]>
Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
6aa8b732 AK |
2 | /* |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * This module enables machines with Intel VT-x extensions to run virtual | |
6 | * machines without emulation or binary translation. | |
7 | * | |
8 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 9 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
10 | * |
11 | * Authors: | |
12 | * Avi Kivity <avi@qumranet.com> | |
13 | * Yaniv Kamay <yaniv@qumranet.com> | |
6aa8b732 AK |
14 | */ |
15 | ||
af669ac6 | 16 | #include <kvm/iodev.h> |
6aa8b732 | 17 | |
edf88417 | 18 | #include <linux/kvm_host.h> |
6aa8b732 AK |
19 | #include <linux/kvm.h> |
20 | #include <linux/module.h> | |
21 | #include <linux/errno.h> | |
6aa8b732 | 22 | #include <linux/percpu.h> |
6aa8b732 AK |
23 | #include <linux/mm.h> |
24 | #include <linux/miscdevice.h> | |
25 | #include <linux/vmalloc.h> | |
6aa8b732 | 26 | #include <linux/reboot.h> |
6aa8b732 AK |
27 | #include <linux/debugfs.h> |
28 | #include <linux/highmem.h> | |
29 | #include <linux/file.h> | |
fb3600cc | 30 | #include <linux/syscore_ops.h> |
774c47f1 | 31 | #include <linux/cpu.h> |
174cd4b1 | 32 | #include <linux/sched/signal.h> |
6e84f315 | 33 | #include <linux/sched/mm.h> |
03441a34 | 34 | #include <linux/sched/stat.h> |
d9e368d6 AK |
35 | #include <linux/cpumask.h> |
36 | #include <linux/smp.h> | |
d6d28168 | 37 | #include <linux/anon_inodes.h> |
04d2cc77 | 38 | #include <linux/profile.h> |
7aa81cc0 | 39 | #include <linux/kvm_para.h> |
6fc138d2 | 40 | #include <linux/pagemap.h> |
8d4e1288 | 41 | #include <linux/mman.h> |
35149e21 | 42 | #include <linux/swap.h> |
e56d532f | 43 | #include <linux/bitops.h> |
547de29e | 44 | #include <linux/spinlock.h> |
6ff5894c | 45 | #include <linux/compat.h> |
bc6678a3 | 46 | #include <linux/srcu.h> |
8f0b1ab6 | 47 | #include <linux/hugetlb.h> |
5a0e3ad6 | 48 | #include <linux/slab.h> |
743eeb0b SL |
49 | #include <linux/sort.h> |
50 | #include <linux/bsearch.h> | |
c011d23b | 51 | #include <linux/io.h> |
2eb06c30 | 52 | #include <linux/lockdep.h> |
c57c8046 | 53 | #include <linux/kthread.h> |
2fdef3a2 | 54 | #include <linux/suspend.h> |
6aa8b732 | 55 | |
e495606d | 56 | #include <asm/processor.h> |
2ea75be3 | 57 | #include <asm/ioctl.h> |
7c0f6ba6 | 58 | #include <linux/uaccess.h> |
6aa8b732 | 59 | |
5f94c174 | 60 | #include "coalesced_mmio.h" |
af585b92 | 61 | #include "async_pf.h" |
531810ca | 62 | #include "mmu_lock.h" |
3c3c29fd | 63 | #include "vfio.h" |
5f94c174 | 64 | |
229456fc MT |
65 | #define CREATE_TRACE_POINTS |
66 | #include <trace/events/kvm.h> | |
67 | ||
fb04a1ed PX |
68 | #include <linux/kvm_dirty_ring.h> |
69 | ||
536a6f88 JF |
70 | /* Worst case buffer size needed for holding an integer. */ |
71 | #define ITOA_MAX_LEN 12 | |
72 | ||
6aa8b732 AK |
73 | MODULE_AUTHOR("Qumranet"); |
74 | MODULE_LICENSE("GPL"); | |
75 | ||
920552b2 | 76 | /* Architectures should define their poll value according to the halt latency */ |
ec76d819 | 77 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
039c5d1b | 78 | module_param(halt_poll_ns, uint, 0644); |
ec76d819 | 79 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
f7819512 | 80 | |
aca6ff29 | 81 | /* Default doubles per-vcpu halt_poll_ns. */ |
ec76d819 | 82 | unsigned int halt_poll_ns_grow = 2; |
039c5d1b | 83 | module_param(halt_poll_ns_grow, uint, 0644); |
ec76d819 | 84 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
aca6ff29 | 85 | |
49113d36 NW |
86 | /* The start value to grow halt_poll_ns from */ |
87 | unsigned int halt_poll_ns_grow_start = 10000; /* 10us */ | |
88 | module_param(halt_poll_ns_grow_start, uint, 0644); | |
89 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start); | |
90 | ||
aca6ff29 | 91 | /* Default resets per-vcpu halt_poll_ns . */ |
ec76d819 | 92 | unsigned int halt_poll_ns_shrink; |
039c5d1b | 93 | module_param(halt_poll_ns_shrink, uint, 0644); |
ec76d819 | 94 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
aca6ff29 | 95 | |
fa40a821 MT |
96 | /* |
97 | * Ordering of locks: | |
98 | * | |
b7d409de | 99 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
fa40a821 MT |
100 | */ |
101 | ||
0d9ce162 | 102 | DEFINE_MUTEX(kvm_lock); |
4a937f96 | 103 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
e9b11c17 | 104 | LIST_HEAD(vm_list); |
133de902 | 105 | |
7f59f492 | 106 | static cpumask_var_t cpus_hardware_enabled; |
f4fee932 | 107 | static int kvm_usage_count; |
10474ae8 | 108 | static atomic_t hardware_enable_failed; |
1b6c0168 | 109 | |
aaba298c | 110 | static struct kmem_cache *kvm_vcpu_cache; |
1165f5fe | 111 | |
15ad7146 | 112 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
7495e22b | 113 | static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu); |
15ad7146 | 114 | |
76f7c879 | 115 | struct dentry *kvm_debugfs_dir; |
e23a808b | 116 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
6aa8b732 | 117 | |
09cbcef6 | 118 | static const struct file_operations stat_fops_per_vm; |
536a6f88 | 119 | |
bccf2150 AK |
120 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
121 | unsigned long arg); | |
de8e5d74 | 122 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
123 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
124 | unsigned long arg); | |
7ddfd3e0 MZ |
125 | #define KVM_COMPAT(c) .compat_ioctl = (c) |
126 | #else | |
9cb09e7c MZ |
127 | /* |
128 | * For architectures that don't implement a compat infrastructure, | |
129 | * adopt a double line of defense: | |
130 | * - Prevent a compat task from opening /dev/kvm | |
131 | * - If the open has been done by a 64bit task, and the KVM fd | |
132 | * passed to a compat task, let the ioctls fail. | |
133 | */ | |
7ddfd3e0 MZ |
134 | static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl, |
135 | unsigned long arg) { return -EINVAL; } | |
b9876e6d MZ |
136 | |
137 | static int kvm_no_compat_open(struct inode *inode, struct file *file) | |
138 | { | |
139 | return is_compat_task() ? -ENODEV : 0; | |
140 | } | |
141 | #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \ | |
142 | .open = kvm_no_compat_open | |
1dda606c | 143 | #endif |
10474ae8 AG |
144 | static int hardware_enable_all(void); |
145 | static void hardware_disable_all(void); | |
bccf2150 | 146 | |
e93f8a0f | 147 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
7940876e | 148 | |
52480137 | 149 | __visible bool kvm_rebooting; |
b7c4145b | 150 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
4ecac3fd | 151 | |
286de8f6 CI |
152 | #define KVM_EVENT_CREATE_VM 0 |
153 | #define KVM_EVENT_DESTROY_VM 1 | |
154 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); | |
155 | static unsigned long long kvm_createvm_count; | |
156 | static unsigned long long kvm_active_vms; | |
157 | ||
e649b3f0 ET |
158 | __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
159 | unsigned long start, unsigned long end) | |
b1394e74 RK |
160 | { |
161 | } | |
162 | ||
a78986aa SC |
163 | bool kvm_is_zone_device_pfn(kvm_pfn_t pfn) |
164 | { | |
165 | /* | |
166 | * The metadata used by is_zone_device_page() to determine whether or | |
167 | * not a page is ZONE_DEVICE is guaranteed to be valid if and only if | |
168 | * the device has been pinned, e.g. by get_user_pages(). WARN if the | |
169 | * page_count() is zero to help detect bad usage of this helper. | |
170 | */ | |
171 | if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn)))) | |
172 | return false; | |
173 | ||
174 | return is_zone_device_page(pfn_to_page(pfn)); | |
175 | } | |
176 | ||
ba049e93 | 177 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
cbff90a7 | 178 | { |
a78986aa SC |
179 | /* |
180 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting | |
181 | * perspective they are "normal" pages, albeit with slightly different | |
182 | * usage rules. | |
183 | */ | |
11feeb49 | 184 | if (pfn_valid(pfn)) |
a78986aa | 185 | return PageReserved(pfn_to_page(pfn)) && |
7df003c8 | 186 | !is_zero_pfn(pfn) && |
a78986aa | 187 | !kvm_is_zone_device_pfn(pfn); |
cbff90a7 BAY |
188 | |
189 | return true; | |
190 | } | |
191 | ||
bccf2150 AK |
192 | /* |
193 | * Switches to specified vcpu, until a matching vcpu_put() | |
194 | */ | |
ec7660cc | 195 | void vcpu_load(struct kvm_vcpu *vcpu) |
6aa8b732 | 196 | { |
ec7660cc | 197 | int cpu = get_cpu(); |
7495e22b PB |
198 | |
199 | __this_cpu_write(kvm_running_vcpu, vcpu); | |
15ad7146 | 200 | preempt_notifier_register(&vcpu->preempt_notifier); |
313a3dc7 | 201 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 | 202 | put_cpu(); |
6aa8b732 | 203 | } |
2f1fe811 | 204 | EXPORT_SYMBOL_GPL(vcpu_load); |
6aa8b732 | 205 | |
313a3dc7 | 206 | void vcpu_put(struct kvm_vcpu *vcpu) |
6aa8b732 | 207 | { |
15ad7146 | 208 | preempt_disable(); |
313a3dc7 | 209 | kvm_arch_vcpu_put(vcpu); |
15ad7146 | 210 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
7495e22b | 211 | __this_cpu_write(kvm_running_vcpu, NULL); |
15ad7146 | 212 | preempt_enable(); |
6aa8b732 | 213 | } |
2f1fe811 | 214 | EXPORT_SYMBOL_GPL(vcpu_put); |
6aa8b732 | 215 | |
7a97cec2 PB |
216 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
217 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) | |
218 | { | |
219 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); | |
220 | ||
221 | /* | |
222 | * We need to wait for the VCPU to reenable interrupts and get out of | |
223 | * READING_SHADOW_PAGE_TABLES mode. | |
224 | */ | |
225 | if (req & KVM_REQUEST_WAIT) | |
226 | return mode != OUTSIDE_GUEST_MODE; | |
227 | ||
228 | /* | |
229 | * Need to kick a running VCPU, but otherwise there is nothing to do. | |
230 | */ | |
231 | return mode == IN_GUEST_MODE; | |
232 | } | |
233 | ||
d9e368d6 AK |
234 | static void ack_flush(void *_completed) |
235 | { | |
d9e368d6 AK |
236 | } |
237 | ||
0bbc2ca8 | 238 | static inline bool kvm_kick_many_cpus(cpumask_var_t tmp, bool wait) |
b49defe8 | 239 | { |
0bbc2ca8 SC |
240 | const struct cpumask *cpus; |
241 | ||
242 | if (likely(cpumask_available(tmp))) | |
243 | cpus = tmp; | |
244 | else | |
b49defe8 PB |
245 | cpus = cpu_online_mask; |
246 | ||
247 | if (cpumask_empty(cpus)) | |
248 | return false; | |
249 | ||
250 | smp_call_function_many(cpus, ack_flush, NULL, wait); | |
251 | return true; | |
252 | } | |
253 | ||
7053df4e | 254 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
54163a34 | 255 | struct kvm_vcpu *except, |
7053df4e | 256 | unsigned long *vcpu_bitmap, cpumask_var_t tmp) |
d9e368d6 | 257 | { |
597a5f55 | 258 | int i, cpu, me; |
d9e368d6 | 259 | struct kvm_vcpu *vcpu; |
7053df4e | 260 | bool called; |
6ef7a1bc | 261 | |
3cba4130 | 262 | me = get_cpu(); |
7053df4e | 263 | |
988a2cae | 264 | kvm_for_each_vcpu(i, vcpu, kvm) { |
54163a34 SS |
265 | if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) || |
266 | vcpu == except) | |
7053df4e VK |
267 | continue; |
268 | ||
3cba4130 | 269 | kvm_make_request(req, vcpu); |
6b7e2d09 | 270 | |
178f02ff RK |
271 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) |
272 | continue; | |
6c6e8360 | 273 | |
0bbc2ca8 SC |
274 | /* |
275 | * tmp can be "unavailable" if cpumasks are allocated off stack | |
276 | * as allocation of the mask is deliberately not fatal and is | |
277 | * handled by falling back to kicking all online CPUs. | |
278 | */ | |
279 | if (!cpumask_available(tmp)) | |
280 | continue; | |
281 | ||
85b64045 SC |
282 | /* |
283 | * Note, the vCPU could get migrated to a different pCPU at any | |
284 | * point after kvm_request_needs_ipi(), which could result in | |
285 | * sending an IPI to the previous pCPU. But, that's ok because | |
286 | * the purpose of the IPI is to ensure the vCPU returns to | |
287 | * OUTSIDE_GUEST_MODE, which is satisfied if the vCPU migrates. | |
288 | * Entering READING_SHADOW_PAGE_TABLES after this point is also | |
289 | * ok, as the requirement is only that KVM wait for vCPUs that | |
290 | * were reading SPTEs _before_ any changes were finalized. See | |
291 | * kvm_vcpu_kick() for more details on handling requests. | |
292 | */ | |
0bbc2ca8 | 293 | if (kvm_request_needs_ipi(vcpu, req)) { |
85b64045 SC |
294 | cpu = READ_ONCE(vcpu->cpu); |
295 | if (cpu != -1 && cpu != me) | |
296 | __cpumask_set_cpu(cpu, tmp); | |
297 | } | |
49846896 | 298 | } |
7053df4e VK |
299 | |
300 | called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT)); | |
3cba4130 | 301 | put_cpu(); |
7053df4e VK |
302 | |
303 | return called; | |
304 | } | |
305 | ||
54163a34 SS |
306 | bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req, |
307 | struct kvm_vcpu *except) | |
7053df4e VK |
308 | { |
309 | cpumask_var_t cpus; | |
310 | bool called; | |
7053df4e VK |
311 | |
312 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); | |
313 | ||
54163a34 | 314 | called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus); |
7053df4e | 315 | |
6ef7a1bc | 316 | free_cpumask_var(cpus); |
49846896 | 317 | return called; |
d9e368d6 AK |
318 | } |
319 | ||
54163a34 SS |
320 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
321 | { | |
322 | return kvm_make_all_cpus_request_except(kvm, req, NULL); | |
323 | } | |
a2486020 | 324 | EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request); |
54163a34 | 325 | |
a6d51016 | 326 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
49846896 | 327 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
2e53d63a | 328 | { |
4ae3cb3a LT |
329 | /* |
330 | * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in | |
331 | * kvm_make_all_cpus_request. | |
332 | */ | |
333 | long dirty_count = smp_load_acquire(&kvm->tlbs_dirty); | |
334 | ||
3cc4e148 | 335 | ++kvm->stat.generic.remote_tlb_flush_requests; |
4ae3cb3a LT |
336 | /* |
337 | * We want to publish modifications to the page tables before reading | |
338 | * mode. Pairs with a memory barrier in arch-specific code. | |
339 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest | |
340 | * and smp_mb in walk_shadow_page_lockless_begin/end. | |
341 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. | |
342 | * | |
343 | * There is already an smp_mb__after_atomic() before | |
344 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that | |
345 | * barrier here. | |
346 | */ | |
b08660e5 TL |
347 | if (!kvm_arch_flush_remote_tlb(kvm) |
348 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) | |
0193cc90 | 349 | ++kvm->stat.generic.remote_tlb_flush; |
a086f6a1 | 350 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
2e53d63a | 351 | } |
2ba9f0d8 | 352 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
a6d51016 | 353 | #endif |
2e53d63a | 354 | |
49846896 RR |
355 | void kvm_reload_remote_mmus(struct kvm *kvm) |
356 | { | |
445b8236 | 357 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
49846896 | 358 | } |
2e53d63a | 359 | |
6926f95a SC |
360 | #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE |
361 | static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc, | |
362 | gfp_t gfp_flags) | |
363 | { | |
364 | gfp_flags |= mc->gfp_zero; | |
365 | ||
366 | if (mc->kmem_cache) | |
367 | return kmem_cache_alloc(mc->kmem_cache, gfp_flags); | |
368 | else | |
369 | return (void *)__get_free_page(gfp_flags); | |
370 | } | |
371 | ||
372 | int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min) | |
373 | { | |
374 | void *obj; | |
375 | ||
376 | if (mc->nobjs >= min) | |
377 | return 0; | |
378 | while (mc->nobjs < ARRAY_SIZE(mc->objects)) { | |
379 | obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT); | |
380 | if (!obj) | |
381 | return mc->nobjs >= min ? 0 : -ENOMEM; | |
382 | mc->objects[mc->nobjs++] = obj; | |
383 | } | |
384 | return 0; | |
385 | } | |
386 | ||
387 | int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc) | |
388 | { | |
389 | return mc->nobjs; | |
390 | } | |
391 | ||
392 | void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
393 | { | |
394 | while (mc->nobjs) { | |
395 | if (mc->kmem_cache) | |
396 | kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]); | |
397 | else | |
398 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
399 | } | |
400 | } | |
401 | ||
402 | void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
403 | { | |
404 | void *p; | |
405 | ||
406 | if (WARN_ON(!mc->nobjs)) | |
407 | p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT); | |
408 | else | |
409 | p = mc->objects[--mc->nobjs]; | |
410 | BUG_ON(!p); | |
411 | return p; | |
412 | } | |
413 | #endif | |
414 | ||
8bd826d6 | 415 | static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
fb3f0f51 | 416 | { |
fb3f0f51 RR |
417 | mutex_init(&vcpu->mutex); |
418 | vcpu->cpu = -1; | |
fb3f0f51 RR |
419 | vcpu->kvm = kvm; |
420 | vcpu->vcpu_id = id; | |
34bb10b7 | 421 | vcpu->pid = NULL; |
da4ad88c | 422 | rcuwait_init(&vcpu->wait); |
af585b92 | 423 | kvm_async_pf_vcpu_init(vcpu); |
fb3f0f51 | 424 | |
bf9f6ac8 FW |
425 | vcpu->pre_pcpu = -1; |
426 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); | |
427 | ||
4c088493 R |
428 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
429 | kvm_vcpu_set_dy_eligible(vcpu, false); | |
3a08a8f9 | 430 | vcpu->preempted = false; |
d73eb57b | 431 | vcpu->ready = false; |
d5c48deb | 432 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
fe22ed82 | 433 | vcpu->last_used_slot = 0; |
fb3f0f51 | 434 | } |
fb3f0f51 | 435 | |
4543bdc0 SC |
436 | void kvm_vcpu_destroy(struct kvm_vcpu *vcpu) |
437 | { | |
fb04a1ed | 438 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
4543bdc0 | 439 | kvm_arch_vcpu_destroy(vcpu); |
e529ef66 | 440 | |
9941d224 SC |
441 | /* |
442 | * No need for rcu_read_lock as VCPU_RUN is the only place that changes | |
443 | * the vcpu->pid pointer, and at destruction time all file descriptors | |
444 | * are already gone. | |
445 | */ | |
446 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); | |
447 | ||
8bd826d6 | 448 | free_page((unsigned long)vcpu->run); |
e529ef66 | 449 | kmem_cache_free(kvm_vcpu_cache, vcpu); |
4543bdc0 SC |
450 | } |
451 | EXPORT_SYMBOL_GPL(kvm_vcpu_destroy); | |
452 | ||
e930bffe AA |
453 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
454 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) | |
455 | { | |
456 | return container_of(mn, struct kvm, mmu_notifier); | |
457 | } | |
458 | ||
e649b3f0 ET |
459 | static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn, |
460 | struct mm_struct *mm, | |
461 | unsigned long start, unsigned long end) | |
462 | { | |
463 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
464 | int idx; | |
465 | ||
466 | idx = srcu_read_lock(&kvm->srcu); | |
467 | kvm_arch_mmu_notifier_invalidate_range(kvm, start, end); | |
468 | srcu_read_unlock(&kvm->srcu, idx); | |
469 | } | |
470 | ||
3039bcc7 SC |
471 | typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range); |
472 | ||
f922bd9b SC |
473 | typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start, |
474 | unsigned long end); | |
475 | ||
3039bcc7 SC |
476 | struct kvm_hva_range { |
477 | unsigned long start; | |
478 | unsigned long end; | |
479 | pte_t pte; | |
480 | hva_handler_t handler; | |
f922bd9b | 481 | on_lock_fn_t on_lock; |
3039bcc7 SC |
482 | bool flush_on_ret; |
483 | bool may_block; | |
484 | }; | |
485 | ||
f922bd9b SC |
486 | /* |
487 | * Use a dedicated stub instead of NULL to indicate that there is no callback | |
488 | * function/handler. The compiler technically can't guarantee that a real | |
489 | * function will have a non-zero address, and so it will generate code to | |
490 | * check for !NULL, whereas comparing against a stub will be elided at compile | |
491 | * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9). | |
492 | */ | |
493 | static void kvm_null_fn(void) | |
494 | { | |
495 | ||
496 | } | |
497 | #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn) | |
498 | ||
3039bcc7 SC |
499 | static __always_inline int __kvm_handle_hva_range(struct kvm *kvm, |
500 | const struct kvm_hva_range *range) | |
501 | { | |
8931a454 | 502 | bool ret = false, locked = false; |
f922bd9b | 503 | struct kvm_gfn_range gfn_range; |
3039bcc7 SC |
504 | struct kvm_memory_slot *slot; |
505 | struct kvm_memslots *slots; | |
3039bcc7 SC |
506 | int i, idx; |
507 | ||
f922bd9b SC |
508 | /* A null handler is allowed if and only if on_lock() is provided. */ |
509 | if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) && | |
510 | IS_KVM_NULL_FN(range->handler))) | |
511 | return 0; | |
512 | ||
3039bcc7 SC |
513 | idx = srcu_read_lock(&kvm->srcu); |
514 | ||
515 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
516 | slots = __kvm_memslots(kvm, i); | |
517 | kvm_for_each_memslot(slot, slots) { | |
518 | unsigned long hva_start, hva_end; | |
519 | ||
520 | hva_start = max(range->start, slot->userspace_addr); | |
521 | hva_end = min(range->end, slot->userspace_addr + | |
522 | (slot->npages << PAGE_SHIFT)); | |
523 | if (hva_start >= hva_end) | |
524 | continue; | |
525 | ||
526 | /* | |
527 | * To optimize for the likely case where the address | |
528 | * range is covered by zero or one memslots, don't | |
529 | * bother making these conditional (to avoid writes on | |
530 | * the second or later invocation of the handler). | |
531 | */ | |
532 | gfn_range.pte = range->pte; | |
533 | gfn_range.may_block = range->may_block; | |
534 | ||
535 | /* | |
536 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
537 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
538 | */ | |
539 | gfn_range.start = hva_to_gfn_memslot(hva_start, slot); | |
540 | gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot); | |
541 | gfn_range.slot = slot; | |
542 | ||
8931a454 SC |
543 | if (!locked) { |
544 | locked = true; | |
545 | KVM_MMU_LOCK(kvm); | |
071064f1 PB |
546 | if (!IS_KVM_NULL_FN(range->on_lock)) |
547 | range->on_lock(kvm, range->start, range->end); | |
548 | if (IS_KVM_NULL_FN(range->handler)) | |
549 | break; | |
8931a454 | 550 | } |
3039bcc7 SC |
551 | ret |= range->handler(kvm, &gfn_range); |
552 | } | |
553 | } | |
554 | ||
555 | if (range->flush_on_ret && (ret || kvm->tlbs_dirty)) | |
556 | kvm_flush_remote_tlbs(kvm); | |
557 | ||
8931a454 SC |
558 | if (locked) |
559 | KVM_MMU_UNLOCK(kvm); | |
f922bd9b | 560 | |
3039bcc7 SC |
561 | srcu_read_unlock(&kvm->srcu, idx); |
562 | ||
563 | /* The notifiers are averse to booleans. :-( */ | |
564 | return (int)ret; | |
565 | } | |
566 | ||
567 | static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn, | |
568 | unsigned long start, | |
569 | unsigned long end, | |
570 | pte_t pte, | |
571 | hva_handler_t handler) | |
572 | { | |
573 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
574 | const struct kvm_hva_range range = { | |
575 | .start = start, | |
576 | .end = end, | |
577 | .pte = pte, | |
578 | .handler = handler, | |
f922bd9b | 579 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
580 | .flush_on_ret = true, |
581 | .may_block = false, | |
582 | }; | |
3039bcc7 | 583 | |
f922bd9b | 584 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 SC |
585 | } |
586 | ||
587 | static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn, | |
588 | unsigned long start, | |
589 | unsigned long end, | |
590 | hva_handler_t handler) | |
591 | { | |
592 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
593 | const struct kvm_hva_range range = { | |
594 | .start = start, | |
595 | .end = end, | |
596 | .pte = __pte(0), | |
597 | .handler = handler, | |
f922bd9b | 598 | .on_lock = (void *)kvm_null_fn, |
3039bcc7 SC |
599 | .flush_on_ret = false, |
600 | .may_block = false, | |
601 | }; | |
3039bcc7 | 602 | |
f922bd9b | 603 | return __kvm_handle_hva_range(kvm, &range); |
3039bcc7 | 604 | } |
3da0dd43 IE |
605 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
606 | struct mm_struct *mm, | |
607 | unsigned long address, | |
608 | pte_t pte) | |
609 | { | |
610 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
611 | ||
501b9185 SC |
612 | trace_kvm_set_spte_hva(address); |
613 | ||
c13fda23 | 614 | /* |
52ac8b35 | 615 | * .change_pte() must be surrounded by .invalidate_range_{start,end}(). |
071064f1 PB |
616 | * If mmu_notifier_count is zero, then no in-progress invalidations, |
617 | * including this one, found a relevant memslot at start(); rechecking | |
618 | * memslots here is unnecessary. Note, a false positive (count elevated | |
619 | * by a different invalidation) is sub-optimal but functionally ok. | |
c13fda23 | 620 | */ |
52ac8b35 | 621 | WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count)); |
071064f1 PB |
622 | if (!READ_ONCE(kvm->mmu_notifier_count)) |
623 | return; | |
c13fda23 | 624 | |
3039bcc7 | 625 | kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn); |
3da0dd43 IE |
626 | } |
627 | ||
edb298c6 | 628 | void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 629 | unsigned long end) |
e930bffe | 630 | { |
e930bffe AA |
631 | /* |
632 | * The count increase must become visible at unlock time as no | |
633 | * spte can be established without taking the mmu_lock and | |
634 | * count is also read inside the mmu_lock critical section. | |
635 | */ | |
636 | kvm->mmu_notifier_count++; | |
4a42d848 | 637 | if (likely(kvm->mmu_notifier_count == 1)) { |
f922bd9b SC |
638 | kvm->mmu_notifier_range_start = start; |
639 | kvm->mmu_notifier_range_end = end; | |
4a42d848 DS |
640 | } else { |
641 | /* | |
642 | * Fully tracking multiple concurrent ranges has dimishing | |
643 | * returns. Keep things simple and just find the minimal range | |
644 | * which includes the current and new ranges. As there won't be | |
645 | * enough information to subtract a range after its invalidate | |
646 | * completes, any ranges invalidated concurrently will | |
647 | * accumulate and persist until all outstanding invalidates | |
648 | * complete. | |
649 | */ | |
650 | kvm->mmu_notifier_range_start = | |
f922bd9b | 651 | min(kvm->mmu_notifier_range_start, start); |
4a42d848 | 652 | kvm->mmu_notifier_range_end = |
f922bd9b | 653 | max(kvm->mmu_notifier_range_end, end); |
4a42d848 | 654 | } |
f922bd9b | 655 | } |
3039bcc7 | 656 | |
f922bd9b SC |
657 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
658 | const struct mmu_notifier_range *range) | |
659 | { | |
660 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
661 | const struct kvm_hva_range hva_range = { | |
662 | .start = range->start, | |
663 | .end = range->end, | |
664 | .pte = __pte(0), | |
665 | .handler = kvm_unmap_gfn_range, | |
666 | .on_lock = kvm_inc_notifier_count, | |
667 | .flush_on_ret = true, | |
668 | .may_block = mmu_notifier_range_blockable(range), | |
669 | }; | |
565f3be2 | 670 | |
f922bd9b SC |
671 | trace_kvm_unmap_hva_range(range->start, range->end); |
672 | ||
52ac8b35 PB |
673 | /* |
674 | * Prevent memslot modification between range_start() and range_end() | |
675 | * so that conditionally locking provides the same result in both | |
676 | * functions. Without that guarantee, the mmu_notifier_count | |
677 | * adjustments will be imbalanced. | |
678 | * | |
679 | * Pairs with the decrement in range_end(). | |
680 | */ | |
681 | spin_lock(&kvm->mn_invalidate_lock); | |
682 | kvm->mn_active_invalidate_count++; | |
683 | spin_unlock(&kvm->mn_invalidate_lock); | |
684 | ||
f922bd9b | 685 | __kvm_handle_hva_range(kvm, &hva_range); |
93065ac7 | 686 | |
e649b3f0 | 687 | return 0; |
e930bffe AA |
688 | } |
689 | ||
edb298c6 | 690 | void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start, |
f922bd9b | 691 | unsigned long end) |
e930bffe | 692 | { |
e930bffe AA |
693 | /* |
694 | * This sequence increase will notify the kvm page fault that | |
695 | * the page that is going to be mapped in the spte could have | |
696 | * been freed. | |
697 | */ | |
698 | kvm->mmu_notifier_seq++; | |
a355aa54 | 699 | smp_wmb(); |
e930bffe AA |
700 | /* |
701 | * The above sequence increase must be visible before the | |
a355aa54 PM |
702 | * below count decrease, which is ensured by the smp_wmb above |
703 | * in conjunction with the smp_rmb in mmu_notifier_retry(). | |
e930bffe AA |
704 | */ |
705 | kvm->mmu_notifier_count--; | |
f922bd9b SC |
706 | } |
707 | ||
708 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, | |
709 | const struct mmu_notifier_range *range) | |
710 | { | |
711 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
712 | const struct kvm_hva_range hva_range = { | |
713 | .start = range->start, | |
714 | .end = range->end, | |
715 | .pte = __pte(0), | |
716 | .handler = (void *)kvm_null_fn, | |
717 | .on_lock = kvm_dec_notifier_count, | |
718 | .flush_on_ret = false, | |
719 | .may_block = mmu_notifier_range_blockable(range), | |
720 | }; | |
52ac8b35 | 721 | bool wake; |
f922bd9b SC |
722 | |
723 | __kvm_handle_hva_range(kvm, &hva_range); | |
e930bffe | 724 | |
52ac8b35 PB |
725 | /* Pairs with the increment in range_start(). */ |
726 | spin_lock(&kvm->mn_invalidate_lock); | |
727 | wake = (--kvm->mn_active_invalidate_count == 0); | |
728 | spin_unlock(&kvm->mn_invalidate_lock); | |
729 | ||
730 | /* | |
731 | * There can only be one waiter, since the wait happens under | |
732 | * slots_lock. | |
733 | */ | |
734 | if (wake) | |
735 | rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait); | |
736 | ||
e930bffe AA |
737 | BUG_ON(kvm->mmu_notifier_count < 0); |
738 | } | |
739 | ||
740 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, | |
741 | struct mm_struct *mm, | |
57128468 ALC |
742 | unsigned long start, |
743 | unsigned long end) | |
e930bffe | 744 | { |
501b9185 SC |
745 | trace_kvm_age_hva(start, end); |
746 | ||
3039bcc7 | 747 | return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn); |
e930bffe AA |
748 | } |
749 | ||
1d7715c6 VD |
750 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
751 | struct mm_struct *mm, | |
752 | unsigned long start, | |
753 | unsigned long end) | |
754 | { | |
501b9185 SC |
755 | trace_kvm_age_hva(start, end); |
756 | ||
1d7715c6 VD |
757 | /* |
758 | * Even though we do not flush TLB, this will still adversely | |
759 | * affect performance on pre-Haswell Intel EPT, where there is | |
760 | * no EPT Access Bit to clear so that we have to tear down EPT | |
761 | * tables instead. If we find this unacceptable, we can always | |
762 | * add a parameter to kvm_age_hva so that it effectively doesn't | |
763 | * do anything on clear_young. | |
764 | * | |
765 | * Also note that currently we never issue secondary TLB flushes | |
766 | * from clear_young, leaving this job up to the regular system | |
767 | * cadence. If we find this inaccurate, we might come up with a | |
768 | * more sophisticated heuristic later. | |
769 | */ | |
3039bcc7 | 770 | return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn); |
1d7715c6 VD |
771 | } |
772 | ||
8ee53820 AA |
773 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
774 | struct mm_struct *mm, | |
775 | unsigned long address) | |
776 | { | |
501b9185 SC |
777 | trace_kvm_test_age_hva(address); |
778 | ||
3039bcc7 SC |
779 | return kvm_handle_hva_range_no_flush(mn, address, address + 1, |
780 | kvm_test_age_gfn); | |
8ee53820 AA |
781 | } |
782 | ||
85db06e5 MT |
783 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
784 | struct mm_struct *mm) | |
785 | { | |
786 | struct kvm *kvm = mmu_notifier_to_kvm(mn); | |
eda2beda LJ |
787 | int idx; |
788 | ||
789 | idx = srcu_read_lock(&kvm->srcu); | |
2df72e9b | 790 | kvm_arch_flush_shadow_all(kvm); |
eda2beda | 791 | srcu_read_unlock(&kvm->srcu, idx); |
85db06e5 MT |
792 | } |
793 | ||
e930bffe | 794 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
e649b3f0 | 795 | .invalidate_range = kvm_mmu_notifier_invalidate_range, |
e930bffe AA |
796 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
797 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, | |
798 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, | |
1d7715c6 | 799 | .clear_young = kvm_mmu_notifier_clear_young, |
8ee53820 | 800 | .test_young = kvm_mmu_notifier_test_young, |
3da0dd43 | 801 | .change_pte = kvm_mmu_notifier_change_pte, |
85db06e5 | 802 | .release = kvm_mmu_notifier_release, |
e930bffe | 803 | }; |
4c07b0a4 AK |
804 | |
805 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
806 | { | |
807 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; | |
808 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); | |
809 | } | |
810 | ||
811 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ | |
812 | ||
813 | static int kvm_init_mmu_notifier(struct kvm *kvm) | |
814 | { | |
815 | return 0; | |
816 | } | |
817 | ||
e930bffe AA |
818 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
819 | ||
2fdef3a2 SS |
820 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
821 | static int kvm_pm_notifier_call(struct notifier_block *bl, | |
822 | unsigned long state, | |
823 | void *unused) | |
824 | { | |
825 | struct kvm *kvm = container_of(bl, struct kvm, pm_notifier); | |
826 | ||
827 | return kvm_arch_pm_notifier(kvm, state); | |
828 | } | |
829 | ||
830 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
831 | { | |
832 | kvm->pm_notifier.notifier_call = kvm_pm_notifier_call; | |
833 | /* Suspend KVM before we suspend ftrace, RCU, etc. */ | |
834 | kvm->pm_notifier.priority = INT_MAX; | |
835 | register_pm_notifier(&kvm->pm_notifier); | |
836 | } | |
837 | ||
838 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
839 | { | |
840 | unregister_pm_notifier(&kvm->pm_notifier); | |
841 | } | |
842 | #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
843 | static void kvm_init_pm_notifier(struct kvm *kvm) | |
844 | { | |
845 | } | |
846 | ||
847 | static void kvm_destroy_pm_notifier(struct kvm *kvm) | |
848 | { | |
849 | } | |
850 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
851 | ||
a47d2b07 | 852 | static struct kvm_memslots *kvm_alloc_memslots(void) |
bf3e05bc XG |
853 | { |
854 | int i; | |
a47d2b07 | 855 | struct kvm_memslots *slots; |
bf3e05bc | 856 | |
b12ce36a | 857 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
a47d2b07 PB |
858 | if (!slots) |
859 | return NULL; | |
860 | ||
bf3e05bc | 861 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
36947254 | 862 | slots->id_to_index[i] = -1; |
a47d2b07 PB |
863 | |
864 | return slots; | |
865 | } | |
866 | ||
867 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) | |
868 | { | |
869 | if (!memslot->dirty_bitmap) | |
870 | return; | |
871 | ||
872 | kvfree(memslot->dirty_bitmap); | |
873 | memslot->dirty_bitmap = NULL; | |
874 | } | |
875 | ||
e96c81ee | 876 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
a47d2b07 | 877 | { |
e96c81ee | 878 | kvm_destroy_dirty_bitmap(slot); |
a47d2b07 | 879 | |
e96c81ee | 880 | kvm_arch_free_memslot(kvm, slot); |
a47d2b07 | 881 | |
e96c81ee SC |
882 | slot->flags = 0; |
883 | slot->npages = 0; | |
a47d2b07 PB |
884 | } |
885 | ||
886 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) | |
887 | { | |
888 | struct kvm_memory_slot *memslot; | |
889 | ||
890 | if (!slots) | |
891 | return; | |
892 | ||
893 | kvm_for_each_memslot(memslot, slots) | |
e96c81ee | 894 | kvm_free_memslot(kvm, memslot); |
a47d2b07 PB |
895 | |
896 | kvfree(slots); | |
bf3e05bc XG |
897 | } |
898 | ||
bc9e9e67 JZ |
899 | static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc) |
900 | { | |
901 | switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) { | |
902 | case KVM_STATS_TYPE_INSTANT: | |
903 | return 0444; | |
904 | case KVM_STATS_TYPE_CUMULATIVE: | |
905 | case KVM_STATS_TYPE_PEAK: | |
906 | default: | |
907 | return 0644; | |
908 | } | |
909 | } | |
910 | ||
911 | ||
536a6f88 JF |
912 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
913 | { | |
914 | int i; | |
bc9e9e67 JZ |
915 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
916 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
917 | |
918 | if (!kvm->debugfs_dentry) | |
919 | return; | |
920 | ||
921 | debugfs_remove_recursive(kvm->debugfs_dentry); | |
922 | ||
9d5a1dce LC |
923 | if (kvm->debugfs_stat_data) { |
924 | for (i = 0; i < kvm_debugfs_num_entries; i++) | |
925 | kfree(kvm->debugfs_stat_data[i]); | |
926 | kfree(kvm->debugfs_stat_data); | |
927 | } | |
536a6f88 JF |
928 | } |
929 | ||
930 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) | |
931 | { | |
85cd39af PB |
932 | static DEFINE_MUTEX(kvm_debugfs_lock); |
933 | struct dentry *dent; | |
536a6f88 JF |
934 | char dir_name[ITOA_MAX_LEN * 2]; |
935 | struct kvm_stat_data *stat_data; | |
bc9e9e67 | 936 | const struct _kvm_stats_desc *pdesc; |
3165af73 | 937 | int i, ret; |
bc9e9e67 JZ |
938 | int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc + |
939 | kvm_vcpu_stats_header.num_desc; | |
536a6f88 JF |
940 | |
941 | if (!debugfs_initialized()) | |
942 | return 0; | |
943 | ||
944 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); | |
85cd39af PB |
945 | mutex_lock(&kvm_debugfs_lock); |
946 | dent = debugfs_lookup(dir_name, kvm_debugfs_dir); | |
947 | if (dent) { | |
948 | pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name); | |
949 | dput(dent); | |
950 | mutex_unlock(&kvm_debugfs_lock); | |
951 | return 0; | |
952 | } | |
953 | dent = debugfs_create_dir(dir_name, kvm_debugfs_dir); | |
954 | mutex_unlock(&kvm_debugfs_lock); | |
955 | if (IS_ERR(dent)) | |
956 | return 0; | |
536a6f88 | 957 | |
85cd39af | 958 | kvm->debugfs_dentry = dent; |
536a6f88 JF |
959 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
960 | sizeof(*kvm->debugfs_stat_data), | |
b12ce36a | 961 | GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
962 | if (!kvm->debugfs_stat_data) |
963 | return -ENOMEM; | |
964 | ||
bc9e9e67 JZ |
965 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
966 | pdesc = &kvm_vm_stats_desc[i]; | |
b12ce36a | 967 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
968 | if (!stat_data) |
969 | return -ENOMEM; | |
970 | ||
971 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
972 | stat_data->desc = pdesc; |
973 | stat_data->kind = KVM_STAT_VM; | |
974 | kvm->debugfs_stat_data[i] = stat_data; | |
975 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
976 | kvm->debugfs_dentry, stat_data, | |
977 | &stat_fops_per_vm); | |
978 | } | |
979 | ||
980 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
981 | pdesc = &kvm_vcpu_stats_desc[i]; | |
b12ce36a | 982 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
536a6f88 JF |
983 | if (!stat_data) |
984 | return -ENOMEM; | |
985 | ||
986 | stat_data->kvm = kvm; | |
bc9e9e67 JZ |
987 | stat_data->desc = pdesc; |
988 | stat_data->kind = KVM_STAT_VCPU; | |
004d62eb | 989 | kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data; |
bc9e9e67 | 990 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), |
09cbcef6 MP |
991 | kvm->debugfs_dentry, stat_data, |
992 | &stat_fops_per_vm); | |
536a6f88 | 993 | } |
3165af73 PX |
994 | |
995 | ret = kvm_arch_create_vm_debugfs(kvm); | |
996 | if (ret) { | |
997 | kvm_destroy_vm_debugfs(kvm); | |
998 | return i; | |
999 | } | |
1000 | ||
536a6f88 JF |
1001 | return 0; |
1002 | } | |
1003 | ||
1aa9b957 JS |
1004 | /* |
1005 | * Called after the VM is otherwise initialized, but just before adding it to | |
1006 | * the vm_list. | |
1007 | */ | |
1008 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) | |
1009 | { | |
1010 | return 0; | |
1011 | } | |
1012 | ||
1013 | /* | |
1014 | * Called just after removing the VM from the vm_list, but before doing any | |
1015 | * other destruction. | |
1016 | */ | |
1017 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) | |
1018 | { | |
1019 | } | |
1020 | ||
3165af73 PX |
1021 | /* |
1022 | * Called after per-vm debugfs created. When called kvm->debugfs_dentry should | |
1023 | * be setup already, so we can create arch-specific debugfs entries under it. | |
1024 | * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so | |
1025 | * a per-arch destroy interface is not needed. | |
1026 | */ | |
1027 | int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm) | |
1028 | { | |
1029 | return 0; | |
1030 | } | |
1031 | ||
e08b9637 | 1032 | static struct kvm *kvm_create_vm(unsigned long type) |
6aa8b732 | 1033 | { |
d89f5eff | 1034 | struct kvm *kvm = kvm_arch_alloc_vm(); |
9121923c JM |
1035 | int r = -ENOMEM; |
1036 | int i; | |
6aa8b732 | 1037 | |
d89f5eff JK |
1038 | if (!kvm) |
1039 | return ERR_PTR(-ENOMEM); | |
1040 | ||
531810ca | 1041 | KVM_MMU_LOCK_INIT(kvm); |
f1f10076 | 1042 | mmgrab(current->mm); |
e9ad4ec8 PB |
1043 | kvm->mm = current->mm; |
1044 | kvm_eventfd_init(kvm); | |
1045 | mutex_init(&kvm->lock); | |
1046 | mutex_init(&kvm->irq_lock); | |
1047 | mutex_init(&kvm->slots_lock); | |
b10a038e | 1048 | mutex_init(&kvm->slots_arch_lock); |
52ac8b35 PB |
1049 | spin_lock_init(&kvm->mn_invalidate_lock); |
1050 | rcuwait_init(&kvm->mn_memslots_update_rcuwait); | |
1051 | ||
e9ad4ec8 PB |
1052 | INIT_LIST_HEAD(&kvm->devices); |
1053 | ||
1e702d9a AW |
1054 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
1055 | ||
8a44119a PB |
1056 | if (init_srcu_struct(&kvm->srcu)) |
1057 | goto out_err_no_srcu; | |
1058 | if (init_srcu_struct(&kvm->irq_srcu)) | |
1059 | goto out_err_no_irq_srcu; | |
1060 | ||
e2d3fcaf | 1061 | refcount_set(&kvm->users_count, 1); |
f481b069 | 1062 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4bd518f1 | 1063 | struct kvm_memslots *slots = kvm_alloc_memslots(); |
9121923c | 1064 | |
4bd518f1 | 1065 | if (!slots) |
a97b0e77 | 1066 | goto out_err_no_arch_destroy_vm; |
0e32958e | 1067 | /* Generations must be different for each address space. */ |
164bf7e5 | 1068 | slots->generation = i; |
4bd518f1 | 1069 | rcu_assign_pointer(kvm->memslots[i], slots); |
f481b069 | 1070 | } |
00f034a1 | 1071 | |
e93f8a0f | 1072 | for (i = 0; i < KVM_NR_BUSES; i++) { |
4a12f951 | 1073 | rcu_assign_pointer(kvm->buses[i], |
b12ce36a | 1074 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
57e7fbee | 1075 | if (!kvm->buses[i]) |
a97b0e77 | 1076 | goto out_err_no_arch_destroy_vm; |
e93f8a0f | 1077 | } |
e930bffe | 1078 | |
acd05785 DM |
1079 | kvm->max_halt_poll_ns = halt_poll_ns; |
1080 | ||
e08b9637 | 1081 | r = kvm_arch_init_vm(kvm, type); |
d89f5eff | 1082 | if (r) |
a97b0e77 | 1083 | goto out_err_no_arch_destroy_vm; |
10474ae8 AG |
1084 | |
1085 | r = hardware_enable_all(); | |
1086 | if (r) | |
719d93cd | 1087 | goto out_err_no_disable; |
10474ae8 | 1088 | |
c77dcacb | 1089 | #ifdef CONFIG_HAVE_KVM_IRQFD |
136bdfee | 1090 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
75858a84 | 1091 | #endif |
6aa8b732 | 1092 | |
74b5c5bf | 1093 | r = kvm_init_mmu_notifier(kvm); |
1aa9b957 JS |
1094 | if (r) |
1095 | goto out_err_no_mmu_notifier; | |
1096 | ||
1097 | r = kvm_arch_post_init_vm(kvm); | |
74b5c5bf MW |
1098 | if (r) |
1099 | goto out_err; | |
1100 | ||
0d9ce162 | 1101 | mutex_lock(&kvm_lock); |
5e58cfe4 | 1102 | list_add(&kvm->vm_list, &vm_list); |
0d9ce162 | 1103 | mutex_unlock(&kvm_lock); |
d89f5eff | 1104 | |
2ecd9d29 | 1105 | preempt_notifier_inc(); |
2fdef3a2 | 1106 | kvm_init_pm_notifier(kvm); |
2ecd9d29 | 1107 | |
f17abe9a | 1108 | return kvm; |
10474ae8 AG |
1109 | |
1110 | out_err: | |
1aa9b957 JS |
1111 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1112 | if (kvm->mmu_notifier.ops) | |
1113 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); | |
1114 | #endif | |
1115 | out_err_no_mmu_notifier: | |
10474ae8 | 1116 | hardware_disable_all(); |
719d93cd | 1117 | out_err_no_disable: |
a97b0e77 | 1118 | kvm_arch_destroy_vm(kvm); |
a97b0e77 | 1119 | out_err_no_arch_destroy_vm: |
e2d3fcaf | 1120 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
e93f8a0f | 1121 | for (i = 0; i < KVM_NR_BUSES; i++) |
3898da94 | 1122 | kfree(kvm_get_bus(kvm, i)); |
f481b069 | 1123 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 1124 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
8a44119a PB |
1125 | cleanup_srcu_struct(&kvm->irq_srcu); |
1126 | out_err_no_irq_srcu: | |
1127 | cleanup_srcu_struct(&kvm->srcu); | |
1128 | out_err_no_srcu: | |
d89f5eff | 1129 | kvm_arch_free_vm(kvm); |
e9ad4ec8 | 1130 | mmdrop(current->mm); |
10474ae8 | 1131 | return ERR_PTR(r); |
f17abe9a AK |
1132 | } |
1133 | ||
07f0a7bd SW |
1134 | static void kvm_destroy_devices(struct kvm *kvm) |
1135 | { | |
e6e3b5a6 | 1136 | struct kvm_device *dev, *tmp; |
07f0a7bd | 1137 | |
a28ebea2 CD |
1138 | /* |
1139 | * We do not need to take the kvm->lock here, because nobody else | |
1140 | * has a reference to the struct kvm at this point and therefore | |
1141 | * cannot access the devices list anyhow. | |
1142 | */ | |
e6e3b5a6 GT |
1143 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
1144 | list_del(&dev->vm_node); | |
07f0a7bd SW |
1145 | dev->ops->destroy(dev); |
1146 | } | |
1147 | } | |
1148 | ||
f17abe9a AK |
1149 | static void kvm_destroy_vm(struct kvm *kvm) |
1150 | { | |
e93f8a0f | 1151 | int i; |
6d4e4c4f AK |
1152 | struct mm_struct *mm = kvm->mm; |
1153 | ||
2fdef3a2 | 1154 | kvm_destroy_pm_notifier(kvm); |
286de8f6 | 1155 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
536a6f88 | 1156 | kvm_destroy_vm_debugfs(kvm); |
ad8ba2cd | 1157 | kvm_arch_sync_events(kvm); |
0d9ce162 | 1158 | mutex_lock(&kvm_lock); |
133de902 | 1159 | list_del(&kvm->vm_list); |
0d9ce162 | 1160 | mutex_unlock(&kvm_lock); |
1aa9b957 JS |
1161 | kvm_arch_pre_destroy_vm(kvm); |
1162 | ||
399ec807 | 1163 | kvm_free_irq_routing(kvm); |
df630b8c | 1164 | for (i = 0; i < KVM_NR_BUSES; i++) { |
3898da94 | 1165 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
4a12f951 | 1166 | |
4a12f951 CB |
1167 | if (bus) |
1168 | kvm_io_bus_destroy(bus); | |
df630b8c PX |
1169 | kvm->buses[i] = NULL; |
1170 | } | |
980da6ce | 1171 | kvm_coalesced_mmio_free(kvm); |
e930bffe AA |
1172 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
1173 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); | |
52ac8b35 PB |
1174 | /* |
1175 | * At this point, pending calls to invalidate_range_start() | |
1176 | * have completed but no more MMU notifiers will run, so | |
1177 | * mn_active_invalidate_count may remain unbalanced. | |
1178 | * No threads can be waiting in install_new_memslots as the | |
1179 | * last reference on KVM has been dropped, but freeing | |
1180 | * memslots would deadlock without this manual intervention. | |
1181 | */ | |
1182 | WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait)); | |
1183 | kvm->mn_active_invalidate_count = 0; | |
f00be0ca | 1184 | #else |
2df72e9b | 1185 | kvm_arch_flush_shadow_all(kvm); |
5f94c174 | 1186 | #endif |
d19a9cd2 | 1187 | kvm_arch_destroy_vm(kvm); |
07f0a7bd | 1188 | kvm_destroy_devices(kvm); |
f481b069 | 1189 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
3898da94 | 1190 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
820b3fcd | 1191 | cleanup_srcu_struct(&kvm->irq_srcu); |
d89f5eff JK |
1192 | cleanup_srcu_struct(&kvm->srcu); |
1193 | kvm_arch_free_vm(kvm); | |
2ecd9d29 | 1194 | preempt_notifier_dec(); |
10474ae8 | 1195 | hardware_disable_all(); |
6d4e4c4f | 1196 | mmdrop(mm); |
f17abe9a AK |
1197 | } |
1198 | ||
d39f13b0 IE |
1199 | void kvm_get_kvm(struct kvm *kvm) |
1200 | { | |
e3736c3e | 1201 | refcount_inc(&kvm->users_count); |
d39f13b0 IE |
1202 | } |
1203 | EXPORT_SYMBOL_GPL(kvm_get_kvm); | |
1204 | ||
605c7130 PX |
1205 | /* |
1206 | * Make sure the vm is not during destruction, which is a safe version of | |
1207 | * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise. | |
1208 | */ | |
1209 | bool kvm_get_kvm_safe(struct kvm *kvm) | |
1210 | { | |
1211 | return refcount_inc_not_zero(&kvm->users_count); | |
1212 | } | |
1213 | EXPORT_SYMBOL_GPL(kvm_get_kvm_safe); | |
1214 | ||
d39f13b0 IE |
1215 | void kvm_put_kvm(struct kvm *kvm) |
1216 | { | |
e3736c3e | 1217 | if (refcount_dec_and_test(&kvm->users_count)) |
d39f13b0 IE |
1218 | kvm_destroy_vm(kvm); |
1219 | } | |
1220 | EXPORT_SYMBOL_GPL(kvm_put_kvm); | |
1221 | ||
149487bd SC |
1222 | /* |
1223 | * Used to put a reference that was taken on behalf of an object associated | |
1224 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation | |
1225 | * of the new file descriptor fails and the reference cannot be transferred to | |
1226 | * its final owner. In such cases, the caller is still actively using @kvm and | |
1227 | * will fail miserably if the refcount unexpectedly hits zero. | |
1228 | */ | |
1229 | void kvm_put_kvm_no_destroy(struct kvm *kvm) | |
1230 | { | |
1231 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); | |
1232 | } | |
1233 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); | |
d39f13b0 | 1234 | |
f17abe9a AK |
1235 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
1236 | { | |
1237 | struct kvm *kvm = filp->private_data; | |
1238 | ||
721eecbf GH |
1239 | kvm_irqfd_release(kvm); |
1240 | ||
d39f13b0 | 1241 | kvm_put_kvm(kvm); |
6aa8b732 AK |
1242 | return 0; |
1243 | } | |
1244 | ||
515a0127 TY |
1245 | /* |
1246 | * Allocation size is twice as large as the actual dirty bitmap size. | |
0dff0846 | 1247 | * See kvm_vm_ioctl_get_dirty_log() why this is needed. |
515a0127 | 1248 | */ |
3c9bd400 | 1249 | static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot) |
a36a57b1 | 1250 | { |
515a0127 | 1251 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
a36a57b1 | 1252 | |
b12ce36a | 1253 | memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT); |
a36a57b1 TY |
1254 | if (!memslot->dirty_bitmap) |
1255 | return -ENOMEM; | |
1256 | ||
a36a57b1 TY |
1257 | return 0; |
1258 | } | |
1259 | ||
bf3e05bc | 1260 | /* |
0577d1ab SC |
1261 | * Delete a memslot by decrementing the number of used slots and shifting all |
1262 | * other entries in the array forward one spot. | |
bf3e05bc | 1263 | */ |
0577d1ab SC |
1264 | static inline void kvm_memslot_delete(struct kvm_memslots *slots, |
1265 | struct kvm_memory_slot *memslot) | |
bf3e05bc | 1266 | { |
063584d4 | 1267 | struct kvm_memory_slot *mslots = slots->memslots; |
0577d1ab | 1268 | int i; |
f85e2cb5 | 1269 | |
0577d1ab SC |
1270 | if (WARN_ON(slots->id_to_index[memslot->id] == -1)) |
1271 | return; | |
0e60b079 | 1272 | |
0577d1ab SC |
1273 | slots->used_slots--; |
1274 | ||
87689270 DM |
1275 | if (atomic_read(&slots->last_used_slot) >= slots->used_slots) |
1276 | atomic_set(&slots->last_used_slot, 0); | |
0774a964 | 1277 | |
0577d1ab | 1278 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) { |
7f379cff IM |
1279 | mslots[i] = mslots[i + 1]; |
1280 | slots->id_to_index[mslots[i].id] = i; | |
7f379cff | 1281 | } |
0577d1ab SC |
1282 | mslots[i] = *memslot; |
1283 | slots->id_to_index[memslot->id] = -1; | |
1284 | } | |
1285 | ||
1286 | /* | |
1287 | * "Insert" a new memslot by incrementing the number of used slots. Returns | |
1288 | * the new slot's initial index into the memslots array. | |
1289 | */ | |
1290 | static inline int kvm_memslot_insert_back(struct kvm_memslots *slots) | |
1291 | { | |
1292 | return slots->used_slots++; | |
1293 | } | |
1294 | ||
1295 | /* | |
1296 | * Move a changed memslot backwards in the array by shifting existing slots | |
1297 | * with a higher GFN toward the front of the array. Note, the changed memslot | |
1298 | * itself is not preserved in the array, i.e. not swapped at this time, only | |
1299 | * its new index into the array is tracked. Returns the changed memslot's | |
1300 | * current index into the memslots array. | |
1301 | */ | |
1302 | static inline int kvm_memslot_move_backward(struct kvm_memslots *slots, | |
1303 | struct kvm_memory_slot *memslot) | |
1304 | { | |
1305 | struct kvm_memory_slot *mslots = slots->memslots; | |
1306 | int i; | |
1307 | ||
1308 | if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) || | |
1309 | WARN_ON_ONCE(!slots->used_slots)) | |
1310 | return -1; | |
efbeec70 PB |
1311 | |
1312 | /* | |
0577d1ab SC |
1313 | * Move the target memslot backward in the array by shifting existing |
1314 | * memslots with a higher GFN (than the target memslot) towards the | |
1315 | * front of the array. | |
efbeec70 | 1316 | */ |
0577d1ab SC |
1317 | for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) { |
1318 | if (memslot->base_gfn > mslots[i + 1].base_gfn) | |
1319 | break; | |
1320 | ||
1321 | WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn); | |
f85e2cb5 | 1322 | |
0577d1ab SC |
1323 | /* Shift the next memslot forward one and update its index. */ |
1324 | mslots[i] = mslots[i + 1]; | |
1325 | slots->id_to_index[mslots[i].id] = i; | |
1326 | } | |
1327 | return i; | |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * Move a changed memslot forwards in the array by shifting existing slots with | |
1332 | * a lower GFN toward the back of the array. Note, the changed memslot itself | |
1333 | * is not preserved in the array, i.e. not swapped at this time, only its new | |
1334 | * index into the array is tracked. Returns the changed memslot's final index | |
1335 | * into the memslots array. | |
1336 | */ | |
1337 | static inline int kvm_memslot_move_forward(struct kvm_memslots *slots, | |
1338 | struct kvm_memory_slot *memslot, | |
1339 | int start) | |
1340 | { | |
1341 | struct kvm_memory_slot *mslots = slots->memslots; | |
1342 | int i; | |
1343 | ||
1344 | for (i = start; i > 0; i--) { | |
1345 | if (memslot->base_gfn < mslots[i - 1].base_gfn) | |
1346 | break; | |
1347 | ||
1348 | WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn); | |
1349 | ||
1350 | /* Shift the next memslot back one and update its index. */ | |
1351 | mslots[i] = mslots[i - 1]; | |
1352 | slots->id_to_index[mslots[i].id] = i; | |
1353 | } | |
1354 | return i; | |
1355 | } | |
1356 | ||
1357 | /* | |
1358 | * Re-sort memslots based on their GFN to account for an added, deleted, or | |
1359 | * moved memslot. Sorting memslots by GFN allows using a binary search during | |
1360 | * memslot lookup. | |
1361 | * | |
1362 | * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry | |
1363 | * at memslots[0] has the highest GFN. | |
1364 | * | |
1365 | * The sorting algorithm takes advantage of having initially sorted memslots | |
1366 | * and knowing the position of the changed memslot. Sorting is also optimized | |
1367 | * by not swapping the updated memslot and instead only shifting other memslots | |
1368 | * and tracking the new index for the update memslot. Only once its final | |
1369 | * index is known is the updated memslot copied into its position in the array. | |
1370 | * | |
1371 | * - When deleting a memslot, the deleted memslot simply needs to be moved to | |
1372 | * the end of the array. | |
1373 | * | |
1374 | * - When creating a memslot, the algorithm "inserts" the new memslot at the | |
1375 | * end of the array and then it forward to its correct location. | |
1376 | * | |
1377 | * - When moving a memslot, the algorithm first moves the updated memslot | |
1378 | * backward to handle the scenario where the memslot's GFN was changed to a | |
1379 | * lower value. update_memslots() then falls through and runs the same flow | |
1380 | * as creating a memslot to move the memslot forward to handle the scenario | |
1381 | * where its GFN was changed to a higher value. | |
1382 | * | |
1383 | * Note, slots are sorted from highest->lowest instead of lowest->highest for | |
1384 | * historical reasons. Originally, invalid memslots where denoted by having | |
1385 | * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots | |
1386 | * to the end of the array. The current algorithm uses dedicated logic to | |
1387 | * delete a memslot and thus does not rely on invalid memslots having GFN=0. | |
1388 | * | |
1389 | * The other historical motiviation for highest->lowest was to improve the | |
1390 | * performance of memslot lookup. KVM originally used a linear search starting | |
1391 | * at memslots[0]. On x86, the largest memslot usually has one of the highest, | |
1392 | * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a | |
1393 | * single memslot above the 4gb boundary. As the largest memslot is also the | |
1394 | * most likely to be referenced, sorting it to the front of the array was | |
1395 | * advantageous. The current binary search starts from the middle of the array | |
1396 | * and uses an LRU pointer to improve performance for all memslots and GFNs. | |
1397 | */ | |
1398 | static void update_memslots(struct kvm_memslots *slots, | |
1399 | struct kvm_memory_slot *memslot, | |
1400 | enum kvm_mr_change change) | |
1401 | { | |
1402 | int i; | |
1403 | ||
1404 | if (change == KVM_MR_DELETE) { | |
1405 | kvm_memslot_delete(slots, memslot); | |
1406 | } else { | |
1407 | if (change == KVM_MR_CREATE) | |
1408 | i = kvm_memslot_insert_back(slots); | |
1409 | else | |
1410 | i = kvm_memslot_move_backward(slots, memslot); | |
1411 | i = kvm_memslot_move_forward(slots, memslot, i); | |
1412 | ||
1413 | /* | |
1414 | * Copy the memslot to its new position in memslots and update | |
1415 | * its index accordingly. | |
1416 | */ | |
1417 | slots->memslots[i] = *memslot; | |
1418 | slots->id_to_index[memslot->id] = i; | |
1419 | } | |
bf3e05bc XG |
1420 | } |
1421 | ||
09170a49 | 1422 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
a50d64d6 | 1423 | { |
4d8b81ab XG |
1424 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
1425 | ||
0f8a4de3 | 1426 | #ifdef __KVM_HAVE_READONLY_MEM |
4d8b81ab XG |
1427 | valid_flags |= KVM_MEM_READONLY; |
1428 | #endif | |
1429 | ||
1430 | if (mem->flags & ~valid_flags) | |
a50d64d6 XG |
1431 | return -EINVAL; |
1432 | ||
1433 | return 0; | |
1434 | } | |
1435 | ||
7ec4fb44 | 1436 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
f481b069 | 1437 | int as_id, struct kvm_memslots *slots) |
7ec4fb44 | 1438 | { |
f481b069 | 1439 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
361209e0 | 1440 | u64 gen = old_memslots->generation; |
7ec4fb44 | 1441 | |
361209e0 SC |
1442 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
1443 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
ee3d1570 | 1444 | |
52ac8b35 PB |
1445 | /* |
1446 | * Do not store the new memslots while there are invalidations in | |
071064f1 PB |
1447 | * progress, otherwise the locking in invalidate_range_start and |
1448 | * invalidate_range_end will be unbalanced. | |
52ac8b35 PB |
1449 | */ |
1450 | spin_lock(&kvm->mn_invalidate_lock); | |
1451 | prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
1452 | while (kvm->mn_active_invalidate_count) { | |
1453 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1454 | spin_unlock(&kvm->mn_invalidate_lock); | |
1455 | schedule(); | |
1456 | spin_lock(&kvm->mn_invalidate_lock); | |
1457 | } | |
1458 | finish_rcuwait(&kvm->mn_memslots_update_rcuwait); | |
f481b069 | 1459 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
52ac8b35 | 1460 | spin_unlock(&kvm->mn_invalidate_lock); |
b10a038e BG |
1461 | |
1462 | /* | |
1463 | * Acquired in kvm_set_memslot. Must be released before synchronize | |
1464 | * SRCU below in order to avoid deadlock with another thread | |
1465 | * acquiring the slots_arch_lock in an srcu critical section. | |
1466 | */ | |
1467 | mutex_unlock(&kvm->slots_arch_lock); | |
1468 | ||
7ec4fb44 | 1469 | synchronize_srcu_expedited(&kvm->srcu); |
e59dbe09 | 1470 | |
ee3d1570 | 1471 | /* |
361209e0 | 1472 | * Increment the new memslot generation a second time, dropping the |
00116795 | 1473 | * update in-progress flag and incrementing the generation based on |
361209e0 SC |
1474 | * the number of address spaces. This provides a unique and easily |
1475 | * identifiable generation number while the memslots are in flux. | |
1476 | */ | |
1477 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; | |
1478 | ||
1479 | /* | |
4bd518f1 PB |
1480 | * Generations must be unique even across address spaces. We do not need |
1481 | * a global counter for that, instead the generation space is evenly split | |
1482 | * across address spaces. For example, with two address spaces, address | |
164bf7e5 SC |
1483 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
1484 | * use generations 1, 3, 5, ... | |
ee3d1570 | 1485 | */ |
164bf7e5 | 1486 | gen += KVM_ADDRESS_SPACE_NUM; |
ee3d1570 | 1487 | |
15248258 | 1488 | kvm_arch_memslots_updated(kvm, gen); |
ee3d1570 | 1489 | |
15248258 | 1490 | slots->generation = gen; |
e59dbe09 TY |
1491 | |
1492 | return old_memslots; | |
7ec4fb44 GN |
1493 | } |
1494 | ||
ddc12f2a BG |
1495 | static size_t kvm_memslots_size(int slots) |
1496 | { | |
1497 | return sizeof(struct kvm_memslots) + | |
1498 | (sizeof(struct kvm_memory_slot) * slots); | |
1499 | } | |
1500 | ||
1501 | static void kvm_copy_memslots(struct kvm_memslots *to, | |
1502 | struct kvm_memslots *from) | |
1503 | { | |
1504 | memcpy(to, from, kvm_memslots_size(from->used_slots)); | |
1505 | } | |
1506 | ||
36947254 SC |
1507 | /* |
1508 | * Note, at a minimum, the current number of used slots must be allocated, even | |
1509 | * when deleting a memslot, as we need a complete duplicate of the memslots for | |
1510 | * use when invalidating a memslot prior to deleting/moving the memslot. | |
1511 | */ | |
1512 | static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old, | |
1513 | enum kvm_mr_change change) | |
1514 | { | |
1515 | struct kvm_memslots *slots; | |
ddc12f2a | 1516 | size_t new_size; |
36947254 SC |
1517 | |
1518 | if (change == KVM_MR_CREATE) | |
ddc12f2a | 1519 | new_size = kvm_memslots_size(old->used_slots + 1); |
36947254 | 1520 | else |
ddc12f2a | 1521 | new_size = kvm_memslots_size(old->used_slots); |
36947254 SC |
1522 | |
1523 | slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT); | |
1524 | if (likely(slots)) | |
ddc12f2a | 1525 | kvm_copy_memslots(slots, old); |
36947254 SC |
1526 | |
1527 | return slots; | |
1528 | } | |
1529 | ||
cf47f50b SC |
1530 | static int kvm_set_memslot(struct kvm *kvm, |
1531 | const struct kvm_userspace_memory_region *mem, | |
9d4c197c | 1532 | struct kvm_memory_slot *old, |
cf47f50b SC |
1533 | struct kvm_memory_slot *new, int as_id, |
1534 | enum kvm_mr_change change) | |
1535 | { | |
1536 | struct kvm_memory_slot *slot; | |
1537 | struct kvm_memslots *slots; | |
1538 | int r; | |
1539 | ||
b10a038e BG |
1540 | /* |
1541 | * Released in install_new_memslots. | |
1542 | * | |
1543 | * Must be held from before the current memslots are copied until | |
1544 | * after the new memslots are installed with rcu_assign_pointer, | |
1545 | * then released before the synchronize srcu in install_new_memslots. | |
1546 | * | |
1547 | * When modifying memslots outside of the slots_lock, must be held | |
1548 | * before reading the pointer to the current memslots until after all | |
1549 | * changes to those memslots are complete. | |
1550 | * | |
1551 | * These rules ensure that installing new memslots does not lose | |
1552 | * changes made to the previous memslots. | |
1553 | */ | |
1554 | mutex_lock(&kvm->slots_arch_lock); | |
1555 | ||
36947254 | 1556 | slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change); |
b10a038e BG |
1557 | if (!slots) { |
1558 | mutex_unlock(&kvm->slots_arch_lock); | |
cf47f50b | 1559 | return -ENOMEM; |
b10a038e | 1560 | } |
cf47f50b SC |
1561 | |
1562 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { | |
1563 | /* | |
1564 | * Note, the INVALID flag needs to be in the appropriate entry | |
1565 | * in the freshly allocated memslots, not in @old or @new. | |
1566 | */ | |
1567 | slot = id_to_memslot(slots, old->id); | |
1568 | slot->flags |= KVM_MEMSLOT_INVALID; | |
1569 | ||
1570 | /* | |
b10a038e BG |
1571 | * We can re-use the memory from the old memslots. |
1572 | * It will be overwritten with a copy of the new memslots | |
1573 | * after reacquiring the slots_arch_lock below. | |
cf47f50b SC |
1574 | */ |
1575 | slots = install_new_memslots(kvm, as_id, slots); | |
1576 | ||
1577 | /* From this point no new shadow pages pointing to a deleted, | |
1578 | * or moved, memslot will be created. | |
1579 | * | |
1580 | * validation of sp->gfn happens in: | |
1581 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) | |
1582 | * - kvm_is_visible_gfn (mmu_check_root) | |
1583 | */ | |
1584 | kvm_arch_flush_shadow_memslot(kvm, slot); | |
b10a038e BG |
1585 | |
1586 | /* Released in install_new_memslots. */ | |
1587 | mutex_lock(&kvm->slots_arch_lock); | |
1588 | ||
1589 | /* | |
1590 | * The arch-specific fields of the memslots could have changed | |
1591 | * between releasing the slots_arch_lock in | |
1592 | * install_new_memslots and here, so get a fresh copy of the | |
1593 | * slots. | |
1594 | */ | |
1595 | kvm_copy_memslots(slots, __kvm_memslots(kvm, as_id)); | |
cf47f50b SC |
1596 | } |
1597 | ||
1598 | r = kvm_arch_prepare_memory_region(kvm, new, mem, change); | |
1599 | if (r) | |
1600 | goto out_slots; | |
1601 | ||
1602 | update_memslots(slots, new, change); | |
1603 | slots = install_new_memslots(kvm, as_id, slots); | |
1604 | ||
1605 | kvm_arch_commit_memory_region(kvm, mem, old, new, change); | |
1606 | ||
1607 | kvfree(slots); | |
1608 | return 0; | |
1609 | ||
1610 | out_slots: | |
b10a038e BG |
1611 | if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) { |
1612 | slot = id_to_memslot(slots, old->id); | |
1613 | slot->flags &= ~KVM_MEMSLOT_INVALID; | |
cf47f50b | 1614 | slots = install_new_memslots(kvm, as_id, slots); |
b10a038e BG |
1615 | } else { |
1616 | mutex_unlock(&kvm->slots_arch_lock); | |
1617 | } | |
cf47f50b SC |
1618 | kvfree(slots); |
1619 | return r; | |
1620 | } | |
1621 | ||
5c0b4f3d SC |
1622 | static int kvm_delete_memslot(struct kvm *kvm, |
1623 | const struct kvm_userspace_memory_region *mem, | |
1624 | struct kvm_memory_slot *old, int as_id) | |
1625 | { | |
1626 | struct kvm_memory_slot new; | |
1627 | int r; | |
1628 | ||
1629 | if (!old->npages) | |
1630 | return -EINVAL; | |
1631 | ||
1632 | memset(&new, 0, sizeof(new)); | |
1633 | new.id = old->id; | |
9e9eb226 PX |
1634 | /* |
1635 | * This is only for debugging purpose; it should never be referenced | |
1636 | * for a removed memslot. | |
1637 | */ | |
1638 | new.as_id = as_id; | |
5c0b4f3d SC |
1639 | |
1640 | r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE); | |
1641 | if (r) | |
1642 | return r; | |
1643 | ||
e96c81ee | 1644 | kvm_free_memslot(kvm, old); |
5c0b4f3d SC |
1645 | return 0; |
1646 | } | |
1647 | ||
6aa8b732 AK |
1648 | /* |
1649 | * Allocate some memory and give it an address in the guest physical address | |
1650 | * space. | |
1651 | * | |
1652 | * Discontiguous memory is allowed, mostly for framebuffers. | |
f78e0e2e | 1653 | * |
02d5d55b | 1654 | * Must be called holding kvm->slots_lock for write. |
6aa8b732 | 1655 | */ |
f78e0e2e | 1656 | int __kvm_set_memory_region(struct kvm *kvm, |
09170a49 | 1657 | const struct kvm_userspace_memory_region *mem) |
6aa8b732 | 1658 | { |
6aa8b732 | 1659 | struct kvm_memory_slot old, new; |
163da372 | 1660 | struct kvm_memory_slot *tmp; |
f64c0398 | 1661 | enum kvm_mr_change change; |
163da372 SC |
1662 | int as_id, id; |
1663 | int r; | |
6aa8b732 | 1664 | |
a50d64d6 XG |
1665 | r = check_memory_region_flags(mem); |
1666 | if (r) | |
71a4c30b | 1667 | return r; |
a50d64d6 | 1668 | |
f481b069 PB |
1669 | as_id = mem->slot >> 16; |
1670 | id = (u16)mem->slot; | |
1671 | ||
6aa8b732 AK |
1672 | /* General sanity checks */ |
1673 | if (mem->memory_size & (PAGE_SIZE - 1)) | |
71a4c30b | 1674 | return -EINVAL; |
6aa8b732 | 1675 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
71a4c30b | 1676 | return -EINVAL; |
fa3d315a | 1677 | /* We can read the guest memory with __xxx_user() later on. */ |
09d952c9 | 1678 | if ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
139bc8a6 | 1679 | (mem->userspace_addr != untagged_addr(mem->userspace_addr)) || |
96d4f267 | 1680 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
09d952c9 | 1681 | mem->memory_size)) |
71a4c30b | 1682 | return -EINVAL; |
f481b069 | 1683 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
71a4c30b | 1684 | return -EINVAL; |
6aa8b732 | 1685 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
71a4c30b | 1686 | return -EINVAL; |
6aa8b732 | 1687 | |
5c0b4f3d SC |
1688 | /* |
1689 | * Make a full copy of the old memslot, the pointer will become stale | |
1690 | * when the memslots are re-sorted by update_memslots(), and the old | |
1691 | * memslot needs to be referenced after calling update_memslots(), e.g. | |
0dff0846 | 1692 | * to free its resources and for arch specific behavior. |
5c0b4f3d | 1693 | */ |
0577d1ab SC |
1694 | tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
1695 | if (tmp) { | |
1696 | old = *tmp; | |
1697 | tmp = NULL; | |
1698 | } else { | |
1699 | memset(&old, 0, sizeof(old)); | |
1700 | old.id = id; | |
1701 | } | |
163da372 | 1702 | |
5c0b4f3d SC |
1703 | if (!mem->memory_size) |
1704 | return kvm_delete_memslot(kvm, mem, &old, as_id); | |
1705 | ||
9e9eb226 | 1706 | new.as_id = as_id; |
f481b069 | 1707 | new.id = id; |
163da372 SC |
1708 | new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
1709 | new.npages = mem->memory_size >> PAGE_SHIFT; | |
6aa8b732 | 1710 | new.flags = mem->flags; |
414de7ab | 1711 | new.userspace_addr = mem->userspace_addr; |
6aa8b732 | 1712 | |
163da372 SC |
1713 | if (new.npages > KVM_MEM_MAX_NR_PAGES) |
1714 | return -EINVAL; | |
1715 | ||
5c0b4f3d SC |
1716 | if (!old.npages) { |
1717 | change = KVM_MR_CREATE; | |
163da372 SC |
1718 | new.dirty_bitmap = NULL; |
1719 | memset(&new.arch, 0, sizeof(new.arch)); | |
5c0b4f3d SC |
1720 | } else { /* Modify an existing slot. */ |
1721 | if ((new.userspace_addr != old.userspace_addr) || | |
163da372 | 1722 | (new.npages != old.npages) || |
5c0b4f3d | 1723 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) |
71a4c30b | 1724 | return -EINVAL; |
09170a49 | 1725 | |
163da372 | 1726 | if (new.base_gfn != old.base_gfn) |
5c0b4f3d SC |
1727 | change = KVM_MR_MOVE; |
1728 | else if (new.flags != old.flags) | |
1729 | change = KVM_MR_FLAGS_ONLY; | |
1730 | else /* Nothing to change. */ | |
1731 | return 0; | |
163da372 SC |
1732 | |
1733 | /* Copy dirty_bitmap and arch from the current memslot. */ | |
1734 | new.dirty_bitmap = old.dirty_bitmap; | |
1735 | memcpy(&new.arch, &old.arch, sizeof(new.arch)); | |
09170a49 | 1736 | } |
6aa8b732 | 1737 | |
f64c0398 | 1738 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
0a706bee | 1739 | /* Check for overlaps */ |
163da372 SC |
1740 | kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) { |
1741 | if (tmp->id == id) | |
0a706bee | 1742 | continue; |
163da372 SC |
1743 | if (!((new.base_gfn + new.npages <= tmp->base_gfn) || |
1744 | (new.base_gfn >= tmp->base_gfn + tmp->npages))) | |
71a4c30b | 1745 | return -EEXIST; |
0a706bee | 1746 | } |
6aa8b732 | 1747 | } |
6aa8b732 | 1748 | |
414de7ab SC |
1749 | /* Allocate/free page dirty bitmap as needed */ |
1750 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
1751 | new.dirty_bitmap = NULL; | |
044c59c4 | 1752 | else if (!new.dirty_bitmap && !kvm->dirty_ring_size) { |
3c9bd400 | 1753 | r = kvm_alloc_dirty_bitmap(&new); |
71a4c30b SC |
1754 | if (r) |
1755 | return r; | |
3c9bd400 JZ |
1756 | |
1757 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
1758 | bitmap_set(new.dirty_bitmap, 0, new.npages); | |
6aa8b732 AK |
1759 | } |
1760 | ||
cf47f50b SC |
1761 | r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change); |
1762 | if (r) | |
1763 | goto out_bitmap; | |
82ce2c96 | 1764 | |
5c0b4f3d SC |
1765 | if (old.dirty_bitmap && !new.dirty_bitmap) |
1766 | kvm_destroy_dirty_bitmap(&old); | |
6aa8b732 AK |
1767 | return 0; |
1768 | ||
bd0e96fd SC |
1769 | out_bitmap: |
1770 | if (new.dirty_bitmap && !old.dirty_bitmap) | |
1771 | kvm_destroy_dirty_bitmap(&new); | |
6aa8b732 | 1772 | return r; |
210c7c4d | 1773 | } |
f78e0e2e SY |
1774 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
1775 | ||
1776 | int kvm_set_memory_region(struct kvm *kvm, | |
09170a49 | 1777 | const struct kvm_userspace_memory_region *mem) |
f78e0e2e SY |
1778 | { |
1779 | int r; | |
1780 | ||
79fac95e | 1781 | mutex_lock(&kvm->slots_lock); |
47ae31e2 | 1782 | r = __kvm_set_memory_region(kvm, mem); |
79fac95e | 1783 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
1784 | return r; |
1785 | } | |
210c7c4d IE |
1786 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
1787 | ||
7940876e SH |
1788 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
1789 | struct kvm_userspace_memory_region *mem) | |
210c7c4d | 1790 | { |
f481b069 | 1791 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
e0d62c7f | 1792 | return -EINVAL; |
09170a49 | 1793 | |
47ae31e2 | 1794 | return kvm_set_memory_region(kvm, mem); |
6aa8b732 AK |
1795 | } |
1796 | ||
0dff0846 | 1797 | #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
2a49f61d SC |
1798 | /** |
1799 | * kvm_get_dirty_log - get a snapshot of dirty pages | |
1800 | * @kvm: pointer to kvm instance | |
1801 | * @log: slot id and address to which we copy the log | |
1802 | * @is_dirty: set to '1' if any dirty pages were found | |
1803 | * @memslot: set to the associated memslot, always valid on success | |
1804 | */ | |
1805 | int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log, | |
1806 | int *is_dirty, struct kvm_memory_slot **memslot) | |
6aa8b732 | 1807 | { |
9f6b8029 | 1808 | struct kvm_memslots *slots; |
843574a3 | 1809 | int i, as_id, id; |
87bf6e7d | 1810 | unsigned long n; |
6aa8b732 AK |
1811 | unsigned long any = 0; |
1812 | ||
b2cc64c4 PX |
1813 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1814 | if (kvm->dirty_ring_size) | |
1815 | return -ENXIO; | |
1816 | ||
2a49f61d SC |
1817 | *memslot = NULL; |
1818 | *is_dirty = 0; | |
1819 | ||
f481b069 PB |
1820 | as_id = log->slot >> 16; |
1821 | id = (u16)log->slot; | |
1822 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
843574a3 | 1823 | return -EINVAL; |
6aa8b732 | 1824 | |
f481b069 | 1825 | slots = __kvm_memslots(kvm, as_id); |
2a49f61d | 1826 | *memslot = id_to_memslot(slots, id); |
0577d1ab | 1827 | if (!(*memslot) || !(*memslot)->dirty_bitmap) |
843574a3 | 1828 | return -ENOENT; |
6aa8b732 | 1829 | |
2a49f61d SC |
1830 | kvm_arch_sync_dirty_log(kvm, *memslot); |
1831 | ||
1832 | n = kvm_dirty_bitmap_bytes(*memslot); | |
6aa8b732 | 1833 | |
cd1a4a98 | 1834 | for (i = 0; !any && i < n/sizeof(long); ++i) |
2a49f61d | 1835 | any = (*memslot)->dirty_bitmap[i]; |
6aa8b732 | 1836 | |
2a49f61d | 1837 | if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n)) |
843574a3 | 1838 | return -EFAULT; |
6aa8b732 | 1839 | |
5bb064dc ZX |
1840 | if (any) |
1841 | *is_dirty = 1; | |
843574a3 | 1842 | return 0; |
6aa8b732 | 1843 | } |
2ba9f0d8 | 1844 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
6aa8b732 | 1845 | |
0dff0846 | 1846 | #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ |
ba0513b5 | 1847 | /** |
b8b00220 | 1848 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
2a31b9db | 1849 | * and reenable dirty page tracking for the corresponding pages. |
ba0513b5 MS |
1850 | * @kvm: pointer to kvm instance |
1851 | * @log: slot id and address to which we copy the log | |
ba0513b5 MS |
1852 | * |
1853 | * We need to keep it in mind that VCPU threads can write to the bitmap | |
1854 | * concurrently. So, to avoid losing track of dirty pages we keep the | |
1855 | * following order: | |
1856 | * | |
1857 | * 1. Take a snapshot of the bit and clear it if needed. | |
1858 | * 2. Write protect the corresponding page. | |
1859 | * 3. Copy the snapshot to the userspace. | |
1860 | * 4. Upon return caller flushes TLB's if needed. | |
1861 | * | |
1862 | * Between 2 and 4, the guest may write to the page using the remaining TLB | |
1863 | * entry. This is not a problem because the page is reported dirty using | |
1864 | * the snapshot taken before and step 4 ensures that writes done after | |
1865 | * exiting to userspace will be logged for the next call. | |
1866 | * | |
1867 | */ | |
0dff0846 | 1868 | static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log) |
ba0513b5 | 1869 | { |
9f6b8029 | 1870 | struct kvm_memslots *slots; |
ba0513b5 | 1871 | struct kvm_memory_slot *memslot; |
58d6db34 | 1872 | int i, as_id, id; |
ba0513b5 MS |
1873 | unsigned long n; |
1874 | unsigned long *dirty_bitmap; | |
1875 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 1876 | bool flush; |
ba0513b5 | 1877 | |
b2cc64c4 PX |
1878 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1879 | if (kvm->dirty_ring_size) | |
1880 | return -ENXIO; | |
1881 | ||
f481b069 PB |
1882 | as_id = log->slot >> 16; |
1883 | id = (u16)log->slot; | |
1884 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
58d6db34 | 1885 | return -EINVAL; |
ba0513b5 | 1886 | |
f481b069 PB |
1887 | slots = __kvm_memslots(kvm, as_id); |
1888 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
1889 | if (!memslot || !memslot->dirty_bitmap) |
1890 | return -ENOENT; | |
ba0513b5 MS |
1891 | |
1892 | dirty_bitmap = memslot->dirty_bitmap; | |
ba0513b5 | 1893 | |
0dff0846 SC |
1894 | kvm_arch_sync_dirty_log(kvm, memslot); |
1895 | ||
ba0513b5 | 1896 | n = kvm_dirty_bitmap_bytes(memslot); |
0dff0846 | 1897 | flush = false; |
2a31b9db PB |
1898 | if (kvm->manual_dirty_log_protect) { |
1899 | /* | |
1900 | * Unlike kvm_get_dirty_log, we always return false in *flush, | |
1901 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There | |
1902 | * is some code duplication between this function and | |
1903 | * kvm_get_dirty_log, but hopefully all architecture | |
1904 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log | |
1905 | * can be eliminated. | |
1906 | */ | |
1907 | dirty_bitmap_buffer = dirty_bitmap; | |
1908 | } else { | |
1909 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); | |
1910 | memset(dirty_bitmap_buffer, 0, n); | |
ba0513b5 | 1911 | |
531810ca | 1912 | KVM_MMU_LOCK(kvm); |
2a31b9db PB |
1913 | for (i = 0; i < n / sizeof(long); i++) { |
1914 | unsigned long mask; | |
1915 | gfn_t offset; | |
ba0513b5 | 1916 | |
2a31b9db PB |
1917 | if (!dirty_bitmap[i]) |
1918 | continue; | |
1919 | ||
0dff0846 | 1920 | flush = true; |
2a31b9db PB |
1921 | mask = xchg(&dirty_bitmap[i], 0); |
1922 | dirty_bitmap_buffer[i] = mask; | |
1923 | ||
a67794ca LT |
1924 | offset = i * BITS_PER_LONG; |
1925 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, | |
1926 | offset, mask); | |
2a31b9db | 1927 | } |
531810ca | 1928 | KVM_MMU_UNLOCK(kvm); |
2a31b9db PB |
1929 | } |
1930 | ||
0dff0846 SC |
1931 | if (flush) |
1932 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
1933 | ||
2a31b9db PB |
1934 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
1935 | return -EFAULT; | |
1936 | return 0; | |
1937 | } | |
0dff0846 SC |
1938 | |
1939 | ||
1940 | /** | |
1941 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot | |
1942 | * @kvm: kvm instance | |
1943 | * @log: slot id and address to which we copy the log | |
1944 | * | |
1945 | * Steps 1-4 below provide general overview of dirty page logging. See | |
1946 | * kvm_get_dirty_log_protect() function description for additional details. | |
1947 | * | |
1948 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
1949 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
1950 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
1951 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
1952 | * writes will be marked dirty for next log read. | |
1953 | * | |
1954 | * 1. Take a snapshot of the bit and clear it if needed. | |
1955 | * 2. Write protect the corresponding page. | |
1956 | * 3. Copy the snapshot to the userspace. | |
1957 | * 4. Flush TLB's if needed. | |
1958 | */ | |
1959 | static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, | |
1960 | struct kvm_dirty_log *log) | |
1961 | { | |
1962 | int r; | |
1963 | ||
1964 | mutex_lock(&kvm->slots_lock); | |
1965 | ||
1966 | r = kvm_get_dirty_log_protect(kvm, log); | |
1967 | ||
1968 | mutex_unlock(&kvm->slots_lock); | |
1969 | return r; | |
1970 | } | |
2a31b9db PB |
1971 | |
1972 | /** | |
1973 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap | |
1974 | * and reenable dirty page tracking for the corresponding pages. | |
1975 | * @kvm: pointer to kvm instance | |
1976 | * @log: slot id and address from which to fetch the bitmap of dirty pages | |
1977 | */ | |
0dff0846 SC |
1978 | static int kvm_clear_dirty_log_protect(struct kvm *kvm, |
1979 | struct kvm_clear_dirty_log *log) | |
2a31b9db PB |
1980 | { |
1981 | struct kvm_memslots *slots; | |
1982 | struct kvm_memory_slot *memslot; | |
98938aa8 | 1983 | int as_id, id; |
2a31b9db | 1984 | gfn_t offset; |
98938aa8 | 1985 | unsigned long i, n; |
2a31b9db PB |
1986 | unsigned long *dirty_bitmap; |
1987 | unsigned long *dirty_bitmap_buffer; | |
0dff0846 | 1988 | bool flush; |
2a31b9db | 1989 | |
b2cc64c4 PX |
1990 | /* Dirty ring tracking is exclusive to dirty log tracking */ |
1991 | if (kvm->dirty_ring_size) | |
1992 | return -ENXIO; | |
1993 | ||
2a31b9db PB |
1994 | as_id = log->slot >> 16; |
1995 | id = (u16)log->slot; | |
1996 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) | |
1997 | return -EINVAL; | |
1998 | ||
76d58e0f | 1999 | if (log->first_page & 63) |
2a31b9db PB |
2000 | return -EINVAL; |
2001 | ||
2002 | slots = __kvm_memslots(kvm, as_id); | |
2003 | memslot = id_to_memslot(slots, id); | |
0577d1ab SC |
2004 | if (!memslot || !memslot->dirty_bitmap) |
2005 | return -ENOENT; | |
2a31b9db PB |
2006 | |
2007 | dirty_bitmap = memslot->dirty_bitmap; | |
2a31b9db | 2008 | |
4ddc9204 | 2009 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
98938aa8 TB |
2010 | |
2011 | if (log->first_page > memslot->npages || | |
76d58e0f PB |
2012 | log->num_pages > memslot->npages - log->first_page || |
2013 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) | |
2014 | return -EINVAL; | |
98938aa8 | 2015 | |
0dff0846 SC |
2016 | kvm_arch_sync_dirty_log(kvm, memslot); |
2017 | ||
2018 | flush = false; | |
2a31b9db PB |
2019 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
2020 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) | |
2021 | return -EFAULT; | |
ba0513b5 | 2022 | |
531810ca | 2023 | KVM_MMU_LOCK(kvm); |
53eac7a8 PX |
2024 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
2025 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; | |
2a31b9db PB |
2026 | i++, offset += BITS_PER_LONG) { |
2027 | unsigned long mask = *dirty_bitmap_buffer++; | |
2028 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; | |
2029 | if (!mask) | |
ba0513b5 MS |
2030 | continue; |
2031 | ||
2a31b9db | 2032 | mask &= atomic_long_fetch_andnot(mask, p); |
ba0513b5 | 2033 | |
2a31b9db PB |
2034 | /* |
2035 | * mask contains the bits that really have been cleared. This | |
2036 | * never includes any bits beyond the length of the memslot (if | |
2037 | * the length is not aligned to 64 pages), therefore it is not | |
2038 | * a problem if userspace sets them in log->dirty_bitmap. | |
2039 | */ | |
58d2930f | 2040 | if (mask) { |
0dff0846 | 2041 | flush = true; |
58d2930f TY |
2042 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
2043 | offset, mask); | |
2044 | } | |
ba0513b5 | 2045 | } |
531810ca | 2046 | KVM_MMU_UNLOCK(kvm); |
2a31b9db | 2047 | |
0dff0846 SC |
2048 | if (flush) |
2049 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); | |
2050 | ||
58d6db34 | 2051 | return 0; |
ba0513b5 | 2052 | } |
0dff0846 SC |
2053 | |
2054 | static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, | |
2055 | struct kvm_clear_dirty_log *log) | |
2056 | { | |
2057 | int r; | |
2058 | ||
2059 | mutex_lock(&kvm->slots_lock); | |
2060 | ||
2061 | r = kvm_clear_dirty_log_protect(kvm, log); | |
2062 | ||
2063 | mutex_unlock(&kvm->slots_lock); | |
2064 | return r; | |
2065 | } | |
2066 | #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */ | |
ba0513b5 | 2067 | |
49c7754c GN |
2068 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
2069 | { | |
2070 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
2071 | } | |
a1f4d395 | 2072 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
6aa8b732 | 2073 | |
8e73485c PB |
2074 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
2075 | { | |
fe22ed82 DM |
2076 | struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu); |
2077 | struct kvm_memory_slot *slot; | |
2078 | int slot_index; | |
2079 | ||
2080 | slot = try_get_memslot(slots, vcpu->last_used_slot, gfn); | |
2081 | if (slot) | |
2082 | return slot; | |
2083 | ||
2084 | /* | |
2085 | * Fall back to searching all memslots. We purposely use | |
2086 | * search_memslots() instead of __gfn_to_memslot() to avoid | |
2087 | * thrashing the VM-wide last_used_index in kvm_memslots. | |
2088 | */ | |
2089 | slot = search_memslots(slots, gfn, &slot_index); | |
2090 | if (slot) { | |
2091 | vcpu->last_used_slot = slot_index; | |
2092 | return slot; | |
2093 | } | |
2094 | ||
2095 | return NULL; | |
8e73485c | 2096 | } |
e72436bc | 2097 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot); |
8e73485c | 2098 | |
33e94154 | 2099 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
e0d62c7f | 2100 | { |
bf3e05bc | 2101 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
e0d62c7f | 2102 | |
c36b7150 | 2103 | return kvm_is_visible_memslot(memslot); |
e0d62c7f IE |
2104 | } |
2105 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); | |
2106 | ||
995decb6 VK |
2107 | bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
2108 | { | |
2109 | struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2110 | ||
2111 | return kvm_is_visible_memslot(memslot); | |
2112 | } | |
2113 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn); | |
2114 | ||
f9b84e19 | 2115 | unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn) |
8f0b1ab6 JR |
2116 | { |
2117 | struct vm_area_struct *vma; | |
2118 | unsigned long addr, size; | |
2119 | ||
2120 | size = PAGE_SIZE; | |
2121 | ||
42cde48b | 2122 | addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL); |
8f0b1ab6 JR |
2123 | if (kvm_is_error_hva(addr)) |
2124 | return PAGE_SIZE; | |
2125 | ||
d8ed45c5 | 2126 | mmap_read_lock(current->mm); |
8f0b1ab6 JR |
2127 | vma = find_vma(current->mm, addr); |
2128 | if (!vma) | |
2129 | goto out; | |
2130 | ||
2131 | size = vma_kernel_pagesize(vma); | |
2132 | ||
2133 | out: | |
d8ed45c5 | 2134 | mmap_read_unlock(current->mm); |
8f0b1ab6 JR |
2135 | |
2136 | return size; | |
2137 | } | |
2138 | ||
4d8b81ab XG |
2139 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
2140 | { | |
2141 | return slot->flags & KVM_MEM_READONLY; | |
2142 | } | |
2143 | ||
4d8b81ab XG |
2144 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2145 | gfn_t *nr_pages, bool write) | |
539cb660 | 2146 | { |
bc6678a3 | 2147 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
ca3a490c | 2148 | return KVM_HVA_ERR_BAD; |
48987781 | 2149 | |
4d8b81ab XG |
2150 | if (memslot_is_readonly(slot) && write) |
2151 | return KVM_HVA_ERR_RO_BAD; | |
48987781 XG |
2152 | |
2153 | if (nr_pages) | |
2154 | *nr_pages = slot->npages - (gfn - slot->base_gfn); | |
2155 | ||
4d8b81ab | 2156 | return __gfn_to_hva_memslot(slot, gfn); |
539cb660 | 2157 | } |
48987781 | 2158 | |
4d8b81ab XG |
2159 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
2160 | gfn_t *nr_pages) | |
2161 | { | |
2162 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); | |
539cb660 | 2163 | } |
48987781 | 2164 | |
4d8b81ab | 2165 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
7940876e | 2166 | gfn_t gfn) |
4d8b81ab XG |
2167 | { |
2168 | return gfn_to_hva_many(slot, gfn, NULL); | |
2169 | } | |
2170 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); | |
2171 | ||
48987781 XG |
2172 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
2173 | { | |
49c7754c | 2174 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
48987781 | 2175 | } |
0d150298 | 2176 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
539cb660 | 2177 | |
8e73485c PB |
2178 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
2179 | { | |
2180 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); | |
2181 | } | |
2182 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); | |
2183 | ||
86ab8cff | 2184 | /* |
970c0d4b WY |
2185 | * Return the hva of a @gfn and the R/W attribute if possible. |
2186 | * | |
2187 | * @slot: the kvm_memory_slot which contains @gfn | |
2188 | * @gfn: the gfn to be translated | |
2189 | * @writable: used to return the read/write attribute of the @slot if the hva | |
2190 | * is valid and @writable is not NULL | |
86ab8cff | 2191 | */ |
64d83126 CD |
2192 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
2193 | gfn_t gfn, bool *writable) | |
86ab8cff | 2194 | { |
a2ac07fe GN |
2195 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
2196 | ||
2197 | if (!kvm_is_error_hva(hva) && writable) | |
ba6a3541 PB |
2198 | *writable = !memslot_is_readonly(slot); |
2199 | ||
a2ac07fe | 2200 | return hva; |
86ab8cff XG |
2201 | } |
2202 | ||
64d83126 CD |
2203 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
2204 | { | |
2205 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2206 | ||
2207 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2208 | } | |
2209 | ||
8e73485c PB |
2210 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
2211 | { | |
2212 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2213 | ||
2214 | return gfn_to_hva_memslot_prot(slot, gfn, writable); | |
2215 | } | |
2216 | ||
fafc3dba HY |
2217 | static inline int check_user_page_hwpoison(unsigned long addr) |
2218 | { | |
0d731759 | 2219 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
fafc3dba | 2220 | |
0d731759 | 2221 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
fafc3dba HY |
2222 | return rc == -EHWPOISON; |
2223 | } | |
2224 | ||
2fc84311 | 2225 | /* |
b9b33da2 PB |
2226 | * The fast path to get the writable pfn which will be stored in @pfn, |
2227 | * true indicates success, otherwise false is returned. It's also the | |
311497e0 | 2228 | * only part that runs if we can in atomic context. |
2fc84311 | 2229 | */ |
b9b33da2 PB |
2230 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
2231 | bool *writable, kvm_pfn_t *pfn) | |
954bbbc2 | 2232 | { |
8d4e1288 | 2233 | struct page *page[1]; |
954bbbc2 | 2234 | |
12ce13fe XG |
2235 | /* |
2236 | * Fast pin a writable pfn only if it is a write fault request | |
2237 | * or the caller allows to map a writable pfn for a read fault | |
2238 | * request. | |
2239 | */ | |
2240 | if (!(write_fault || writable)) | |
2241 | return false; | |
612819c3 | 2242 | |
dadbb612 | 2243 | if (get_user_page_fast_only(addr, FOLL_WRITE, page)) { |
2fc84311 | 2244 | *pfn = page_to_pfn(page[0]); |
612819c3 | 2245 | |
2fc84311 XG |
2246 | if (writable) |
2247 | *writable = true; | |
2248 | return true; | |
2249 | } | |
af585b92 | 2250 | |
2fc84311 XG |
2251 | return false; |
2252 | } | |
612819c3 | 2253 | |
2fc84311 XG |
2254 | /* |
2255 | * The slow path to get the pfn of the specified host virtual address, | |
2256 | * 1 indicates success, -errno is returned if error is detected. | |
2257 | */ | |
2258 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, | |
ba049e93 | 2259 | bool *writable, kvm_pfn_t *pfn) |
2fc84311 | 2260 | { |
ce53053c AV |
2261 | unsigned int flags = FOLL_HWPOISON; |
2262 | struct page *page; | |
2fc84311 | 2263 | int npages = 0; |
612819c3 | 2264 | |
2fc84311 XG |
2265 | might_sleep(); |
2266 | ||
2267 | if (writable) | |
2268 | *writable = write_fault; | |
2269 | ||
ce53053c AV |
2270 | if (write_fault) |
2271 | flags |= FOLL_WRITE; | |
2272 | if (async) | |
2273 | flags |= FOLL_NOWAIT; | |
d4944b0e | 2274 | |
ce53053c | 2275 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
2fc84311 XG |
2276 | if (npages != 1) |
2277 | return npages; | |
2278 | ||
2279 | /* map read fault as writable if possible */ | |
12ce13fe | 2280 | if (unlikely(!write_fault) && writable) { |
ce53053c | 2281 | struct page *wpage; |
2fc84311 | 2282 | |
dadbb612 | 2283 | if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) { |
2fc84311 | 2284 | *writable = true; |
ce53053c AV |
2285 | put_page(page); |
2286 | page = wpage; | |
612819c3 | 2287 | } |
887c08ac | 2288 | } |
ce53053c | 2289 | *pfn = page_to_pfn(page); |
2fc84311 XG |
2290 | return npages; |
2291 | } | |
539cb660 | 2292 | |
4d8b81ab XG |
2293 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
2294 | { | |
2295 | if (unlikely(!(vma->vm_flags & VM_READ))) | |
2296 | return false; | |
2e2e3738 | 2297 | |
4d8b81ab XG |
2298 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
2299 | return false; | |
887c08ac | 2300 | |
4d8b81ab XG |
2301 | return true; |
2302 | } | |
bf998156 | 2303 | |
f8be156b NP |
2304 | static int kvm_try_get_pfn(kvm_pfn_t pfn) |
2305 | { | |
2306 | if (kvm_is_reserved_pfn(pfn)) | |
2307 | return 1; | |
2308 | return get_page_unless_zero(pfn_to_page(pfn)); | |
2309 | } | |
2310 | ||
92176a8e PB |
2311 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
2312 | unsigned long addr, bool *async, | |
a340b3e2 KA |
2313 | bool write_fault, bool *writable, |
2314 | kvm_pfn_t *p_pfn) | |
92176a8e | 2315 | { |
a9545779 | 2316 | kvm_pfn_t pfn; |
bd2fae8d PB |
2317 | pte_t *ptep; |
2318 | spinlock_t *ptl; | |
add6a0cd PB |
2319 | int r; |
2320 | ||
9fd6dad1 | 2321 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2322 | if (r) { |
2323 | /* | |
2324 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does | |
2325 | * not call the fault handler, so do it here. | |
2326 | */ | |
2327 | bool unlocked = false; | |
64019a2e | 2328 | r = fixup_user_fault(current->mm, addr, |
add6a0cd PB |
2329 | (write_fault ? FAULT_FLAG_WRITE : 0), |
2330 | &unlocked); | |
a8387d0b PB |
2331 | if (unlocked) |
2332 | return -EAGAIN; | |
add6a0cd PB |
2333 | if (r) |
2334 | return r; | |
2335 | ||
9fd6dad1 | 2336 | r = follow_pte(vma->vm_mm, addr, &ptep, &ptl); |
add6a0cd PB |
2337 | if (r) |
2338 | return r; | |
bd2fae8d | 2339 | } |
add6a0cd | 2340 | |
bd2fae8d PB |
2341 | if (write_fault && !pte_write(*ptep)) { |
2342 | pfn = KVM_PFN_ERR_RO_FAULT; | |
2343 | goto out; | |
add6a0cd PB |
2344 | } |
2345 | ||
a340b3e2 | 2346 | if (writable) |
bd2fae8d PB |
2347 | *writable = pte_write(*ptep); |
2348 | pfn = pte_pfn(*ptep); | |
add6a0cd PB |
2349 | |
2350 | /* | |
2351 | * Get a reference here because callers of *hva_to_pfn* and | |
2352 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the | |
2353 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP | |
36c3ce6c | 2354 | * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will |
add6a0cd PB |
2355 | * simply do nothing for reserved pfns. |
2356 | * | |
2357 | * Whoever called remap_pfn_range is also going to call e.g. | |
2358 | * unmap_mapping_range before the underlying pages are freed, | |
2359 | * causing a call to our MMU notifier. | |
f8be156b NP |
2360 | * |
2361 | * Certain IO or PFNMAP mappings can be backed with valid | |
2362 | * struct pages, but be allocated without refcounting e.g., | |
2363 | * tail pages of non-compound higher order allocations, which | |
2364 | * would then underflow the refcount when the caller does the | |
2365 | * required put_page. Don't allow those pages here. | |
add6a0cd | 2366 | */ |
f8be156b NP |
2367 | if (!kvm_try_get_pfn(pfn)) |
2368 | r = -EFAULT; | |
add6a0cd | 2369 | |
bd2fae8d PB |
2370 | out: |
2371 | pte_unmap_unlock(ptep, ptl); | |
add6a0cd | 2372 | *p_pfn = pfn; |
f8be156b NP |
2373 | |
2374 | return r; | |
92176a8e PB |
2375 | } |
2376 | ||
12ce13fe XG |
2377 | /* |
2378 | * Pin guest page in memory and return its pfn. | |
2379 | * @addr: host virtual address which maps memory to the guest | |
2380 | * @atomic: whether this function can sleep | |
2381 | * @async: whether this function need to wait IO complete if the | |
2382 | * host page is not in the memory | |
2383 | * @write_fault: whether we should get a writable host page | |
2384 | * @writable: whether it allows to map a writable host page for !@write_fault | |
2385 | * | |
2386 | * The function will map a writable host page for these two cases: | |
2387 | * 1): @write_fault = true | |
2388 | * 2): @write_fault = false && @writable, @writable will tell the caller | |
2389 | * whether the mapping is writable. | |
2390 | */ | |
ba049e93 | 2391 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
2fc84311 XG |
2392 | bool write_fault, bool *writable) |
2393 | { | |
2394 | struct vm_area_struct *vma; | |
ba049e93 | 2395 | kvm_pfn_t pfn = 0; |
92176a8e | 2396 | int npages, r; |
2e2e3738 | 2397 | |
2fc84311 XG |
2398 | /* we can do it either atomically or asynchronously, not both */ |
2399 | BUG_ON(atomic && async); | |
8d4e1288 | 2400 | |
b9b33da2 | 2401 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
2fc84311 XG |
2402 | return pfn; |
2403 | ||
2404 | if (atomic) | |
2405 | return KVM_PFN_ERR_FAULT; | |
2406 | ||
2407 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); | |
2408 | if (npages == 1) | |
2409 | return pfn; | |
8d4e1288 | 2410 | |
d8ed45c5 | 2411 | mmap_read_lock(current->mm); |
2fc84311 XG |
2412 | if (npages == -EHWPOISON || |
2413 | (!async && check_user_page_hwpoison(addr))) { | |
2414 | pfn = KVM_PFN_ERR_HWPOISON; | |
2415 | goto exit; | |
2416 | } | |
2417 | ||
a8387d0b | 2418 | retry: |
fc98c03b | 2419 | vma = vma_lookup(current->mm, addr); |
2fc84311 XG |
2420 | |
2421 | if (vma == NULL) | |
2422 | pfn = KVM_PFN_ERR_FAULT; | |
92176a8e | 2423 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
a340b3e2 | 2424 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn); |
a8387d0b PB |
2425 | if (r == -EAGAIN) |
2426 | goto retry; | |
92176a8e PB |
2427 | if (r < 0) |
2428 | pfn = KVM_PFN_ERR_FAULT; | |
2fc84311 | 2429 | } else { |
4d8b81ab | 2430 | if (async && vma_is_valid(vma, write_fault)) |
2fc84311 XG |
2431 | *async = true; |
2432 | pfn = KVM_PFN_ERR_FAULT; | |
2433 | } | |
2434 | exit: | |
d8ed45c5 | 2435 | mmap_read_unlock(current->mm); |
2e2e3738 | 2436 | return pfn; |
35149e21 AL |
2437 | } |
2438 | ||
ba049e93 DW |
2439 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
2440 | bool atomic, bool *async, bool write_fault, | |
4a42d848 | 2441 | bool *writable, hva_t *hva) |
887c08ac | 2442 | { |
4d8b81ab XG |
2443 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
2444 | ||
4a42d848 DS |
2445 | if (hva) |
2446 | *hva = addr; | |
2447 | ||
b2740d35 PB |
2448 | if (addr == KVM_HVA_ERR_RO_BAD) { |
2449 | if (writable) | |
2450 | *writable = false; | |
4d8b81ab | 2451 | return KVM_PFN_ERR_RO_FAULT; |
b2740d35 | 2452 | } |
4d8b81ab | 2453 | |
b2740d35 PB |
2454 | if (kvm_is_error_hva(addr)) { |
2455 | if (writable) | |
2456 | *writable = false; | |
81c52c56 | 2457 | return KVM_PFN_NOSLOT; |
b2740d35 | 2458 | } |
4d8b81ab XG |
2459 | |
2460 | /* Do not map writable pfn in the readonly memslot. */ | |
2461 | if (writable && memslot_is_readonly(slot)) { | |
2462 | *writable = false; | |
2463 | writable = NULL; | |
2464 | } | |
2465 | ||
2466 | return hva_to_pfn(addr, atomic, async, write_fault, | |
2467 | writable); | |
887c08ac | 2468 | } |
3520469d | 2469 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
887c08ac | 2470 | |
ba049e93 | 2471 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
612819c3 MT |
2472 | bool *writable) |
2473 | { | |
e37afc6e | 2474 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
4a42d848 | 2475 | write_fault, writable, NULL); |
612819c3 MT |
2476 | } |
2477 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); | |
2478 | ||
ba049e93 | 2479 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2480 | { |
4a42d848 | 2481 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL); |
506f0d6f | 2482 | } |
e37afc6e | 2483 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
506f0d6f | 2484 | |
ba049e93 | 2485 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
506f0d6f | 2486 | { |
4a42d848 | 2487 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL); |
506f0d6f | 2488 | } |
037d92dc | 2489 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
506f0d6f | 2490 | |
ba049e93 | 2491 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2492 | { |
2493 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2494 | } | |
2495 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); | |
2496 | ||
ba049e93 | 2497 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
e37afc6e PB |
2498 | { |
2499 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); | |
2500 | } | |
2501 | EXPORT_SYMBOL_GPL(gfn_to_pfn); | |
2502 | ||
ba049e93 | 2503 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
8e73485c PB |
2504 | { |
2505 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); | |
2506 | } | |
2507 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); | |
2508 | ||
d9ef13c2 PB |
2509 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2510 | struct page **pages, int nr_pages) | |
48987781 XG |
2511 | { |
2512 | unsigned long addr; | |
076b925d | 2513 | gfn_t entry = 0; |
48987781 | 2514 | |
d9ef13c2 | 2515 | addr = gfn_to_hva_many(slot, gfn, &entry); |
48987781 XG |
2516 | if (kvm_is_error_hva(addr)) |
2517 | return -1; | |
2518 | ||
2519 | if (entry < nr_pages) | |
2520 | return 0; | |
2521 | ||
dadbb612 | 2522 | return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages); |
48987781 XG |
2523 | } |
2524 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); | |
2525 | ||
ba049e93 | 2526 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
a2766325 | 2527 | { |
81c52c56 | 2528 | if (is_error_noslot_pfn(pfn)) |
cb9aaa30 | 2529 | return KVM_ERR_PTR_BAD_PAGE; |
a2766325 | 2530 | |
bf4bea8e | 2531 | if (kvm_is_reserved_pfn(pfn)) { |
cb9aaa30 | 2532 | WARN_ON(1); |
6cede2e6 | 2533 | return KVM_ERR_PTR_BAD_PAGE; |
cb9aaa30 | 2534 | } |
a2766325 XG |
2535 | |
2536 | return pfn_to_page(pfn); | |
2537 | } | |
2538 | ||
35149e21 AL |
2539 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
2540 | { | |
ba049e93 | 2541 | kvm_pfn_t pfn; |
2e2e3738 AL |
2542 | |
2543 | pfn = gfn_to_pfn(kvm, gfn); | |
2e2e3738 | 2544 | |
a2766325 | 2545 | return kvm_pfn_to_page(pfn); |
954bbbc2 AK |
2546 | } |
2547 | EXPORT_SYMBOL_GPL(gfn_to_page); | |
2548 | ||
91724814 BO |
2549 | void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache) |
2550 | { | |
2551 | if (pfn == 0) | |
2552 | return; | |
2553 | ||
2554 | if (cache) | |
2555 | cache->pfn = cache->gfn = 0; | |
2556 | ||
2557 | if (dirty) | |
2558 | kvm_release_pfn_dirty(pfn); | |
2559 | else | |
2560 | kvm_release_pfn_clean(pfn); | |
2561 | } | |
2562 | ||
2563 | static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn, | |
2564 | struct gfn_to_pfn_cache *cache, u64 gen) | |
2565 | { | |
2566 | kvm_release_pfn(cache->pfn, cache->dirty, cache); | |
2567 | ||
2568 | cache->pfn = gfn_to_pfn_memslot(slot, gfn); | |
2569 | cache->gfn = gfn; | |
2570 | cache->dirty = false; | |
2571 | cache->generation = gen; | |
2572 | } | |
2573 | ||
1eff70a9 | 2574 | static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn, |
91724814 BO |
2575 | struct kvm_host_map *map, |
2576 | struct gfn_to_pfn_cache *cache, | |
2577 | bool atomic) | |
e45adf66 KA |
2578 | { |
2579 | kvm_pfn_t pfn; | |
2580 | void *hva = NULL; | |
2581 | struct page *page = KVM_UNMAPPED_PAGE; | |
1eff70a9 | 2582 | struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn); |
91724814 | 2583 | u64 gen = slots->generation; |
e45adf66 KA |
2584 | |
2585 | if (!map) | |
2586 | return -EINVAL; | |
2587 | ||
91724814 BO |
2588 | if (cache) { |
2589 | if (!cache->pfn || cache->gfn != gfn || | |
2590 | cache->generation != gen) { | |
2591 | if (atomic) | |
2592 | return -EAGAIN; | |
2593 | kvm_cache_gfn_to_pfn(slot, gfn, cache, gen); | |
2594 | } | |
2595 | pfn = cache->pfn; | |
2596 | } else { | |
2597 | if (atomic) | |
2598 | return -EAGAIN; | |
2599 | pfn = gfn_to_pfn_memslot(slot, gfn); | |
2600 | } | |
e45adf66 KA |
2601 | if (is_error_noslot_pfn(pfn)) |
2602 | return -EINVAL; | |
2603 | ||
2604 | if (pfn_valid(pfn)) { | |
2605 | page = pfn_to_page(pfn); | |
91724814 BO |
2606 | if (atomic) |
2607 | hva = kmap_atomic(page); | |
2608 | else | |
2609 | hva = kmap(page); | |
d30b214d | 2610 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2611 | } else if (!atomic) { |
e45adf66 | 2612 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
91724814 BO |
2613 | } else { |
2614 | return -EINVAL; | |
d30b214d | 2615 | #endif |
e45adf66 KA |
2616 | } |
2617 | ||
2618 | if (!hva) | |
2619 | return -EFAULT; | |
2620 | ||
2621 | map->page = page; | |
2622 | map->hva = hva; | |
2623 | map->pfn = pfn; | |
2624 | map->gfn = gfn; | |
2625 | ||
2626 | return 0; | |
2627 | } | |
2628 | ||
91724814 BO |
2629 | int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map, |
2630 | struct gfn_to_pfn_cache *cache, bool atomic) | |
1eff70a9 | 2631 | { |
91724814 BO |
2632 | return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map, |
2633 | cache, atomic); | |
1eff70a9 BO |
2634 | } |
2635 | EXPORT_SYMBOL_GPL(kvm_map_gfn); | |
2636 | ||
e45adf66 KA |
2637 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
2638 | { | |
91724814 BO |
2639 | return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map, |
2640 | NULL, false); | |
e45adf66 KA |
2641 | } |
2642 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); | |
2643 | ||
28bd726a PX |
2644 | static void __kvm_unmap_gfn(struct kvm *kvm, |
2645 | struct kvm_memory_slot *memslot, | |
91724814 BO |
2646 | struct kvm_host_map *map, |
2647 | struct gfn_to_pfn_cache *cache, | |
2648 | bool dirty, bool atomic) | |
e45adf66 KA |
2649 | { |
2650 | if (!map) | |
2651 | return; | |
2652 | ||
2653 | if (!map->hva) | |
2654 | return; | |
2655 | ||
91724814 BO |
2656 | if (map->page != KVM_UNMAPPED_PAGE) { |
2657 | if (atomic) | |
2658 | kunmap_atomic(map->hva); | |
2659 | else | |
2660 | kunmap(map->page); | |
2661 | } | |
eb1f2f38 | 2662 | #ifdef CONFIG_HAS_IOMEM |
91724814 | 2663 | else if (!atomic) |
e45adf66 | 2664 | memunmap(map->hva); |
91724814 BO |
2665 | else |
2666 | WARN_ONCE(1, "Unexpected unmapping in atomic context"); | |
eb1f2f38 | 2667 | #endif |
e45adf66 | 2668 | |
91724814 | 2669 | if (dirty) |
28bd726a | 2670 | mark_page_dirty_in_slot(kvm, memslot, map->gfn); |
91724814 BO |
2671 | |
2672 | if (cache) | |
2673 | cache->dirty |= dirty; | |
2674 | else | |
2675 | kvm_release_pfn(map->pfn, dirty, NULL); | |
e45adf66 KA |
2676 | |
2677 | map->hva = NULL; | |
2678 | map->page = NULL; | |
2679 | } | |
1eff70a9 | 2680 | |
91724814 BO |
2681 | int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map, |
2682 | struct gfn_to_pfn_cache *cache, bool dirty, bool atomic) | |
1eff70a9 | 2683 | { |
28bd726a | 2684 | __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map, |
91724814 | 2685 | cache, dirty, atomic); |
1eff70a9 BO |
2686 | return 0; |
2687 | } | |
2688 | EXPORT_SYMBOL_GPL(kvm_unmap_gfn); | |
2689 | ||
2690 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty) | |
2691 | { | |
28bd726a PX |
2692 | __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), |
2693 | map, NULL, dirty, false); | |
1eff70a9 | 2694 | } |
e45adf66 KA |
2695 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); |
2696 | ||
8e73485c PB |
2697 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
2698 | { | |
ba049e93 | 2699 | kvm_pfn_t pfn; |
8e73485c PB |
2700 | |
2701 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); | |
2702 | ||
2703 | return kvm_pfn_to_page(pfn); | |
2704 | } | |
2705 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); | |
2706 | ||
b4231d61 IE |
2707 | void kvm_release_page_clean(struct page *page) |
2708 | { | |
32cad84f XG |
2709 | WARN_ON(is_error_page(page)); |
2710 | ||
35149e21 | 2711 | kvm_release_pfn_clean(page_to_pfn(page)); |
b4231d61 IE |
2712 | } |
2713 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); | |
2714 | ||
ba049e93 | 2715 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
35149e21 | 2716 | { |
bf4bea8e | 2717 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
2e2e3738 | 2718 | put_page(pfn_to_page(pfn)); |
35149e21 AL |
2719 | } |
2720 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); | |
2721 | ||
b4231d61 | 2722 | void kvm_release_page_dirty(struct page *page) |
8a7ae055 | 2723 | { |
a2766325 XG |
2724 | WARN_ON(is_error_page(page)); |
2725 | ||
35149e21 AL |
2726 | kvm_release_pfn_dirty(page_to_pfn(page)); |
2727 | } | |
2728 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); | |
2729 | ||
f7a6509f | 2730 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
35149e21 AL |
2731 | { |
2732 | kvm_set_pfn_dirty(pfn); | |
2733 | kvm_release_pfn_clean(pfn); | |
2734 | } | |
f7a6509f | 2735 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
35149e21 | 2736 | |
ba049e93 | 2737 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
35149e21 | 2738 | { |
d29c03a5 ML |
2739 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2740 | SetPageDirty(pfn_to_page(pfn)); | |
8a7ae055 | 2741 | } |
35149e21 AL |
2742 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
2743 | ||
ba049e93 | 2744 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
35149e21 | 2745 | { |
a78986aa | 2746 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
2e2e3738 | 2747 | mark_page_accessed(pfn_to_page(pfn)); |
35149e21 AL |
2748 | } |
2749 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); | |
2750 | ||
195aefde IE |
2751 | static int next_segment(unsigned long len, int offset) |
2752 | { | |
2753 | if (len > PAGE_SIZE - offset) | |
2754 | return PAGE_SIZE - offset; | |
2755 | else | |
2756 | return len; | |
2757 | } | |
2758 | ||
8e73485c PB |
2759 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
2760 | void *data, int offset, int len) | |
195aefde | 2761 | { |
e0506bcb IE |
2762 | int r; |
2763 | unsigned long addr; | |
195aefde | 2764 | |
8e73485c | 2765 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
e0506bcb IE |
2766 | if (kvm_is_error_hva(addr)) |
2767 | return -EFAULT; | |
3180a7fc | 2768 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
e0506bcb | 2769 | if (r) |
195aefde | 2770 | return -EFAULT; |
195aefde IE |
2771 | return 0; |
2772 | } | |
8e73485c PB |
2773 | |
2774 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, | |
2775 | int len) | |
2776 | { | |
2777 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2778 | ||
2779 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2780 | } | |
195aefde IE |
2781 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
2782 | ||
8e73485c PB |
2783 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
2784 | int offset, int len) | |
2785 | { | |
2786 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2787 | ||
2788 | return __kvm_read_guest_page(slot, gfn, data, offset, len); | |
2789 | } | |
2790 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); | |
2791 | ||
195aefde IE |
2792 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
2793 | { | |
2794 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2795 | int seg; | |
2796 | int offset = offset_in_page(gpa); | |
2797 | int ret; | |
2798 | ||
2799 | while ((seg = next_segment(len, offset)) != 0) { | |
2800 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); | |
2801 | if (ret < 0) | |
2802 | return ret; | |
2803 | offset = 0; | |
2804 | len -= seg; | |
2805 | data += seg; | |
2806 | ++gfn; | |
2807 | } | |
2808 | return 0; | |
2809 | } | |
2810 | EXPORT_SYMBOL_GPL(kvm_read_guest); | |
2811 | ||
8e73485c | 2812 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
7ec54588 | 2813 | { |
7ec54588 | 2814 | gfn_t gfn = gpa >> PAGE_SHIFT; |
8e73485c | 2815 | int seg; |
7ec54588 | 2816 | int offset = offset_in_page(gpa); |
8e73485c PB |
2817 | int ret; |
2818 | ||
2819 | while ((seg = next_segment(len, offset)) != 0) { | |
2820 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); | |
2821 | if (ret < 0) | |
2822 | return ret; | |
2823 | offset = 0; | |
2824 | len -= seg; | |
2825 | data += seg; | |
2826 | ++gfn; | |
2827 | } | |
2828 | return 0; | |
2829 | } | |
2830 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); | |
7ec54588 | 2831 | |
8e73485c PB |
2832 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
2833 | void *data, int offset, unsigned long len) | |
2834 | { | |
2835 | int r; | |
2836 | unsigned long addr; | |
2837 | ||
2838 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); | |
7ec54588 MT |
2839 | if (kvm_is_error_hva(addr)) |
2840 | return -EFAULT; | |
0aac03f0 | 2841 | pagefault_disable(); |
3180a7fc | 2842 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
0aac03f0 | 2843 | pagefault_enable(); |
7ec54588 MT |
2844 | if (r) |
2845 | return -EFAULT; | |
2846 | return 0; | |
2847 | } | |
7ec54588 | 2848 | |
8e73485c PB |
2849 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
2850 | void *data, unsigned long len) | |
2851 | { | |
2852 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2853 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2854 | int offset = offset_in_page(gpa); | |
2855 | ||
2856 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); | |
2857 | } | |
2858 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); | |
2859 | ||
28bd726a PX |
2860 | static int __kvm_write_guest_page(struct kvm *kvm, |
2861 | struct kvm_memory_slot *memslot, gfn_t gfn, | |
8e73485c | 2862 | const void *data, int offset, int len) |
195aefde | 2863 | { |
e0506bcb IE |
2864 | int r; |
2865 | unsigned long addr; | |
195aefde | 2866 | |
251eb841 | 2867 | addr = gfn_to_hva_memslot(memslot, gfn); |
e0506bcb IE |
2868 | if (kvm_is_error_hva(addr)) |
2869 | return -EFAULT; | |
8b0cedff | 2870 | r = __copy_to_user((void __user *)addr + offset, data, len); |
e0506bcb | 2871 | if (r) |
195aefde | 2872 | return -EFAULT; |
28bd726a | 2873 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
195aefde IE |
2874 | return 0; |
2875 | } | |
8e73485c PB |
2876 | |
2877 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, | |
2878 | const void *data, int offset, int len) | |
2879 | { | |
2880 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); | |
2881 | ||
28bd726a | 2882 | return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len); |
8e73485c | 2883 | } |
195aefde IE |
2884 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
2885 | ||
8e73485c PB |
2886 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
2887 | const void *data, int offset, int len) | |
2888 | { | |
2889 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
2890 | ||
28bd726a | 2891 | return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len); |
8e73485c PB |
2892 | } |
2893 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); | |
2894 | ||
195aefde IE |
2895 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
2896 | unsigned long len) | |
2897 | { | |
2898 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2899 | int seg; | |
2900 | int offset = offset_in_page(gpa); | |
2901 | int ret; | |
2902 | ||
2903 | while ((seg = next_segment(len, offset)) != 0) { | |
2904 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); | |
2905 | if (ret < 0) | |
2906 | return ret; | |
2907 | offset = 0; | |
2908 | len -= seg; | |
2909 | data += seg; | |
2910 | ++gfn; | |
2911 | } | |
2912 | return 0; | |
2913 | } | |
ff651cb6 | 2914 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
195aefde | 2915 | |
8e73485c PB |
2916 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
2917 | unsigned long len) | |
2918 | { | |
2919 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
2920 | int seg; | |
2921 | int offset = offset_in_page(gpa); | |
2922 | int ret; | |
2923 | ||
2924 | while ((seg = next_segment(len, offset)) != 0) { | |
2925 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); | |
2926 | if (ret < 0) | |
2927 | return ret; | |
2928 | offset = 0; | |
2929 | len -= seg; | |
2930 | data += seg; | |
2931 | ++gfn; | |
2932 | } | |
2933 | return 0; | |
2934 | } | |
2935 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); | |
2936 | ||
5a2d4365 PB |
2937 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
2938 | struct gfn_to_hva_cache *ghc, | |
2939 | gpa_t gpa, unsigned long len) | |
49c7754c | 2940 | { |
49c7754c | 2941 | int offset = offset_in_page(gpa); |
8f964525 AH |
2942 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
2943 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; | |
2944 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; | |
2945 | gfn_t nr_pages_avail; | |
49c7754c | 2946 | |
6ad1e29f | 2947 | /* Update ghc->generation before performing any error checks. */ |
49c7754c | 2948 | ghc->generation = slots->generation; |
6ad1e29f SC |
2949 | |
2950 | if (start_gfn > end_gfn) { | |
2951 | ghc->hva = KVM_HVA_ERR_BAD; | |
2952 | return -EINVAL; | |
2953 | } | |
f1b9dd5e JM |
2954 | |
2955 | /* | |
2956 | * If the requested region crosses two memslots, we still | |
2957 | * verify that the entire region is valid here. | |
2958 | */ | |
6ad1e29f | 2959 | for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) { |
f1b9dd5e JM |
2960 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
2961 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, | |
2962 | &nr_pages_avail); | |
2963 | if (kvm_is_error_hva(ghc->hva)) | |
6ad1e29f | 2964 | return -EFAULT; |
f1b9dd5e JM |
2965 | } |
2966 | ||
2967 | /* Use the slow path for cross page reads and writes. */ | |
6ad1e29f | 2968 | if (nr_pages_needed == 1) |
49c7754c | 2969 | ghc->hva += offset; |
f1b9dd5e | 2970 | else |
8f964525 | 2971 | ghc->memslot = NULL; |
f1b9dd5e | 2972 | |
6ad1e29f SC |
2973 | ghc->gpa = gpa; |
2974 | ghc->len = len; | |
2975 | return 0; | |
49c7754c | 2976 | } |
5a2d4365 | 2977 | |
4e335d9e | 2978 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
5a2d4365 PB |
2979 | gpa_t gpa, unsigned long len) |
2980 | { | |
4e335d9e | 2981 | struct kvm_memslots *slots = kvm_memslots(kvm); |
5a2d4365 PB |
2982 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
2983 | } | |
4e335d9e | 2984 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
49c7754c | 2985 | |
4e335d9e | 2986 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
7a86dab8 JM |
2987 | void *data, unsigned int offset, |
2988 | unsigned long len) | |
49c7754c | 2989 | { |
4e335d9e | 2990 | struct kvm_memslots *slots = kvm_memslots(kvm); |
49c7754c | 2991 | int r; |
4ec6e863 | 2992 | gpa_t gpa = ghc->gpa + offset; |
49c7754c | 2993 | |
4ec6e863 | 2994 | BUG_ON(len + offset > ghc->len); |
8f964525 | 2995 | |
dc9ce71e SC |
2996 | if (slots->generation != ghc->generation) { |
2997 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
2998 | return -EFAULT; | |
2999 | } | |
8f964525 | 3000 | |
49c7754c GN |
3001 | if (kvm_is_error_hva(ghc->hva)) |
3002 | return -EFAULT; | |
3003 | ||
fcfbc617 SC |
3004 | if (unlikely(!ghc->memslot)) |
3005 | return kvm_write_guest(kvm, gpa, data, len); | |
3006 | ||
4ec6e863 | 3007 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
49c7754c GN |
3008 | if (r) |
3009 | return -EFAULT; | |
28bd726a | 3010 | mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT); |
49c7754c GN |
3011 | |
3012 | return 0; | |
3013 | } | |
4e335d9e | 3014 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
4ec6e863 | 3015 | |
4e335d9e PB |
3016 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3017 | void *data, unsigned long len) | |
4ec6e863 | 3018 | { |
4e335d9e | 3019 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
4ec6e863 | 3020 | } |
4e335d9e | 3021 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
49c7754c | 3022 | |
0958f0ce VK |
3023 | int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
3024 | void *data, unsigned int offset, | |
3025 | unsigned long len) | |
e03b644f | 3026 | { |
4e335d9e | 3027 | struct kvm_memslots *slots = kvm_memslots(kvm); |
e03b644f | 3028 | int r; |
0958f0ce | 3029 | gpa_t gpa = ghc->gpa + offset; |
e03b644f | 3030 | |
0958f0ce | 3031 | BUG_ON(len + offset > ghc->len); |
8f964525 | 3032 | |
dc9ce71e SC |
3033 | if (slots->generation != ghc->generation) { |
3034 | if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len)) | |
3035 | return -EFAULT; | |
3036 | } | |
8f964525 | 3037 | |
e03b644f GN |
3038 | if (kvm_is_error_hva(ghc->hva)) |
3039 | return -EFAULT; | |
3040 | ||
fcfbc617 | 3041 | if (unlikely(!ghc->memslot)) |
0958f0ce | 3042 | return kvm_read_guest(kvm, gpa, data, len); |
fcfbc617 | 3043 | |
0958f0ce | 3044 | r = __copy_from_user(data, (void __user *)ghc->hva + offset, len); |
e03b644f GN |
3045 | if (r) |
3046 | return -EFAULT; | |
3047 | ||
3048 | return 0; | |
3049 | } | |
0958f0ce VK |
3050 | EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached); |
3051 | ||
3052 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, | |
3053 | void *data, unsigned long len) | |
3054 | { | |
3055 | return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len); | |
3056 | } | |
4e335d9e | 3057 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
e03b644f | 3058 | |
195aefde IE |
3059 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
3060 | { | |
2f541442 | 3061 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
195aefde IE |
3062 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3063 | int seg; | |
3064 | int offset = offset_in_page(gpa); | |
3065 | int ret; | |
3066 | ||
bfda0e84 | 3067 | while ((seg = next_segment(len, offset)) != 0) { |
2f541442 | 3068 | ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
195aefde IE |
3069 | if (ret < 0) |
3070 | return ret; | |
3071 | offset = 0; | |
3072 | len -= seg; | |
3073 | ++gfn; | |
3074 | } | |
3075 | return 0; | |
3076 | } | |
3077 | EXPORT_SYMBOL_GPL(kvm_clear_guest); | |
3078 | ||
28bd726a PX |
3079 | void mark_page_dirty_in_slot(struct kvm *kvm, |
3080 | struct kvm_memory_slot *memslot, | |
3081 | gfn_t gfn) | |
6aa8b732 | 3082 | { |
044c59c4 | 3083 | if (memslot && kvm_slot_dirty_track_enabled(memslot)) { |
7e9d619d | 3084 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
fb04a1ed | 3085 | u32 slot = (memslot->as_id << 16) | memslot->id; |
6aa8b732 | 3086 | |
fb04a1ed PX |
3087 | if (kvm->dirty_ring_size) |
3088 | kvm_dirty_ring_push(kvm_dirty_ring_get(kvm), | |
3089 | slot, rel_gfn); | |
3090 | else | |
3091 | set_bit_le(rel_gfn, memslot->dirty_bitmap); | |
6aa8b732 AK |
3092 | } |
3093 | } | |
a6a0b05d | 3094 | EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot); |
6aa8b732 | 3095 | |
49c7754c GN |
3096 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
3097 | { | |
3098 | struct kvm_memory_slot *memslot; | |
3099 | ||
3100 | memslot = gfn_to_memslot(kvm, gfn); | |
28bd726a | 3101 | mark_page_dirty_in_slot(kvm, memslot, gfn); |
49c7754c | 3102 | } |
2ba9f0d8 | 3103 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
49c7754c | 3104 | |
8e73485c PB |
3105 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
3106 | { | |
3107 | struct kvm_memory_slot *memslot; | |
3108 | ||
3109 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
28bd726a | 3110 | mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn); |
8e73485c PB |
3111 | } |
3112 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); | |
3113 | ||
20b7035c JS |
3114 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
3115 | { | |
3116 | if (!vcpu->sigset_active) | |
3117 | return; | |
3118 | ||
3119 | /* | |
3120 | * This does a lockless modification of ->real_blocked, which is fine | |
3121 | * because, only current can change ->real_blocked and all readers of | |
3122 | * ->real_blocked don't care as long ->real_blocked is always a subset | |
3123 | * of ->blocked. | |
3124 | */ | |
3125 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); | |
3126 | } | |
3127 | ||
3128 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) | |
3129 | { | |
3130 | if (!vcpu->sigset_active) | |
3131 | return; | |
3132 | ||
3133 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); | |
3134 | sigemptyset(¤t->real_blocked); | |
3135 | } | |
3136 | ||
aca6ff29 WL |
3137 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
3138 | { | |
dee339b5 | 3139 | unsigned int old, val, grow, grow_start; |
aca6ff29 | 3140 | |
2cbd7824 | 3141 | old = val = vcpu->halt_poll_ns; |
dee339b5 | 3142 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3143 | grow = READ_ONCE(halt_poll_ns_grow); |
7fa08e71 NW |
3144 | if (!grow) |
3145 | goto out; | |
3146 | ||
dee339b5 NW |
3147 | val *= grow; |
3148 | if (val < grow_start) | |
3149 | val = grow_start; | |
aca6ff29 | 3150 | |
258785ef DM |
3151 | if (val > vcpu->kvm->max_halt_poll_ns) |
3152 | val = vcpu->kvm->max_halt_poll_ns; | |
313f636d | 3153 | |
aca6ff29 | 3154 | vcpu->halt_poll_ns = val; |
7fa08e71 | 3155 | out: |
2cbd7824 | 3156 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3157 | } |
3158 | ||
3159 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) | |
3160 | { | |
ae232ea4 | 3161 | unsigned int old, val, shrink, grow_start; |
aca6ff29 | 3162 | |
2cbd7824 | 3163 | old = val = vcpu->halt_poll_ns; |
6b6de68c | 3164 | shrink = READ_ONCE(halt_poll_ns_shrink); |
ae232ea4 | 3165 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
6b6de68c | 3166 | if (shrink == 0) |
aca6ff29 WL |
3167 | val = 0; |
3168 | else | |
6b6de68c | 3169 | val /= shrink; |
aca6ff29 | 3170 | |
ae232ea4 SS |
3171 | if (val < grow_start) |
3172 | val = 0; | |
3173 | ||
aca6ff29 | 3174 | vcpu->halt_poll_ns = val; |
2cbd7824 | 3175 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
aca6ff29 WL |
3176 | } |
3177 | ||
f7819512 PB |
3178 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
3179 | { | |
50c28f21 JS |
3180 | int ret = -EINTR; |
3181 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
3182 | ||
f7819512 PB |
3183 | if (kvm_arch_vcpu_runnable(vcpu)) { |
3184 | kvm_make_request(KVM_REQ_UNHALT, vcpu); | |
50c28f21 | 3185 | goto out; |
f7819512 PB |
3186 | } |
3187 | if (kvm_cpu_has_pending_timer(vcpu)) | |
50c28f21 | 3188 | goto out; |
f7819512 | 3189 | if (signal_pending(current)) |
50c28f21 | 3190 | goto out; |
084071d5 MT |
3191 | if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu)) |
3192 | goto out; | |
f7819512 | 3193 | |
50c28f21 JS |
3194 | ret = 0; |
3195 | out: | |
3196 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
3197 | return ret; | |
f7819512 PB |
3198 | } |
3199 | ||
cb953129 DM |
3200 | static inline void |
3201 | update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited) | |
3202 | { | |
3203 | if (waited) | |
0193cc90 | 3204 | vcpu->stat.generic.halt_poll_fail_ns += poll_ns; |
cb953129 | 3205 | else |
0193cc90 | 3206 | vcpu->stat.generic.halt_poll_success_ns += poll_ns; |
cb953129 DM |
3207 | } |
3208 | ||
b6958ce4 ED |
3209 | /* |
3210 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. | |
3211 | */ | |
8776e519 | 3212 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
d3bef15f | 3213 | { |
cb953129 | 3214 | ktime_t start, cur, poll_end; |
f7819512 | 3215 | bool waited = false; |
aca6ff29 | 3216 | u64 block_ns; |
f7819512 | 3217 | |
07ab0f8d MZ |
3218 | kvm_arch_vcpu_blocking(vcpu); |
3219 | ||
cb953129 | 3220 | start = cur = poll_end = ktime_get(); |
cdd6ad3a | 3221 | if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) { |
19020f8a | 3222 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); |
f95ef0cd | 3223 | |
0193cc90 | 3224 | ++vcpu->stat.generic.halt_attempted_poll; |
f7819512 PB |
3225 | do { |
3226 | /* | |
3227 | * This sets KVM_REQ_UNHALT if an interrupt | |
3228 | * arrives. | |
3229 | */ | |
3230 | if (kvm_vcpu_check_block(vcpu) < 0) { | |
0193cc90 | 3231 | ++vcpu->stat.generic.halt_successful_poll; |
3491caf2 | 3232 | if (!vcpu_valid_wakeup(vcpu)) |
0193cc90 | 3233 | ++vcpu->stat.generic.halt_poll_invalid; |
8ccba534 JZ |
3234 | |
3235 | KVM_STATS_LOG_HIST_UPDATE( | |
3236 | vcpu->stat.generic.halt_poll_success_hist, | |
3237 | ktime_to_ns(ktime_get()) - | |
3238 | ktime_to_ns(start)); | |
f7819512 PB |
3239 | goto out; |
3240 | } | |
74775654 | 3241 | cpu_relax(); |
cb953129 | 3242 | poll_end = cur = ktime_get(); |
6bd5b743 | 3243 | } while (kvm_vcpu_can_poll(cur, stop)); |
8ccba534 JZ |
3244 | |
3245 | KVM_STATS_LOG_HIST_UPDATE( | |
3246 | vcpu->stat.generic.halt_poll_fail_hist, | |
3247 | ktime_to_ns(ktime_get()) - ktime_to_ns(start)); | |
f7819512 | 3248 | } |
e5c239cf | 3249 | |
8ccba534 | 3250 | |
da4ad88c | 3251 | prepare_to_rcuwait(&vcpu->wait); |
e5c239cf | 3252 | for (;;) { |
da4ad88c | 3253 | set_current_state(TASK_INTERRUPTIBLE); |
e5c239cf | 3254 | |
f7819512 | 3255 | if (kvm_vcpu_check_block(vcpu) < 0) |
e5c239cf MT |
3256 | break; |
3257 | ||
f7819512 | 3258 | waited = true; |
b6958ce4 | 3259 | schedule(); |
b6958ce4 | 3260 | } |
da4ad88c | 3261 | finish_rcuwait(&vcpu->wait); |
f7819512 | 3262 | cur = ktime_get(); |
87bcc5fa JZ |
3263 | if (waited) { |
3264 | vcpu->stat.generic.halt_wait_ns += | |
3265 | ktime_to_ns(cur) - ktime_to_ns(poll_end); | |
8ccba534 JZ |
3266 | KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist, |
3267 | ktime_to_ns(cur) - ktime_to_ns(poll_end)); | |
87bcc5fa | 3268 | } |
f7819512 | 3269 | out: |
07ab0f8d | 3270 | kvm_arch_vcpu_unblocking(vcpu); |
aca6ff29 WL |
3271 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
3272 | ||
cb953129 DM |
3273 | update_halt_poll_stats( |
3274 | vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited); | |
3275 | ||
44551b2f WL |
3276 | if (!kvm_arch_no_poll(vcpu)) { |
3277 | if (!vcpu_valid_wakeup(vcpu)) { | |
aca6ff29 | 3278 | shrink_halt_poll_ns(vcpu); |
acd05785 | 3279 | } else if (vcpu->kvm->max_halt_poll_ns) { |
44551b2f WL |
3280 | if (block_ns <= vcpu->halt_poll_ns) |
3281 | ; | |
3282 | /* we had a long block, shrink polling */ | |
acd05785 DM |
3283 | else if (vcpu->halt_poll_ns && |
3284 | block_ns > vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
3285 | shrink_halt_poll_ns(vcpu); |
3286 | /* we had a short halt and our poll time is too small */ | |
acd05785 DM |
3287 | else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && |
3288 | block_ns < vcpu->kvm->max_halt_poll_ns) | |
44551b2f WL |
3289 | grow_halt_poll_ns(vcpu); |
3290 | } else { | |
3291 | vcpu->halt_poll_ns = 0; | |
3292 | } | |
3293 | } | |
aca6ff29 | 3294 | |
3491caf2 CB |
3295 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
3296 | kvm_arch_vcpu_block_finish(vcpu); | |
b6958ce4 | 3297 | } |
2ba9f0d8 | 3298 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
b6958ce4 | 3299 | |
178f02ff | 3300 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
b6d33834 | 3301 | { |
da4ad88c | 3302 | struct rcuwait *waitp; |
b6d33834 | 3303 | |
da4ad88c DB |
3304 | waitp = kvm_arch_vcpu_get_wait(vcpu); |
3305 | if (rcuwait_wake_up(waitp)) { | |
d73eb57b | 3306 | WRITE_ONCE(vcpu->ready, true); |
0193cc90 | 3307 | ++vcpu->stat.generic.halt_wakeup; |
178f02ff | 3308 | return true; |
b6d33834 CD |
3309 | } |
3310 | ||
178f02ff | 3311 | return false; |
dd1a4cc1 RK |
3312 | } |
3313 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); | |
3314 | ||
0266c894 | 3315 | #ifndef CONFIG_S390 |
dd1a4cc1 RK |
3316 | /* |
3317 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. | |
3318 | */ | |
3319 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) | |
3320 | { | |
85b64045 | 3321 | int me, cpu; |
dd1a4cc1 | 3322 | |
178f02ff RK |
3323 | if (kvm_vcpu_wake_up(vcpu)) |
3324 | return; | |
3325 | ||
85b64045 SC |
3326 | /* |
3327 | * Note, the vCPU could get migrated to a different pCPU at any point | |
3328 | * after kvm_arch_vcpu_should_kick(), which could result in sending an | |
3329 | * IPI to the previous pCPU. But, that's ok because the purpose of the | |
3330 | * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the | |
3331 | * vCPU also requires it to leave IN_GUEST_MODE. | |
3332 | */ | |
b6d33834 | 3333 | me = get_cpu(); |
85b64045 SC |
3334 | if (kvm_arch_vcpu_should_kick(vcpu)) { |
3335 | cpu = READ_ONCE(vcpu->cpu); | |
3336 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) | |
b6d33834 | 3337 | smp_send_reschedule(cpu); |
85b64045 | 3338 | } |
b6d33834 CD |
3339 | put_cpu(); |
3340 | } | |
a20ed54d | 3341 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
0266c894 | 3342 | #endif /* !CONFIG_S390 */ |
b6d33834 | 3343 | |
fa93384f | 3344 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
41628d33 KW |
3345 | { |
3346 | struct pid *pid; | |
3347 | struct task_struct *task = NULL; | |
fa93384f | 3348 | int ret = 0; |
41628d33 KW |
3349 | |
3350 | rcu_read_lock(); | |
3351 | pid = rcu_dereference(target->pid); | |
3352 | if (pid) | |
27fbe64b | 3353 | task = get_pid_task(pid, PIDTYPE_PID); |
41628d33 KW |
3354 | rcu_read_unlock(); |
3355 | if (!task) | |
c45c528e | 3356 | return ret; |
c45c528e | 3357 | ret = yield_to(task, 1); |
41628d33 | 3358 | put_task_struct(task); |
c45c528e R |
3359 | |
3360 | return ret; | |
41628d33 KW |
3361 | } |
3362 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); | |
3363 | ||
06e48c51 R |
3364 | /* |
3365 | * Helper that checks whether a VCPU is eligible for directed yield. | |
3366 | * Most eligible candidate to yield is decided by following heuristics: | |
3367 | * | |
3368 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently | |
3369 | * (preempted lock holder), indicated by @in_spin_loop. | |
656012c7 | 3370 | * Set at the beginning and cleared at the end of interception/PLE handler. |
06e48c51 R |
3371 | * |
3372 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get | |
3373 | * chance last time (mostly it has become eligible now since we have probably | |
3374 | * yielded to lockholder in last iteration. This is done by toggling | |
3375 | * @dy_eligible each time a VCPU checked for eligibility.) | |
3376 | * | |
3377 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding | |
3378 | * to preempted lock-holder could result in wrong VCPU selection and CPU | |
3379 | * burning. Giving priority for a potential lock-holder increases lock | |
3380 | * progress. | |
3381 | * | |
3382 | * Since algorithm is based on heuristics, accessing another VCPU data without | |
3383 | * locking does not harm. It may result in trying to yield to same VCPU, fail | |
3384 | * and continue with next VCPU and so on. | |
3385 | */ | |
7940876e | 3386 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
06e48c51 | 3387 | { |
4a55dd72 | 3388 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
06e48c51 R |
3389 | bool eligible; |
3390 | ||
3391 | eligible = !vcpu->spin_loop.in_spin_loop || | |
34656113 | 3392 | vcpu->spin_loop.dy_eligible; |
06e48c51 R |
3393 | |
3394 | if (vcpu->spin_loop.in_spin_loop) | |
3395 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); | |
3396 | ||
3397 | return eligible; | |
4a55dd72 SW |
3398 | #else |
3399 | return true; | |
06e48c51 | 3400 | #endif |
4a55dd72 | 3401 | } |
c45c528e | 3402 | |
17e433b5 WL |
3403 | /* |
3404 | * Unlike kvm_arch_vcpu_runnable, this function is called outside | |
3405 | * a vcpu_load/vcpu_put pair. However, for most architectures | |
3406 | * kvm_arch_vcpu_runnable does not require vcpu_load. | |
3407 | */ | |
3408 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) | |
3409 | { | |
3410 | return kvm_arch_vcpu_runnable(vcpu); | |
3411 | } | |
3412 | ||
3413 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) | |
3414 | { | |
3415 | if (kvm_arch_dy_runnable(vcpu)) | |
3416 | return true; | |
3417 | ||
3418 | #ifdef CONFIG_KVM_ASYNC_PF | |
3419 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
3420 | return true; | |
3421 | #endif | |
3422 | ||
3423 | return false; | |
3424 | } | |
3425 | ||
52acd22f WL |
3426 | bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
3427 | { | |
3428 | return false; | |
3429 | } | |
3430 | ||
199b5763 | 3431 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
d255f4f2 | 3432 | { |
217ece61 RR |
3433 | struct kvm *kvm = me->kvm; |
3434 | struct kvm_vcpu *vcpu; | |
3435 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; | |
3436 | int yielded = 0; | |
c45c528e | 3437 | int try = 3; |
217ece61 RR |
3438 | int pass; |
3439 | int i; | |
d255f4f2 | 3440 | |
4c088493 | 3441 | kvm_vcpu_set_in_spin_loop(me, true); |
217ece61 RR |
3442 | /* |
3443 | * We boost the priority of a VCPU that is runnable but not | |
3444 | * currently running, because it got preempted by something | |
3445 | * else and called schedule in __vcpu_run. Hopefully that | |
3446 | * VCPU is holding the lock that we need and will release it. | |
3447 | * We approximate round-robin by starting at the last boosted VCPU. | |
3448 | */ | |
c45c528e | 3449 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
217ece61 | 3450 | kvm_for_each_vcpu(i, vcpu, kvm) { |
5cfc2aab | 3451 | if (!pass && i <= last_boosted_vcpu) { |
217ece61 RR |
3452 | i = last_boosted_vcpu; |
3453 | continue; | |
3454 | } else if (pass && i > last_boosted_vcpu) | |
3455 | break; | |
d73eb57b | 3456 | if (!READ_ONCE(vcpu->ready)) |
7bc7ae25 | 3457 | continue; |
217ece61 RR |
3458 | if (vcpu == me) |
3459 | continue; | |
da4ad88c DB |
3460 | if (rcuwait_active(&vcpu->wait) && |
3461 | !vcpu_dy_runnable(vcpu)) | |
217ece61 | 3462 | continue; |
046ddeed | 3463 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
52acd22f WL |
3464 | !kvm_arch_dy_has_pending_interrupt(vcpu) && |
3465 | !kvm_arch_vcpu_in_kernel(vcpu)) | |
199b5763 | 3466 | continue; |
06e48c51 R |
3467 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
3468 | continue; | |
c45c528e R |
3469 | |
3470 | yielded = kvm_vcpu_yield_to(vcpu); | |
3471 | if (yielded > 0) { | |
217ece61 | 3472 | kvm->last_boosted_vcpu = i; |
217ece61 | 3473 | break; |
c45c528e R |
3474 | } else if (yielded < 0) { |
3475 | try--; | |
3476 | if (!try) | |
3477 | break; | |
217ece61 | 3478 | } |
217ece61 RR |
3479 | } |
3480 | } | |
4c088493 | 3481 | kvm_vcpu_set_in_spin_loop(me, false); |
06e48c51 R |
3482 | |
3483 | /* Ensure vcpu is not eligible during next spinloop */ | |
3484 | kvm_vcpu_set_dy_eligible(me, false); | |
d255f4f2 ZE |
3485 | } |
3486 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); | |
3487 | ||
fb04a1ed PX |
3488 | static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff) |
3489 | { | |
3490 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
3491 | return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) && | |
3492 | (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET + | |
3493 | kvm->dirty_ring_size / PAGE_SIZE); | |
3494 | #else | |
3495 | return false; | |
3496 | #endif | |
3497 | } | |
3498 | ||
1499fa80 | 3499 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
9a2bb7f4 | 3500 | { |
11bac800 | 3501 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
9a2bb7f4 AK |
3502 | struct page *page; |
3503 | ||
e4a533a4 | 3504 | if (vmf->pgoff == 0) |
039576c0 | 3505 | page = virt_to_page(vcpu->run); |
09566765 | 3506 | #ifdef CONFIG_X86 |
e4a533a4 | 3507 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
ad312c7c | 3508 | page = virt_to_page(vcpu->arch.pio_data); |
5f94c174 | 3509 | #endif |
4b4357e0 | 3510 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
3511 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
3512 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); | |
09566765 | 3513 | #endif |
fb04a1ed PX |
3514 | else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff)) |
3515 | page = kvm_dirty_ring_get_page( | |
3516 | &vcpu->dirty_ring, | |
3517 | vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET); | |
039576c0 | 3518 | else |
5b1c1493 | 3519 | return kvm_arch_vcpu_fault(vcpu, vmf); |
9a2bb7f4 | 3520 | get_page(page); |
e4a533a4 NP |
3521 | vmf->page = page; |
3522 | return 0; | |
9a2bb7f4 AK |
3523 | } |
3524 | ||
f0f37e2f | 3525 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
e4a533a4 | 3526 | .fault = kvm_vcpu_fault, |
9a2bb7f4 AK |
3527 | }; |
3528 | ||
3529 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) | |
3530 | { | |
fb04a1ed PX |
3531 | struct kvm_vcpu *vcpu = file->private_data; |
3532 | unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; | |
3533 | ||
3534 | if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) || | |
3535 | kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) && | |
3536 | ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED))) | |
3537 | return -EINVAL; | |
3538 | ||
9a2bb7f4 AK |
3539 | vma->vm_ops = &kvm_vcpu_vm_ops; |
3540 | return 0; | |
3541 | } | |
3542 | ||
bccf2150 AK |
3543 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
3544 | { | |
3545 | struct kvm_vcpu *vcpu = filp->private_data; | |
3546 | ||
66c0b394 | 3547 | kvm_put_kvm(vcpu->kvm); |
bccf2150 AK |
3548 | return 0; |
3549 | } | |
3550 | ||
3d3aab1b | 3551 | static struct file_operations kvm_vcpu_fops = { |
bccf2150 AK |
3552 | .release = kvm_vcpu_release, |
3553 | .unlocked_ioctl = kvm_vcpu_ioctl, | |
9a2bb7f4 | 3554 | .mmap = kvm_vcpu_mmap, |
6038f373 | 3555 | .llseek = noop_llseek, |
7ddfd3e0 | 3556 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
bccf2150 AK |
3557 | }; |
3558 | ||
3559 | /* | |
3560 | * Allocates an inode for the vcpu. | |
3561 | */ | |
3562 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) | |
3563 | { | |
e46b4692 MY |
3564 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
3565 | ||
3566 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); | |
3567 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); | |
bccf2150 AK |
3568 | } |
3569 | ||
3e7093d0 | 3570 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
45b5939e | 3571 | { |
741cbbae | 3572 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
d56f5136 | 3573 | struct dentry *debugfs_dentry; |
45b5939e | 3574 | char dir_name[ITOA_MAX_LEN * 2]; |
45b5939e | 3575 | |
45b5939e | 3576 | if (!debugfs_initialized()) |
3e7093d0 | 3577 | return; |
45b5939e LC |
3578 | |
3579 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); | |
d56f5136 PB |
3580 | debugfs_dentry = debugfs_create_dir(dir_name, |
3581 | vcpu->kvm->debugfs_dentry); | |
45b5939e | 3582 | |
d56f5136 | 3583 | kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry); |
741cbbae | 3584 | #endif |
45b5939e LC |
3585 | } |
3586 | ||
c5ea7660 AK |
3587 | /* |
3588 | * Creates some virtual cpus. Good luck creating more than one. | |
3589 | */ | |
73880c80 | 3590 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
c5ea7660 AK |
3591 | { |
3592 | int r; | |
e09fefde | 3593 | struct kvm_vcpu *vcpu; |
8bd826d6 | 3594 | struct page *page; |
c5ea7660 | 3595 | |
0b1b1dfd | 3596 | if (id >= KVM_MAX_VCPU_ID) |
338c7dba AH |
3597 | return -EINVAL; |
3598 | ||
6c7caebc PB |
3599 | mutex_lock(&kvm->lock); |
3600 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { | |
3601 | mutex_unlock(&kvm->lock); | |
3602 | return -EINVAL; | |
3603 | } | |
3604 | ||
3605 | kvm->created_vcpus++; | |
3606 | mutex_unlock(&kvm->lock); | |
3607 | ||
897cc38e SC |
3608 | r = kvm_arch_vcpu_precreate(kvm, id); |
3609 | if (r) | |
3610 | goto vcpu_decrement; | |
3611 | ||
85f47930 | 3612 | vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT); |
e529ef66 SC |
3613 | if (!vcpu) { |
3614 | r = -ENOMEM; | |
6c7caebc PB |
3615 | goto vcpu_decrement; |
3616 | } | |
c5ea7660 | 3617 | |
fcd97ad5 | 3618 | BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE); |
93bb59ca | 3619 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
8bd826d6 SC |
3620 | if (!page) { |
3621 | r = -ENOMEM; | |
e529ef66 | 3622 | goto vcpu_free; |
8bd826d6 SC |
3623 | } |
3624 | vcpu->run = page_address(page); | |
3625 | ||
3626 | kvm_vcpu_init(vcpu, kvm, id); | |
e529ef66 SC |
3627 | |
3628 | r = kvm_arch_vcpu_create(vcpu); | |
3629 | if (r) | |
8bd826d6 | 3630 | goto vcpu_free_run_page; |
e529ef66 | 3631 | |
fb04a1ed PX |
3632 | if (kvm->dirty_ring_size) { |
3633 | r = kvm_dirty_ring_alloc(&vcpu->dirty_ring, | |
3634 | id, kvm->dirty_ring_size); | |
3635 | if (r) | |
3636 | goto arch_vcpu_destroy; | |
3637 | } | |
3638 | ||
11ec2804 | 3639 | mutex_lock(&kvm->lock); |
e09fefde DH |
3640 | if (kvm_get_vcpu_by_id(kvm, id)) { |
3641 | r = -EEXIST; | |
3642 | goto unlock_vcpu_destroy; | |
3643 | } | |
73880c80 | 3644 | |
8750e72a RK |
3645 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
3646 | BUG_ON(kvm->vcpus[vcpu->vcpu_idx]); | |
c5ea7660 | 3647 | |
ce55c049 JZ |
3648 | /* Fill the stats id string for the vcpu */ |
3649 | snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d", | |
3650 | task_pid_nr(current), id); | |
3651 | ||
fb3f0f51 | 3652 | /* Now it's all set up, let userspace reach it */ |
66c0b394 | 3653 | kvm_get_kvm(kvm); |
bccf2150 | 3654 | r = create_vcpu_fd(vcpu); |
73880c80 | 3655 | if (r < 0) { |
149487bd | 3656 | kvm_put_kvm_no_destroy(kvm); |
d780592b | 3657 | goto unlock_vcpu_destroy; |
73880c80 GN |
3658 | } |
3659 | ||
8750e72a | 3660 | kvm->vcpus[vcpu->vcpu_idx] = vcpu; |
dd489240 PB |
3661 | |
3662 | /* | |
3663 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus | |
3664 | * before kvm->online_vcpu's incremented value. | |
3665 | */ | |
73880c80 GN |
3666 | smp_wmb(); |
3667 | atomic_inc(&kvm->online_vcpus); | |
3668 | ||
73880c80 | 3669 | mutex_unlock(&kvm->lock); |
42897d86 | 3670 | kvm_arch_vcpu_postcreate(vcpu); |
63d04348 | 3671 | kvm_create_vcpu_debugfs(vcpu); |
fb3f0f51 | 3672 | return r; |
39c3b86e | 3673 | |
d780592b | 3674 | unlock_vcpu_destroy: |
7d8fece6 | 3675 | mutex_unlock(&kvm->lock); |
fb04a1ed PX |
3676 | kvm_dirty_ring_free(&vcpu->dirty_ring); |
3677 | arch_vcpu_destroy: | |
d40ccc62 | 3678 | kvm_arch_vcpu_destroy(vcpu); |
8bd826d6 SC |
3679 | vcpu_free_run_page: |
3680 | free_page((unsigned long)vcpu->run); | |
e529ef66 SC |
3681 | vcpu_free: |
3682 | kmem_cache_free(kvm_vcpu_cache, vcpu); | |
6c7caebc PB |
3683 | vcpu_decrement: |
3684 | mutex_lock(&kvm->lock); | |
3685 | kvm->created_vcpus--; | |
3686 | mutex_unlock(&kvm->lock); | |
c5ea7660 AK |
3687 | return r; |
3688 | } | |
3689 | ||
1961d276 AK |
3690 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
3691 | { | |
3692 | if (sigset) { | |
3693 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); | |
3694 | vcpu->sigset_active = 1; | |
3695 | vcpu->sigset = *sigset; | |
3696 | } else | |
3697 | vcpu->sigset_active = 0; | |
3698 | return 0; | |
3699 | } | |
3700 | ||
ce55c049 JZ |
3701 | static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer, |
3702 | size_t size, loff_t *offset) | |
3703 | { | |
3704 | struct kvm_vcpu *vcpu = file->private_data; | |
3705 | ||
3706 | return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header, | |
3707 | &kvm_vcpu_stats_desc[0], &vcpu->stat, | |
3708 | sizeof(vcpu->stat), user_buffer, size, offset); | |
3709 | } | |
3710 | ||
3711 | static const struct file_operations kvm_vcpu_stats_fops = { | |
3712 | .read = kvm_vcpu_stats_read, | |
3713 | .llseek = noop_llseek, | |
3714 | }; | |
3715 | ||
3716 | static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu) | |
3717 | { | |
3718 | int fd; | |
3719 | struct file *file; | |
3720 | char name[15 + ITOA_MAX_LEN + 1]; | |
3721 | ||
3722 | snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id); | |
3723 | ||
3724 | fd = get_unused_fd_flags(O_CLOEXEC); | |
3725 | if (fd < 0) | |
3726 | return fd; | |
3727 | ||
3728 | file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY); | |
3729 | if (IS_ERR(file)) { | |
3730 | put_unused_fd(fd); | |
3731 | return PTR_ERR(file); | |
3732 | } | |
3733 | file->f_mode |= FMODE_PREAD; | |
3734 | fd_install(fd, file); | |
3735 | ||
3736 | return fd; | |
3737 | } | |
3738 | ||
bccf2150 AK |
3739 | static long kvm_vcpu_ioctl(struct file *filp, |
3740 | unsigned int ioctl, unsigned long arg) | |
6aa8b732 | 3741 | { |
bccf2150 | 3742 | struct kvm_vcpu *vcpu = filp->private_data; |
2f366987 | 3743 | void __user *argp = (void __user *)arg; |
313a3dc7 | 3744 | int r; |
fa3795a7 DH |
3745 | struct kvm_fpu *fpu = NULL; |
3746 | struct kvm_sregs *kvm_sregs = NULL; | |
6aa8b732 | 3747 | |
0b8f1173 | 3748 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged) |
6d4e4c4f | 3749 | return -EIO; |
2122ff5e | 3750 | |
2ea75be3 DM |
3751 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
3752 | return -EINVAL; | |
3753 | ||
2122ff5e | 3754 | /* |
5cb0944c PB |
3755 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
3756 | * execution; mutex_lock() would break them. | |
2122ff5e | 3757 | */ |
5cb0944c PB |
3758 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
3759 | if (r != -ENOIOCTLCMD) | |
9fc77441 | 3760 | return r; |
2122ff5e | 3761 | |
ec7660cc CD |
3762 | if (mutex_lock_killable(&vcpu->mutex)) |
3763 | return -EINTR; | |
6aa8b732 | 3764 | switch (ioctl) { |
0e4524a5 CB |
3765 | case KVM_RUN: { |
3766 | struct pid *oldpid; | |
f0fe5108 AK |
3767 | r = -EINVAL; |
3768 | if (arg) | |
3769 | goto out; | |
0e4524a5 | 3770 | oldpid = rcu_access_pointer(vcpu->pid); |
71dbc8a9 | 3771 | if (unlikely(oldpid != task_pid(current))) { |
7a72f7a1 | 3772 | /* The thread running this VCPU changed. */ |
bd2a6394 | 3773 | struct pid *newpid; |
f95ef0cd | 3774 | |
bd2a6394 CD |
3775 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
3776 | if (r) | |
3777 | break; | |
3778 | ||
3779 | newpid = get_task_pid(current, PIDTYPE_PID); | |
7a72f7a1 CB |
3780 | rcu_assign_pointer(vcpu->pid, newpid); |
3781 | if (oldpid) | |
3782 | synchronize_rcu(); | |
3783 | put_pid(oldpid); | |
3784 | } | |
1b94f6f8 | 3785 | r = kvm_arch_vcpu_ioctl_run(vcpu); |
64be5007 | 3786 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
6aa8b732 | 3787 | break; |
0e4524a5 | 3788 | } |
6aa8b732 | 3789 | case KVM_GET_REGS: { |
3e4bb3ac | 3790 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3791 | |
3e4bb3ac | 3792 | r = -ENOMEM; |
b12ce36a | 3793 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
3e4bb3ac | 3794 | if (!kvm_regs) |
6aa8b732 | 3795 | goto out; |
3e4bb3ac XZ |
3796 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
3797 | if (r) | |
3798 | goto out_free1; | |
6aa8b732 | 3799 | r = -EFAULT; |
3e4bb3ac XZ |
3800 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
3801 | goto out_free1; | |
6aa8b732 | 3802 | r = 0; |
3e4bb3ac XZ |
3803 | out_free1: |
3804 | kfree(kvm_regs); | |
6aa8b732 AK |
3805 | break; |
3806 | } | |
3807 | case KVM_SET_REGS: { | |
3e4bb3ac | 3808 | struct kvm_regs *kvm_regs; |
6aa8b732 | 3809 | |
ff5c2c03 SL |
3810 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
3811 | if (IS_ERR(kvm_regs)) { | |
3812 | r = PTR_ERR(kvm_regs); | |
6aa8b732 | 3813 | goto out; |
ff5c2c03 | 3814 | } |
3e4bb3ac | 3815 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
3e4bb3ac | 3816 | kfree(kvm_regs); |
6aa8b732 AK |
3817 | break; |
3818 | } | |
3819 | case KVM_GET_SREGS: { | |
b12ce36a BG |
3820 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
3821 | GFP_KERNEL_ACCOUNT); | |
fa3795a7 DH |
3822 | r = -ENOMEM; |
3823 | if (!kvm_sregs) | |
3824 | goto out; | |
3825 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); | |
6aa8b732 AK |
3826 | if (r) |
3827 | goto out; | |
3828 | r = -EFAULT; | |
fa3795a7 | 3829 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
6aa8b732 AK |
3830 | goto out; |
3831 | r = 0; | |
3832 | break; | |
3833 | } | |
3834 | case KVM_SET_SREGS: { | |
ff5c2c03 SL |
3835 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
3836 | if (IS_ERR(kvm_sregs)) { | |
3837 | r = PTR_ERR(kvm_sregs); | |
18595411 | 3838 | kvm_sregs = NULL; |
6aa8b732 | 3839 | goto out; |
ff5c2c03 | 3840 | } |
fa3795a7 | 3841 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
6aa8b732 AK |
3842 | break; |
3843 | } | |
62d9f0db MT |
3844 | case KVM_GET_MP_STATE: { |
3845 | struct kvm_mp_state mp_state; | |
3846 | ||
3847 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); | |
3848 | if (r) | |
3849 | goto out; | |
3850 | r = -EFAULT; | |
893bdbf1 | 3851 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
62d9f0db MT |
3852 | goto out; |
3853 | r = 0; | |
3854 | break; | |
3855 | } | |
3856 | case KVM_SET_MP_STATE: { | |
3857 | struct kvm_mp_state mp_state; | |
3858 | ||
3859 | r = -EFAULT; | |
893bdbf1 | 3860 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
62d9f0db MT |
3861 | goto out; |
3862 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); | |
62d9f0db MT |
3863 | break; |
3864 | } | |
6aa8b732 AK |
3865 | case KVM_TRANSLATE: { |
3866 | struct kvm_translation tr; | |
3867 | ||
3868 | r = -EFAULT; | |
893bdbf1 | 3869 | if (copy_from_user(&tr, argp, sizeof(tr))) |
6aa8b732 | 3870 | goto out; |
8b006791 | 3871 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
6aa8b732 AK |
3872 | if (r) |
3873 | goto out; | |
3874 | r = -EFAULT; | |
893bdbf1 | 3875 | if (copy_to_user(argp, &tr, sizeof(tr))) |
6aa8b732 AK |
3876 | goto out; |
3877 | r = 0; | |
3878 | break; | |
3879 | } | |
d0bfb940 JK |
3880 | case KVM_SET_GUEST_DEBUG: { |
3881 | struct kvm_guest_debug dbg; | |
6aa8b732 AK |
3882 | |
3883 | r = -EFAULT; | |
893bdbf1 | 3884 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
6aa8b732 | 3885 | goto out; |
d0bfb940 | 3886 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
6aa8b732 AK |
3887 | break; |
3888 | } | |
1961d276 AK |
3889 | case KVM_SET_SIGNAL_MASK: { |
3890 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3891 | struct kvm_signal_mask kvm_sigmask; | |
3892 | sigset_t sigset, *p; | |
3893 | ||
3894 | p = NULL; | |
3895 | if (argp) { | |
3896 | r = -EFAULT; | |
3897 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3898 | sizeof(kvm_sigmask))) |
1961d276 AK |
3899 | goto out; |
3900 | r = -EINVAL; | |
893bdbf1 | 3901 | if (kvm_sigmask.len != sizeof(sigset)) |
1961d276 AK |
3902 | goto out; |
3903 | r = -EFAULT; | |
3904 | if (copy_from_user(&sigset, sigmask_arg->sigset, | |
893bdbf1 | 3905 | sizeof(sigset))) |
1961d276 AK |
3906 | goto out; |
3907 | p = &sigset; | |
3908 | } | |
376d41ff | 3909 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
1961d276 AK |
3910 | break; |
3911 | } | |
b8836737 | 3912 | case KVM_GET_FPU: { |
b12ce36a | 3913 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
fa3795a7 DH |
3914 | r = -ENOMEM; |
3915 | if (!fpu) | |
3916 | goto out; | |
3917 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); | |
b8836737 AK |
3918 | if (r) |
3919 | goto out; | |
3920 | r = -EFAULT; | |
fa3795a7 | 3921 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
b8836737 AK |
3922 | goto out; |
3923 | r = 0; | |
3924 | break; | |
3925 | } | |
3926 | case KVM_SET_FPU: { | |
ff5c2c03 SL |
3927 | fpu = memdup_user(argp, sizeof(*fpu)); |
3928 | if (IS_ERR(fpu)) { | |
3929 | r = PTR_ERR(fpu); | |
18595411 | 3930 | fpu = NULL; |
b8836737 | 3931 | goto out; |
ff5c2c03 | 3932 | } |
fa3795a7 | 3933 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
b8836737 AK |
3934 | break; |
3935 | } | |
ce55c049 JZ |
3936 | case KVM_GET_STATS_FD: { |
3937 | r = kvm_vcpu_ioctl_get_stats_fd(vcpu); | |
3938 | break; | |
3939 | } | |
bccf2150 | 3940 | default: |
313a3dc7 | 3941 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
bccf2150 AK |
3942 | } |
3943 | out: | |
ec7660cc | 3944 | mutex_unlock(&vcpu->mutex); |
fa3795a7 DH |
3945 | kfree(fpu); |
3946 | kfree(kvm_sregs); | |
bccf2150 AK |
3947 | return r; |
3948 | } | |
3949 | ||
de8e5d74 | 3950 | #ifdef CONFIG_KVM_COMPAT |
1dda606c AG |
3951 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
3952 | unsigned int ioctl, unsigned long arg) | |
3953 | { | |
3954 | struct kvm_vcpu *vcpu = filp->private_data; | |
3955 | void __user *argp = compat_ptr(arg); | |
3956 | int r; | |
3957 | ||
0b8f1173 | 3958 | if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged) |
1dda606c AG |
3959 | return -EIO; |
3960 | ||
3961 | switch (ioctl) { | |
3962 | case KVM_SET_SIGNAL_MASK: { | |
3963 | struct kvm_signal_mask __user *sigmask_arg = argp; | |
3964 | struct kvm_signal_mask kvm_sigmask; | |
1dda606c AG |
3965 | sigset_t sigset; |
3966 | ||
3967 | if (argp) { | |
3968 | r = -EFAULT; | |
3969 | if (copy_from_user(&kvm_sigmask, argp, | |
893bdbf1 | 3970 | sizeof(kvm_sigmask))) |
1dda606c AG |
3971 | goto out; |
3972 | r = -EINVAL; | |
3968cf62 | 3973 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
1dda606c AG |
3974 | goto out; |
3975 | r = -EFAULT; | |
1393b4aa PB |
3976 | if (get_compat_sigset(&sigset, |
3977 | (compat_sigset_t __user *)sigmask_arg->sigset)) | |
1dda606c | 3978 | goto out; |
760a9a30 AC |
3979 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
3980 | } else | |
3981 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); | |
1dda606c AG |
3982 | break; |
3983 | } | |
3984 | default: | |
3985 | r = kvm_vcpu_ioctl(filp, ioctl, arg); | |
3986 | } | |
3987 | ||
3988 | out: | |
3989 | return r; | |
3990 | } | |
3991 | #endif | |
3992 | ||
a1cd3f08 CLG |
3993 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
3994 | { | |
3995 | struct kvm_device *dev = filp->private_data; | |
3996 | ||
3997 | if (dev->ops->mmap) | |
3998 | return dev->ops->mmap(dev, vma); | |
3999 | ||
4000 | return -ENODEV; | |
4001 | } | |
4002 | ||
852b6d57 SW |
4003 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
4004 | int (*accessor)(struct kvm_device *dev, | |
4005 | struct kvm_device_attr *attr), | |
4006 | unsigned long arg) | |
4007 | { | |
4008 | struct kvm_device_attr attr; | |
4009 | ||
4010 | if (!accessor) | |
4011 | return -EPERM; | |
4012 | ||
4013 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) | |
4014 | return -EFAULT; | |
4015 | ||
4016 | return accessor(dev, &attr); | |
4017 | } | |
4018 | ||
4019 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, | |
4020 | unsigned long arg) | |
4021 | { | |
4022 | struct kvm_device *dev = filp->private_data; | |
4023 | ||
0b8f1173 | 4024 | if (dev->kvm->mm != current->mm || dev->kvm->vm_bugged) |
ddba9180 SC |
4025 | return -EIO; |
4026 | ||
852b6d57 SW |
4027 | switch (ioctl) { |
4028 | case KVM_SET_DEVICE_ATTR: | |
4029 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); | |
4030 | case KVM_GET_DEVICE_ATTR: | |
4031 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); | |
4032 | case KVM_HAS_DEVICE_ATTR: | |
4033 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); | |
4034 | default: | |
4035 | if (dev->ops->ioctl) | |
4036 | return dev->ops->ioctl(dev, ioctl, arg); | |
4037 | ||
4038 | return -ENOTTY; | |
4039 | } | |
4040 | } | |
4041 | ||
852b6d57 SW |
4042 | static int kvm_device_release(struct inode *inode, struct file *filp) |
4043 | { | |
4044 | struct kvm_device *dev = filp->private_data; | |
4045 | struct kvm *kvm = dev->kvm; | |
4046 | ||
2bde9b3e CLG |
4047 | if (dev->ops->release) { |
4048 | mutex_lock(&kvm->lock); | |
4049 | list_del(&dev->vm_node); | |
4050 | dev->ops->release(dev); | |
4051 | mutex_unlock(&kvm->lock); | |
4052 | } | |
4053 | ||
852b6d57 SW |
4054 | kvm_put_kvm(kvm); |
4055 | return 0; | |
4056 | } | |
4057 | ||
4058 | static const struct file_operations kvm_device_fops = { | |
4059 | .unlocked_ioctl = kvm_device_ioctl, | |
4060 | .release = kvm_device_release, | |
7ddfd3e0 | 4061 | KVM_COMPAT(kvm_device_ioctl), |
a1cd3f08 | 4062 | .mmap = kvm_device_mmap, |
852b6d57 SW |
4063 | }; |
4064 | ||
4065 | struct kvm_device *kvm_device_from_filp(struct file *filp) | |
4066 | { | |
4067 | if (filp->f_op != &kvm_device_fops) | |
4068 | return NULL; | |
4069 | ||
4070 | return filp->private_data; | |
4071 | } | |
4072 | ||
8538cb22 | 4073 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
5df554ad | 4074 | #ifdef CONFIG_KVM_MPIC |
d60eacb0 WD |
4075 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
4076 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, | |
5975a2e0 | 4077 | #endif |
d60eacb0 WD |
4078 | }; |
4079 | ||
8538cb22 | 4080 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
d60eacb0 WD |
4081 | { |
4082 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) | |
4083 | return -ENOSPC; | |
4084 | ||
4085 | if (kvm_device_ops_table[type] != NULL) | |
4086 | return -EEXIST; | |
4087 | ||
4088 | kvm_device_ops_table[type] = ops; | |
4089 | return 0; | |
4090 | } | |
4091 | ||
571ee1b6 WL |
4092 | void kvm_unregister_device_ops(u32 type) |
4093 | { | |
4094 | if (kvm_device_ops_table[type] != NULL) | |
4095 | kvm_device_ops_table[type] = NULL; | |
4096 | } | |
4097 | ||
852b6d57 SW |
4098 | static int kvm_ioctl_create_device(struct kvm *kvm, |
4099 | struct kvm_create_device *cd) | |
4100 | { | |
8538cb22 | 4101 | const struct kvm_device_ops *ops = NULL; |
852b6d57 SW |
4102 | struct kvm_device *dev; |
4103 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; | |
1d487e9b | 4104 | int type; |
852b6d57 SW |
4105 | int ret; |
4106 | ||
d60eacb0 WD |
4107 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
4108 | return -ENODEV; | |
4109 | ||
1d487e9b PB |
4110 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
4111 | ops = kvm_device_ops_table[type]; | |
d60eacb0 | 4112 | if (ops == NULL) |
852b6d57 | 4113 | return -ENODEV; |
852b6d57 SW |
4114 | |
4115 | if (test) | |
4116 | return 0; | |
4117 | ||
b12ce36a | 4118 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
852b6d57 SW |
4119 | if (!dev) |
4120 | return -ENOMEM; | |
4121 | ||
4122 | dev->ops = ops; | |
4123 | dev->kvm = kvm; | |
852b6d57 | 4124 | |
a28ebea2 | 4125 | mutex_lock(&kvm->lock); |
1d487e9b | 4126 | ret = ops->create(dev, type); |
852b6d57 | 4127 | if (ret < 0) { |
a28ebea2 | 4128 | mutex_unlock(&kvm->lock); |
852b6d57 SW |
4129 | kfree(dev); |
4130 | return ret; | |
4131 | } | |
a28ebea2 CD |
4132 | list_add(&dev->vm_node, &kvm->devices); |
4133 | mutex_unlock(&kvm->lock); | |
852b6d57 | 4134 | |
023e9fdd CD |
4135 | if (ops->init) |
4136 | ops->init(dev); | |
4137 | ||
cfa39381 | 4138 | kvm_get_kvm(kvm); |
24009b05 | 4139 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
852b6d57 | 4140 | if (ret < 0) { |
149487bd | 4141 | kvm_put_kvm_no_destroy(kvm); |
a28ebea2 CD |
4142 | mutex_lock(&kvm->lock); |
4143 | list_del(&dev->vm_node); | |
4144 | mutex_unlock(&kvm->lock); | |
a0f1d21c | 4145 | ops->destroy(dev); |
852b6d57 SW |
4146 | return ret; |
4147 | } | |
4148 | ||
852b6d57 SW |
4149 | cd->fd = ret; |
4150 | return 0; | |
4151 | } | |
4152 | ||
92b591a4 AG |
4153 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
4154 | { | |
4155 | switch (arg) { | |
4156 | case KVM_CAP_USER_MEMORY: | |
4157 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: | |
4158 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: | |
92b591a4 AG |
4159 | case KVM_CAP_INTERNAL_ERROR_DATA: |
4160 | #ifdef CONFIG_HAVE_KVM_MSI | |
4161 | case KVM_CAP_SIGNAL_MSI: | |
4162 | #endif | |
297e2105 | 4163 | #ifdef CONFIG_HAVE_KVM_IRQFD |
dc9be0fa | 4164 | case KVM_CAP_IRQFD: |
92b591a4 AG |
4165 | case KVM_CAP_IRQFD_RESAMPLE: |
4166 | #endif | |
e9ea5069 | 4167 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
92b591a4 | 4168 | case KVM_CAP_CHECK_EXTENSION_VM: |
e5d83c74 | 4169 | case KVM_CAP_ENABLE_CAP_VM: |
acd05785 | 4170 | case KVM_CAP_HALT_POLL: |
92b591a4 | 4171 | return 1; |
4b4357e0 | 4172 | #ifdef CONFIG_KVM_MMIO |
30422558 PB |
4173 | case KVM_CAP_COALESCED_MMIO: |
4174 | return KVM_COALESCED_MMIO_PAGE_OFFSET; | |
0804c849 PH |
4175 | case KVM_CAP_COALESCED_PIO: |
4176 | return 1; | |
30422558 | 4177 | #endif |
3c9bd400 JZ |
4178 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4179 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: | |
4180 | return KVM_DIRTY_LOG_MANUAL_CAPS; | |
4181 | #endif | |
92b591a4 AG |
4182 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4183 | case KVM_CAP_IRQ_ROUTING: | |
4184 | return KVM_MAX_IRQ_ROUTES; | |
f481b069 PB |
4185 | #endif |
4186 | #if KVM_ADDRESS_SPACE_NUM > 1 | |
4187 | case KVM_CAP_MULTI_ADDRESS_SPACE: | |
4188 | return KVM_ADDRESS_SPACE_NUM; | |
92b591a4 | 4189 | #endif |
c110ae57 PB |
4190 | case KVM_CAP_NR_MEMSLOTS: |
4191 | return KVM_USER_MEM_SLOTS; | |
fb04a1ed PX |
4192 | case KVM_CAP_DIRTY_LOG_RING: |
4193 | #if KVM_DIRTY_LOG_PAGE_OFFSET > 0 | |
4194 | return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn); | |
4195 | #else | |
4196 | return 0; | |
4197 | #endif | |
ce55c049 JZ |
4198 | case KVM_CAP_BINARY_STATS_FD: |
4199 | return 1; | |
92b591a4 AG |
4200 | default: |
4201 | break; | |
4202 | } | |
4203 | return kvm_vm_ioctl_check_extension(kvm, arg); | |
4204 | } | |
4205 | ||
fb04a1ed PX |
4206 | static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size) |
4207 | { | |
4208 | int r; | |
4209 | ||
4210 | if (!KVM_DIRTY_LOG_PAGE_OFFSET) | |
4211 | return -EINVAL; | |
4212 | ||
4213 | /* the size should be power of 2 */ | |
4214 | if (!size || (size & (size - 1))) | |
4215 | return -EINVAL; | |
4216 | ||
4217 | /* Should be bigger to keep the reserved entries, or a page */ | |
4218 | if (size < kvm_dirty_ring_get_rsvd_entries() * | |
4219 | sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE) | |
4220 | return -EINVAL; | |
4221 | ||
4222 | if (size > KVM_DIRTY_RING_MAX_ENTRIES * | |
4223 | sizeof(struct kvm_dirty_gfn)) | |
4224 | return -E2BIG; | |
4225 | ||
4226 | /* We only allow it to set once */ | |
4227 | if (kvm->dirty_ring_size) | |
4228 | return -EINVAL; | |
4229 | ||
4230 | mutex_lock(&kvm->lock); | |
4231 | ||
4232 | if (kvm->created_vcpus) { | |
4233 | /* We don't allow to change this value after vcpu created */ | |
4234 | r = -EINVAL; | |
4235 | } else { | |
4236 | kvm->dirty_ring_size = size; | |
4237 | r = 0; | |
4238 | } | |
4239 | ||
4240 | mutex_unlock(&kvm->lock); | |
4241 | return r; | |
4242 | } | |
4243 | ||
4244 | static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm) | |
4245 | { | |
4246 | int i; | |
4247 | struct kvm_vcpu *vcpu; | |
4248 | int cleared = 0; | |
4249 | ||
4250 | if (!kvm->dirty_ring_size) | |
4251 | return -EINVAL; | |
4252 | ||
4253 | mutex_lock(&kvm->slots_lock); | |
4254 | ||
4255 | kvm_for_each_vcpu(i, vcpu, kvm) | |
4256 | cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring); | |
4257 | ||
4258 | mutex_unlock(&kvm->slots_lock); | |
4259 | ||
4260 | if (cleared) | |
4261 | kvm_flush_remote_tlbs(kvm); | |
4262 | ||
4263 | return cleared; | |
4264 | } | |
4265 | ||
e5d83c74 PB |
4266 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4267 | struct kvm_enable_cap *cap) | |
4268 | { | |
4269 | return -EINVAL; | |
4270 | } | |
4271 | ||
4272 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, | |
4273 | struct kvm_enable_cap *cap) | |
4274 | { | |
4275 | switch (cap->cap) { | |
2a31b9db | 4276 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
3c9bd400 JZ |
4277 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: { |
4278 | u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE; | |
4279 | ||
4280 | if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE) | |
4281 | allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS; | |
4282 | ||
4283 | if (cap->flags || (cap->args[0] & ~allowed_options)) | |
2a31b9db PB |
4284 | return -EINVAL; |
4285 | kvm->manual_dirty_log_protect = cap->args[0]; | |
4286 | return 0; | |
3c9bd400 | 4287 | } |
2a31b9db | 4288 | #endif |
acd05785 DM |
4289 | case KVM_CAP_HALT_POLL: { |
4290 | if (cap->flags || cap->args[0] != (unsigned int)cap->args[0]) | |
4291 | return -EINVAL; | |
4292 | ||
4293 | kvm->max_halt_poll_ns = cap->args[0]; | |
4294 | return 0; | |
4295 | } | |
fb04a1ed PX |
4296 | case KVM_CAP_DIRTY_LOG_RING: |
4297 | return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]); | |
e5d83c74 PB |
4298 | default: |
4299 | return kvm_vm_ioctl_enable_cap(kvm, cap); | |
4300 | } | |
4301 | } | |
4302 | ||
fcfe1bae JZ |
4303 | static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer, |
4304 | size_t size, loff_t *offset) | |
4305 | { | |
4306 | struct kvm *kvm = file->private_data; | |
4307 | ||
4308 | return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header, | |
4309 | &kvm_vm_stats_desc[0], &kvm->stat, | |
4310 | sizeof(kvm->stat), user_buffer, size, offset); | |
4311 | } | |
4312 | ||
4313 | static const struct file_operations kvm_vm_stats_fops = { | |
4314 | .read = kvm_vm_stats_read, | |
4315 | .llseek = noop_llseek, | |
4316 | }; | |
4317 | ||
4318 | static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm) | |
4319 | { | |
4320 | int fd; | |
4321 | struct file *file; | |
4322 | ||
4323 | fd = get_unused_fd_flags(O_CLOEXEC); | |
4324 | if (fd < 0) | |
4325 | return fd; | |
4326 | ||
4327 | file = anon_inode_getfile("kvm-vm-stats", | |
4328 | &kvm_vm_stats_fops, kvm, O_RDONLY); | |
4329 | if (IS_ERR(file)) { | |
4330 | put_unused_fd(fd); | |
4331 | return PTR_ERR(file); | |
4332 | } | |
4333 | file->f_mode |= FMODE_PREAD; | |
4334 | fd_install(fd, file); | |
4335 | ||
4336 | return fd; | |
4337 | } | |
4338 | ||
bccf2150 AK |
4339 | static long kvm_vm_ioctl(struct file *filp, |
4340 | unsigned int ioctl, unsigned long arg) | |
4341 | { | |
4342 | struct kvm *kvm = filp->private_data; | |
4343 | void __user *argp = (void __user *)arg; | |
1fe779f8 | 4344 | int r; |
bccf2150 | 4345 | |
0b8f1173 | 4346 | if (kvm->mm != current->mm || kvm->vm_bugged) |
6d4e4c4f | 4347 | return -EIO; |
bccf2150 AK |
4348 | switch (ioctl) { |
4349 | case KVM_CREATE_VCPU: | |
4350 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); | |
bccf2150 | 4351 | break; |
e5d83c74 PB |
4352 | case KVM_ENABLE_CAP: { |
4353 | struct kvm_enable_cap cap; | |
4354 | ||
4355 | r = -EFAULT; | |
4356 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4357 | goto out; | |
4358 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); | |
4359 | break; | |
4360 | } | |
6fc138d2 IE |
4361 | case KVM_SET_USER_MEMORY_REGION: { |
4362 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
4363 | ||
4364 | r = -EFAULT; | |
4365 | if (copy_from_user(&kvm_userspace_mem, argp, | |
893bdbf1 | 4366 | sizeof(kvm_userspace_mem))) |
6fc138d2 IE |
4367 | goto out; |
4368 | ||
47ae31e2 | 4369 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
6aa8b732 AK |
4370 | break; |
4371 | } | |
4372 | case KVM_GET_DIRTY_LOG: { | |
4373 | struct kvm_dirty_log log; | |
4374 | ||
4375 | r = -EFAULT; | |
893bdbf1 | 4376 | if (copy_from_user(&log, argp, sizeof(log))) |
6aa8b732 | 4377 | goto out; |
2c6f5df9 | 4378 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
6aa8b732 AK |
4379 | break; |
4380 | } | |
2a31b9db PB |
4381 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4382 | case KVM_CLEAR_DIRTY_LOG: { | |
4383 | struct kvm_clear_dirty_log log; | |
4384 | ||
4385 | r = -EFAULT; | |
4386 | if (copy_from_user(&log, argp, sizeof(log))) | |
4387 | goto out; | |
4388 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4389 | break; | |
4390 | } | |
4391 | #endif | |
4b4357e0 | 4392 | #ifdef CONFIG_KVM_MMIO |
5f94c174 LV |
4393 | case KVM_REGISTER_COALESCED_MMIO: { |
4394 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4395 | |
5f94c174 | 4396 | r = -EFAULT; |
893bdbf1 | 4397 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4398 | goto out; |
5f94c174 | 4399 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4400 | break; |
4401 | } | |
4402 | case KVM_UNREGISTER_COALESCED_MMIO: { | |
4403 | struct kvm_coalesced_mmio_zone zone; | |
f95ef0cd | 4404 | |
5f94c174 | 4405 | r = -EFAULT; |
893bdbf1 | 4406 | if (copy_from_user(&zone, argp, sizeof(zone))) |
5f94c174 | 4407 | goto out; |
5f94c174 | 4408 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
5f94c174 LV |
4409 | break; |
4410 | } | |
4411 | #endif | |
721eecbf GH |
4412 | case KVM_IRQFD: { |
4413 | struct kvm_irqfd data; | |
4414 | ||
4415 | r = -EFAULT; | |
893bdbf1 | 4416 | if (copy_from_user(&data, argp, sizeof(data))) |
721eecbf | 4417 | goto out; |
d4db2935 | 4418 | r = kvm_irqfd(kvm, &data); |
721eecbf GH |
4419 | break; |
4420 | } | |
d34e6b17 GH |
4421 | case KVM_IOEVENTFD: { |
4422 | struct kvm_ioeventfd data; | |
4423 | ||
4424 | r = -EFAULT; | |
893bdbf1 | 4425 | if (copy_from_user(&data, argp, sizeof(data))) |
d34e6b17 GH |
4426 | goto out; |
4427 | r = kvm_ioeventfd(kvm, &data); | |
4428 | break; | |
4429 | } | |
07975ad3 JK |
4430 | #ifdef CONFIG_HAVE_KVM_MSI |
4431 | case KVM_SIGNAL_MSI: { | |
4432 | struct kvm_msi msi; | |
4433 | ||
4434 | r = -EFAULT; | |
893bdbf1 | 4435 | if (copy_from_user(&msi, argp, sizeof(msi))) |
07975ad3 JK |
4436 | goto out; |
4437 | r = kvm_send_userspace_msi(kvm, &msi); | |
4438 | break; | |
4439 | } | |
23d43cf9 CD |
4440 | #endif |
4441 | #ifdef __KVM_HAVE_IRQ_LINE | |
4442 | case KVM_IRQ_LINE_STATUS: | |
4443 | case KVM_IRQ_LINE: { | |
4444 | struct kvm_irq_level irq_event; | |
4445 | ||
4446 | r = -EFAULT; | |
893bdbf1 | 4447 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
23d43cf9 CD |
4448 | goto out; |
4449 | ||
aa2fbe6d YZ |
4450 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
4451 | ioctl == KVM_IRQ_LINE_STATUS); | |
23d43cf9 CD |
4452 | if (r) |
4453 | goto out; | |
4454 | ||
4455 | r = -EFAULT; | |
4456 | if (ioctl == KVM_IRQ_LINE_STATUS) { | |
893bdbf1 | 4457 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
23d43cf9 CD |
4458 | goto out; |
4459 | } | |
4460 | ||
4461 | r = 0; | |
4462 | break; | |
4463 | } | |
73880c80 | 4464 | #endif |
aa8d5944 AG |
4465 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
4466 | case KVM_SET_GSI_ROUTING: { | |
4467 | struct kvm_irq_routing routing; | |
4468 | struct kvm_irq_routing __user *urouting; | |
f8c1b85b | 4469 | struct kvm_irq_routing_entry *entries = NULL; |
aa8d5944 AG |
4470 | |
4471 | r = -EFAULT; | |
4472 | if (copy_from_user(&routing, argp, sizeof(routing))) | |
4473 | goto out; | |
4474 | r = -EINVAL; | |
5c0aea0e DH |
4475 | if (!kvm_arch_can_set_irq_routing(kvm)) |
4476 | goto out; | |
caf1ff26 | 4477 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
aa8d5944 AG |
4478 | goto out; |
4479 | if (routing.flags) | |
4480 | goto out; | |
f8c1b85b | 4481 | if (routing.nr) { |
f8c1b85b | 4482 | urouting = argp; |
7ec28e26 DE |
4483 | entries = vmemdup_user(urouting->entries, |
4484 | array_size(sizeof(*entries), | |
4485 | routing.nr)); | |
4486 | if (IS_ERR(entries)) { | |
4487 | r = PTR_ERR(entries); | |
4488 | goto out; | |
4489 | } | |
f8c1b85b | 4490 | } |
aa8d5944 AG |
4491 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
4492 | routing.flags); | |
7ec28e26 | 4493 | kvfree(entries); |
aa8d5944 AG |
4494 | break; |
4495 | } | |
4496 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ | |
852b6d57 SW |
4497 | case KVM_CREATE_DEVICE: { |
4498 | struct kvm_create_device cd; | |
4499 | ||
4500 | r = -EFAULT; | |
4501 | if (copy_from_user(&cd, argp, sizeof(cd))) | |
4502 | goto out; | |
4503 | ||
4504 | r = kvm_ioctl_create_device(kvm, &cd); | |
4505 | if (r) | |
4506 | goto out; | |
4507 | ||
4508 | r = -EFAULT; | |
4509 | if (copy_to_user(argp, &cd, sizeof(cd))) | |
4510 | goto out; | |
4511 | ||
4512 | r = 0; | |
4513 | break; | |
4514 | } | |
92b591a4 AG |
4515 | case KVM_CHECK_EXTENSION: |
4516 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); | |
4517 | break; | |
fb04a1ed PX |
4518 | case KVM_RESET_DIRTY_RINGS: |
4519 | r = kvm_vm_ioctl_reset_dirty_pages(kvm); | |
4520 | break; | |
fcfe1bae JZ |
4521 | case KVM_GET_STATS_FD: |
4522 | r = kvm_vm_ioctl_get_stats_fd(kvm); | |
4523 | break; | |
f17abe9a | 4524 | default: |
1fe779f8 | 4525 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
f17abe9a AK |
4526 | } |
4527 | out: | |
4528 | return r; | |
4529 | } | |
4530 | ||
de8e5d74 | 4531 | #ifdef CONFIG_KVM_COMPAT |
6ff5894c AB |
4532 | struct compat_kvm_dirty_log { |
4533 | __u32 slot; | |
4534 | __u32 padding1; | |
4535 | union { | |
4536 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4537 | __u64 padding2; | |
4538 | }; | |
4539 | }; | |
4540 | ||
8750f9bb PB |
4541 | struct compat_kvm_clear_dirty_log { |
4542 | __u32 slot; | |
4543 | __u32 num_pages; | |
4544 | __u64 first_page; | |
4545 | union { | |
4546 | compat_uptr_t dirty_bitmap; /* one bit per page */ | |
4547 | __u64 padding2; | |
4548 | }; | |
4549 | }; | |
4550 | ||
6ff5894c AB |
4551 | static long kvm_vm_compat_ioctl(struct file *filp, |
4552 | unsigned int ioctl, unsigned long arg) | |
4553 | { | |
4554 | struct kvm *kvm = filp->private_data; | |
4555 | int r; | |
4556 | ||
0b8f1173 | 4557 | if (kvm->mm != current->mm || kvm->vm_bugged) |
6ff5894c AB |
4558 | return -EIO; |
4559 | switch (ioctl) { | |
8750f9bb PB |
4560 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
4561 | case KVM_CLEAR_DIRTY_LOG: { | |
4562 | struct compat_kvm_clear_dirty_log compat_log; | |
4563 | struct kvm_clear_dirty_log log; | |
4564 | ||
4565 | if (copy_from_user(&compat_log, (void __user *)arg, | |
4566 | sizeof(compat_log))) | |
4567 | return -EFAULT; | |
4568 | log.slot = compat_log.slot; | |
4569 | log.num_pages = compat_log.num_pages; | |
4570 | log.first_page = compat_log.first_page; | |
4571 | log.padding2 = compat_log.padding2; | |
4572 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4573 | ||
4574 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); | |
4575 | break; | |
4576 | } | |
4577 | #endif | |
6ff5894c AB |
4578 | case KVM_GET_DIRTY_LOG: { |
4579 | struct compat_kvm_dirty_log compat_log; | |
4580 | struct kvm_dirty_log log; | |
4581 | ||
6ff5894c AB |
4582 | if (copy_from_user(&compat_log, (void __user *)arg, |
4583 | sizeof(compat_log))) | |
f6a3b168 | 4584 | return -EFAULT; |
6ff5894c AB |
4585 | log.slot = compat_log.slot; |
4586 | log.padding1 = compat_log.padding1; | |
4587 | log.padding2 = compat_log.padding2; | |
4588 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); | |
4589 | ||
4590 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); | |
6ff5894c AB |
4591 | break; |
4592 | } | |
4593 | default: | |
4594 | r = kvm_vm_ioctl(filp, ioctl, arg); | |
4595 | } | |
6ff5894c AB |
4596 | return r; |
4597 | } | |
4598 | #endif | |
4599 | ||
3d3aab1b | 4600 | static struct file_operations kvm_vm_fops = { |
f17abe9a AK |
4601 | .release = kvm_vm_release, |
4602 | .unlocked_ioctl = kvm_vm_ioctl, | |
6038f373 | 4603 | .llseek = noop_llseek, |
7ddfd3e0 | 4604 | KVM_COMPAT(kvm_vm_compat_ioctl), |
f17abe9a AK |
4605 | }; |
4606 | ||
54526d1f NT |
4607 | bool file_is_kvm(struct file *file) |
4608 | { | |
4609 | return file && file->f_op == &kvm_vm_fops; | |
4610 | } | |
4611 | EXPORT_SYMBOL_GPL(file_is_kvm); | |
4612 | ||
e08b9637 | 4613 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
f17abe9a | 4614 | { |
aac87636 | 4615 | int r; |
f17abe9a | 4616 | struct kvm *kvm; |
506cfba9 | 4617 | struct file *file; |
f17abe9a | 4618 | |
e08b9637 | 4619 | kvm = kvm_create_vm(type); |
d6d28168 AK |
4620 | if (IS_ERR(kvm)) |
4621 | return PTR_ERR(kvm); | |
4b4357e0 | 4622 | #ifdef CONFIG_KVM_MMIO |
6ce5a090 | 4623 | r = kvm_coalesced_mmio_init(kvm); |
78588335 ME |
4624 | if (r < 0) |
4625 | goto put_kvm; | |
6ce5a090 | 4626 | #endif |
506cfba9 | 4627 | r = get_unused_fd_flags(O_CLOEXEC); |
78588335 ME |
4628 | if (r < 0) |
4629 | goto put_kvm; | |
4630 | ||
fcfe1bae JZ |
4631 | snprintf(kvm->stats_id, sizeof(kvm->stats_id), |
4632 | "kvm-%d", task_pid_nr(current)); | |
4633 | ||
506cfba9 AV |
4634 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
4635 | if (IS_ERR(file)) { | |
4636 | put_unused_fd(r); | |
78588335 ME |
4637 | r = PTR_ERR(file); |
4638 | goto put_kvm; | |
506cfba9 | 4639 | } |
536a6f88 | 4640 | |
525df861 PB |
4641 | /* |
4642 | * Don't call kvm_put_kvm anymore at this point; file->f_op is | |
4643 | * already set, with ->release() being kvm_vm_release(). In error | |
4644 | * cases it will be called by the final fput(file) and will take | |
4645 | * care of doing kvm_put_kvm(kvm). | |
4646 | */ | |
536a6f88 | 4647 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
506cfba9 AV |
4648 | put_unused_fd(r); |
4649 | fput(file); | |
536a6f88 JF |
4650 | return -ENOMEM; |
4651 | } | |
286de8f6 | 4652 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
f17abe9a | 4653 | |
506cfba9 | 4654 | fd_install(r, file); |
aac87636 | 4655 | return r; |
78588335 ME |
4656 | |
4657 | put_kvm: | |
4658 | kvm_put_kvm(kvm); | |
4659 | return r; | |
f17abe9a AK |
4660 | } |
4661 | ||
4662 | static long kvm_dev_ioctl(struct file *filp, | |
4663 | unsigned int ioctl, unsigned long arg) | |
4664 | { | |
07c45a36 | 4665 | long r = -EINVAL; |
f17abe9a AK |
4666 | |
4667 | switch (ioctl) { | |
4668 | case KVM_GET_API_VERSION: | |
f0fe5108 AK |
4669 | if (arg) |
4670 | goto out; | |
f17abe9a AK |
4671 | r = KVM_API_VERSION; |
4672 | break; | |
4673 | case KVM_CREATE_VM: | |
e08b9637 | 4674 | r = kvm_dev_ioctl_create_vm(arg); |
f17abe9a | 4675 | break; |
018d00d2 | 4676 | case KVM_CHECK_EXTENSION: |
784aa3d7 | 4677 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
5d308f45 | 4678 | break; |
07c45a36 | 4679 | case KVM_GET_VCPU_MMAP_SIZE: |
07c45a36 AK |
4680 | if (arg) |
4681 | goto out; | |
adb1ff46 AK |
4682 | r = PAGE_SIZE; /* struct kvm_run */ |
4683 | #ifdef CONFIG_X86 | |
4684 | r += PAGE_SIZE; /* pio data page */ | |
5f94c174 | 4685 | #endif |
4b4357e0 | 4686 | #ifdef CONFIG_KVM_MMIO |
5f94c174 | 4687 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
adb1ff46 | 4688 | #endif |
07c45a36 | 4689 | break; |
d4c9ff2d FEL |
4690 | case KVM_TRACE_ENABLE: |
4691 | case KVM_TRACE_PAUSE: | |
4692 | case KVM_TRACE_DISABLE: | |
2023a29c | 4693 | r = -EOPNOTSUPP; |
d4c9ff2d | 4694 | break; |
6aa8b732 | 4695 | default: |
043405e1 | 4696 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
6aa8b732 AK |
4697 | } |
4698 | out: | |
4699 | return r; | |
4700 | } | |
4701 | ||
6aa8b732 | 4702 | static struct file_operations kvm_chardev_ops = { |
6aa8b732 | 4703 | .unlocked_ioctl = kvm_dev_ioctl, |
6038f373 | 4704 | .llseek = noop_llseek, |
7ddfd3e0 | 4705 | KVM_COMPAT(kvm_dev_ioctl), |
6aa8b732 AK |
4706 | }; |
4707 | ||
4708 | static struct miscdevice kvm_dev = { | |
bbe4432e | 4709 | KVM_MINOR, |
6aa8b732 AK |
4710 | "kvm", |
4711 | &kvm_chardev_ops, | |
4712 | }; | |
4713 | ||
75b7127c | 4714 | static void hardware_enable_nolock(void *junk) |
1b6c0168 AK |
4715 | { |
4716 | int cpu = raw_smp_processor_id(); | |
10474ae8 | 4717 | int r; |
1b6c0168 | 4718 | |
7f59f492 | 4719 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4720 | return; |
10474ae8 | 4721 | |
7f59f492 | 4722 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
10474ae8 | 4723 | |
13a34e06 | 4724 | r = kvm_arch_hardware_enable(); |
10474ae8 AG |
4725 | |
4726 | if (r) { | |
4727 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); | |
4728 | atomic_inc(&hardware_enable_failed); | |
1170adc6 | 4729 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
10474ae8 | 4730 | } |
1b6c0168 AK |
4731 | } |
4732 | ||
8c18b2d2 | 4733 | static int kvm_starting_cpu(unsigned int cpu) |
75b7127c | 4734 | { |
4a937f96 | 4735 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4736 | if (kvm_usage_count) |
4737 | hardware_enable_nolock(NULL); | |
4a937f96 | 4738 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4739 | return 0; |
75b7127c TY |
4740 | } |
4741 | ||
4742 | static void hardware_disable_nolock(void *junk) | |
1b6c0168 AK |
4743 | { |
4744 | int cpu = raw_smp_processor_id(); | |
4745 | ||
7f59f492 | 4746 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
1b6c0168 | 4747 | return; |
7f59f492 | 4748 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
13a34e06 | 4749 | kvm_arch_hardware_disable(); |
1b6c0168 AK |
4750 | } |
4751 | ||
8c18b2d2 | 4752 | static int kvm_dying_cpu(unsigned int cpu) |
75b7127c | 4753 | { |
4a937f96 | 4754 | raw_spin_lock(&kvm_count_lock); |
4fa92fb2 PB |
4755 | if (kvm_usage_count) |
4756 | hardware_disable_nolock(NULL); | |
4a937f96 | 4757 | raw_spin_unlock(&kvm_count_lock); |
8c18b2d2 | 4758 | return 0; |
75b7127c TY |
4759 | } |
4760 | ||
10474ae8 AG |
4761 | static void hardware_disable_all_nolock(void) |
4762 | { | |
4763 | BUG_ON(!kvm_usage_count); | |
4764 | ||
4765 | kvm_usage_count--; | |
4766 | if (!kvm_usage_count) | |
75b7127c | 4767 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
10474ae8 AG |
4768 | } |
4769 | ||
4770 | static void hardware_disable_all(void) | |
4771 | { | |
4a937f96 | 4772 | raw_spin_lock(&kvm_count_lock); |
10474ae8 | 4773 | hardware_disable_all_nolock(); |
4a937f96 | 4774 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4775 | } |
4776 | ||
4777 | static int hardware_enable_all(void) | |
4778 | { | |
4779 | int r = 0; | |
4780 | ||
4a937f96 | 4781 | raw_spin_lock(&kvm_count_lock); |
10474ae8 AG |
4782 | |
4783 | kvm_usage_count++; | |
4784 | if (kvm_usage_count == 1) { | |
4785 | atomic_set(&hardware_enable_failed, 0); | |
75b7127c | 4786 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
10474ae8 AG |
4787 | |
4788 | if (atomic_read(&hardware_enable_failed)) { | |
4789 | hardware_disable_all_nolock(); | |
4790 | r = -EBUSY; | |
4791 | } | |
4792 | } | |
4793 | ||
4a937f96 | 4794 | raw_spin_unlock(&kvm_count_lock); |
10474ae8 AG |
4795 | |
4796 | return r; | |
4797 | } | |
4798 | ||
9a2b85c6 | 4799 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
d77c26fc | 4800 | void *v) |
9a2b85c6 | 4801 | { |
8e1c1815 SY |
4802 | /* |
4803 | * Some (well, at least mine) BIOSes hang on reboot if | |
4804 | * in vmx root mode. | |
4805 | * | |
4806 | * And Intel TXT required VMX off for all cpu when system shutdown. | |
4807 | */ | |
1170adc6 | 4808 | pr_info("kvm: exiting hardware virtualization\n"); |
8e1c1815 | 4809 | kvm_rebooting = true; |
75b7127c | 4810 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
9a2b85c6 RR |
4811 | return NOTIFY_OK; |
4812 | } | |
4813 | ||
4814 | static struct notifier_block kvm_reboot_notifier = { | |
4815 | .notifier_call = kvm_reboot, | |
4816 | .priority = 0, | |
4817 | }; | |
4818 | ||
e93f8a0f | 4819 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
2eeb2e94 GH |
4820 | { |
4821 | int i; | |
4822 | ||
4823 | for (i = 0; i < bus->dev_count; i++) { | |
743eeb0b | 4824 | struct kvm_io_device *pos = bus->range[i].dev; |
2eeb2e94 GH |
4825 | |
4826 | kvm_iodevice_destructor(pos); | |
4827 | } | |
e93f8a0f | 4828 | kfree(bus); |
2eeb2e94 GH |
4829 | } |
4830 | ||
c21fbff1 | 4831 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
20e87b72 | 4832 | const struct kvm_io_range *r2) |
743eeb0b | 4833 | { |
8f4216c7 JW |
4834 | gpa_t addr1 = r1->addr; |
4835 | gpa_t addr2 = r2->addr; | |
4836 | ||
4837 | if (addr1 < addr2) | |
743eeb0b | 4838 | return -1; |
8f4216c7 JW |
4839 | |
4840 | /* If r2->len == 0, match the exact address. If r2->len != 0, | |
4841 | * accept any overlapping write. Any order is acceptable for | |
4842 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures | |
4843 | * we process all of them. | |
4844 | */ | |
4845 | if (r2->len) { | |
4846 | addr1 += r1->len; | |
4847 | addr2 += r2->len; | |
4848 | } | |
4849 | ||
4850 | if (addr1 > addr2) | |
743eeb0b | 4851 | return 1; |
8f4216c7 | 4852 | |
743eeb0b SL |
4853 | return 0; |
4854 | } | |
4855 | ||
a343c9b7 PB |
4856 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
4857 | { | |
c21fbff1 | 4858 | return kvm_io_bus_cmp(p1, p2); |
a343c9b7 PB |
4859 | } |
4860 | ||
39369f7a | 4861 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
743eeb0b SL |
4862 | gpa_t addr, int len) |
4863 | { | |
4864 | struct kvm_io_range *range, key; | |
4865 | int off; | |
4866 | ||
4867 | key = (struct kvm_io_range) { | |
4868 | .addr = addr, | |
4869 | .len = len, | |
4870 | }; | |
4871 | ||
4872 | range = bsearch(&key, bus->range, bus->dev_count, | |
4873 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); | |
4874 | if (range == NULL) | |
4875 | return -ENOENT; | |
4876 | ||
4877 | off = range - bus->range; | |
4878 | ||
c21fbff1 | 4879 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
743eeb0b SL |
4880 | off--; |
4881 | ||
4882 | return off; | |
4883 | } | |
4884 | ||
e32edf4f | 4885 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
126a5af5 CH |
4886 | struct kvm_io_range *range, const void *val) |
4887 | { | |
4888 | int idx; | |
4889 | ||
4890 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
4891 | if (idx < 0) | |
4892 | return -EOPNOTSUPP; | |
4893 | ||
4894 | while (idx < bus->dev_count && | |
c21fbff1 | 4895 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4896 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4897 | range->len, val)) |
4898 | return idx; | |
4899 | idx++; | |
4900 | } | |
4901 | ||
4902 | return -EOPNOTSUPP; | |
4903 | } | |
4904 | ||
bda9020e | 4905 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
e32edf4f | 4906 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
bda9020e | 4907 | int len, const void *val) |
2eeb2e94 | 4908 | { |
90d83dc3 | 4909 | struct kvm_io_bus *bus; |
743eeb0b | 4910 | struct kvm_io_range range; |
126a5af5 | 4911 | int r; |
743eeb0b SL |
4912 | |
4913 | range = (struct kvm_io_range) { | |
4914 | .addr = addr, | |
4915 | .len = len, | |
4916 | }; | |
90d83dc3 | 4917 | |
e32edf4f | 4918 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4919 | if (!bus) |
4920 | return -ENOMEM; | |
e32edf4f | 4921 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4922 | return r < 0 ? r : 0; |
4923 | } | |
a2420107 | 4924 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
126a5af5 CH |
4925 | |
4926 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ | |
e32edf4f NN |
4927 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
4928 | gpa_t addr, int len, const void *val, long cookie) | |
126a5af5 CH |
4929 | { |
4930 | struct kvm_io_bus *bus; | |
4931 | struct kvm_io_range range; | |
4932 | ||
4933 | range = (struct kvm_io_range) { | |
4934 | .addr = addr, | |
4935 | .len = len, | |
4936 | }; | |
4937 | ||
e32edf4f | 4938 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4939 | if (!bus) |
4940 | return -ENOMEM; | |
126a5af5 CH |
4941 | |
4942 | /* First try the device referenced by cookie. */ | |
4943 | if ((cookie >= 0) && (cookie < bus->dev_count) && | |
c21fbff1 | 4944 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
e32edf4f | 4945 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
126a5af5 CH |
4946 | val)) |
4947 | return cookie; | |
4948 | ||
4949 | /* | |
4950 | * cookie contained garbage; fall back to search and return the | |
4951 | * correct cookie value. | |
4952 | */ | |
e32edf4f | 4953 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
126a5af5 CH |
4954 | } |
4955 | ||
e32edf4f NN |
4956 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
4957 | struct kvm_io_range *range, void *val) | |
126a5af5 CH |
4958 | { |
4959 | int idx; | |
4960 | ||
4961 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); | |
743eeb0b SL |
4962 | if (idx < 0) |
4963 | return -EOPNOTSUPP; | |
4964 | ||
4965 | while (idx < bus->dev_count && | |
c21fbff1 | 4966 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
e32edf4f | 4967 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
126a5af5 CH |
4968 | range->len, val)) |
4969 | return idx; | |
743eeb0b SL |
4970 | idx++; |
4971 | } | |
4972 | ||
bda9020e MT |
4973 | return -EOPNOTSUPP; |
4974 | } | |
2eeb2e94 | 4975 | |
bda9020e | 4976 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
e32edf4f | 4977 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
e93f8a0f | 4978 | int len, void *val) |
bda9020e | 4979 | { |
90d83dc3 | 4980 | struct kvm_io_bus *bus; |
743eeb0b | 4981 | struct kvm_io_range range; |
126a5af5 | 4982 | int r; |
743eeb0b SL |
4983 | |
4984 | range = (struct kvm_io_range) { | |
4985 | .addr = addr, | |
4986 | .len = len, | |
4987 | }; | |
e93f8a0f | 4988 | |
e32edf4f | 4989 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
90db1043 DH |
4990 | if (!bus) |
4991 | return -ENOMEM; | |
e32edf4f | 4992 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
126a5af5 CH |
4993 | return r < 0 ? r : 0; |
4994 | } | |
743eeb0b | 4995 | |
79fac95e | 4996 | /* Caller must hold slots_lock. */ |
743eeb0b SL |
4997 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
4998 | int len, struct kvm_io_device *dev) | |
6c474694 | 4999 | { |
d4c67a7a | 5000 | int i; |
e93f8a0f | 5001 | struct kvm_io_bus *new_bus, *bus; |
d4c67a7a | 5002 | struct kvm_io_range range; |
090b7aff | 5003 | |
4a12f951 | 5004 | bus = kvm_get_bus(kvm, bus_idx); |
90db1043 DH |
5005 | if (!bus) |
5006 | return -ENOMEM; | |
5007 | ||
6ea34c9b AK |
5008 | /* exclude ioeventfd which is limited by maximum fd */ |
5009 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) | |
090b7aff | 5010 | return -ENOSPC; |
2eeb2e94 | 5011 | |
90952cd3 | 5012 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
b12ce36a | 5013 | GFP_KERNEL_ACCOUNT); |
e93f8a0f MT |
5014 | if (!new_bus) |
5015 | return -ENOMEM; | |
d4c67a7a GH |
5016 | |
5017 | range = (struct kvm_io_range) { | |
5018 | .addr = addr, | |
5019 | .len = len, | |
5020 | .dev = dev, | |
5021 | }; | |
5022 | ||
5023 | for (i = 0; i < bus->dev_count; i++) | |
5024 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) | |
5025 | break; | |
5026 | ||
5027 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); | |
5028 | new_bus->dev_count++; | |
5029 | new_bus->range[i] = range; | |
5030 | memcpy(new_bus->range + i + 1, bus->range + i, | |
5031 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); | |
e93f8a0f MT |
5032 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
5033 | synchronize_srcu_expedited(&kvm->srcu); | |
5034 | kfree(bus); | |
090b7aff GH |
5035 | |
5036 | return 0; | |
5037 | } | |
5038 | ||
5d3c4c79 SC |
5039 | int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5040 | struct kvm_io_device *dev) | |
090b7aff | 5041 | { |
f6588660 | 5042 | int i, j; |
e93f8a0f | 5043 | struct kvm_io_bus *new_bus, *bus; |
090b7aff | 5044 | |
7c896d37 SC |
5045 | lockdep_assert_held(&kvm->slots_lock); |
5046 | ||
4a12f951 | 5047 | bus = kvm_get_bus(kvm, bus_idx); |
df630b8c | 5048 | if (!bus) |
5d3c4c79 | 5049 | return 0; |
df630b8c | 5050 | |
7c896d37 | 5051 | for (i = 0; i < bus->dev_count; i++) { |
a1300716 | 5052 | if (bus->range[i].dev == dev) { |
090b7aff GH |
5053 | break; |
5054 | } | |
7c896d37 | 5055 | } |
e93f8a0f | 5056 | |
90db1043 | 5057 | if (i == bus->dev_count) |
5d3c4c79 | 5058 | return 0; |
a1300716 | 5059 | |
90952cd3 | 5060 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
b12ce36a | 5061 | GFP_KERNEL_ACCOUNT); |
f6588660 | 5062 | if (new_bus) { |
871c433b | 5063 | memcpy(new_bus, bus, struct_size(bus, range, i)); |
f6588660 RK |
5064 | new_bus->dev_count--; |
5065 | memcpy(new_bus->range + i, bus->range + i + 1, | |
871c433b | 5066 | flex_array_size(new_bus, range, new_bus->dev_count - i)); |
2ee37574 SC |
5067 | } |
5068 | ||
5069 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); | |
5070 | synchronize_srcu_expedited(&kvm->srcu); | |
5071 | ||
5072 | /* Destroy the old bus _after_ installing the (null) bus. */ | |
5073 | if (!new_bus) { | |
90db1043 | 5074 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
f6588660 RK |
5075 | for (j = 0; j < bus->dev_count; j++) { |
5076 | if (j == i) | |
5077 | continue; | |
5078 | kvm_iodevice_destructor(bus->range[j].dev); | |
5079 | } | |
90db1043 | 5080 | } |
a1300716 | 5081 | |
e93f8a0f | 5082 | kfree(bus); |
5d3c4c79 | 5083 | return new_bus ? 0 : -ENOMEM; |
2eeb2e94 GH |
5084 | } |
5085 | ||
8a39d006 AP |
5086 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
5087 | gpa_t addr) | |
5088 | { | |
5089 | struct kvm_io_bus *bus; | |
5090 | int dev_idx, srcu_idx; | |
5091 | struct kvm_io_device *iodev = NULL; | |
5092 | ||
5093 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
5094 | ||
5095 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); | |
90db1043 DH |
5096 | if (!bus) |
5097 | goto out_unlock; | |
8a39d006 AP |
5098 | |
5099 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); | |
5100 | if (dev_idx < 0) | |
5101 | goto out_unlock; | |
5102 | ||
5103 | iodev = bus->range[dev_idx].dev; | |
5104 | ||
5105 | out_unlock: | |
5106 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
5107 | ||
5108 | return iodev; | |
5109 | } | |
5110 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); | |
5111 | ||
536a6f88 JF |
5112 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
5113 | int (*get)(void *, u64 *), int (*set)(void *, u64), | |
5114 | const char *fmt) | |
5115 | { | |
5116 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5117 | inode->i_private; | |
5118 | ||
605c7130 PX |
5119 | /* |
5120 | * The debugfs files are a reference to the kvm struct which | |
5121 | * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe | |
5122 | * avoids the race between open and the removal of the debugfs directory. | |
536a6f88 | 5123 | */ |
605c7130 | 5124 | if (!kvm_get_kvm_safe(stat_data->kvm)) |
536a6f88 JF |
5125 | return -ENOENT; |
5126 | ||
833b45de | 5127 | if (simple_attr_open(inode, file, get, |
bc9e9e67 | 5128 | kvm_stats_debugfs_mode(stat_data->desc) & 0222 |
09cbcef6 MP |
5129 | ? set : NULL, |
5130 | fmt)) { | |
536a6f88 JF |
5131 | kvm_put_kvm(stat_data->kvm); |
5132 | return -ENOMEM; | |
5133 | } | |
5134 | ||
5135 | return 0; | |
5136 | } | |
5137 | ||
5138 | static int kvm_debugfs_release(struct inode *inode, struct file *file) | |
5139 | { | |
5140 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) | |
5141 | inode->i_private; | |
5142 | ||
5143 | simple_attr_release(inode, file); | |
5144 | kvm_put_kvm(stat_data->kvm); | |
5145 | ||
5146 | return 0; | |
5147 | } | |
5148 | ||
09cbcef6 | 5149 | static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val) |
536a6f88 | 5150 | { |
bc9e9e67 | 5151 | *val = *(u64 *)((void *)(&kvm->stat) + offset); |
536a6f88 | 5152 | |
09cbcef6 MP |
5153 | return 0; |
5154 | } | |
5155 | ||
5156 | static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset) | |
5157 | { | |
bc9e9e67 | 5158 | *(u64 *)((void *)(&kvm->stat) + offset) = 0; |
536a6f88 JF |
5159 | |
5160 | return 0; | |
5161 | } | |
5162 | ||
09cbcef6 | 5163 | static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val) |
ce35ef27 | 5164 | { |
09cbcef6 MP |
5165 | int i; |
5166 | struct kvm_vcpu *vcpu; | |
ce35ef27 | 5167 | |
09cbcef6 | 5168 | *val = 0; |
ce35ef27 | 5169 | |
09cbcef6 | 5170 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5171 | *val += *(u64 *)((void *)(&vcpu->stat) + offset); |
ce35ef27 SJS |
5172 | |
5173 | return 0; | |
5174 | } | |
5175 | ||
09cbcef6 | 5176 | static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset) |
536a6f88 | 5177 | { |
09cbcef6 MP |
5178 | int i; |
5179 | struct kvm_vcpu *vcpu; | |
536a6f88 | 5180 | |
09cbcef6 | 5181 | kvm_for_each_vcpu(i, vcpu, kvm) |
bc9e9e67 | 5182 | *(u64 *)((void *)(&vcpu->stat) + offset) = 0; |
09cbcef6 MP |
5183 | |
5184 | return 0; | |
5185 | } | |
536a6f88 | 5186 | |
09cbcef6 | 5187 | static int kvm_stat_data_get(void *data, u64 *val) |
536a6f88 | 5188 | { |
09cbcef6 | 5189 | int r = -EFAULT; |
536a6f88 | 5190 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
536a6f88 | 5191 | |
bc9e9e67 | 5192 | switch (stat_data->kind) { |
09cbcef6 MP |
5193 | case KVM_STAT_VM: |
5194 | r = kvm_get_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5195 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5196 | break; |
5197 | case KVM_STAT_VCPU: | |
5198 | r = kvm_get_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5199 | stat_data->desc->desc.offset, val); |
09cbcef6 MP |
5200 | break; |
5201 | } | |
536a6f88 | 5202 | |
09cbcef6 | 5203 | return r; |
536a6f88 JF |
5204 | } |
5205 | ||
09cbcef6 | 5206 | static int kvm_stat_data_clear(void *data, u64 val) |
ce35ef27 | 5207 | { |
09cbcef6 | 5208 | int r = -EFAULT; |
ce35ef27 | 5209 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
ce35ef27 SJS |
5210 | |
5211 | if (val) | |
5212 | return -EINVAL; | |
5213 | ||
bc9e9e67 | 5214 | switch (stat_data->kind) { |
09cbcef6 MP |
5215 | case KVM_STAT_VM: |
5216 | r = kvm_clear_stat_per_vm(stat_data->kvm, | |
bc9e9e67 | 5217 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5218 | break; |
5219 | case KVM_STAT_VCPU: | |
5220 | r = kvm_clear_stat_per_vcpu(stat_data->kvm, | |
bc9e9e67 | 5221 | stat_data->desc->desc.offset); |
09cbcef6 MP |
5222 | break; |
5223 | } | |
ce35ef27 | 5224 | |
09cbcef6 | 5225 | return r; |
ce35ef27 SJS |
5226 | } |
5227 | ||
09cbcef6 | 5228 | static int kvm_stat_data_open(struct inode *inode, struct file *file) |
536a6f88 JF |
5229 | { |
5230 | __simple_attr_check_format("%llu\n", 0ull); | |
09cbcef6 MP |
5231 | return kvm_debugfs_open(inode, file, kvm_stat_data_get, |
5232 | kvm_stat_data_clear, "%llu\n"); | |
536a6f88 JF |
5233 | } |
5234 | ||
09cbcef6 MP |
5235 | static const struct file_operations stat_fops_per_vm = { |
5236 | .owner = THIS_MODULE, | |
5237 | .open = kvm_stat_data_open, | |
536a6f88 | 5238 | .release = kvm_debugfs_release, |
09cbcef6 MP |
5239 | .read = simple_attr_read, |
5240 | .write = simple_attr_write, | |
5241 | .llseek = no_llseek, | |
536a6f88 JF |
5242 | }; |
5243 | ||
8b88b099 | 5244 | static int vm_stat_get(void *_offset, u64 *val) |
ba1389b7 AK |
5245 | { |
5246 | unsigned offset = (long)_offset; | |
ba1389b7 | 5247 | struct kvm *kvm; |
536a6f88 | 5248 | u64 tmp_val; |
ba1389b7 | 5249 | |
8b88b099 | 5250 | *val = 0; |
0d9ce162 | 5251 | mutex_lock(&kvm_lock); |
536a6f88 | 5252 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5253 | kvm_get_stat_per_vm(kvm, offset, &tmp_val); |
536a6f88 JF |
5254 | *val += tmp_val; |
5255 | } | |
0d9ce162 | 5256 | mutex_unlock(&kvm_lock); |
8b88b099 | 5257 | return 0; |
ba1389b7 AK |
5258 | } |
5259 | ||
ce35ef27 SJS |
5260 | static int vm_stat_clear(void *_offset, u64 val) |
5261 | { | |
5262 | unsigned offset = (long)_offset; | |
5263 | struct kvm *kvm; | |
ce35ef27 SJS |
5264 | |
5265 | if (val) | |
5266 | return -EINVAL; | |
5267 | ||
0d9ce162 | 5268 | mutex_lock(&kvm_lock); |
ce35ef27 | 5269 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5270 | kvm_clear_stat_per_vm(kvm, offset); |
ce35ef27 | 5271 | } |
0d9ce162 | 5272 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5273 | |
5274 | return 0; | |
5275 | } | |
5276 | ||
5277 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); | |
bc9e9e67 | 5278 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n"); |
ba1389b7 | 5279 | |
8b88b099 | 5280 | static int vcpu_stat_get(void *_offset, u64 *val) |
1165f5fe AK |
5281 | { |
5282 | unsigned offset = (long)_offset; | |
1165f5fe | 5283 | struct kvm *kvm; |
536a6f88 | 5284 | u64 tmp_val; |
1165f5fe | 5285 | |
8b88b099 | 5286 | *val = 0; |
0d9ce162 | 5287 | mutex_lock(&kvm_lock); |
536a6f88 | 5288 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5289 | kvm_get_stat_per_vcpu(kvm, offset, &tmp_val); |
536a6f88 JF |
5290 | *val += tmp_val; |
5291 | } | |
0d9ce162 | 5292 | mutex_unlock(&kvm_lock); |
8b88b099 | 5293 | return 0; |
1165f5fe AK |
5294 | } |
5295 | ||
ce35ef27 SJS |
5296 | static int vcpu_stat_clear(void *_offset, u64 val) |
5297 | { | |
5298 | unsigned offset = (long)_offset; | |
5299 | struct kvm *kvm; | |
ce35ef27 SJS |
5300 | |
5301 | if (val) | |
5302 | return -EINVAL; | |
5303 | ||
0d9ce162 | 5304 | mutex_lock(&kvm_lock); |
ce35ef27 | 5305 | list_for_each_entry(kvm, &vm_list, vm_list) { |
09cbcef6 | 5306 | kvm_clear_stat_per_vcpu(kvm, offset); |
ce35ef27 | 5307 | } |
0d9ce162 | 5308 | mutex_unlock(&kvm_lock); |
ce35ef27 SJS |
5309 | |
5310 | return 0; | |
5311 | } | |
5312 | ||
5313 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, | |
5314 | "%llu\n"); | |
bc9e9e67 | 5315 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n"); |
1165f5fe | 5316 | |
286de8f6 CI |
5317 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
5318 | { | |
5319 | struct kobj_uevent_env *env; | |
286de8f6 CI |
5320 | unsigned long long created, active; |
5321 | ||
5322 | if (!kvm_dev.this_device || !kvm) | |
5323 | return; | |
5324 | ||
0d9ce162 | 5325 | mutex_lock(&kvm_lock); |
286de8f6 CI |
5326 | if (type == KVM_EVENT_CREATE_VM) { |
5327 | kvm_createvm_count++; | |
5328 | kvm_active_vms++; | |
5329 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
5330 | kvm_active_vms--; | |
5331 | } | |
5332 | created = kvm_createvm_count; | |
5333 | active = kvm_active_vms; | |
0d9ce162 | 5334 | mutex_unlock(&kvm_lock); |
286de8f6 | 5335 | |
b12ce36a | 5336 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
286de8f6 CI |
5337 | if (!env) |
5338 | return; | |
5339 | ||
5340 | add_uevent_var(env, "CREATED=%llu", created); | |
5341 | add_uevent_var(env, "COUNT=%llu", active); | |
5342 | ||
fdeaf7e3 | 5343 | if (type == KVM_EVENT_CREATE_VM) { |
286de8f6 | 5344 | add_uevent_var(env, "EVENT=create"); |
fdeaf7e3 CI |
5345 | kvm->userspace_pid = task_pid_nr(current); |
5346 | } else if (type == KVM_EVENT_DESTROY_VM) { | |
286de8f6 | 5347 | add_uevent_var(env, "EVENT=destroy"); |
fdeaf7e3 CI |
5348 | } |
5349 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); | |
286de8f6 | 5350 | |
85cd39af | 5351 | if (kvm->debugfs_dentry) { |
b12ce36a | 5352 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
fdeaf7e3 CI |
5353 | |
5354 | if (p) { | |
5355 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); | |
5356 | if (!IS_ERR(tmp)) | |
5357 | add_uevent_var(env, "STATS_PATH=%s", tmp); | |
5358 | kfree(p); | |
286de8f6 CI |
5359 | } |
5360 | } | |
5361 | /* no need for checks, since we are adding at most only 5 keys */ | |
5362 | env->envp[env->envp_idx++] = NULL; | |
5363 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); | |
5364 | kfree(env); | |
286de8f6 CI |
5365 | } |
5366 | ||
929f45e3 | 5367 | static void kvm_init_debug(void) |
6aa8b732 | 5368 | { |
bc9e9e67 JZ |
5369 | const struct file_operations *fops; |
5370 | const struct _kvm_stats_desc *pdesc; | |
5371 | int i; | |
6aa8b732 | 5372 | |
76f7c879 | 5373 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
4f69b680 | 5374 | |
bc9e9e67 JZ |
5375 | for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) { |
5376 | pdesc = &kvm_vm_stats_desc[i]; | |
5377 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5378 | fops = &vm_stat_fops; | |
5379 | else | |
5380 | fops = &vm_stat_readonly_fops; | |
5381 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5382 | kvm_debugfs_dir, | |
5383 | (void *)(long)pdesc->desc.offset, fops); | |
5384 | } | |
5385 | ||
5386 | for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) { | |
5387 | pdesc = &kvm_vcpu_stats_desc[i]; | |
5388 | if (kvm_stats_debugfs_mode(pdesc) & 0222) | |
5389 | fops = &vcpu_stat_fops; | |
5390 | else | |
5391 | fops = &vcpu_stat_readonly_fops; | |
5392 | debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc), | |
5393 | kvm_debugfs_dir, | |
5394 | (void *)(long)pdesc->desc.offset, fops); | |
4f69b680 | 5395 | } |
6aa8b732 AK |
5396 | } |
5397 | ||
fb3600cc | 5398 | static int kvm_suspend(void) |
59ae6c6b | 5399 | { |
10474ae8 | 5400 | if (kvm_usage_count) |
75b7127c | 5401 | hardware_disable_nolock(NULL); |
59ae6c6b AK |
5402 | return 0; |
5403 | } | |
5404 | ||
fb3600cc | 5405 | static void kvm_resume(void) |
59ae6c6b | 5406 | { |
ca84d1a2 | 5407 | if (kvm_usage_count) { |
2eb06c30 WL |
5408 | #ifdef CONFIG_LOCKDEP |
5409 | WARN_ON(lockdep_is_held(&kvm_count_lock)); | |
5410 | #endif | |
75b7127c | 5411 | hardware_enable_nolock(NULL); |
ca84d1a2 | 5412 | } |
59ae6c6b AK |
5413 | } |
5414 | ||
fb3600cc | 5415 | static struct syscore_ops kvm_syscore_ops = { |
59ae6c6b AK |
5416 | .suspend = kvm_suspend, |
5417 | .resume = kvm_resume, | |
5418 | }; | |
5419 | ||
15ad7146 AK |
5420 | static inline |
5421 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) | |
5422 | { | |
5423 | return container_of(pn, struct kvm_vcpu, preempt_notifier); | |
5424 | } | |
5425 | ||
5426 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) | |
5427 | { | |
5428 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
f95ef0cd | 5429 | |
046ddeed | 5430 | WRITE_ONCE(vcpu->preempted, false); |
d73eb57b | 5431 | WRITE_ONCE(vcpu->ready, false); |
15ad7146 | 5432 | |
7495e22b | 5433 | __this_cpu_write(kvm_running_vcpu, vcpu); |
e790d9ef | 5434 | kvm_arch_sched_in(vcpu, cpu); |
e9b11c17 | 5435 | kvm_arch_vcpu_load(vcpu, cpu); |
15ad7146 AK |
5436 | } |
5437 | ||
5438 | static void kvm_sched_out(struct preempt_notifier *pn, | |
5439 | struct task_struct *next) | |
5440 | { | |
5441 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); | |
5442 | ||
3ba9f93b | 5443 | if (current->on_rq) { |
046ddeed | 5444 | WRITE_ONCE(vcpu->preempted, true); |
d73eb57b WL |
5445 | WRITE_ONCE(vcpu->ready, true); |
5446 | } | |
e9b11c17 | 5447 | kvm_arch_vcpu_put(vcpu); |
7495e22b PB |
5448 | __this_cpu_write(kvm_running_vcpu, NULL); |
5449 | } | |
5450 | ||
5451 | /** | |
5452 | * kvm_get_running_vcpu - get the vcpu running on the current CPU. | |
1f03b2bc MZ |
5453 | * |
5454 | * We can disable preemption locally around accessing the per-CPU variable, | |
5455 | * and use the resolved vcpu pointer after enabling preemption again, | |
5456 | * because even if the current thread is migrated to another CPU, reading | |
5457 | * the per-CPU value later will give us the same value as we update the | |
5458 | * per-CPU variable in the preempt notifier handlers. | |
7495e22b PB |
5459 | */ |
5460 | struct kvm_vcpu *kvm_get_running_vcpu(void) | |
5461 | { | |
1f03b2bc MZ |
5462 | struct kvm_vcpu *vcpu; |
5463 | ||
5464 | preempt_disable(); | |
5465 | vcpu = __this_cpu_read(kvm_running_vcpu); | |
5466 | preempt_enable(); | |
5467 | ||
5468 | return vcpu; | |
7495e22b | 5469 | } |
379a3c8e | 5470 | EXPORT_SYMBOL_GPL(kvm_get_running_vcpu); |
7495e22b PB |
5471 | |
5472 | /** | |
5473 | * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus. | |
5474 | */ | |
5475 | struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void) | |
5476 | { | |
5477 | return &kvm_running_vcpu; | |
15ad7146 AK |
5478 | } |
5479 | ||
b9904085 SC |
5480 | struct kvm_cpu_compat_check { |
5481 | void *opaque; | |
5482 | int *ret; | |
5483 | }; | |
5484 | ||
5485 | static void check_processor_compat(void *data) | |
f257d6dc | 5486 | { |
b9904085 SC |
5487 | struct kvm_cpu_compat_check *c = data; |
5488 | ||
5489 | *c->ret = kvm_arch_check_processor_compat(c->opaque); | |
f257d6dc SC |
5490 | } |
5491 | ||
0ee75bea | 5492 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
c16f862d | 5493 | struct module *module) |
6aa8b732 | 5494 | { |
b9904085 | 5495 | struct kvm_cpu_compat_check c; |
6aa8b732 | 5496 | int r; |
002c7f7c | 5497 | int cpu; |
6aa8b732 | 5498 | |
f8c16bba ZX |
5499 | r = kvm_arch_init(opaque); |
5500 | if (r) | |
d2308784 | 5501 | goto out_fail; |
cb498ea2 | 5502 | |
7dac16c3 AH |
5503 | /* |
5504 | * kvm_arch_init makes sure there's at most one caller | |
5505 | * for architectures that support multiple implementations, | |
5506 | * like intel and amd on x86. | |
36343f6e PB |
5507 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
5508 | * conflicts in case kvm is already setup for another implementation. | |
7dac16c3 | 5509 | */ |
36343f6e PB |
5510 | r = kvm_irqfd_init(); |
5511 | if (r) | |
5512 | goto out_irqfd; | |
7dac16c3 | 5513 | |
8437a617 | 5514 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
7f59f492 RR |
5515 | r = -ENOMEM; |
5516 | goto out_free_0; | |
5517 | } | |
5518 | ||
b9904085 | 5519 | r = kvm_arch_hardware_setup(opaque); |
6aa8b732 | 5520 | if (r < 0) |
faf0be22 | 5521 | goto out_free_1; |
6aa8b732 | 5522 | |
b9904085 SC |
5523 | c.ret = &r; |
5524 | c.opaque = opaque; | |
002c7f7c | 5525 | for_each_online_cpu(cpu) { |
b9904085 | 5526 | smp_call_function_single(cpu, check_processor_compat, &c, 1); |
002c7f7c | 5527 | if (r < 0) |
faf0be22 | 5528 | goto out_free_2; |
002c7f7c YS |
5529 | } |
5530 | ||
73c1b41e | 5531 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
8c18b2d2 | 5532 | kvm_starting_cpu, kvm_dying_cpu); |
774c47f1 | 5533 | if (r) |
d2308784 | 5534 | goto out_free_2; |
6aa8b732 AK |
5535 | register_reboot_notifier(&kvm_reboot_notifier); |
5536 | ||
c16f862d | 5537 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
0ee75bea AK |
5538 | if (!vcpu_align) |
5539 | vcpu_align = __alignof__(struct kvm_vcpu); | |
46515736 PB |
5540 | kvm_vcpu_cache = |
5541 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, | |
5542 | SLAB_ACCOUNT, | |
5543 | offsetof(struct kvm_vcpu, arch), | |
ce55c049 JZ |
5544 | offsetofend(struct kvm_vcpu, stats_id) |
5545 | - offsetof(struct kvm_vcpu, arch), | |
46515736 | 5546 | NULL); |
c16f862d RR |
5547 | if (!kvm_vcpu_cache) { |
5548 | r = -ENOMEM; | |
fb3600cc | 5549 | goto out_free_3; |
c16f862d RR |
5550 | } |
5551 | ||
af585b92 GN |
5552 | r = kvm_async_pf_init(); |
5553 | if (r) | |
5554 | goto out_free; | |
5555 | ||
6aa8b732 | 5556 | kvm_chardev_ops.owner = module; |
3d3aab1b CB |
5557 | kvm_vm_fops.owner = module; |
5558 | kvm_vcpu_fops.owner = module; | |
6aa8b732 AK |
5559 | |
5560 | r = misc_register(&kvm_dev); | |
5561 | if (r) { | |
1170adc6 | 5562 | pr_err("kvm: misc device register failed\n"); |
af585b92 | 5563 | goto out_unreg; |
6aa8b732 AK |
5564 | } |
5565 | ||
fb3600cc RW |
5566 | register_syscore_ops(&kvm_syscore_ops); |
5567 | ||
15ad7146 AK |
5568 | kvm_preempt_ops.sched_in = kvm_sched_in; |
5569 | kvm_preempt_ops.sched_out = kvm_sched_out; | |
5570 | ||
929f45e3 | 5571 | kvm_init_debug(); |
0ea4ed8e | 5572 | |
3c3c29fd PB |
5573 | r = kvm_vfio_ops_init(); |
5574 | WARN_ON(r); | |
5575 | ||
c7addb90 | 5576 | return 0; |
6aa8b732 | 5577 | |
af585b92 GN |
5578 | out_unreg: |
5579 | kvm_async_pf_deinit(); | |
6aa8b732 | 5580 | out_free: |
c16f862d | 5581 | kmem_cache_destroy(kvm_vcpu_cache); |
d2308784 | 5582 | out_free_3: |
6aa8b732 | 5583 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5584 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
d2308784 | 5585 | out_free_2: |
e9b11c17 | 5586 | kvm_arch_hardware_unsetup(); |
faf0be22 | 5587 | out_free_1: |
7f59f492 | 5588 | free_cpumask_var(cpus_hardware_enabled); |
d2308784 | 5589 | out_free_0: |
a0f155e9 | 5590 | kvm_irqfd_exit(); |
36343f6e | 5591 | out_irqfd: |
7dac16c3 AH |
5592 | kvm_arch_exit(); |
5593 | out_fail: | |
6aa8b732 AK |
5594 | return r; |
5595 | } | |
cb498ea2 | 5596 | EXPORT_SYMBOL_GPL(kvm_init); |
6aa8b732 | 5597 | |
cb498ea2 | 5598 | void kvm_exit(void) |
6aa8b732 | 5599 | { |
4bd33b56 | 5600 | debugfs_remove_recursive(kvm_debugfs_dir); |
6aa8b732 | 5601 | misc_deregister(&kvm_dev); |
c16f862d | 5602 | kmem_cache_destroy(kvm_vcpu_cache); |
af585b92 | 5603 | kvm_async_pf_deinit(); |
fb3600cc | 5604 | unregister_syscore_ops(&kvm_syscore_ops); |
6aa8b732 | 5605 | unregister_reboot_notifier(&kvm_reboot_notifier); |
8c18b2d2 | 5606 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
75b7127c | 5607 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
e9b11c17 | 5608 | kvm_arch_hardware_unsetup(); |
f8c16bba | 5609 | kvm_arch_exit(); |
a0f155e9 | 5610 | kvm_irqfd_exit(); |
7f59f492 | 5611 | free_cpumask_var(cpus_hardware_enabled); |
571ee1b6 | 5612 | kvm_vfio_ops_exit(); |
6aa8b732 | 5613 | } |
cb498ea2 | 5614 | EXPORT_SYMBOL_GPL(kvm_exit); |
c57c8046 JS |
5615 | |
5616 | struct kvm_vm_worker_thread_context { | |
5617 | struct kvm *kvm; | |
5618 | struct task_struct *parent; | |
5619 | struct completion init_done; | |
5620 | kvm_vm_thread_fn_t thread_fn; | |
5621 | uintptr_t data; | |
5622 | int err; | |
5623 | }; | |
5624 | ||
5625 | static int kvm_vm_worker_thread(void *context) | |
5626 | { | |
5627 | /* | |
5628 | * The init_context is allocated on the stack of the parent thread, so | |
5629 | * we have to locally copy anything that is needed beyond initialization | |
5630 | */ | |
5631 | struct kvm_vm_worker_thread_context *init_context = context; | |
5632 | struct kvm *kvm = init_context->kvm; | |
5633 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; | |
5634 | uintptr_t data = init_context->data; | |
5635 | int err; | |
5636 | ||
5637 | err = kthread_park(current); | |
5638 | /* kthread_park(current) is never supposed to return an error */ | |
5639 | WARN_ON(err != 0); | |
5640 | if (err) | |
5641 | goto init_complete; | |
5642 | ||
5643 | err = cgroup_attach_task_all(init_context->parent, current); | |
5644 | if (err) { | |
5645 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", | |
5646 | __func__, err); | |
5647 | goto init_complete; | |
5648 | } | |
5649 | ||
5650 | set_user_nice(current, task_nice(init_context->parent)); | |
5651 | ||
5652 | init_complete: | |
5653 | init_context->err = err; | |
5654 | complete(&init_context->init_done); | |
5655 | init_context = NULL; | |
5656 | ||
5657 | if (err) | |
5658 | return err; | |
5659 | ||
5660 | /* Wait to be woken up by the spawner before proceeding. */ | |
5661 | kthread_parkme(); | |
5662 | ||
5663 | if (!kthread_should_stop()) | |
5664 | err = thread_fn(kvm, data); | |
5665 | ||
5666 | return err; | |
5667 | } | |
5668 | ||
5669 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, | |
5670 | uintptr_t data, const char *name, | |
5671 | struct task_struct **thread_ptr) | |
5672 | { | |
5673 | struct kvm_vm_worker_thread_context init_context = {}; | |
5674 | struct task_struct *thread; | |
5675 | ||
5676 | *thread_ptr = NULL; | |
5677 | init_context.kvm = kvm; | |
5678 | init_context.parent = current; | |
5679 | init_context.thread_fn = thread_fn; | |
5680 | init_context.data = data; | |
5681 | init_completion(&init_context.init_done); | |
5682 | ||
5683 | thread = kthread_run(kvm_vm_worker_thread, &init_context, | |
5684 | "%s-%d", name, task_pid_nr(current)); | |
5685 | if (IS_ERR(thread)) | |
5686 | return PTR_ERR(thread); | |
5687 | ||
5688 | /* kthread_run is never supposed to return NULL */ | |
5689 | WARN_ON(thread == NULL); | |
5690 | ||
5691 | wait_for_completion(&init_context.init_done); | |
5692 | ||
5693 | if (!init_context.err) | |
5694 | *thread_ptr = thread; | |
5695 | ||
5696 | return init_context.err; | |
5697 | } |