drivers/pci/iova.c

   1 /*
   2  * Copyright © 2006-2009, Intel Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
  18  */
  19
  20 #include <linux/iova.h>
  21
  22 void
  23 init_iova_domain(struct iova_domain *iovad, unsigned long pfn_32bit)
  24 {
  25         spin_lock_init(&iovad->iova_alloc_lock);
  26         spin_lock_init(&iovad->iova_rbtree_lock);
  27         iovad->rbroot = RB_ROOT;
  28         iovad->cached32_node = NULL;
  29         iovad->dma_32bit_pfn = pfn_32bit;
  30 }
  31
  32 static struct rb_node *
  33 __get_cached_rbnode(struct iova_domain *iovad, unsigned long *limit_pfn)
  34 {
  35         if ((*limit_pfn != iovad->dma_32bit_pfn) ||
  36                 (iovad->cached32_node == NULL))
  37                 return rb_last(&iovad->rbroot);
  38         else {
  39                 struct rb_node *prev_node = rb_prev(iovad->cached32_node);
  40                 struct iova *curr_iova =
  41                         container_of(iovad->cached32_node, struct iova, node);
  42                 *limit_pfn = curr_iova->pfn_lo - 1;
  43                 return prev_node;
  44         }
  45 }
  46
  47 static void
  48 __cached_rbnode_insert_update(struct iova_domain *iovad,
  49         unsigned long limit_pfn, struct iova *new)
  50 {
  51         if (limit_pfn != iovad->dma_32bit_pfn)
  52                 return;
  53         iovad->cached32_node = &new->node;
  54 }
  55
  56 static void
  57 __cached_rbnode_delete_update(struct iova_domain *iovad, struct iova *free)
  58 {
  59         struct iova *cached_iova;
  60         struct rb_node *curr;
  61
  62         if (!iovad->cached32_node)
  63                 return;
  64         curr = iovad->cached32_node;
  65         cached_iova = container_of(curr, struct iova, node);
  66
  67         if (free->pfn_lo >= cached_iova->pfn_lo)
  68                 iovad->cached32_node = rb_next(&free->node);
  69 }
  70
  71 /* Computes the padding size required, to make the
  72  * the start address naturally aligned on its size
  73  */
  74 static int
  75 iova_get_pad_size(int size, unsigned int limit_pfn)
  76 {
  77         unsigned int pad_size = 0;
  78         unsigned int order = ilog2(size);
  79
  80         if (order)
  81                 pad_size = (limit_pfn + 1) % (1 << order);
  82
  83         return pad_size;
  84 }
  85
  86 static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
  87                 unsigned long size, unsigned long limit_pfn,
  88                         struct iova *new, bool size_aligned)
  89 {
  90         struct rb_node *prev, *curr = NULL;
  91         unsigned long flags;
  92         unsigned long saved_pfn;
  93         unsigned int pad_size = 0;
  94
  95         /* Walk the tree backwards */
  96         spin_lock_irqsave(&iovad->iova_rbtree_lock, flags);
  97         saved_pfn = limit_pfn;
  98         curr = __get_cached_rbnode(iovad, &limit_pfn);
  99         prev = curr;
 100         while (curr) {
 101                 struct iova *curr_iova = container_of(curr, struct iova, node);
 102
 103                 if (limit_pfn < curr_iova->pfn_lo)
 104                         goto move_left;
 105                 else if (limit_pfn < curr_iova->pfn_hi)
 106                         goto adjust_limit_pfn;
 107                 else {
 108                         if (size_aligned)
 109                                 pad_size = iova_get_pad_size(size, limit_pfn);
 110                         if ((curr_iova->pfn_hi + size + pad_size) <= limit_pfn)
 111                                 break;  /* found a free slot */
 112                 }
 113 adjust_limit_pfn:
 114                 limit_pfn = curr_iova->pfn_lo - 1;
 115 move_left:
 116                 prev = curr;
 117                 curr = rb_prev(curr);
 118         }
 119
 120         if (!curr) {
 121                 if (size_aligned)
 122                         pad_size = iova_get_pad_size(size, limit_pfn);
 123                 if ((IOVA_START_PFN + size + pad_size) > limit_pfn) {
 124                         spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
 125                         return -ENOMEM;
 126                 }
 127         }
 128
 129         /* pfn_lo will point to size aligned address if size_aligned is set */
 130         new->pfn_lo = limit_pfn - (size + pad_size) + 1;
 131         new->pfn_hi = new->pfn_lo + size - 1;
 132
 133         /* Insert the new_iova into domain rbtree by holding writer lock */
 134         /* Add new node and rebalance tree. */
 135         {
 136                 struct rb_node **entry, *parent = NULL;
 137
 138                 /* If we have 'prev', it's a valid place to start the
 139                    insertion. Otherwise, start from the root. */
 140                 if (prev)
 141                         entry = &prev;
 142                 else
 143                         entry = &iovad->rbroot.rb_node;
 144
 145                 /* Figure out where to put new node */
 146                 while (*entry) {
 147                         struct iova *this = container_of(*entry,
 148                                                         struct iova, node);
 149                         parent = *entry;
 150
 151                         if (new->pfn_lo < this->pfn_lo)
 152                                 entry = &((*entry)->rb_left);
 153                         else if (new->pfn_lo > this->pfn_lo)
 154                                 entry = &((*entry)->rb_right);
 155                         else
 156                                 BUG(); /* this should not happen */
 157                 }
 158
 159                 /* Add new node and rebalance tree. */
 160                 rb_link_node(&new->node, parent, entry);
 161                 rb_insert_color(&new->node, &iovad->rbroot);
 162         }
 163         __cached_rbnode_insert_update(iovad, saved_pfn, new);
 164
 165         spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
 166
 167
 168         return 0;
 169 }
 170
 171 static void
 172 iova_insert_rbtree(struct rb_root *root, struct iova *iova)
 173 {
 174         struct rb_node **new = &(root->rb_node), *parent = NULL;
 175         /* Figure out where to put new node */
 176         while (*new) {
 177                 struct iova *this = container_of(*new, struct iova, node);
 178                 parent = *new;
 179
 180                 if (iova->pfn_lo < this->pfn_lo)
 181                         new = &((*new)->rb_left);
 182                 else if (iova->pfn_lo > this->pfn_lo)
 183                         new = &((*new)->rb_right);
 184                 else
 185                         BUG(); /* this should not happen */
 186         }
 187         /* Add new node and rebalance tree. */
 188         rb_link_node(&iova->node, parent, new);
 189         rb_insert_color(&iova->node, root);
 190 }
 191
 192 /**
 193  * alloc_iova - allocates an iova
 194  * @iovad - iova domain in question
 195  * @size - size of page frames to allocate
 196  * @limit_pfn - max limit address
 197  * @size_aligned - set if size_aligned address range is required
 198  * This function allocates an iova in the range limit_pfn to IOVA_START_PFN
 199  * looking from limit_pfn instead from IOVA_START_PFN. If the size_aligned
 200  * flag is set then the allocated address iova->pfn_lo will be naturally
 201  * aligned on roundup_power_of_two(size).
 202  */
 203 struct iova *
 204 alloc_iova(struct iova_domain *iovad, unsigned long size,
 205         unsigned long limit_pfn,
 206         bool size_aligned)
 207 {
 208         unsigned long flags;
 209         struct iova *new_iova;
 210         int ret;
 211
 212         new_iova = alloc_iova_mem();
 213         if (!new_iova)
 214                 return NULL;
 215
 216         /* If size aligned is set then round the size to
 217          * to next power of two.
 218          */
 219         if (size_aligned)
 220                 size = __roundup_pow_of_two(size);
 221
 222         spin_lock_irqsave(&iovad->iova_alloc_lock, flags);
 223         ret = __alloc_and_insert_iova_range(iovad, size, limit_pfn,
 224                         new_iova, size_aligned);
 225
 226         spin_unlock_irqrestore(&iovad->iova_alloc_lock, flags);
 227         if (ret) {
 228                 free_iova_mem(new_iova);
 229                 return NULL;
 230         }
 231
 232         return new_iova;
 233 }
 234
 235 /**
 236  * find_iova - find's an iova for a given pfn
 237  * @iovad - iova domain in question.
 238  * pfn - page frame number
 239  * This function finds and returns an iova belonging to the
 240  * given doamin which matches the given pfn.
 241  */
 242 struct iova *find_iova(struct iova_domain *iovad, unsigned long pfn)
 243 {
 244         unsigned long flags;
 245         struct rb_node *node;
 246
 247         /* Take the lock so that no other thread is manipulating the rbtree */
 248         spin_lock_irqsave(&iovad->iova_rbtree_lock, flags);
 249         node = iovad->rbroot.rb_node;
 250         while (node) {
 251                 struct iova *iova = container_of(node, struct iova, node);
 252
 253                 /* If pfn falls within iova's range, return iova */
 254                 if ((pfn >= iova->pfn_lo) && (pfn <= iova->pfn_hi)) {
 255                         spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
 256                         /* We are not holding the lock while this iova
 257                          * is referenced by the caller as the same thread
 258                          * which called this function also calls __free_iova()
 259                          * and it is by desing that only one thread can possibly
 260                          * reference a particular iova and hence no conflict.
 261                          */
 262                         return iova;
 263                 }
 264
 265                 if (pfn < iova->pfn_lo)
 266                         node = node->rb_left;
 267                 else if (pfn > iova->pfn_lo)
 268                         node = node->rb_right;
 269         }
 270
 271         spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
 272         return NULL;
 273 }
 274
 275 /**
 276  * __free_iova - frees the given iova
 277  * @iovad: iova domain in question.
 278  * @iova: iova in question.
 279  * Frees the given iova belonging to the giving domain
 280  */
 281 void
 282 __free_iova(struct iova_domain *iovad, struct iova *iova)
 283 {
 284         unsigned long flags;
 285
 286         spin_lock_irqsave(&iovad->iova_rbtree_lock, flags);
 287         __cached_rbnode_delete_update(iovad, iova);
 288         rb_erase(&iova->node, &iovad->rbroot);
 289         spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
 290         free_iova_mem(iova);
 291 }
 292
 293 /**
 294  * free_iova - finds and frees the iova for a given pfn
 295  * @iovad: - iova domain in question.
 296  * @pfn: - pfn that is allocated previously
 297  * This functions finds an iova for a given pfn and then
 298  * frees the iova from that domain.
 299  */
 300 void
 301 free_iova(struct iova_domain *iovad, unsigned long pfn)
 302 {
 303         struct iova *iova = find_iova(iovad, pfn);
 304         if (iova)
 305                 __free_iova(iovad, iova);
 306
 307 }
 308
 309 /**
 310  * put_iova_domain - destroys the iova doamin
 311  * @iovad: - iova domain in question.
 312  * All the iova's in that domain are destroyed.
 313  */
 314 void put_iova_domain(struct iova_domain *iovad)
 315 {
 316         struct rb_node *node;
 317         unsigned long flags;
 318
 319         spin_lock_irqsave(&iovad->iova_rbtree_lock, flags);
 320         node = rb_first(&iovad->rbroot);
 321         while (node) {
 322                 struct iova *iova = container_of(node, struct iova, node);
 323                 rb_erase(node, &iovad->rbroot);
 324                 free_iova_mem(iova);
 325                 node = rb_first(&iovad->rbroot);
 326         }
 327         spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
 328 }
 329
 330 static int
 331 __is_range_overlap(struct rb_node *node,
 332         unsigned long pfn_lo, unsigned long pfn_hi)
 333 {
 334         struct iova *iova = container_of(node, struct iova, node);
 335
 336         if ((pfn_lo <= iova->pfn_hi) && (pfn_hi >= iova->pfn_lo))
 337                 return 1;
 338         return 0;
 339 }
 340
 341 static struct iova *
 342 __insert_new_range(struct iova_domain *iovad,
 343         unsigned long pfn_lo, unsigned long pfn_hi)
 344 {
 345         struct iova *iova;
 346
 347         iova = alloc_iova_mem();
 348         if (!iova)
 349                 return iova;
 350
 351         iova->pfn_hi = pfn_hi;
 352         iova->pfn_lo = pfn_lo;
 353         iova_insert_rbtree(&iovad->rbroot, iova);
 354         return iova;
 355 }
 356
 357 static void
 358 __adjust_overlap_range(struct iova *iova,
 359         unsigned long *pfn_lo, unsigned long *pfn_hi)
 360 {
 361         if (*pfn_lo < iova->pfn_lo)
 362                 iova->pfn_lo = *pfn_lo;
 363         if (*pfn_hi > iova->pfn_hi)
 364                 *pfn_lo = iova->pfn_hi + 1;
 365 }
 366
 367 /**
 368  * reserve_iova - reserves an iova in the given range
 369  * @iovad: - iova domain pointer
 370  * @pfn_lo: - lower page frame address
 371  * @pfn_hi:- higher pfn adderss
 372  * This function allocates reserves the address range from pfn_lo to pfn_hi so
 373  * that this address is not dished out as part of alloc_iova.
 374  */
 375 struct iova *
 376 reserve_iova(struct iova_domain *iovad,
 377         unsigned long pfn_lo, unsigned long pfn_hi)
 378 {
 379         struct rb_node *node;
 380         unsigned long flags;
 381         struct iova *iova;
 382         unsigned int overlap = 0;
 383
 384         spin_lock_irqsave(&iovad->iova_alloc_lock, flags);
 385         spin_lock(&iovad->iova_rbtree_lock);
 386         for (node = rb_first(&iovad->rbroot); node; node = rb_next(node)) {
 387                 if (__is_range_overlap(node, pfn_lo, pfn_hi)) {
 388                         iova = container_of(node, struct iova, node);
 389                         __adjust_overlap_range(iova, &pfn_lo, &pfn_hi);
 390                         if ((pfn_lo >= iova->pfn_lo) &&
 391                                 (pfn_hi <= iova->pfn_hi))
 392                                 goto finish;
 393                         overlap = 1;
 394
 395                 } else if (overlap)
 396                                 break;
 397         }
 398
 399         /* We are here either becasue this is the first reserver node
 400          * or need to insert remaining non overlap addr range
 401          */
 402         iova = __insert_new_range(iovad, pfn_lo, pfn_hi);
 403 finish:
 404
 405         spin_unlock(&iovad->iova_rbtree_lock);
 406         spin_unlock_irqrestore(&iovad->iova_alloc_lock, flags);
 407         return iova;
 408 }
 409
 410 /**
 411  * copy_reserved_iova - copies the reserved between domains
 412  * @from: - source doamin from where to copy
 413  * @to: - destination domin where to copy
 414  * This function copies reserved iova's from one doamin to
 415  * other.
 416  */
 417 void
 418 copy_reserved_iova(struct iova_domain *from, struct iova_domain *to)
 419 {
 420         unsigned long flags;
 421         struct rb_node *node;
 422
 423         spin_lock_irqsave(&from->iova_alloc_lock, flags);
 424         spin_lock(&from->iova_rbtree_lock);
 425         for (node = rb_first(&from->rbroot); node; node = rb_next(node)) {
 426                 struct iova *iova = container_of(node, struct iova, node);
 427                 struct iova *new_iova;
 428                 new_iova = reserve_iova(to, iova->pfn_lo, iova->pfn_hi);
 429                 if (!new_iova)
 430                         printk(KERN_ERR "Reserve iova range %lx@%lx failed\n",
 431                                 iova->pfn_lo, iova->pfn_lo);
 432         }
 433         spin_unlock(&from->iova_rbtree_lock);
 434         spin_unlock_irqrestore(&from->iova_alloc_lock, flags);
 435 }