct kvm_io_device *dev) | |
090b7aff | 5239 | { |
f6588660 | 5240 | int i, j; |
e93f8a0f | 5241 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 5242 | |
7c896d37 SC |
5243 | lockdep_assert_held(&kvm->slots_lock); |
5244 | ||
4a12f951 | 5245 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 5246 | if (!bus) |
5d3c4c79 | 5247 | return 0; |
df630b8c | 5248 | |
7c896d37 | 5249 | for (i = 0; i < bus->dev_count; i++) { |
a1300716 | 5250 | if (bus->range[i].dev == dev) { |
090b7aff GH |
5251 | break; |
5252 | } | |
7c896d37 | 5253 | } |
e93f8a0f | 5254 | |
90db1043 | 5255 | if (i == bus->dev_count) |
5d3c4c79 | 5256 | return 0; |
a1300716 | 5257 | |
90952cd3 | 5258 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 5259 | GFP_KERNEL_ACCOUNT); |
f6588660 | 5260 | if (new_bus) { |
871c433b | 5261 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
5262 | new_bus->dev_count--; |
5263 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 5264 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
2ee37574 SC |
5265 | } |
5266 | ||
5267 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
5268 | synchronize_srcu_expedited(&kvm->srcu); | |
5269 | ||
5270 | /* Destroy the old bus _after_ installing the (null) bus. */ | |
5271 | if (!new_bus) { | |
90db1043 | 5272 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
f6588660 RK |
5273 | for (j = 0; j < bus->dev_count; j++) { |
5274 | if (j == i) | |
5275 | continue; | |
5276 | kvm_iodevice_destructor(bus->range[j].dev); | |
5277 | } | |
90db1043 | 5278 | } |
a1300716 | 5279 | |
e93f8a0f | 5280 | kfree(bus); |
5d3c4c79 | 5281 | return new_bus ? 0 : -ENOMEM; |
2eeb2e94 GH |
5282 | } |
5283 | ||
8a39d006 AP |
5284 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5285 | gpa_t addr) | |
5286 | { | |
5287 | struct kvm_io_bus *bus; | |
5288 | int dev_idx, srcu_idx; | |
5289 | struct kvm_io_device *iodev = NULL; | |
5290 | ||
5291 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
5292 | ||
5293 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
5294 | if (!bus) |
5295 | goto out_unlock; | |
8a39d006 AP |
5296 | |
5297 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
5298 | if (dev_idx < 0) | |
5299 | goto out_unlock; | |
5300 | ||
5301 | iodev = bus->range[dev_idx].dev; | |
5302 | ||
5303 | out_unlock: | |
5304 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
5305 | ||
5306 | return iodev; | |
5307 | } | |
5308 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
5309 | ||
536a6f88 JF |
5310 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
5311 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
5312 | const char *fmt) | |
5313 | { | |
5314 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5315 | inode->i_private; | |
5316 | ||
605c7130 PX |
5317 | /* |
5318 | * The debugfs files are a reference to the kvm struct which | |
5319 | * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe | |
5320 | * avoids the race between open and the removal of the debugfs directory. | |
536a6f88 | 5321 | */ |
605c7130 | 5322 | if (!kvm_get_kvm_safe(stat_data->kvm)) |
536a6f88 JF |
5323 | return -ENOENT; |
5324 | ||
833b45de | 5325 | if (simple_attr_open(inode, file, get, |
bc9e9e67 | 5326 | kvm_stats_debugfs_mode(stat_data->desc) & 0222 |
09cbcef6 MP |
5327 | ? set : NULL, |
5328 | fmt)) { | |
536a6f88 JF |
5329 | kvm_put_kvm(stat_data->kvm); |
5330 | return -ENOMEM; | |
5331 | } | |
5332 | ||
5333 | return 0; | |
5334 | } | |
5335 | ||
5336 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
5337 | { | |
5338 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5339 | inode->i_private; | |
5340 | ||
5341 | simple_attr_release(inode, file); | |
5342 | kvm_put_kvm(stat_data->kvm); | |
5343 | ||
5344 | return 0; | |
5345 | } | |
5346 | ||
09cbcef6 | 5347 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 5348 | { |
bc9e9e67 | 5349 | *val = *(u64 *)((void *)(&kvm->stat) + offset); |
536a6f88 | 5350 | |
09cbcef6 MP |
5351 | return 0; |
5352 | } | |
5353 | ||
5354 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
5355 | { | |
bc9e9e67 | 5356 | *(u64 *)((void *)(&kvm->stat) + offset) = 0; |
536a6f88 JF |
5357 | |
5358 | return 0; | |
5359 | } | |
5360 | ||
09cbcef6 | 5361 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 5362 | { |
46808a4c | 5363 | unsigned long i; |
09cbcef6 | 5364 | struct kvm_vcpu *vcpu; |
ce35ef27 | 5365 | |
09cbcef6 | 5366 | *val = 0; |
ce35ef27 | 5367 | |
09cbcef6 | 5368 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5369 | *val += *(u64 *)((void *)(&vcpu->stat) + offset); |
ce35ef27 SJS |
5370 | |
5371 | return 0; | |
5372 | } | |
5373 | ||
09cbcef6 | 5374 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 5375 | { |
46808a4c | 5376 | unsigned long i; |
09cbcef6 | 5377 | struct kvm_vcpu *vcpu; |
536a6f88 | 5378 | |
09cbcef6 | 5379 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5380 | *(u64 *)((void *)(&vcpu->stat) + offset) = 0; |
09cbcef6 MP |
5381 | |
5382 | return 0; | |
5383 | } | |
536a6f88 | 5384 | |
09cbcef6 | 5385 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 5386 | { |
09cbcef6 | 5387 | int r = -EFAULT; |
536a6f88 | 5388 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
536a6f88 | 5389 | |
bc9e9e67 | 5390 | switch (stat_data->kind) { |
09cbcef6 MP |
5391 | case KVM_STAT_VM: |
5392 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5393 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5394 | break; |
5395 | case KVM_STAT_VCPU: | |
5396 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5397 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5398 | break; |
5399 | } | |
536a6f88 | 5400 | |
09cbcef6 | 5401 | return r; |
536a6f88 JF |
5402 | } |
5403 | ||
09cbcef6 | 5404 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 5405 | { |
09cbcef6 | 5406 | int r = -EFAULT; |
ce35ef27 | 5407 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
ce35ef27 SJS |
5408 | |
5409 | if (val) | |
5410 | return -EINVAL; | |
5411 | ||
bc9e9e67 | 5412 | switch (stat_data->kind) { |
09cbcef6 MP |
5413 | case KVM_STAT_VM: |
5414 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5415 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5416 | break; |
5417 | case KVM_STAT_VCPU: | |
5418 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5419 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5420 | break; |
5421 | } | |
ce35ef27 | 5422 | |
09cbcef6 | 5423 | return r; |
ce35ef27 SJS |
5424 | } |
5425 | ||
09cbcef6 | 5426 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
5427 | { |
5428 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
5429 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
5430 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
5431 | } |
5432 | ||
09cbcef6 MP |
5433 | static const struct file_operations stat_fops_per_vm = { |
5434 | .owner = THIS_MODULE, | |
5435 | .open = kvm_stat_data_open, | |
536a6f88 | 5436 | .release = kvm_debugfs_release, |
09cbcef6 MP |
5437 | .read = simple_attr_read, |
5438 | .write = simple_attr_write, | |
5439 | .llseek = no_llseek, | |
536a6f88 JF |
5440 | }; |
5441 | ||
8b88b099 | 5442 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
5443 | { |
5444 | unsigned offset = (long)_offset; | |
ba1389b7 | 5445 | struct kvm *kvm; |
536a6f88 | 5446 | u64 tmp_val; |
ba1389b7 | 5447 | |
8b88b099 | 5448 | *val = 0; |
0d9ce162 | 5449 | mutex_lock(&kvm_lock); |
536a6f88 | 5450 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5451 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
5452 | *val += tmp_val; |
5453 | } | |
0d9ce162 | 5454 | mutex_unlock(&kvm_lock); |
8b88b099 | 5455 | return 0; |
ba1389b7 AK |
5456 | } |
5457 | ||
ce35ef27 SJS |
5458 | static int vm_stat_clear(void *_offset, u64 val) |
5459 | { | |
5460 | unsigned offset = (long)_offset; | |
5461 | struct kvm *kvm; | |
ce35ef27 SJS |
5462 | |
5463 | if (val) | |
5464 | return -EINVAL; | |
5465 | ||
0d9ce162 | 5466 | mutex_lock(&kvm_lock); |
ce35ef27 | 5467 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5468 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 5469 | } |
0d9ce162 | 5470 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5471 | |
5472 | return 0; | |
5473 | } | |
5474 | ||
5475 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
bc9e9e67 | 5476 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n"); |
ba1389b7 | 5477 | |
8b88b099 | 5478 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
5479 | { |
5480 | unsigned offset = (long)_offset; | |
1165f5fe | 5481 | struct kvm *kvm; |
536a6f88 | 5482 | u64 tmp_val; |
1165f5fe | 5483 | |
8b88b099 | 5484 | *val = 0; |
0d9ce162 | 5485 | mutex_lock(&kvm_lock); |
536a6f88 | 5486 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5487 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
5488 | *val += tmp_val; |
5489 | } | |
0d9ce162 | 5490 | mutex_unlock(&kvm_lock); |
8b88b099 | 5491 | return 0; |
1165f5fe AK |
5492 | } |
5493 | ||
ce35ef27 SJS |
5494 | static int vcpu_stat_clear(void *_offset, u64 val) |
5495 | { | |
5496 | unsigned offset = (long)_offset; | |
5497 | struct kvm *kvm; | |
ce35ef27 SJS |
5498 | |
5499 | if (val) | |
5500 | return -EINVAL; | |
5501 | ||
0d9ce162 | 5502 | mutex_lock(&kvm_lock); |
ce35ef27 | 5503 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5504 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 5505 | } |
0d9ce162 | 5506 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5507 | |
5508 | return 0; | |
5509 | } | |
5510 | ||
5511 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
5512 | "%llu\n"); | |
bc9e9e67 | 5513 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n"); |
1165f5fe | 5514 | |
286de8f6 CI |
5515 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
5516 | { | |
5517 | struct kobj_uevent_env *env; | |
286de8f6 CI |
5518 | unsigned long long created, active; |
5519 | ||
5520 | if (!kvm_dev.this_device || !kvm) | |
5521 | return; | |
5522 | ||
0d9ce162 | 5523 | mutex_lock(&kvm_lock); |
286de8f6 CI |
5524 | if (type == KVM_EVENT_CREATE_VM) { |
5525 | kvm_createvm_count++; | |
5526 | kvm_active_vms++; | |
5527 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
5528 | kvm_active_vms--; | |
5529 | } | |
5530 | created = kvm_createvm_count; | |
5531 | active = kvm_active_vms; | |
0d9ce162 | 5532 | mutex_unlock(&kvm_lock); |
286de8f6 | 5533 | |
b12ce36a | 5534 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
5535 | if (!env) |
5536 | return; | |
5537 | ||
5538 | add_uevent_var(env, "CREATED=%llu", created); | |
5539 | add_uevent_var(env, "COUNT=%llu", active); | |
5540 | ||
fdeaf7e3 | 5541 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 5542 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
5543 | kvm->userspace_pid = task_pid_nr(current); |
5544 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 5545 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
5546 | } |
5547 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 5548 | |
a44a4cc1 | 5549 | if (!IS_ERR(kvm->debugfs_dentry)) { |
b12ce36a | 5550 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
5551 | |
5552 | if (p) { | |
5553 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
5554 | if (!IS_ERR(tmp)) | |
5555 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
5556 | kfree(p); | |
286de8f6 CI |
5557 | } |
5558 | } | |
5559 | /* no need for checks, since we are adding at most only 5 keys */ | |
5560 | env->envp[env->envp_idx++] = NULL; | |
5561 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
5562 | kfree(env); | |
286de8f6 CI |
5563 | } |
5564 | ||
929f45e3 | 5565 | static void kvm_init_debug(void) |
6aa8b732 | 5566 | { |
bc9e9e67 JZ |
5567 | const struct file_operations *fops; |
5568 | const struct _kvm_stats_desc *pdesc; | |
5569 | int i; | |
6aa8b732 | 5570 | |
76f7c879 | 5571 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 5572 | |
bc9e9e67 JZ |
5573 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
5574 | pdesc = &kvm_vm_stats_desc[i]; | |
5575 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5576 | fops = &vm_stat_fops; | |
5577 | else | |
5578 | fops = &vm_stat_readonly_fops; | |
5579 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5580 | kvm_debugfs_dir, | |
5581 | (void *)(long)pdesc->desc.offset, fops); | |
5582 | } | |
5583 | ||
5584 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
5585 | pdesc = &kvm_vcpu_stats_desc[i]; | |
5586 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5587 | fops = &vcpu_stat_fops; | |
5588 | else | |
5589 | fops = &vcpu_stat_readonly_fops; | |
5590 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5591 | kvm_debugfs_dir, | |
5592 | (void *)(long)pdesc->desc.offset, fops); | |
4f69b680 | 5593 | } |
6aa8b732 AK |
5594 | } |
5595 | ||
fb3600cc | 5596 | static int kvm_suspend(void) |
59ae6c6b | 5597 | { |
10474ae8 | 5598 | if (kvm_usage_count) |
75b7127c | 5599 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
5600 | return 0; |
5601 | } | |
5602 | ||
fb3600cc | 5603 | static void kvm_resume(void) |
59ae6c6b | 5604 | { |
ca84d1a2 | 5605 | if (kvm_usage_count) { |
4cb9a998 | 5606 | lockdep_assert_not_held(&kvm_count_lock); |
75b7127c | 5607 | hardware_enable_nolock(NULL); |
ca84d1a2 | 5608 | } |
59ae6c6b AK |
5609 | } |
5610 | ||
fb3600cc | 5611 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
5612 | .suspend = kvm_suspend, |
5613 | .resume = kvm_resume, | |
5614 | }; | |
5615 | ||
15ad7146 AK |
5616 | static inline |
5617 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
5618 | { | |
5619 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
5620 | } | |
5621 | ||
5622 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
5623 | { | |
5624 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 5625 | |
046ddeed | 5626 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 5627 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 5628 | |
7495e22b | 5629 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 5630 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 5631 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
5632 | } |
5633 | ||
5634 | static void kvm_sched_out(struct preempt_notifier *pn, | |
5635 | struct task_struct *next) | |
5636 | { | |
5637 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
5638 | ||
3ba9f93b | 5639 | if (current->on_rq) { |
046ddeed | 5640 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
5641 | WRITE_ONCE(vcpu->ready, true); |
5642 | } | |
e9b11c17 | 5643 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
5644 | __this_cpu_write(kvm_running_vcpu, NULL); |
5645 | } | |
5646 | ||
5647 | /** | |
5648 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
5649 | * |
5650 | * We can disable preemption locally around accessing the per-CPU variable, | |
5651 | * and use the resolved vcpu pointer after enabling preemption again, | |
5652 | * because even if the current thread is migrated to another CPU, reading | |
5653 | * the per-CPU value later will give us the same value as we update the | |
5654 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
5655 | */ |
5656 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
5657 | { | |
1f03b2bc MZ |
5658 | struct kvm_vcpu *vcpu; |
5659 | ||
5660 | preempt_disable(); | |
5661 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
5662 | preempt_enable(); | |
5663 | ||
5664 | return vcpu; | |
7495e22b | 5665 | } |
379a3c8e | 5666 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
5667 | |
5668 | /** | |
5669 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
5670 | */ | |
5671 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
5672 | { | |
5673 | return &kvm_running_vcpu; | |
15ad7146 AK |
5674 | } |
5675 | ||
e1bfc245 SC |
5676 | #ifdef CONFIG_GUEST_PERF_EVENTS |
5677 | static unsigned int kvm_guest_state(void) | |
5678 | { | |
5679 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
5680 | unsigned int state; | |
5681 | ||
5682 | if (!kvm_arch_pmi_in_guest(vcpu)) | |
5683 | return 0; | |
5684 | ||
5685 | state = PERF_GUEST_ACTIVE; | |
5686 | if (!kvm_arch_vcpu_in_kernel(vcpu)) | |
5687 | state |= PERF_GUEST_USER; | |
5688 | ||
5689 | return state; | |
5690 | } | |
5691 | ||
5692 | static unsigned long kvm_guest_get_ip(void) | |
5693 | { | |
5694 | struct kvm_vcpu *vcpu = kvm_get_running_vcpu(); | |
5695 | ||
5696 | /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */ | |
5697 | if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu))) | |
5698 | return 0; | |
5699 | ||
5700 | return kvm_arch_vcpu_get_ip(vcpu); | |
5701 | } | |
5702 | ||
5703 | static struct perf_guest_info_callbacks kvm_guest_cbs = { | |
5704 | .state = kvm_guest_state, | |
5705 | .get_ip = kvm_guest_get_ip, | |
5706 | .handle_intel_pt_intr = NULL, | |
5707 | }; | |
5708 | ||
5709 | void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void)) | |
5710 | { | |
5711 | kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler; | |
5712 | perf_register_guest_info_callbacks(&kvm_guest_cbs); | |
5713 | } | |
5714 | void kvm_unregister_perf_callbacks(void) | |
5715 | { | |
5716 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); | |
5717 | } | |
5718 | #endif | |
5719 | ||
b9904085 SC |
5720 | struct kvm_cpu_compat_check { |
5721 | void *opaque; | |
5722 | int *ret; | |
5723 | }; | |
5724 | ||
5725 | static void check_processor_compat(void *data) | |
f257d6dc | 5726 | { |
b9904085 SC |
5727 | struct kvm_cpu_compat_check *c = data; |
5728 | ||
5729 | *c->ret = kvm_arch_check_processor_compat(c->opaque); | |
f257d6dc SC |
5730 | } |
5731 | ||
0ee75bea | 5732 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 5733 | struct module *module) |
6aa8b732 | 5734 | { |
b9904085 | 5735 | struct kvm_cpu_compat_check c; |
6aa8b732 | 5736 | int r; |
002c7f7c | 5737 | int cpu; |
6aa8b732 | 5738 | |
f8c16bba ZX |
5739 | r = kvm_arch_init(opaque); |
5740 | if (r) | |
d2308784 | 5741 | goto out_fail; |
cb498ea2 | 5742 | |
7dac16c3 AH |
5743 | /* |
5744 | * kvm_arch_init makes sure there's at most one caller | |
5745 | * for architectures that support multiple implementations, | |
5746 | * like intel and amd on x86. | |
36343f6e PB |
5747 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
5748 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 5749 | */ |
36343f6e PB |
5750 | r = kvm_irqfd_init(); |
5751 | if (r) | |
5752 | goto out_irqfd; | |
7dac16c3 | 5753 | |
8437a617 | 5754 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
5755 | r = -ENOMEM; |
5756 | goto out_free_0; | |
5757 | } | |
5758 | ||
b9904085 | 5759 | r = kvm_arch_hardware_setup(opaque); |
6aa8b732 | 5760 | if (r < 0) |
faf0be22 | 5761 | goto out_free_1; |
6aa8b732 | 5762 | |
b9904085 SC |
5763 | c.ret = &r; |
5764 | c.opaque = opaque; | |
002c7f7c | 5765 | for_each_online_cpu(cpu) { |
b9904085 | 5766 | smp_call_function_single(cpu, check_processor_compat, &c, 1); |
002c7f7c | 5767 | if (r < 0) |
faf0be22 | 5768 | goto out_free_2; |
002c7f7c YS |
5769 | } |
5770 | ||
73c1b41e | 5771 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 5772 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 5773 | if (r) |
d2308784 | 5774 | goto out_free_2; |
6aa8b732 AK |
5775 | register_reboot_notifier(&kvm_reboot_notifier); |
5776 | ||
c16f862d | 5777 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
5778 | if (!vcpu_align) |
5779 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
5780 | kvm_vcpu_cache = |
5781 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
5782 | SLAB_ACCOUNT, | |
5783 | offsetof(struct kvm_vcpu, arch), | |
ce55c049 JZ |
5784 | offsetofend(struct kvm_vcpu, stats_id) |
5785 | - offsetof(struct kvm_vcpu, arch), | |
46515736 | 5786 | NULL); |
c16f862d RR |
5787 | if (!kvm_vcpu_cache) { |
5788 | r = -ENOMEM; | |
fb3600cc | 5789 | goto out_free_3; |
c16f862d RR |
5790 | } |
5791 | ||
baff59cc VK |
5792 | for_each_possible_cpu(cpu) { |
5793 | if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu), | |
5794 | GFP_KERNEL, cpu_to_node(cpu))) { | |
5795 | r = -ENOMEM; | |
5796 | goto out_free_4; | |
5797 | } | |
5798 | } | |
5799 | ||
af585b92 GN |
5800 | r = kvm_async_pf_init(); |
5801 | if (r) | |
baff59cc | 5802 | goto out_free_5; |
af585b92 | 5803 | |
6aa8b732 AK |
5804 | kvm_chardev_ops.owner = module; |
5805 | ||
5806 | r = misc_register(&kvm_dev); | |
5807 | if (r) { | |
1170adc6 | 5808 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 5809 | goto out_unreg; |
6aa8b732 AK |
5810 | } |
5811 | ||
fb3600cc RW |
5812 | register_syscore_ops(&kvm_syscore_ops); |
5813 | ||
15ad7146 AK |
5814 | kvm_preempt_ops.sched_in = kvm_sched_in; |
5815 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
5816 | ||
929f45e3 | 5817 | kvm_init_debug(); |
0ea4ed8e | 5818 | |
3c3c29fd PB |
5819 | r = kvm_vfio_ops_init(); |
5820 | WARN_ON(r); | |
5821 | ||
c7addb90 | 5822 | return 0; |
6aa8b732 | 5823 | |
af585b92 GN |
5824 | out_unreg: |
5825 | kvm_async_pf_deinit(); | |
baff59cc VK |
5826 | out_free_5: |
5827 | for_each_possible_cpu(cpu) | |
5828 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
5829 | out_free_4: | |
c16f862d | 5830 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 5831 | out_free_3: |
6aa8b732 | 5832 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5833 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 5834 | out_free_2: |
e9b11c17 | 5835 | kvm_arch_hardware_unsetup(); |
faf0be22 | 5836 | out_free_1: |
7f59f492 | 5837 | free_cpumask_var(cpus_hardware_enabled); |
d2308784 | 5838 | out_free_0: |
a0f155e9 | 5839 | kvm_irqfd_exit(); |
36343f6e | 5840 | out_irqfd: |
7dac16c3 AH |
5841 | kvm_arch_exit(); |
5842 | out_fail: | |
6aa8b732 AK |
5843 | return r; |
5844 | } | |
cb498ea2 | 5845 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 5846 | |
cb498ea2 | 5847 | void kvm_exit(void) |
6aa8b732 | 5848 | { |
baff59cc VK |
5849 | int cpu; |
5850 | ||
4bd33b56 | 5851 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 5852 | misc_deregister(&kvm_dev); |
baff59cc VK |
5853 | for_each_possible_cpu(cpu) |
5854 | free_cpumask_var(per_cpu(cpu_kick_mask, cpu)); | |
c16f862d | 5855 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 5856 | kvm_async_pf_deinit(); |
fb3600cc | 5857 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 5858 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5859 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 5860 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 5861 | kvm_arch_hardware_unsetup(); |
f8c16bba | 5862 | kvm_arch_exit(); |
a0f155e9 | 5863 | kvm_irqfd_exit(); |
7f59f492 | 5864 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 5865 | kvm_vfio_ops_exit(); |
6aa8b732 | 5866 | } |
cb498ea2 | 5867 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
5868 | |
5869 | struct kvm_vm_worker_thread_context { | |
5870 | struct kvm *kvm; | |
5871 | struct task_struct *parent; | |
5872 | struct completion init_done; | |
5873 | kvm_vm_thread_fn_t thread_fn; | |
5874 | uintptr_t data; | |
5875 | int err; | |
5876 | }; | |
5877 | ||
5878 | static int kvm_vm_worker_thread(void *context) | |
5879 | { | |
5880 | /* | |
5881 | * The init_context is allocated on the stack of the parent thread, so | |
5882 | * we have to locally copy anything that is needed beyond initialization | |
5883 | */ | |
5884 | struct kvm_vm_worker_thread_context *init_context = context; | |
e45cce30 | 5885 | struct task_struct *parent; |
c57c8046 JS |
5886 | struct kvm *kvm = init_context->kvm; |
5887 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
5888 | uintptr_t data = init_context->data; | |
5889 | int err; | |
5890 | ||
5891 | err = kthread_park(current); | |
5892 | /* kthread_park(current) is never supposed to return an error */ | |
5893 | WARN_ON(err != 0); | |
5894 | if (err) | |
5895 | goto init_complete; | |
5896 | ||
5897 | err = cgroup_attach_task_all(init_context->parent, current); | |
5898 | if (err) { | |
5899 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
5900 | __func__, err); | |
5901 | goto init_complete; | |
5902 | } | |
5903 | ||
5904 | set_user_nice(current, task_nice(init_context->parent)); | |
5905 | ||
5906 | init_complete: | |
5907 | init_context->err = err; | |
5908 | complete(&init_context->init_done); | |
5909 | init_context = NULL; | |
5910 | ||
5911 | if (err) | |
e45cce30 | 5912 | goto out; |
c57c8046 JS |
5913 | |
5914 | /* Wait to be woken up by the spawner before proceeding. */ | |
5915 | kthread_parkme(); | |
5916 | ||
5917 | if (!kthread_should_stop()) | |
5918 | err = thread_fn(kvm, data); | |
5919 | ||
e45cce30 VS |
5920 | out: |
5921 | /* | |
5922 | * Move kthread back to its original cgroup to prevent it lingering in | |
5923 | * the cgroup of the VM process, after the latter finishes its | |
5924 | * execution. | |
5925 | * | |
5926 | * kthread_stop() waits on the 'exited' completion condition which is | |
5927 | * set in exit_mm(), via mm_release(), in do_exit(). However, the | |
5928 | * kthread is removed from the cgroup in the cgroup_exit() which is | |
5929 | * called after the exit_mm(). This causes the kthread_stop() to return | |
5930 | * before the kthread actually quits the cgroup. | |
5931 | */ | |
5932 | rcu_read_lock(); | |
5933 | parent = rcu_dereference(current->real_parent); | |
5934 | get_task_struct(parent); | |
5935 | rcu_read_unlock(); | |
5936 | cgroup_attach_task_all(parent, current); | |
5937 | put_task_struct(parent); | |
5938 | ||
c57c8046 JS |
5939 | return err; |
5940 | } | |
5941 | ||
5942 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
5943 | uintptr_t data, const char *name, | |
5944 | struct task_struct **thread_ptr) | |
5945 | { | |
5946 | struct kvm_vm_worker_thread_context init_context = {}; | |
5947 | struct task_struct *thread; | |
5948 | ||
5949 | *thread_ptr = NULL; | |
5950 | init_context.kvm = kvm; | |
5951 | init_context.parent = current; | |
5952 | init_context.thread_fn = thread_fn; | |
5953 | init_context.data = data; | |
5954 | init_completion(&init_context.init_done); | |
5955 | ||
5956 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
5957 | "%s-%d", name, task_pid_nr(current)); | |
5958 | if (IS_ERR(thread)) | |
5959 | return PTR_ERR(thread); | |
5960 | ||
5961 | /* kthread_run is never supposed to return NULL */ | |
5962 | WARN_ON(thread == NULL); | |
5963 | ||
5964 | wait_for_completion(&init_context.init_done); | |
5965 | ||
5966 | if (!init_context.err) | |
5967 | *thread_ptr = thread; | |
5968 | ||
5969 | return init_context.err; | |
5970 | } |