xref: /openbmc/linux/arch/x86/kvm/mmu/page_track.c (revision 42641042)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Support KVM gust page tracking
4  *
5  * This feature allows us to track page access in guest. Currently, only
6  * write access is tracked.
7  *
8  * Copyright(C) 2015 Intel Corporation.
9  *
10  * Author:
11  *   Xiao Guangrong <guangrong.xiao@linux.intel.com>
12  */
13 
14 #include <linux/kvm_host.h>
15 #include <linux/rculist.h>
16 
17 #include <asm/kvm_page_track.h>
18 
19 #include "mmu.h"
20 #include "mmu_internal.h"
21 
22 void kvm_page_track_free_memslot(struct kvm_memory_slot *slot)
23 {
24 	int i;
25 
26 	for (i = 0; i < KVM_PAGE_TRACK_MAX; i++) {
27 		kvfree(slot->arch.gfn_track[i]);
28 		slot->arch.gfn_track[i] = NULL;
29 	}
30 }
31 
32 int kvm_page_track_create_memslot(struct kvm_memory_slot *slot,
33 				  unsigned long npages)
34 {
35 	int  i;
36 
37 	for (i = 0; i < KVM_PAGE_TRACK_MAX; i++) {
38 		slot->arch.gfn_track[i] =
39 			kvcalloc(npages, sizeof(*slot->arch.gfn_track[i]),
40 				 GFP_KERNEL_ACCOUNT);
41 		if (!slot->arch.gfn_track[i])
42 			goto track_free;
43 	}
44 
45 	return 0;
46 
47 track_free:
48 	kvm_page_track_free_memslot(slot);
49 	return -ENOMEM;
50 }
51 
52 static inline bool page_track_mode_is_valid(enum kvm_page_track_mode mode)
53 {
54 	if (mode < 0 || mode >= KVM_PAGE_TRACK_MAX)
55 		return false;
56 
57 	return true;
58 }
59 
60 static void update_gfn_track(struct kvm_memory_slot *slot, gfn_t gfn,
61 			     enum kvm_page_track_mode mode, short count)
62 {
63 	int index, val;
64 
65 	index = gfn_to_index(gfn, slot->base_gfn, PG_LEVEL_4K);
66 
67 	val = slot->arch.gfn_track[mode][index];
68 
69 	if (WARN_ON(val + count < 0 || val + count > USHRT_MAX))
70 		return;
71 
72 	slot->arch.gfn_track[mode][index] += count;
73 }
74 
75 /*
76  * add guest page to the tracking pool so that corresponding access on that
77  * page will be intercepted.
78  *
79  * It should be called under the protection both of mmu-lock and kvm->srcu
80  * or kvm->slots_lock.
81  *
82  * @kvm: the guest instance we are interested in.
83  * @slot: the @gfn belongs to.
84  * @gfn: the guest page.
85  * @mode: tracking mode, currently only write track is supported.
86  */
87 void kvm_slot_page_track_add_page(struct kvm *kvm,
88 				  struct kvm_memory_slot *slot, gfn_t gfn,
89 				  enum kvm_page_track_mode mode)
90 {
91 
92 	if (WARN_ON(!page_track_mode_is_valid(mode)))
93 		return;
94 
95 	update_gfn_track(slot, gfn, mode, 1);
96 
97 	/*
98 	 * new track stops large page mapping for the
99 	 * tracked page.
100 	 */
101 	kvm_mmu_gfn_disallow_lpage(slot, gfn);
102 
103 	if (mode == KVM_PAGE_TRACK_WRITE)
104 		if (kvm_mmu_slot_gfn_write_protect(kvm, slot, gfn, PG_LEVEL_4K))
105 			kvm_flush_remote_tlbs(kvm);
106 }
107 EXPORT_SYMBOL_GPL(kvm_slot_page_track_add_page);
108 
109 /*
110  * remove the guest page from the tracking pool which stops the interception
111  * of corresponding access on that page. It is the opposed operation of
112  * kvm_slot_page_track_add_page().
113  *
114  * It should be called under the protection both of mmu-lock and kvm->srcu
115  * or kvm->slots_lock.
116  *
117  * @kvm: the guest instance we are interested in.
118  * @slot: the @gfn belongs to.
119  * @gfn: the guest page.
120  * @mode: tracking mode, currently only write track is supported.
121  */
122 void kvm_slot_page_track_remove_page(struct kvm *kvm,
123 				     struct kvm_memory_slot *slot, gfn_t gfn,
124 				     enum kvm_page_track_mode mode)
125 {
126 	if (WARN_ON(!page_track_mode_is_valid(mode)))
127 		return;
128 
129 	update_gfn_track(slot, gfn, mode, -1);
130 
131 	/*
132 	 * allow large page mapping for the tracked page
133 	 * after the tracker is gone.
134 	 */
135 	kvm_mmu_gfn_allow_lpage(slot, gfn);
136 }
137 EXPORT_SYMBOL_GPL(kvm_slot_page_track_remove_page);
138 
139 /*
140  * check if the corresponding access on the specified guest page is tracked.
141  */
142 bool kvm_page_track_is_active(struct kvm_vcpu *vcpu, gfn_t gfn,
143 			      enum kvm_page_track_mode mode)
144 {
145 	struct kvm_memory_slot *slot;
146 	int index;
147 
148 	if (WARN_ON(!page_track_mode_is_valid(mode)))
149 		return false;
150 
151 	slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
152 	if (!slot)
153 		return false;
154 
155 	index = gfn_to_index(gfn, slot->base_gfn, PG_LEVEL_4K);
156 	return !!READ_ONCE(slot->arch.gfn_track[mode][index]);
157 }
158 
159 void kvm_page_track_cleanup(struct kvm *kvm)
160 {
161 	struct kvm_page_track_notifier_head *head;
162 
163 	head = &kvm->arch.track_notifier_head;
164 	cleanup_srcu_struct(&head->track_srcu);
165 }
166 
167 int kvm_page_track_init(struct kvm *kvm)
168 {
169 	struct kvm_page_track_notifier_head *head;
170 
171 	head = &kvm->arch.track_notifier_head;
172 	INIT_HLIST_HEAD(&head->track_notifier_list);
173 	return init_srcu_struct(&head->track_srcu);
174 }
175 
176 /*
177  * register the notifier so that event interception for the tracked guest
178  * pages can be received.
179  */
180 void
181 kvm_page_track_register_notifier(struct kvm *kvm,
182 				 struct kvm_page_track_notifier_node *n)
183 {
184 	struct kvm_page_track_notifier_head *head;
185 
186 	head = &kvm->arch.track_notifier_head;
187 
188 	write_lock(&kvm->mmu_lock);
189 	hlist_add_head_rcu(&n->node, &head->track_notifier_list);
190 	write_unlock(&kvm->mmu_lock);
191 }
192 EXPORT_SYMBOL_GPL(kvm_page_track_register_notifier);
193 
194 /*
195  * stop receiving the event interception. It is the opposed operation of
196  * kvm_page_track_register_notifier().
197  */
198 void
199 kvm_page_track_unregister_notifier(struct kvm *kvm,
200 				   struct kvm_page_track_notifier_node *n)
201 {
202 	struct kvm_page_track_notifier_head *head;
203 
204 	head = &kvm->arch.track_notifier_head;
205 
206 	write_lock(&kvm->mmu_lock);
207 	hlist_del_rcu(&n->node);
208 	write_unlock(&kvm->mmu_lock);
209 	synchronize_srcu(&head->track_srcu);
210 }
211 EXPORT_SYMBOL_GPL(kvm_page_track_unregister_notifier);
212 
213 /*
214  * Notify the node that write access is intercepted and write emulation is
215  * finished at this time.
216  *
217  * The node should figure out if the written page is the one that node is
218  * interested in by itself.
219  */
220 void kvm_page_track_write(struct kvm_vcpu *vcpu, gpa_t gpa, const u8 *new,
221 			  int bytes)
222 {
223 	struct kvm_page_track_notifier_head *head;
224 	struct kvm_page_track_notifier_node *n;
225 	int idx;
226 
227 	head = &vcpu->kvm->arch.track_notifier_head;
228 
229 	if (hlist_empty(&head->track_notifier_list))
230 		return;
231 
232 	idx = srcu_read_lock(&head->track_srcu);
233 	hlist_for_each_entry_srcu(n, &head->track_notifier_list, node,
234 				srcu_read_lock_held(&head->track_srcu))
235 		if (n->track_write)
236 			n->track_write(vcpu, gpa, new, bytes, n);
237 	srcu_read_unlock(&head->track_srcu, idx);
238 }
239 
240 /*
241  * Notify the node that memory slot is being removed or moved so that it can
242  * drop write-protection for the pages in the memory slot.
243  *
244  * The node should figure out it has any write-protected pages in this slot
245  * by itself.
246  */
247 void kvm_page_track_flush_slot(struct kvm *kvm, struct kvm_memory_slot *slot)
248 {
249 	struct kvm_page_track_notifier_head *head;
250 	struct kvm_page_track_notifier_node *n;
251 	int idx;
252 
253 	head = &kvm->arch.track_notifier_head;
254 
255 	if (hlist_empty(&head->track_notifier_list))
256 		return;
257 
258 	idx = srcu_read_lock(&head->track_srcu);
259 	hlist_for_each_entry_srcu(n, &head->track_notifier_list, node,
260 				srcu_read_lock_held(&head->track_srcu))
261 		if (n->track_flush_slot)
262 			n->track_flush_slot(kvm, slot, n);
263 	srcu_read_unlock(&head->track_srcu, idx);
264 }
265