drivers/infiniband/hw/hfi1/mmu_rb.c

   1 /*
   2  * Copyright(c) 2016 - 2017 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_cached 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 int mmu_notifier_range_start(struct mmu_notifier *,
  71                 const struct mmu_notifier_range *);
  72 static struct mmu_rb_node *__mmu_rb_search(struct mmu_rb_handler *,
  73                                            unsigned long, unsigned long);
  74 static void do_remove(struct mmu_rb_handler *handler,
  75                       struct list_head *del_list);
  76 static void handle_remove(struct work_struct *work);
  77
  78 static const struct mmu_notifier_ops mn_opts = {
  79         .invalidate_range_start = mmu_notifier_range_start,
  80 };
  81
  82 INTERVAL_TREE_DEFINE(struct mmu_rb_node, node, unsigned long, __last,
  83                      mmu_node_start, mmu_node_last, static, __mmu_int_rb);
  84
  85 static unsigned long mmu_node_start(struct mmu_rb_node *node)
  86 {
  87         return node->addr & PAGE_MASK;
  88 }
  89
  90 static unsigned long mmu_node_last(struct mmu_rb_node *node)
  91 {
  92         return PAGE_ALIGN(node->addr + node->len) - 1;
  93 }
  94
  95 int hfi1_mmu_rb_register(void *ops_arg, struct mm_struct *mm,
  96                          struct mmu_rb_ops *ops,
  97                          struct workqueue_struct *wq,
  98                          struct mmu_rb_handler **handler)
  99 {
 100         struct mmu_rb_handler *handlr;
 101         int ret;
 102
 103         handlr = kmalloc(sizeof(*handlr), GFP_KERNEL);
 104         if (!handlr)
 105                 return -ENOMEM;
 106
 107         handlr->root = RB_ROOT_CACHED;
 108         handlr->ops = ops;
 109         handlr->ops_arg = ops_arg;
 110         INIT_HLIST_NODE(&handlr->mn.hlist);
 111         spin_lock_init(&handlr->lock);
 112         handlr->mn.ops = &mn_opts;
 113         handlr->mm = mm;
 114         INIT_WORK(&handlr->del_work, handle_remove);
 115         INIT_LIST_HEAD(&handlr->del_list);
 116         INIT_LIST_HEAD(&handlr->lru_list);
 117         handlr->wq = wq;
 118
 119         ret = mmu_notifier_register(&handlr->mn, handlr->mm);
 120         if (ret) {
 121                 kfree(handlr);
 122                 return ret;
 123         }
 124
 125         *handler = handlr;
 126         return 0;
 127 }
 128
 129 void hfi1_mmu_rb_unregister(struct mmu_rb_handler *handler)
 130 {
 131         struct mmu_rb_node *rbnode;
 132         struct rb_node *node;
 133         unsigned long flags;
 134         struct list_head del_list;
 135
 136         /* Unregister first so we don't get any more notifications. */
 137         mmu_notifier_unregister(&handler->mn, handler->mm);
 138
 139         /*
 140          * Make sure the wq delete handler is finished running.  It will not
 141          * be triggered once the mmu notifiers are unregistered above.
 142          */
 143         flush_work(&handler->del_work);
 144
 145         INIT_LIST_HEAD(&del_list);
 146
 147         spin_lock_irqsave(&handler->lock, flags);
 148         while ((node = rb_first_cached(&handler->root))) {
 149                 rbnode = rb_entry(node, struct mmu_rb_node, node);
 150                 rb_erase_cached(node, &handler->root);
 151                 /* move from LRU list to delete list */
 152                 list_move(&rbnode->list, &del_list);
 153         }
 154         spin_unlock_irqrestore(&handler->lock, flags);
 155
 156         do_remove(handler, &del_list);
 157
 158         kfree(handler);
 159 }
 160
 161 int hfi1_mmu_rb_insert(struct mmu_rb_handler *handler,
 162                        struct mmu_rb_node *mnode)
 163 {
 164         struct mmu_rb_node *node;
 165         unsigned long flags;
 166         int ret = 0;
 167
 168         trace_hfi1_mmu_rb_insert(mnode->addr, mnode->len);
 169         spin_lock_irqsave(&handler->lock, flags);
 170         node = __mmu_rb_search(handler, mnode->addr, mnode->len);
 171         if (node) {
 172                 ret = -EINVAL;
 173                 goto unlock;
 174         }
 175         __mmu_int_rb_insert(mnode, &handler->root);
 176         list_add(&mnode->list, &handler->lru_list);
 177
 178         ret = handler->ops->insert(handler->ops_arg, mnode);
 179         if (ret) {
 180                 __mmu_int_rb_remove(mnode, &handler->root);
 181                 list_del(&mnode->list); /* remove from LRU list */
 182         }
 183 unlock:
 184         spin_unlock_irqrestore(&handler->lock, flags);
 185         return ret;
 186 }
 187
 188 /* Caller must hold handler lock */
 189 static struct mmu_rb_node *__mmu_rb_search(struct mmu_rb_handler *handler,
 190                                            unsigned long addr,
 191                                            unsigned long len)
 192 {
 193         struct mmu_rb_node *node = NULL;
 194
 195         trace_hfi1_mmu_rb_search(addr, len);
 196         if (!handler->ops->filter) {
 197                 node = __mmu_int_rb_iter_first(&handler->root, addr,
 198                                                (addr + len) - 1);
 199         } else {
 200                 for (node = __mmu_int_rb_iter_first(&handler->root, addr,
 201                                                     (addr + len) - 1);
 202                      node;
 203                      node = __mmu_int_rb_iter_next(node, addr,
 204                                                    (addr + len) - 1)) {
 205                         if (handler->ops->filter(node, addr, len))
 206                                 return node;
 207                 }
 208         }
 209         return node;
 210 }
 211
 212 bool hfi1_mmu_rb_remove_unless_exact(struct mmu_rb_handler *handler,
 213                                      unsigned long addr, unsigned long len,
 214                                      struct mmu_rb_node **rb_node)
 215 {
 216         struct mmu_rb_node *node;
 217         unsigned long flags;
 218         bool ret = false;
 219
 220         spin_lock_irqsave(&handler->lock, flags);
 221         node = __mmu_rb_search(handler, addr, len);
 222         if (node) {
 223                 if (node->addr == addr && node->len == len)
 224                         goto unlock;
 225                 __mmu_int_rb_remove(node, &handler->root);
 226                 list_del(&node->list); /* remove from LRU list */
 227                 ret = true;
 228         }
 229 unlock:
 230         spin_unlock_irqrestore(&handler->lock, flags);
 231         *rb_node = node;
 232         return ret;
 233 }
 234
 235 void hfi1_mmu_rb_evict(struct mmu_rb_handler *handler, void *evict_arg)
 236 {
 237         struct mmu_rb_node *rbnode, *ptr;
 238         struct list_head del_list;
 239         unsigned long flags;
 240         bool stop = false;
 241
 242         INIT_LIST_HEAD(&del_list);
 243
 244         spin_lock_irqsave(&handler->lock, flags);
 245         list_for_each_entry_safe_reverse(rbnode, ptr, &handler->lru_list,
 246                                          list) {
 247                 if (handler->ops->evict(handler->ops_arg, rbnode, evict_arg,
 248                                         &stop)) {
 249                         __mmu_int_rb_remove(rbnode, &handler->root);
 250                         /* move from LRU list to delete list */
 251                         list_move(&rbnode->list, &del_list);
 252                 }
 253                 if (stop)
 254                         break;
 255         }
 256         spin_unlock_irqrestore(&handler->lock, flags);
 257
 258         while (!list_empty(&del_list)) {
 259                 rbnode = list_first_entry(&del_list, struct mmu_rb_node, list);
 260                 list_del(&rbnode->list);
 261                 handler->ops->remove(handler->ops_arg, rbnode);
 262         }
 263 }
 264
 265 /*
 266  * It is up to the caller to ensure that this function does not race with the
 267  * mmu invalidate notifier which may be calling the users remove callback on
 268  * 'node'.
 269  */
 270 void hfi1_mmu_rb_remove(struct mmu_rb_handler *handler,
 271                         struct mmu_rb_node *node)
 272 {
 273         unsigned long flags;
 274
 275         /* Validity of handler and node pointers has been checked by caller. */
 276         trace_hfi1_mmu_rb_remove(node->addr, node->len);
 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);
 281
 282         handler->ops->remove(handler->ops_arg, node);
 283 }
 284
 285 static int mmu_notifier_range_start(struct mmu_notifier *mn,
 286                 const struct mmu_notifier_range *range)
 287 {
 288         struct mmu_rb_handler *handler =
 289                 container_of(mn, struct mmu_rb_handler, mn);
 290         struct rb_root_cached *root = &handler->root;
 291         struct mmu_rb_node *node, *ptr = NULL;
 292         unsigned long flags;
 293         bool added = false;
 294
 295         spin_lock_irqsave(&handler->lock, flags);
 296         for (node = __mmu_int_rb_iter_first(root, range->start, range->end-1);
 297              node; node = ptr) {
 298                 /* Guard against node removal. */
 299                 ptr = __mmu_int_rb_iter_next(node, range->start,
 300                                              range->end - 1);
 301                 trace_hfi1_mmu_mem_invalidate(node->addr, node->len);
 302                 if (handler->ops->invalidate(handler->ops_arg, node)) {
 303                         __mmu_int_rb_remove(node, root);
 304                         /* move from LRU list to delete list */
 305                         list_move(&node->list, &handler->del_list);
 306                         added = true;
 307                 }
 308         }
 309         spin_unlock_irqrestore(&handler->lock, flags);
 310
 311         if (added)
 312                 queue_work(handler->wq, &handler->del_work);
 313
 314         return 0;
 315 }
 316
 317 /*
 318  * Call the remove function for the given handler and the list.  This
 319  * is expected to be called with a delete list extracted from handler.
 320  * The caller should not be holding the handler lock.
 321  */
 322 static void do_remove(struct mmu_rb_handler *handler,
 323                       struct list_head *del_list)
 324 {
 325         struct mmu_rb_node *node;
 326
 327         while (!list_empty(del_list)) {
 328                 node = list_first_entry(del_list, struct mmu_rb_node, list);
 329                 list_del(&node->list);
 330                 handler->ops->remove(handler->ops_arg, node);
 331         }
 332 }
 333
 334 /*
 335  * Work queue function to remove all nodes that have been queued up to
 336  * be removed.  The key feature is that mm->mmap_lock is not being held
 337  * and the remove callback can sleep while taking it, if needed.
 338  */
 339 static void handle_remove(struct work_struct *work)
 340 {
 341         struct mmu_rb_handler *handler = container_of(work,
 342                                                 struct mmu_rb_handler,
 343                                                 del_work);
 344         struct list_head del_list;
 345         unsigned long flags;
 346
 347         /* remove anything that is queued to get removed */
 348         spin_lock_irqsave(&handler->lock, flags);
 349         list_replace_init(&handler->del_list, &del_list);
 350         spin_unlock_irqrestore(&handler->lock, flags);
 351
 352         do_remove(handler, &del_list);
 353 }