2 * Copyright © 2006-2009, Intel Corporation.
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.
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
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.
17 * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
20 #include <linux/iova.h>
21 #include <linux/module.h>
22 #include <linux/slab.h>
25 init_iova_domain(struct iova_domain
*iovad
, unsigned long granule
,
26 unsigned long start_pfn
, unsigned long pfn_32bit
)
29 * IOVA granularity will normally be equal to the smallest
30 * supported IOMMU page size; both *must* be capable of
31 * representing individual CPU pages exactly.
33 BUG_ON((granule
> PAGE_SIZE
) || !is_power_of_2(granule
));
35 spin_lock_init(&iovad
->iova_rbtree_lock
);
36 iovad
->rbroot
= RB_ROOT
;
37 iovad
->cached32_node
= NULL
;
38 iovad
->granule
= granule
;
39 iovad
->start_pfn
= start_pfn
;
40 iovad
->dma_32bit_pfn
= pfn_32bit
;
42 EXPORT_SYMBOL_GPL(init_iova_domain
);
44 static struct rb_node
*
45 __get_cached_rbnode(struct iova_domain
*iovad
, unsigned long *limit_pfn
)
47 if ((*limit_pfn
!= iovad
->dma_32bit_pfn
) ||
48 (iovad
->cached32_node
== NULL
))
49 return rb_last(&iovad
->rbroot
);
51 struct rb_node
*prev_node
= rb_prev(iovad
->cached32_node
);
52 struct iova
*curr_iova
=
53 container_of(iovad
->cached32_node
, struct iova
, node
);
54 *limit_pfn
= curr_iova
->pfn_lo
- 1;
60 __cached_rbnode_insert_update(struct iova_domain
*iovad
,
61 unsigned long limit_pfn
, struct iova
*new)
63 if (limit_pfn
!= iovad
->dma_32bit_pfn
)
65 iovad
->cached32_node
= &new->node
;
69 __cached_rbnode_delete_update(struct iova_domain
*iovad
, struct iova
*free
)
71 struct iova
*cached_iova
;
74 if (!iovad
->cached32_node
)
76 curr
= iovad
->cached32_node
;
77 cached_iova
= container_of(curr
, struct iova
, node
);
79 if (free
->pfn_lo
>= cached_iova
->pfn_lo
) {
80 struct rb_node
*node
= rb_next(&free
->node
);
81 struct iova
*iova
= container_of(node
, struct iova
, node
);
83 /* only cache if it's below 32bit pfn */
84 if (node
&& iova
->pfn_lo
< iovad
->dma_32bit_pfn
)
85 iovad
->cached32_node
= node
;
87 iovad
->cached32_node
= NULL
;
92 * Computes the padding size required, to make the start address
93 * naturally aligned on the power-of-two order of its size
96 iova_get_pad_size(unsigned int size
, unsigned int limit_pfn
)
98 return (limit_pfn
+ 1 - size
) & (__roundup_pow_of_two(size
) - 1);
101 static int __alloc_and_insert_iova_range(struct iova_domain
*iovad
,
102 unsigned long size
, unsigned long limit_pfn
,
103 struct iova
*new, bool size_aligned
)
105 struct rb_node
*prev
, *curr
= NULL
;
107 unsigned long saved_pfn
;
108 unsigned int pad_size
= 0;
110 /* Walk the tree backwards */
111 spin_lock_irqsave(&iovad
->iova_rbtree_lock
, flags
);
112 saved_pfn
= limit_pfn
;
113 curr
= __get_cached_rbnode(iovad
, &limit_pfn
);
116 struct iova
*curr_iova
= container_of(curr
, struct iova
, node
);
118 if (limit_pfn
< curr_iova
->pfn_lo
)
120 else if (limit_pfn
< curr_iova
->pfn_hi
)
121 goto adjust_limit_pfn
;
124 pad_size
= iova_get_pad_size(size
, limit_pfn
);
125 if ((curr_iova
->pfn_hi
+ size
+ pad_size
) <= limit_pfn
)
126 break; /* found a free slot */
129 limit_pfn
= curr_iova
->pfn_lo
- 1;
132 curr
= rb_prev(curr
);
137 pad_size
= iova_get_pad_size(size
, limit_pfn
);
138 if ((iovad
->start_pfn
+ size
+ pad_size
) > limit_pfn
) {
139 spin_unlock_irqrestore(&iovad
->iova_rbtree_lock
, flags
);
144 /* pfn_lo will point to size aligned address if size_aligned is set */
145 new->pfn_lo
= limit_pfn
- (size
+ pad_size
) + 1;
146 new->pfn_hi
= new->pfn_lo
+ size
- 1;
148 /* Insert the new_iova into domain rbtree by holding writer lock */
149 /* Add new node and rebalance tree. */
151 struct rb_node
**entry
, *parent
= NULL
;
153 /* If we have 'prev', it's a valid place to start the
154 insertion. Otherwise, start from the root. */
158 entry
= &iovad
->rbroot
.rb_node
;
160 /* Figure out where to put new node */
162 struct iova
*this = container_of(*entry
,
166 if (new->pfn_lo
< this->pfn_lo
)
167 entry
= &((*entry
)->rb_left
);
168 else if (new->pfn_lo
> this->pfn_lo
)
169 entry
= &((*entry
)->rb_right
);
171 BUG(); /* this should not happen */
174 /* Add new node and rebalance tree. */
175 rb_link_node(&new->node
, parent
, entry
);
176 rb_insert_color(&new->node
, &iovad
->rbroot
);
178 __cached_rbnode_insert_update(iovad
, saved_pfn
, new);
180 spin_unlock_irqrestore(&iovad
->iova_rbtree_lock
, flags
);
187 iova_insert_rbtree(struct rb_root
*root
, struct iova
*iova
)
189 struct rb_node
**new = &(root
->rb_node
), *parent
= NULL
;
190 /* Figure out where to put new node */
192 struct iova
*this = container_of(*new, struct iova
, node
);
196 if (iova
->pfn_lo
< this->pfn_lo
)
197 new = &((*new)->rb_left
);
198 else if (iova
->pfn_lo
> this->pfn_lo
)
199 new = &((*new)->rb_right
);
201 BUG(); /* this should not happen */
203 /* Add new node and rebalance tree. */
204 rb_link_node(&iova
->node
, parent
, new);
205 rb_insert_color(&iova
->node
, root
);
208 static struct kmem_cache
*iova_cache
;
209 static unsigned int iova_cache_users
;
210 static DEFINE_MUTEX(iova_cache_mutex
);
212 struct iova
*alloc_iova_mem(void)
214 return kmem_cache_alloc(iova_cache
, GFP_ATOMIC
);
216 EXPORT_SYMBOL(alloc_iova_mem
);
218 void free_iova_mem(struct iova
*iova
)
220 kmem_cache_free(iova_cache
, iova
);
222 EXPORT_SYMBOL(free_iova_mem
);
224 int iova_cache_get(void)
226 mutex_lock(&iova_cache_mutex
);
227 if (!iova_cache_users
) {
228 iova_cache
= kmem_cache_create(
229 "iommu_iova", sizeof(struct iova
), 0,
230 SLAB_HWCACHE_ALIGN
, NULL
);
232 mutex_unlock(&iova_cache_mutex
);
233 printk(KERN_ERR
"Couldn't create iova cache\n");
239 mutex_unlock(&iova_cache_mutex
);
243 EXPORT_SYMBOL_GPL(iova_cache_get
);
245 void iova_cache_put(void)
247 mutex_lock(&iova_cache_mutex
);
248 if (WARN_ON(!iova_cache_users
)) {
249 mutex_unlock(&iova_cache_mutex
);
253 if (!iova_cache_users
)
254 kmem_cache_destroy(iova_cache
);
255 mutex_unlock(&iova_cache_mutex
);
257 EXPORT_SYMBOL_GPL(iova_cache_put
);
260 * alloc_iova - allocates an iova
261 * @iovad: - iova domain in question
262 * @size: - size of page frames to allocate
263 * @limit_pfn: - max limit address
264 * @size_aligned: - set if size_aligned address range is required
265 * This function allocates an iova in the range iovad->start_pfn to limit_pfn,
266 * searching top-down from limit_pfn to iovad->start_pfn. If the size_aligned
267 * flag is set then the allocated address iova->pfn_lo will be naturally
268 * aligned on roundup_power_of_two(size).
271 alloc_iova(struct iova_domain
*iovad
, unsigned long size
,
272 unsigned long limit_pfn
,
275 struct iova
*new_iova
;
278 new_iova
= alloc_iova_mem();
282 ret
= __alloc_and_insert_iova_range(iovad
, size
, limit_pfn
,
283 new_iova
, size_aligned
);
286 free_iova_mem(new_iova
);
292 EXPORT_SYMBOL_GPL(alloc_iova
);
295 * find_iova - find's an iova for a given pfn
296 * @iovad: - iova domain in question.
297 * @pfn: - page frame number
298 * This function finds and returns an iova belonging to the
299 * given doamin which matches the given pfn.
301 struct iova
*find_iova(struct iova_domain
*iovad
, unsigned long pfn
)
304 struct rb_node
*node
;
306 /* Take the lock so that no other thread is manipulating the rbtree */
307 spin_lock_irqsave(&iovad
->iova_rbtree_lock
, flags
);
308 node
= iovad
->rbroot
.rb_node
;
310 struct iova
*iova
= container_of(node
, struct iova
, node
);
312 /* If pfn falls within iova's range, return iova */
313 if ((pfn
>= iova
->pfn_lo
) && (pfn
<= iova
->pfn_hi
)) {
314 spin_unlock_irqrestore(&iovad
->iova_rbtree_lock
, flags
);
315 /* We are not holding the lock while this iova
316 * is referenced by the caller as the same thread
317 * which called this function also calls __free_iova()
318 * and it is by design that only one thread can possibly
319 * reference a particular iova and hence no conflict.
324 if (pfn
< iova
->pfn_lo
)
325 node
= node
->rb_left
;
326 else if (pfn
> iova
->pfn_lo
)
327 node
= node
->rb_right
;
330 spin_unlock_irqrestore(&iovad
->iova_rbtree_lock
, flags
);
333 EXPORT_SYMBOL_GPL(find_iova
);
336 * __free_iova - frees the given iova
337 * @iovad: iova domain in question.
338 * @iova: iova in question.
339 * Frees the given iova belonging to the giving domain
342 __free_iova(struct iova_domain
*iovad
, struct iova
*iova
)
346 spin_lock_irqsave(&iovad
->iova_rbtree_lock
, flags
);
347 __cached_rbnode_delete_update(iovad
, iova
);
348 rb_erase(&iova
->node
, &iovad
->rbroot
);
349 spin_unlock_irqrestore(&iovad
->iova_rbtree_lock
, flags
);
352 EXPORT_SYMBOL_GPL(__free_iova
);
355 * free_iova - finds and frees the iova for a given pfn
356 * @iovad: - iova domain in question.
357 * @pfn: - pfn that is allocated previously
358 * This functions finds an iova for a given pfn and then
359 * frees the iova from that domain.
362 free_iova(struct iova_domain
*iovad
, unsigned long pfn
)
364 struct iova
*iova
= find_iova(iovad
, pfn
);
367 __free_iova(iovad
, iova
);
370 EXPORT_SYMBOL_GPL(free_iova
);
373 * put_iova_domain - destroys the iova doamin
374 * @iovad: - iova domain in question.
375 * All the iova's in that domain are destroyed.
377 void put_iova_domain(struct iova_domain
*iovad
)
379 struct rb_node
*node
;
382 spin_lock_irqsave(&iovad
->iova_rbtree_lock
, flags
);
383 node
= rb_first(&iovad
->rbroot
);
385 struct iova
*iova
= container_of(node
, struct iova
, node
);
387 rb_erase(node
, &iovad
->rbroot
);
389 node
= rb_first(&iovad
->rbroot
);
391 spin_unlock_irqrestore(&iovad
->iova_rbtree_lock
, flags
);
393 EXPORT_SYMBOL_GPL(put_iova_domain
);
396 __is_range_overlap(struct rb_node
*node
,
397 unsigned long pfn_lo
, unsigned long pfn_hi
)
399 struct iova
*iova
= container_of(node
, struct iova
, node
);
401 if ((pfn_lo
<= iova
->pfn_hi
) && (pfn_hi
>= iova
->pfn_lo
))
406 static inline struct iova
*
407 alloc_and_init_iova(unsigned long pfn_lo
, unsigned long pfn_hi
)
411 iova
= alloc_iova_mem();
413 iova
->pfn_lo
= pfn_lo
;
414 iova
->pfn_hi
= pfn_hi
;
421 __insert_new_range(struct iova_domain
*iovad
,
422 unsigned long pfn_lo
, unsigned long pfn_hi
)
426 iova
= alloc_and_init_iova(pfn_lo
, pfn_hi
);
428 iova_insert_rbtree(&iovad
->rbroot
, iova
);
434 __adjust_overlap_range(struct iova
*iova
,
435 unsigned long *pfn_lo
, unsigned long *pfn_hi
)
437 if (*pfn_lo
< iova
->pfn_lo
)
438 iova
->pfn_lo
= *pfn_lo
;
439 if (*pfn_hi
> iova
->pfn_hi
)
440 *pfn_lo
= iova
->pfn_hi
+ 1;
444 * reserve_iova - reserves an iova in the given range
445 * @iovad: - iova domain pointer
446 * @pfn_lo: - lower page frame address
447 * @pfn_hi:- higher pfn adderss
448 * This function allocates reserves the address range from pfn_lo to pfn_hi so
449 * that this address is not dished out as part of alloc_iova.
452 reserve_iova(struct iova_domain
*iovad
,
453 unsigned long pfn_lo
, unsigned long pfn_hi
)
455 struct rb_node
*node
;
458 unsigned int overlap
= 0;
460 spin_lock_irqsave(&iovad
->iova_rbtree_lock
, flags
);
461 for (node
= rb_first(&iovad
->rbroot
); node
; node
= rb_next(node
)) {
462 if (__is_range_overlap(node
, pfn_lo
, pfn_hi
)) {
463 iova
= container_of(node
, struct iova
, node
);
464 __adjust_overlap_range(iova
, &pfn_lo
, &pfn_hi
);
465 if ((pfn_lo
>= iova
->pfn_lo
) &&
466 (pfn_hi
<= iova
->pfn_hi
))
474 /* We are here either because this is the first reserver node
475 * or need to insert remaining non overlap addr range
477 iova
= __insert_new_range(iovad
, pfn_lo
, pfn_hi
);
480 spin_unlock_irqrestore(&iovad
->iova_rbtree_lock
, flags
);
483 EXPORT_SYMBOL_GPL(reserve_iova
);
486 * copy_reserved_iova - copies the reserved between domains
487 * @from: - source doamin from where to copy
488 * @to: - destination domin where to copy
489 * This function copies reserved iova's from one doamin to
493 copy_reserved_iova(struct iova_domain
*from
, struct iova_domain
*to
)
496 struct rb_node
*node
;
498 spin_lock_irqsave(&from
->iova_rbtree_lock
, flags
);
499 for (node
= rb_first(&from
->rbroot
); node
; node
= rb_next(node
)) {
500 struct iova
*iova
= container_of(node
, struct iova
, node
);
501 struct iova
*new_iova
;
503 new_iova
= reserve_iova(to
, iova
->pfn_lo
, iova
->pfn_hi
);
505 printk(KERN_ERR
"Reserve iova range %lx@%lx failed\n",
506 iova
->pfn_lo
, iova
->pfn_lo
);
508 spin_unlock_irqrestore(&from
->iova_rbtree_lock
, flags
);
510 EXPORT_SYMBOL_GPL(copy_reserved_iova
);
513 split_and_remove_iova(struct iova_domain
*iovad
, struct iova
*iova
,
514 unsigned long pfn_lo
, unsigned long pfn_hi
)
517 struct iova
*prev
= NULL
, *next
= NULL
;
519 spin_lock_irqsave(&iovad
->iova_rbtree_lock
, flags
);
520 if (iova
->pfn_lo
< pfn_lo
) {
521 prev
= alloc_and_init_iova(iova
->pfn_lo
, pfn_lo
- 1);
525 if (iova
->pfn_hi
> pfn_hi
) {
526 next
= alloc_and_init_iova(pfn_hi
+ 1, iova
->pfn_hi
);
531 __cached_rbnode_delete_update(iovad
, iova
);
532 rb_erase(&iova
->node
, &iovad
->rbroot
);
535 iova_insert_rbtree(&iovad
->rbroot
, prev
);
536 iova
->pfn_lo
= pfn_lo
;
539 iova_insert_rbtree(&iovad
->rbroot
, next
);
540 iova
->pfn_hi
= pfn_hi
;
542 spin_unlock_irqrestore(&iovad
->iova_rbtree_lock
, flags
);
547 spin_unlock_irqrestore(&iovad
->iova_rbtree_lock
, flags
);
553 MODULE_AUTHOR("Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>");
554 MODULE_LICENSE("GPL");