drivers/infiniband/hw/hfi1/mmu_rb.c

   1 /*
   2  * Copyright(c) 2020 Cornelis Networks, Inc.
   3  * Copyright(c) 2016 - 2017 Intel Corporation.
   4  *
   5  * This file is provided under a dual BSD/GPLv2 license.  When using or
   6  * redistributing this file, you may do so under either license.
   7  *
   8  * GPL LICENSE SUMMARY
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of version 2 of the GNU General Public License as
  12  * published by the Free Software Foundation.
  13  *
  14  * This program is distributed in the hope that it will be useful, but
  15  * WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * General Public License for more details.
  18  *
  19  * BSD LICENSE
  20  *
  21  * Redistribution and use in source and binary forms, with or without
  22  * modification, are permitted provided that the following conditions
  23  * are met:
  24  *
  25  *  - Redistributions of source code must retain the above copyright
  26  *    notice, this list of conditions and the following disclaimer.
  27  *  - Redistributions in binary form must reproduce the above copyright
  28  *    notice, this list of conditions and the following disclaimer in
  29  *    the documentation and/or other materials provided with the
  30  *    distribution.
  31  *  - Neither the name of Intel Corporation nor the names of its
  32  *    contributors may be used to endorse or promote products derived
  33  *    from this software without specific prior written permission.
  34  *
  35  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  36  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  37  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  38  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  39  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  40  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  41  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  42  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  43  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  44  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  45  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  46  *
  47  */
  48 #include <linux/list.h>
  49 #include <linux/rculist.h>
  50 #include <linux/mmu_notifier.h>
  51 #include <linux/interval_tree_generic.h>
  52 #include <linux/sched/mm.h>
  53
  54 #include "mmu_rb.h"
  55 #include "trace.h"
  56
  57 static unsigned long mmu_node_start(struct mmu_rb_node *);
  58 static unsigned long mmu_node_last(struct mmu_rb_node *);
  59 static int mmu_notifier_range_start(struct mmu_notifier *,
  60                 const struct mmu_notifier_range *);
  61 static struct mmu_rb_node *__mmu_rb_search(struct mmu_rb_handler *,
  62                                            unsigned long, unsigned long);
  63 static void do_remove(struct mmu_rb_handler *handler,
  64                       struct list_head *del_list);
  65 static void handle_remove(struct work_struct *work);
  66
  67 static const struct mmu_notifier_ops mn_opts = {
  68         .invalidate_range_start = mmu_notifier_range_start,
  69 };
  70
  71 INTERVAL_TREE_DEFINE(struct mmu_rb_node, node, unsigned long, __last,
  72                      mmu_node_start, mmu_node_last, static, __mmu_int_rb);
  73
  74 static unsigned long mmu_node_start(struct mmu_rb_node *node)
  75 {
  76         return node->addr & PAGE_MASK;
  77 }
  78
  79 static unsigned long mmu_node_last(struct mmu_rb_node *node)
  80 {
  81         return PAGE_ALIGN(node->addr + node->len) - 1;
  82 }
  83
  84 int hfi1_mmu_rb_register(void *ops_arg,
  85                          struct mmu_rb_ops *ops,
  86                          struct workqueue_struct *wq,
  87                          struct mmu_rb_handler **handler)
  88 {
  89         struct mmu_rb_handler *h;
  90         int ret;
  91
  92         h = kmalloc(sizeof(*h), GFP_KERNEL);
  93         if (!h)
  94                 return -ENOMEM;
  95
  96         h->root = RB_ROOT_CACHED;
  97         h->ops = ops;
  98         h->ops_arg = ops_arg;
  99         INIT_HLIST_NODE(&h->mn.hlist);
 100         spin_lock_init(&h->lock);
 101         h->mn.ops = &mn_opts;
 102         INIT_WORK(&h->del_work, handle_remove);
 103         INIT_LIST_HEAD(&h->del_list);
 104         INIT_LIST_HEAD(&h->lru_list);
 105         h->wq = wq;
 106
 107         ret = mmu_notifier_register(&h->mn, current->mm);
 108         if (ret) {
 109                 kfree(h);
 110                 return ret;
 111         }
 112
 113         *handler = h;
 114         return 0;
 115 }
 116
 117 void hfi1_mmu_rb_unregister(struct mmu_rb_handler *handler)
 118 {
 119         struct mmu_rb_node *rbnode;
 120         struct rb_node *node;
 121         unsigned long flags;
 122         struct list_head del_list;
 123
 124         /* Unregister first so we don't get any more notifications. */
 125         mmu_notifier_unregister(&handler->mn, handler->mn.mm);
 126
 127         /*
 128          * Make sure the wq delete handler is finished running.  It will not
 129          * be triggered once the mmu notifiers are unregistered above.
 130          */
 131         flush_work(&handler->del_work);
 132
 133         INIT_LIST_HEAD(&del_list);
 134
 135         spin_lock_irqsave(&handler->lock, flags);
 136         while ((node = rb_first_cached(&handler->root))) {
 137                 rbnode = rb_entry(node, struct mmu_rb_node, node);
 138                 rb_erase_cached(node, &handler->root);
 139                 /* move from LRU list to delete list */
 140                 list_move(&rbnode->list, &del_list);
 141         }
 142         spin_unlock_irqrestore(&handler->lock, flags);
 143
 144         do_remove(handler, &del_list);
 145
 146         kfree(handler);
 147 }
 148
 149 int hfi1_mmu_rb_insert(struct mmu_rb_handler *handler,
 150                        struct mmu_rb_node *mnode)
 151 {
 152         struct mmu_rb_node *node;
 153         unsigned long flags;
 154         int ret = 0;
 155
 156         trace_hfi1_mmu_rb_insert(mnode->addr, mnode->len);
 157
 158         if (current->mm != handler->mn.mm)
 159                 return -EPERM;
 160
 161         spin_lock_irqsave(&handler->lock, flags);
 162         node = __mmu_rb_search(handler, mnode->addr, mnode->len);
 163         if (node) {
 164                 ret = -EINVAL;
 165                 goto unlock;
 166         }
 167         __mmu_int_rb_insert(mnode, &handler->root);
 168         list_add(&mnode->list, &handler->lru_list);
 169
 170         ret = handler->ops->insert(handler->ops_arg, mnode);
 171         if (ret) {
 172                 __mmu_int_rb_remove(mnode, &handler->root);
 173                 list_del(&mnode->list); /* remove from LRU list */
 174         }
 175         mnode->handler = handler;
 176 unlock:
 177         spin_unlock_irqrestore(&handler->lock, flags);
 178         return ret;
 179 }
 180
 181 /* Caller must hold handler lock */
 182 static struct mmu_rb_node *__mmu_rb_search(struct mmu_rb_handler *handler,
 183                                            unsigned long addr,
 184                                            unsigned long len)
 185 {
 186         struct mmu_rb_node *node = NULL;
 187
 188         trace_hfi1_mmu_rb_search(addr, len);
 189         if (!handler->ops->filter) {
 190                 node = __mmu_int_rb_iter_first(&handler->root, addr,
 191                                                (addr + len) - 1);
 192         } else {
 193                 for (node = __mmu_int_rb_iter_first(&handler->root, addr,
 194                                                     (addr + len) - 1);
 195                      node;
 196                      node = __mmu_int_rb_iter_next(node, addr,
 197                                                    (addr + len) - 1)) {
 198                         if (handler->ops->filter(node, addr, len))
 199                                 return node;
 200                 }
 201         }
 202         return node;
 203 }
 204
 205 bool hfi1_mmu_rb_remove_unless_exact(struct mmu_rb_handler *handler,
 206                                      unsigned long addr, unsigned long len,
 207                                      struct mmu_rb_node **rb_node)
 208 {
 209         struct mmu_rb_node *node;
 210         unsigned long flags;
 211         bool ret = false;
 212
 213         if (current->mm != handler->mn.mm)
 214                 return ret;
 215
 216         spin_lock_irqsave(&handler->lock, flags);
 217         node = __mmu_rb_search(handler, addr, len);
 218         if (node) {
 219                 if (node->addr == addr && node->len == len)
 220                         goto unlock;
 221                 __mmu_int_rb_remove(node, &handler->root);
 222                 list_del(&node->list); /* remove from LRU list */
 223                 ret = true;
 224         }
 225 unlock:
 226         spin_unlock_irqrestore(&handler->lock, flags);
 227         *rb_node = node;
 228         return ret;
 229 }
 230
 231 void hfi1_mmu_rb_evict(struct mmu_rb_handler *handler, void *evict_arg)
 232 {
 233         struct mmu_rb_node *rbnode, *ptr;
 234         struct list_head del_list;
 235         unsigned long flags;
 236         bool stop = false;
 237
 238         if (current->mm != handler->mn.mm)
 239                 return;
 240
 241         INIT_LIST_HEAD(&del_list);
 242
 243         spin_lock_irqsave(&handler->lock, flags);
 244         list_for_each_entry_safe_reverse(rbnode, ptr, &handler->lru_list,
 245                                          list) {
 246                 if (handler->ops->evict(handler->ops_arg, rbnode, evict_arg,
 247                                         &stop)) {
 248                         __mmu_int_rb_remove(rbnode, &handler->root);
 249                         /* move from LRU list to delete list */
 250                         list_move(&rbnode->list, &del_list);
 251                 }
 252                 if (stop)
 253                         break;
 254         }
 255         spin_unlock_irqrestore(&handler->lock, flags);
 256
 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);
 261         }
 262 }
 263
 264 /*
 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
 267  * 'node'.
 268  */
 269 void hfi1_mmu_rb_remove(struct mmu_rb_handler *handler,
 270                         struct mmu_rb_node *node)
 271 {
 272         unsigned long flags;
 273
 274         if (current->mm != handler->mn.mm)
 275                 return;
 276
 277         /* Validity of handler and node pointers has been checked by caller. */
 278         trace_hfi1_mmu_rb_remove(node->addr, node->len);
 279         spin_lock_irqsave(&handler->lock, flags);
 280         __mmu_int_rb_remove(node, &handler->root);
 281         list_del(&node->list); /* remove from LRU list */
 282         spin_unlock_irqrestore(&handler->lock, flags);
 283
 284         handler->ops->remove(handler->ops_arg, node);
 285 }
 286
 287 static int mmu_notifier_range_start(struct mmu_notifier *mn,
 288                 const struct mmu_notifier_range *range)
 289 {
 290         struct mmu_rb_handler *handler =
 291                 container_of(mn, struct mmu_rb_handler, mn);
 292         struct rb_root_cached *root = &handler->root;
 293         struct mmu_rb_node *node, *ptr = NULL;
 294         unsigned long flags;
 295         bool added = false;
 296
 297         spin_lock_irqsave(&handler->lock, flags);
 298         for (node = __mmu_int_rb_iter_first(root, range->start, range->end-1);
 299              node; node = ptr) {
 300                 /* Guard against node removal. */
 301                 ptr = __mmu_int_rb_iter_next(node, range->start,
 302                                              range->end - 1);
 303                 trace_hfi1_mmu_mem_invalidate(node->addr, node->len);
 304                 if (handler->ops->invalidate(handler->ops_arg, node)) {
 305                         __mmu_int_rb_remove(node, root);
 306                         /* move from LRU list to delete list */
 307                         list_move(&node->list, &handler->del_list);
 308                         added = true;
 309                 }
 310         }
 311         spin_unlock_irqrestore(&handler->lock, flags);
 312
 313         if (added)
 314                 queue_work(handler->wq, &handler->del_work);
 315
 316         return 0;
 317 }
 318
 319 /*
 320  * Call the remove function for the given handler and the list.  This
 321  * is expected to be called with a delete list extracted from handler.
 322  * The caller should not be holding the handler lock.
 323  */
 324 static void do_remove(struct mmu_rb_handler *handler,
 325                       struct list_head *del_list)
 326 {
 327         struct mmu_rb_node *node;
 328
 329         while (!list_empty(del_list)) {
 330                 node = list_first_entry(del_list, struct mmu_rb_node, list);
 331                 list_del(&node->list);
 332                 handler->ops->remove(handler->ops_arg, node);
 333         }
 334 }
 335
 336 /*
 337  * Work queue function to remove all nodes that have been queued up to
 338  * be removed.  The key feature is that mm->mmap_lock is not being held
 339  * and the remove callback can sleep while taking it, if needed.
 340  */
 341 static void handle_remove(struct work_struct *work)
 342 {
 343         struct mmu_rb_handler *handler = container_of(work,
 344                                                 struct mmu_rb_handler,
 345                                                 del_work);
 346         struct list_head del_list;
 347         unsigned long flags;
 348
 349         /* remove anything that is queued to get removed */
 350         spin_lock_irqsave(&handler->lock, flags);
 351         list_replace_init(&handler->del_list, &del_list);
 352         spin_unlock_irqrestore(&handler->lock, flags);
 353
 354         do_remove(handler, &del_list);
 355 }