2 * Copyright(c) 2016 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;
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 inline void mmu_notifier_range_start(struct mmu_notifier *,
72 unsigned long, unsigned long);
73 static void mmu_notifier_mem_invalidate(struct mmu_notifier *,
75 unsigned long, unsigned long);
76 static struct mmu_rb_node *__mmu_rb_search(struct mmu_rb_handler *,
77 unsigned long, unsigned long);
78 static void do_remove(struct mmu_rb_handler *handler,
79 struct list_head *del_list);
80 static void handle_remove(struct work_struct *work);
82 static const struct mmu_notifier_ops mn_opts = {
83 .invalidate_range_start = mmu_notifier_range_start,
86 INTERVAL_TREE_DEFINE(struct mmu_rb_node, node, unsigned long, __last,
87 mmu_node_start, mmu_node_last, static, __mmu_int_rb);
89 static unsigned long mmu_node_start(struct mmu_rb_node *node)
91 return node->addr & PAGE_MASK;
94 static unsigned long mmu_node_last(struct mmu_rb_node *node)
96 return PAGE_ALIGN(node->addr + node->len) - 1;
99 int hfi1_mmu_rb_register(void *ops_arg, struct mm_struct *mm,
100 struct mmu_rb_ops *ops,
101 struct workqueue_struct *wq,
102 struct mmu_rb_handler **handler)
104 struct mmu_rb_handler *handlr;
107 handlr = kmalloc(sizeof(*handlr), GFP_KERNEL);
111 handlr->root = RB_ROOT;
113 handlr->ops_arg = ops_arg;
114 INIT_HLIST_NODE(&handlr->mn.hlist);
115 spin_lock_init(&handlr->lock);
116 handlr->mn.ops = &mn_opts;
118 INIT_WORK(&handlr->del_work, handle_remove);
119 INIT_LIST_HEAD(&handlr->del_list);
120 INIT_LIST_HEAD(&handlr->lru_list);
123 ret = mmu_notifier_register(&handlr->mn, handlr->mm);
133 void hfi1_mmu_rb_unregister(struct mmu_rb_handler *handler)
135 struct mmu_rb_node *rbnode;
136 struct rb_node *node;
138 struct list_head del_list;
140 /* Unregister first so we don't get any more notifications. */
141 mmu_notifier_unregister(&handler->mn, handler->mm);
144 * Make sure the wq delete handler is finished running. It will not
145 * be triggered once the mmu notifiers are unregistered above.
147 flush_work(&handler->del_work);
149 INIT_LIST_HEAD(&del_list);
151 spin_lock_irqsave(&handler->lock, flags);
152 while ((node = rb_first(&handler->root))) {
153 rbnode = rb_entry(node, struct mmu_rb_node, node);
154 rb_erase(node, &handler->root);
155 /* move from LRU list to delete list */
156 list_move(&rbnode->list, &del_list);
158 spin_unlock_irqrestore(&handler->lock, flags);
160 do_remove(handler, &del_list);
165 int hfi1_mmu_rb_insert(struct mmu_rb_handler *handler,
166 struct mmu_rb_node *mnode)
168 struct mmu_rb_node *node;
172 spin_lock_irqsave(&handler->lock, flags);
173 hfi1_cdbg(MMU, "Inserting node addr 0x%llx, len %u", mnode->addr,
175 node = __mmu_rb_search(handler, mnode->addr, mnode->len);
180 __mmu_int_rb_insert(mnode, &handler->root);
181 list_add(&mnode->list, &handler->lru_list);
183 ret = handler->ops->insert(handler->ops_arg, mnode);
185 __mmu_int_rb_remove(mnode, &handler->root);
186 list_del(&mnode->list); /* remove from LRU list */
189 spin_unlock_irqrestore(&handler->lock, flags);
193 /* Caller must hold handler lock */
194 static struct mmu_rb_node *__mmu_rb_search(struct mmu_rb_handler *handler,
198 struct mmu_rb_node *node = NULL;
200 hfi1_cdbg(MMU, "Searching for addr 0x%llx, len %u", addr, len);
201 if (!handler->ops->filter) {
202 node = __mmu_int_rb_iter_first(&handler->root, addr,
205 for (node = __mmu_int_rb_iter_first(&handler->root, addr,
208 node = __mmu_int_rb_iter_next(node, addr,
210 if (handler->ops->filter(node, addr, len))
217 struct mmu_rb_node *hfi1_mmu_rb_extract(struct mmu_rb_handler *handler,
218 unsigned long addr, unsigned long len)
220 struct mmu_rb_node *node;
223 spin_lock_irqsave(&handler->lock, flags);
224 node = __mmu_rb_search(handler, addr, len);
226 __mmu_int_rb_remove(node, &handler->root);
227 list_del(&node->list); /* remove from LRU list */
229 spin_unlock_irqrestore(&handler->lock, flags);
234 void hfi1_mmu_rb_evict(struct mmu_rb_handler *handler, void *evict_arg)
236 struct mmu_rb_node *rbnode, *ptr;
237 struct list_head del_list;
241 INIT_LIST_HEAD(&del_list);
243 spin_lock_irqsave(&handler->lock, flags);
244 list_for_each_entry_safe_reverse(rbnode, ptr, &handler->lru_list,
246 if (handler->ops->evict(handler->ops_arg, rbnode, evict_arg,
248 __mmu_int_rb_remove(rbnode, &handler->root);
249 /* move from LRU list to delete list */
250 list_move(&rbnode->list, &del_list);
255 spin_unlock_irqrestore(&handler->lock, flags);
257 while (!list_empty(&del_list)) {
258 rbnode = list_first_entry(&del_list, struct mmu_rb_node, list);
259 list_del(&rbnode->list);
260 handler->ops->remove(handler->ops_arg, rbnode);
265 * It is up to the caller to ensure that this function does not race with the
266 * mmu invalidate notifier which may be calling the users remove callback on
269 void hfi1_mmu_rb_remove(struct mmu_rb_handler *handler,
270 struct mmu_rb_node *node)
274 /* Validity of handler and node pointers has been checked by caller. */
275 hfi1_cdbg(MMU, "Removing node addr 0x%llx, len %u", node->addr,
277 spin_lock_irqsave(&handler->lock, flags);
278 __mmu_int_rb_remove(node, &handler->root);
279 list_del(&node->list); /* remove from LRU list */
280 spin_unlock_irqrestore(&handler->lock, flags);
282 handler->ops->remove(handler->ops_arg, node);
285 static inline void mmu_notifier_range_start(struct mmu_notifier *mn,
286 struct mm_struct *mm,
290 mmu_notifier_mem_invalidate(mn, mm, start, end);
293 static void mmu_notifier_mem_invalidate(struct mmu_notifier *mn,
294 struct mm_struct *mm,
295 unsigned long start, unsigned long end)
297 struct mmu_rb_handler *handler =
298 container_of(mn, struct mmu_rb_handler, mn);
299 struct rb_root *root = &handler->root;
300 struct mmu_rb_node *node, *ptr = NULL;
304 spin_lock_irqsave(&handler->lock, flags);
305 for (node = __mmu_int_rb_iter_first(root, start, end - 1);
307 /* Guard against node removal. */
308 ptr = __mmu_int_rb_iter_next(node, start, end - 1);
309 hfi1_cdbg(MMU, "Invalidating node addr 0x%llx, len %u",
310 node->addr, node->len);
311 if (handler->ops->invalidate(handler->ops_arg, node)) {
312 __mmu_int_rb_remove(node, root);
313 /* move from LRU list to delete list */
314 list_move(&node->list, &handler->del_list);
318 spin_unlock_irqrestore(&handler->lock, flags);
321 queue_work(handler->wq, &handler->del_work);
325 * Call the remove function for the given handler and the list. This
326 * is expected to be called with a delete list extracted from handler.
327 * The caller should not be holding the handler lock.
329 static void do_remove(struct mmu_rb_handler *handler,
330 struct list_head *del_list)
332 struct mmu_rb_node *node;
334 while (!list_empty(del_list)) {
335 node = list_first_entry(del_list, struct mmu_rb_node, list);
336 list_del(&node->list);
337 handler->ops->remove(handler->ops_arg, node);
342 * Work queue function to remove all nodes that have been queued up to
343 * be removed. The key feature is that mm->mmap_sem is not being held
344 * and the remove callback can sleep while taking it, if needed.
346 static void handle_remove(struct work_struct *work)
348 struct mmu_rb_handler *handler = container_of(work,
349 struct mmu_rb_handler,
351 struct list_head del_list;
354 /* remove anything that is queued to get removed */
355 spin_lock_irqsave(&handler->lock, flags);
356 list_replace_init(&handler->del_list, &del_list);
357 spin_unlock_irqrestore(&handler->lock, flags);
359 do_remove(handler, &del_list);