2 * Copyright(c) 2016 - 2017 Intel Corporation.
4 * This file is provided under a dual BSD/GPLv2 license. When using or
5 * redistributing this file, you may do so under either license.
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of version 2 of the GNU General Public License as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
20 * Redistribution and use in source and binary forms, with or without
21 * modification, are permitted provided that the following conditions
24 * - Redistributions of source code must retain the above copyright
25 * notice, this list of conditions and the following disclaimer.
26 * - Redistributions in binary form must reproduce the above copyright
27 * notice, this list of conditions and the following disclaimer in
28 * the documentation and/or other materials provided with the
30 * - Neither the name of Intel Corporation nor the names of its
31 * contributors may be used to endorse or promote products derived
32 * from this software without specific prior written permission.
34 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
36 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
38 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
39 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
40 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
42 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
43 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
47 #include <linux/list.h>
48 #include <linux/rculist.h>
49 #include <linux/mmu_notifier.h>
50 #include <linux/interval_tree_generic.h>
55 struct mmu_rb_handler {
56 struct mmu_notifier mn;
57 struct rb_root_cached root;
59 spinlock_t lock; /* protect the RB tree */
60 struct mmu_rb_ops *ops;
62 struct list_head lru_list;
63 struct work_struct del_work;
64 struct list_head del_list;
65 struct workqueue_struct *wq;
68 static unsigned long mmu_node_start(struct mmu_rb_node *);
69 static unsigned long mmu_node_last(struct mmu_rb_node *);
70 static int mmu_notifier_range_start(struct mmu_notifier *,
72 unsigned long, unsigned long, bool);
73 static struct mmu_rb_node *__mmu_rb_search(struct mmu_rb_handler *,
74 unsigned long, unsigned long);
75 static void do_remove(struct mmu_rb_handler *handler,
76 struct list_head *del_list);
77 static void handle_remove(struct work_struct *work);
79 static const struct mmu_notifier_ops mn_opts = {
80 .flags = MMU_INVALIDATE_DOES_NOT_BLOCK,
81 .invalidate_range_start = mmu_notifier_range_start,
84 INTERVAL_TREE_DEFINE(struct mmu_rb_node, node, unsigned long, __last,
85 mmu_node_start, mmu_node_last, static, __mmu_int_rb);
87 static unsigned long mmu_node_start(struct mmu_rb_node *node)
89 return node->addr & PAGE_MASK;
92 static unsigned long mmu_node_last(struct mmu_rb_node *node)
94 return PAGE_ALIGN(node->addr + node->len) - 1;
97 int hfi1_mmu_rb_register(void *ops_arg, struct mm_struct *mm,
98 struct mmu_rb_ops *ops,
99 struct workqueue_struct *wq,
100 struct mmu_rb_handler **handler)
102 struct mmu_rb_handler *handlr;
105 handlr = kmalloc(sizeof(*handlr), GFP_KERNEL);
109 handlr->root = RB_ROOT_CACHED;
111 handlr->ops_arg = ops_arg;
112 INIT_HLIST_NODE(&handlr->mn.hlist);
113 spin_lock_init(&handlr->lock);
114 handlr->mn.ops = &mn_opts;
116 INIT_WORK(&handlr->del_work, handle_remove);
117 INIT_LIST_HEAD(&handlr->del_list);
118 INIT_LIST_HEAD(&handlr->lru_list);
121 ret = mmu_notifier_register(&handlr->mn, handlr->mm);
131 void hfi1_mmu_rb_unregister(struct mmu_rb_handler *handler)
133 struct mmu_rb_node *rbnode;
134 struct rb_node *node;
136 struct list_head del_list;
138 /* Unregister first so we don't get any more notifications. */
139 mmu_notifier_unregister(&handler->mn, handler->mm);
142 * Make sure the wq delete handler is finished running. It will not
143 * be triggered once the mmu notifiers are unregistered above.
145 flush_work(&handler->del_work);
147 INIT_LIST_HEAD(&del_list);
149 spin_lock_irqsave(&handler->lock, flags);
150 while ((node = rb_first_cached(&handler->root))) {
151 rbnode = rb_entry(node, struct mmu_rb_node, node);
152 rb_erase_cached(node, &handler->root);
153 /* move from LRU list to delete list */
154 list_move(&rbnode->list, &del_list);
156 spin_unlock_irqrestore(&handler->lock, flags);
158 do_remove(handler, &del_list);
163 int hfi1_mmu_rb_insert(struct mmu_rb_handler *handler,
164 struct mmu_rb_node *mnode)
166 struct mmu_rb_node *node;
170 trace_hfi1_mmu_rb_insert(mnode->addr, mnode->len);
171 spin_lock_irqsave(&handler->lock, flags);
172 node = __mmu_rb_search(handler, mnode->addr, mnode->len);
177 __mmu_int_rb_insert(mnode, &handler->root);
178 list_add_tail(&mnode->list, &handler->lru_list);
180 ret = handler->ops->insert(handler->ops_arg, mnode);
182 __mmu_int_rb_remove(mnode, &handler->root);
183 list_del(&mnode->list); /* remove from LRU list */
186 spin_unlock_irqrestore(&handler->lock, flags);
190 /* Caller must hold handler lock */
191 static struct mmu_rb_node *__mmu_rb_search(struct mmu_rb_handler *handler,
195 struct mmu_rb_node *node = NULL;
197 trace_hfi1_mmu_rb_search(addr, len);
198 if (!handler->ops->filter) {
199 node = __mmu_int_rb_iter_first(&handler->root, addr,
202 for (node = __mmu_int_rb_iter_first(&handler->root, addr,
205 node = __mmu_int_rb_iter_next(node, addr,
207 if (handler->ops->filter(node, addr, len))
214 bool hfi1_mmu_rb_remove_unless_exact(struct mmu_rb_handler *handler,
215 unsigned long addr, unsigned long len,
216 struct mmu_rb_node **rb_node)
218 struct mmu_rb_node *node;
222 spin_lock_irqsave(&handler->lock, flags);
223 node = __mmu_rb_search(handler, addr, len);
225 if (node->addr == addr && node->len == len) {
226 list_move_tail(&node->list, &handler->lru_list);
229 __mmu_int_rb_remove(node, &handler->root);
230 list_del(&node->list); /* remove from LRU list */
234 spin_unlock_irqrestore(&handler->lock, flags);
239 void hfi1_mmu_rb_evict(struct mmu_rb_handler *handler, void *evict_arg)
241 struct mmu_rb_node *rbnode, *ptr;
242 struct list_head del_list;
246 INIT_LIST_HEAD(&del_list);
248 spin_lock_irqsave(&handler->lock, flags);
249 list_for_each_entry_safe(rbnode, ptr, &handler->lru_list, list) {
250 if (handler->ops->evict(handler->ops_arg, rbnode, evict_arg,
252 __mmu_int_rb_remove(rbnode, &handler->root);
253 /* move from LRU list to delete list */
254 list_move(&rbnode->list, &del_list);
259 spin_unlock_irqrestore(&handler->lock, flags);
261 list_for_each_entry_safe(rbnode, ptr, &del_list, list) {
262 handler->ops->remove(handler->ops_arg, rbnode);
267 * It is up to the caller to ensure that this function does not race with the
268 * mmu invalidate notifier which may be calling the users remove callback on
271 void hfi1_mmu_rb_remove(struct mmu_rb_handler *handler,
272 struct mmu_rb_node *node)
276 /* Validity of handler and node pointers has been checked by caller. */
277 trace_hfi1_mmu_rb_remove(node->addr, node->len);
278 spin_lock_irqsave(&handler->lock, flags);
279 __mmu_int_rb_remove(node, &handler->root);
280 list_del(&node->list); /* remove from LRU list */
281 spin_unlock_irqrestore(&handler->lock, flags);
283 handler->ops->remove(handler->ops_arg, node);
286 static int mmu_notifier_range_start(struct mmu_notifier *mn,
287 struct mm_struct *mm,
292 struct mmu_rb_handler *handler =
293 container_of(mn, struct mmu_rb_handler, mn);
294 struct rb_root_cached *root = &handler->root;
295 struct mmu_rb_node *node, *ptr = NULL;
299 spin_lock_irqsave(&handler->lock, flags);
300 for (node = __mmu_int_rb_iter_first(root, start, end - 1);
302 /* Guard against node removal. */
303 ptr = __mmu_int_rb_iter_next(node, start, end - 1);
304 trace_hfi1_mmu_mem_invalidate(node->addr, node->len);
305 if (handler->ops->invalidate(handler->ops_arg, node)) {
306 __mmu_int_rb_remove(node, root);
307 /* move from LRU list to delete list */
308 list_move(&node->list, &handler->del_list);
312 spin_unlock_irqrestore(&handler->lock, flags);
315 queue_work(handler->wq, &handler->del_work);
321 * Call the remove function for the given handler and the list. This
322 * is expected to be called with a delete list extracted from handler.
323 * The caller should not be holding the handler lock.
325 static void do_remove(struct mmu_rb_handler *handler,
326 struct list_head *del_list)
328 struct mmu_rb_node *node;
330 while (!list_empty(del_list)) {
331 node = list_first_entry(del_list, struct mmu_rb_node, list);
332 list_del(&node->list);
333 handler->ops->remove(handler->ops_arg, node);
338 * Work queue function to remove all nodes that have been queued up to
339 * be removed. The key feature is that mm->mmap_sem is not being held
340 * and the remove callback can sleep while taking it, if needed.
342 static void handle_remove(struct work_struct *work)
344 struct mmu_rb_handler *handler = container_of(work,
345 struct mmu_rb_handler,
347 struct list_head del_list;
350 /* remove anything that is queued to get removed */
351 spin_lock_irqsave(&handler->lock, flags);
352 list_replace_init(&handler->del_list, &del_list);
353 spin_unlock_irqrestore(&handler->lock, flags);
355 do_remove(handler, &del_list);