drivers/infiniband/hw/hfi1/mmu_rb.c

   1 /*
   2  * Copyright(c) 2016 Intel Corporation.
   3  *
   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.
   6  *
   7  * GPL LICENSE SUMMARY
   8  *
   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.
  12  *
  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.
  17  *
  18  * BSD LICENSE
  19  *
  20  * Redistribution and use in source and binary forms, with or without
  21  * modification, are permitted provided that the following conditions
  22  * are met:
  23  *
  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
  29  *    distribution.
  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.
  33  *
  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.
  45  *
  46  */
  47 #include <linux/list.h>
  48 #include <linux/rculist.h>
  49 #include <linux/mmu_notifier.h>
  50 #include <linux/interval_tree_generic.h>
  51
  52 #include "mmu_rb.h"
  53 #include "trace.h"
  54
  55 struct mmu_rb_handler {
  56         struct mmu_notifier mn;
  57         struct rb_root root;
  58         void *ops_arg;
  59         spinlock_t lock;        /* protect the RB tree */
  60         struct mmu_rb_ops *ops;
  61         struct mm_struct *mm;
  62         struct list_head lru_list;
  63         struct work_struct del_work;
  64         struct list_head del_list;
  65         struct workqueue_struct *wq;
  66 };
  67
  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_page(struct mmu_notifier *, struct mm_struct *,
  71                                      unsigned long);
  72 static inline void mmu_notifier_range_start(struct mmu_notifier *,
  73                                             struct mm_struct *,
  74                                             unsigned long, unsigned long);
  75 static void mmu_notifier_mem_invalidate(struct mmu_notifier *,
  76                                         struct mm_struct *,
  77                                         unsigned long, unsigned long);
  78 static struct mmu_rb_node *__mmu_rb_search(struct mmu_rb_handler *,
  79                                            unsigned long, unsigned long);
  80 static void do_remove(struct mmu_rb_handler *handler,
  81                       struct list_head *del_list);
  82 static void handle_remove(struct work_struct *work);
  83
  84 static const struct mmu_notifier_ops mn_opts = {
  85         .invalidate_page = mmu_notifier_page,
  86         .invalidate_range_start = mmu_notifier_range_start,
  87 };
  88
  89 INTERVAL_TREE_DEFINE(struct mmu_rb_node, node, unsigned long, __last,
  90                      mmu_node_start, mmu_node_last, static, __mmu_int_rb);
  91
  92 static unsigned long mmu_node_start(struct mmu_rb_node *node)
  93 {
  94         return node->addr & PAGE_MASK;
  95 }
  96
  97 static unsigned long mmu_node_last(struct mmu_rb_node *node)
  98 {
  99         return PAGE_ALIGN(node->addr + node->len) - 1;
 100 }
 101
 102 int hfi1_mmu_rb_register(void *ops_arg, struct mm_struct *mm,
 103                          struct mmu_rb_ops *ops,
 104                          struct workqueue_struct *wq,
 105                          struct mmu_rb_handler **handler)
 106 {
 107         struct mmu_rb_handler *handlr;
 108         int ret;
 109
 110         handlr = kmalloc(sizeof(*handlr), GFP_KERNEL);
 111         if (!handlr)
 112                 return -ENOMEM;
 113
 114         handlr->root = RB_ROOT;
 115         handlr->ops = ops;
 116         handlr->ops_arg = ops_arg;
 117         INIT_HLIST_NODE(&handlr->mn.hlist);
 118         spin_lock_init(&handlr->lock);
 119         handlr->mn.ops = &mn_opts;
 120         handlr->mm = mm;
 121         INIT_WORK(&handlr->del_work, handle_remove);
 122         INIT_LIST_HEAD(&handlr->del_list);
 123         INIT_LIST_HEAD(&handlr->lru_list);
 124         handlr->wq = wq;
 125
 126         ret = mmu_notifier_register(&handlr->mn, handlr->mm);
 127         if (ret) {
 128                 kfree(handlr);
 129                 return ret;
 130         }
 131
 132         *handler = handlr;
 133         return 0;
 134 }
 135
 136 void hfi1_mmu_rb_unregister(struct mmu_rb_handler *handler)
 137 {
 138         struct mmu_rb_node *rbnode;
 139         struct rb_node *node;
 140         unsigned long flags;
 141         struct list_head del_list;
 142
 143         /* Unregister first so we don't get any more notifications. */
 144         mmu_notifier_unregister(&handler->mn, handler->mm);
 145
 146         /*
 147          * Make sure the wq delete handler is finished running.  It will not
 148          * be triggered once the mmu notifiers are unregistered above.
 149          */
 150         flush_work(&handler->del_work);
 151
 152         INIT_LIST_HEAD(&del_list);
 153
 154         spin_lock_irqsave(&handler->lock, flags);
 155         while ((node = rb_first(&handler->root))) {
 156                 rbnode = rb_entry(node, struct mmu_rb_node, node);
 157                 rb_erase(node, &handler->root);
 158                 /* move from LRU list to delete list */
 159                 list_move(&rbnode->list, &del_list);
 160         }
 161         spin_unlock_irqrestore(&handler->lock, flags);
 162
 163         do_remove(handler, &del_list);
 164
 165         kfree(handler);
 166 }
 167
 168 int hfi1_mmu_rb_insert(struct mmu_rb_handler *handler,
 169                        struct mmu_rb_node *mnode)
 170 {
 171         struct mmu_rb_node *node;
 172         unsigned long flags;
 173         int ret = 0;
 174
 175         spin_lock_irqsave(&handler->lock, flags);
 176         hfi1_cdbg(MMU, "Inserting node addr 0x%llx, len %u", mnode->addr,
 177                   mnode->len);
 178         node = __mmu_rb_search(handler, mnode->addr, mnode->len);
 179         if (node) {
 180                 ret = -EINVAL;
 181                 goto unlock;
 182         }
 183         __mmu_int_rb_insert(mnode, &handler->root);
 184         list_add(&mnode->list, &handler->lru_list);
 185
 186         ret = handler->ops->insert(handler->ops_arg, mnode);
 187         if (ret) {
 188                 __mmu_int_rb_remove(mnode, &handler->root);
 189                 list_del(&mnode->list); /* remove from LRU list */
 190         }
 191 unlock:
 192         spin_unlock_irqrestore(&handler->lock, flags);
 193         return ret;
 194 }
 195
 196 /* Caller must hold handler lock */
 197 static struct mmu_rb_node *__mmu_rb_search(struct mmu_rb_handler *handler,
 198                                            unsigned long addr,
 199                                            unsigned long len)
 200 {
 201         struct mmu_rb_node *node = NULL;
 202
 203         hfi1_cdbg(MMU, "Searching for addr 0x%llx, len %u", addr, len);
 204         if (!handler->ops->filter) {
 205                 node = __mmu_int_rb_iter_first(&handler->root, addr,
 206                                                (addr + len) - 1);
 207         } else {
 208                 for (node = __mmu_int_rb_iter_first(&handler->root, addr,
 209                                                     (addr + len) - 1);
 210                      node;
 211                      node = __mmu_int_rb_iter_next(node, addr,
 212                                                    (addr + len) - 1)) {
 213                         if (handler->ops->filter(node, addr, len))
 214                                 return node;
 215                 }
 216         }
 217         return node;
 218 }
 219
 220 struct mmu_rb_node *hfi1_mmu_rb_extract(struct mmu_rb_handler *handler,
 221                                         unsigned long addr, unsigned long len)
 222 {
 223         struct mmu_rb_node *node;
 224         unsigned long flags;
 225
 226         spin_lock_irqsave(&handler->lock, flags);
 227         node = __mmu_rb_search(handler, addr, len);
 228         if (node) {
 229                 __mmu_int_rb_remove(node, &handler->root);
 230                 list_del(&node->list); /* remove from LRU list */
 231         }
 232         spin_unlock_irqrestore(&handler->lock, flags);
 233
 234         return node;
 235 }
 236
 237 void hfi1_mmu_rb_evict(struct mmu_rb_handler *handler, void *evict_arg)
 238 {
 239         struct mmu_rb_node *rbnode, *ptr;
 240         struct list_head del_list;
 241         unsigned long flags;
 242         bool stop = false;
 243
 244         INIT_LIST_HEAD(&del_list);
 245
 246         spin_lock_irqsave(&handler->lock, flags);
 247         list_for_each_entry_safe_reverse(rbnode, ptr, &handler->lru_list,
 248                                          list) {
 249                 if (handler->ops->evict(handler->ops_arg, rbnode, evict_arg,
 250                                         &stop)) {
 251                         __mmu_int_rb_remove(rbnode, &handler->root);
 252                         /* move from LRU list to delete list */
 253                         list_move(&rbnode->list, &del_list);
 254                 }
 255                 if (stop)
 256                         break;
 257         }
 258         spin_unlock_irqrestore(&handler->lock, flags);
 259
 260         while (!list_empty(&del_list)) {
 261                 rbnode = list_first_entry(&del_list, struct mmu_rb_node, list);
 262                 list_del(&rbnode->list);
 263                 handler->ops->remove(handler->ops_arg, rbnode);
 264         }
 265 }
 266
 267 /*
 268  * It is up to the caller to ensure that this function does not race with the
 269  * mmu invalidate notifier which may be calling the users remove callback on
 270  * 'node'.
 271  */
 272 void hfi1_mmu_rb_remove(struct mmu_rb_handler *handler,
 273                         struct mmu_rb_node *node)
 274 {
 275         unsigned long flags;
 276
 277         /* Validity of handler and node pointers has been checked by caller. */
 278         hfi1_cdbg(MMU, "Removing node addr 0x%llx, len %u", node->addr,
 279                   node->len);
 280         spin_lock_irqsave(&handler->lock, flags);
 281         __mmu_int_rb_remove(node, &handler->root);
 282         list_del(&node->list); /* remove from LRU list */
 283         spin_unlock_irqrestore(&handler->lock, flags);
 284
 285         handler->ops->remove(handler->ops_arg, node);
 286 }
 287
 288 static inline void mmu_notifier_page(struct mmu_notifier *mn,
 289                                      struct mm_struct *mm, unsigned long addr)
 290 {
 291         mmu_notifier_mem_invalidate(mn, mm, addr, addr + PAGE_SIZE);
 292 }
 293
 294 static inline void mmu_notifier_range_start(struct mmu_notifier *mn,
 295                                             struct mm_struct *mm,
 296                                             unsigned long start,
 297                                             unsigned long end)
 298 {
 299         mmu_notifier_mem_invalidate(mn, mm, start, end);
 300 }
 301
 302 static void mmu_notifier_mem_invalidate(struct mmu_notifier *mn,
 303                                         struct mm_struct *mm,
 304                                         unsigned long start, unsigned long end)
 305 {
 306         struct mmu_rb_handler *handler =
 307                 container_of(mn, struct mmu_rb_handler, mn);
 308         struct rb_root *root = &handler->root;
 309         struct mmu_rb_node *node, *ptr = NULL;
 310         unsigned long flags;
 311         bool added = false;
 312
 313         spin_lock_irqsave(&handler->lock, flags);
 314         for (node = __mmu_int_rb_iter_first(root, start, end - 1);
 315              node; node = ptr) {
 316                 /* Guard against node removal. */
 317                 ptr = __mmu_int_rb_iter_next(node, start, end - 1);
 318                 hfi1_cdbg(MMU, "Invalidating node addr 0x%llx, len %u",
 319                           node->addr, node->len);
 320                 if (handler->ops->invalidate(handler->ops_arg, node)) {
 321                         __mmu_int_rb_remove(node, root);
 322                         /* move from LRU list to delete list */
 323                         list_move(&node->list, &handler->del_list);
 324                         added = true;
 325                 }
 326         }
 327         spin_unlock_irqrestore(&handler->lock, flags);
 328
 329         if (added)
 330                 queue_work(handler->wq, &handler->del_work);
 331 }
 332
 333 /*
 334  * Call the remove function for the given handler and the list.  This
 335  * is expected to be called with a delete list extracted from handler.
 336  * The caller should not be holding the handler lock.
 337  */
 338 static void do_remove(struct mmu_rb_handler *handler,
 339                       struct list_head *del_list)
 340 {
 341         struct mmu_rb_node *node;
 342
 343         while (!list_empty(del_list)) {
 344                 node = list_first_entry(del_list, struct mmu_rb_node, list);
 345                 list_del(&node->list);
 346                 handler->ops->remove(handler->ops_arg, node);
 347         }
 348 }
 349
 350 /*
 351  * Work queue function to remove all nodes that have been queued up to
 352  * be removed.  The key feature is that mm->mmap_sem is not being held
 353  * and the remove callback can sleep while taking it, if needed.
 354  */
 355 static void handle_remove(struct work_struct *work)
 356 {
 357         struct mmu_rb_handler *handler = container_of(work,
 358                                                 struct mmu_rb_handler,
 359                                                 del_work);
 360         struct list_head del_list;
 361         unsigned long flags;
 362
 363         /* remove anything that is queued to get removed */
 364         spin_lock_irqsave(&handler->lock, flags);
 365         list_replace_init(&handler->del_list, &del_list);
 366         spin_unlock_irqrestore(&handler->lock, flags);
 367
 368         do_remove(handler, &del_list);
 369 }