xref: /openbmc/linux/arch/x86/kvm/mmu/page_track.c (revision 7b574863e71833b5a4907245c1d95e76d507b984)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 
15 #include <linux/kvm_host.h>
16 #include <linux/rculist.h>
17 
18 #include "mmu.h"
19 #include "mmu_internal.h"
20 #include "page_track.h"
21 
22 bool kvm_page_track_write_tracking_enabled(struct kvm *kvm)
23 {
24 	return IS_ENABLED(CONFIG_KVM_EXTERNAL_WRITE_TRACKING) ||
25 	       !tdp_enabled || kvm_shadow_root_allocated(kvm);
26 }
27 
28 void kvm_page_track_free_memslot(struct kvm_memory_slot *slot)
29 {
30 	kvfree(slot->arch.gfn_write_track);
31 	slot->arch.gfn_write_track = NULL;
32 }
33 
34 static int __kvm_page_track_write_tracking_alloc(struct kvm_memory_slot *slot,
35 						 unsigned long npages)
36 {
37 	const size_t size = sizeof(*slot->arch.gfn_write_track);
38 
39 	if (!slot->arch.gfn_write_track)
40 		slot->arch.gfn_write_track = __vcalloc(npages, size,
41 						       GFP_KERNEL_ACCOUNT);
42 
43 	return slot->arch.gfn_write_track ? 0 : -ENOMEM;
44 }
45 
46 int kvm_page_track_create_memslot(struct kvm *kvm,
47 				  struct kvm_memory_slot *slot,
48 				  unsigned long npages)
49 {
50 	if (!kvm_page_track_write_tracking_enabled(kvm))
51 		return 0;
52 
53 	return __kvm_page_track_write_tracking_alloc(slot, npages);
54 }
55 
56 int kvm_page_track_write_tracking_alloc(struct kvm_memory_slot *slot)
57 {
58 	return __kvm_page_track_write_tracking_alloc(slot, slot->npages);
59 }
60 
61 static void update_gfn_write_track(struct kvm_memory_slot *slot, gfn_t gfn,
62 				   short count)
63 {
64 	int index, val;
65 
66 	index = gfn_to_index(gfn, slot->base_gfn, PG_LEVEL_4K);
67 
68 	val = slot->arch.gfn_write_track[index];
69 
70 	if (WARN_ON_ONCE(val + count < 0 || val + count > USHRT_MAX))
71 		return;
72 
73 	slot->arch.gfn_write_track[index] += count;
74 }
75 
76 /*
77  * add guest page to the tracking pool so that corresponding access on that
78  * page will be intercepted.
79  *
80  * It should be called under the protection both of mmu-lock and kvm->srcu
81  * or kvm->slots_lock.
82  *
83  * @kvm: the guest instance we are interested in.
84  * @slot: the @gfn belongs to.
85  * @gfn: the guest page.
86  */
87 void kvm_write_track_add_gfn(struct kvm *kvm, struct kvm_memory_slot *slot,
88 			     gfn_t gfn)
89 {
90 	if (WARN_ON_ONCE(!kvm_page_track_write_tracking_enabled(kvm)))
91 		return;
92 
93 	update_gfn_write_track(slot, gfn, 1);
94 
95 	/*
96 	 * new track stops large page mapping for the
97 	 * tracked page.
98 	 */
99 	kvm_mmu_gfn_disallow_lpage(slot, gfn);
100 
101 	if (kvm_mmu_slot_gfn_write_protect(kvm, slot, gfn, PG_LEVEL_4K))
102 		kvm_flush_remote_tlbs(kvm);
103 }
104 EXPORT_SYMBOL_GPL(kvm_write_track_add_gfn);
105 
106 /*
107  * remove the guest page from the tracking pool which stops the interception
108  * of corresponding access on that page.
109  *
110  * It should be called under the protection both of mmu-lock and kvm->srcu
111  * or kvm->slots_lock.
112  *
113  * @kvm: the guest instance we are interested in.
114  * @slot: the @gfn belongs to.
115  * @gfn: the guest page.
116  */
117 void kvm_write_track_remove_gfn(struct kvm *kvm,
118 				struct kvm_memory_slot *slot, gfn_t gfn)
119 {
120 	if (WARN_ON_ONCE(!kvm_page_track_write_tracking_enabled(kvm)))
121 		return;
122 
123 	update_gfn_write_track(slot, gfn, -1);
124 
125 	/*
126 	 * allow large page mapping for the tracked page
127 	 * after the tracker is gone.
128 	 */
129 	kvm_mmu_gfn_allow_lpage(slot, gfn);
130 }
131 EXPORT_SYMBOL_GPL(kvm_write_track_remove_gfn);
132 
133 /*
134  * check if the corresponding access on the specified guest page is tracked.
135  */
136 bool kvm_gfn_is_write_tracked(struct kvm *kvm,
137 			      const struct kvm_memory_slot *slot, gfn_t gfn)
138 {
139 	int index;
140 
141 	if (!slot)
142 		return false;
143 
144 	if (!kvm_page_track_write_tracking_enabled(kvm))
145 		return false;
146 
147 	index = gfn_to_index(gfn, slot->base_gfn, PG_LEVEL_4K);
148 	return !!READ_ONCE(slot->arch.gfn_write_track[index]);
149 }
150 
151 #ifdef CONFIG_KVM_EXTERNAL_WRITE_TRACKING
152 void kvm_page_track_cleanup(struct kvm *kvm)
153 {
154 	struct kvm_page_track_notifier_head *head;
155 
156 	head = &kvm->arch.track_notifier_head;
157 	cleanup_srcu_struct(&head->track_srcu);
158 }
159 
160 int kvm_page_track_init(struct kvm *kvm)
161 {
162 	struct kvm_page_track_notifier_head *head;
163 
164 	head = &kvm->arch.track_notifier_head;
165 	INIT_HLIST_HEAD(&head->track_notifier_list);
166 	return init_srcu_struct(&head->track_srcu);
167 }
168 
169 /*
170  * register the notifier so that event interception for the tracked guest
171  * pages can be received.
172  */
173 void
174 kvm_page_track_register_notifier(struct kvm *kvm,
175 				 struct kvm_page_track_notifier_node *n)
176 {
177 	struct kvm_page_track_notifier_head *head;
178 
179 	head = &kvm->arch.track_notifier_head;
180 
181 	write_lock(&kvm->mmu_lock);
182 	hlist_add_head_rcu(&n->node, &head->track_notifier_list);
183 	write_unlock(&kvm->mmu_lock);
184 }
185 EXPORT_SYMBOL_GPL(kvm_page_track_register_notifier);
186 
187 /*
188  * stop receiving the event interception. It is the opposed operation of
189  * kvm_page_track_register_notifier().
190  */
191 void
192 kvm_page_track_unregister_notifier(struct kvm *kvm,
193 				   struct kvm_page_track_notifier_node *n)
194 {
195 	struct kvm_page_track_notifier_head *head;
196 
197 	head = &kvm->arch.track_notifier_head;
198 
199 	write_lock(&kvm->mmu_lock);
200 	hlist_del_rcu(&n->node);
201 	write_unlock(&kvm->mmu_lock);
202 	synchronize_srcu(&head->track_srcu);
203 }
204 EXPORT_SYMBOL_GPL(kvm_page_track_unregister_notifier);
205 
206 /*
207  * Notify the node that write access is intercepted and write emulation is
208  * finished at this time.
209  *
210  * The node should figure out if the written page is the one that node is
211  * interested in by itself.
212  */
213 void __kvm_page_track_write(struct kvm *kvm, gpa_t gpa, const u8 *new, int bytes)
214 {
215 	struct kvm_page_track_notifier_head *head;
216 	struct kvm_page_track_notifier_node *n;
217 	int idx;
218 
219 	head = &kvm->arch.track_notifier_head;
220 
221 	if (hlist_empty(&head->track_notifier_list))
222 		return;
223 
224 	idx = srcu_read_lock(&head->track_srcu);
225 	hlist_for_each_entry_srcu(n, &head->track_notifier_list, node,
226 				  srcu_read_lock_held(&head->track_srcu))
227 		if (n->track_write)
228 			n->track_write(gpa, new, bytes, n);
229 	srcu_read_unlock(&head->track_srcu, idx);
230 }
231 
232 /*
233  * Notify external page track nodes that a memory region is being removed from
234  * the VM, e.g. so that users can free any associated metadata.
235  */
236 void kvm_page_track_delete_slot(struct kvm *kvm, struct kvm_memory_slot *slot)
237 {
238 	struct kvm_page_track_notifier_head *head;
239 	struct kvm_page_track_notifier_node *n;
240 	int idx;
241 
242 	head = &kvm->arch.track_notifier_head;
243 
244 	if (hlist_empty(&head->track_notifier_list))
245 		return;
246 
247 	idx = srcu_read_lock(&head->track_srcu);
248 	hlist_for_each_entry_srcu(n, &head->track_notifier_list, node,
249 				  srcu_read_lock_held(&head->track_srcu))
250 		if (n->track_remove_region)
251 			n->track_remove_region(slot->base_gfn, slot->npages, n);
252 	srcu_read_unlock(&head->track_srcu, idx);
253 }
254 
255 #endif
256