2 * In-kernel transcendent memory (generic implementation)
4 * Copyright (c) 2009-2011, Dan Magenheimer, Oracle Corp.
6 * The primary purpose of Transcedent Memory ("tmem") is to map object-oriented
7 * "handles" (triples containing a pool id, and object id, and an index), to
8 * pages in a page-accessible memory (PAM). Tmem references the PAM pages via
9 * an abstract "pampd" (PAM page-descriptor), which can be operated on by a
10 * set of functions (pamops). Each pampd contains some representation of
11 * PAGE_SIZE bytes worth of data. Tmem must support potentially millions of
12 * pages and must be able to insert, find, and delete these pages at a
13 * potential frequency of thousands per second concurrently across many CPUs,
14 * (and, if used with KVM, across many vcpus across many guests).
15 * Tmem is tracked with a hierarchy of data structures, organized by
16 * the elements in a handle-tuple: pool_id, object_id, and page index.
17 * One or more "clients" (e.g. guests) each provide one or more tmem_pools.
18 * Each pool, contains a hash table of rb_trees of tmem_objs. Each
19 * tmem_obj contains a radix-tree-like tree of pointers, with intermediate
20 * nodes called tmem_objnodes. Each leaf pointer in this tree points to
21 * a pampd, which is accessible only through a small set of callbacks
22 * registered by the PAM implementation (see tmem_register_pamops). Tmem
23 * does all memory allocation via a set of callbacks registered by the tmem
24 * host implementation (e.g. see tmem_register_hostops).
27 #include <linux/list.h>
28 #include <linux/spinlock.h>
29 #include <linux/atomic.h>
33 /* data structure sentinels used for debugging... see tmem.h */
34 #define POOL_SENTINEL 0x87658765
35 #define OBJ_SENTINEL 0x12345678
36 #define OBJNODE_SENTINEL 0xfedcba09
39 * A tmem host implementation must use this function to register callbacks
40 * for memory allocation.
42 static struct tmem_hostops tmem_hostops
;
44 static void tmem_objnode_tree_init(void);
46 void tmem_register_hostops(struct tmem_hostops
*m
)
48 tmem_objnode_tree_init();
53 * A tmem host implementation must use this function to register
54 * callbacks for a page-accessible memory (PAM) implementation
56 static struct tmem_pamops tmem_pamops
;
58 void tmem_register_pamops(struct tmem_pamops
*m
)
64 * Oid's are potentially very sparse and tmem_objs may have an indeterminately
65 * short life, being added and deleted at a relatively high frequency.
66 * So an rb_tree is an ideal data structure to manage tmem_objs. But because
67 * of the potentially huge number of tmem_objs, each pool manages a hashtable
68 * of rb_trees to reduce search, insert, delete, and rebalancing time.
69 * Each hashbucket also has a lock to manage concurrent access.
71 * The following routines manage tmem_objs. When any tmem_obj is accessed,
72 * the hashbucket lock must be held.
75 static struct tmem_obj
76 *__tmem_obj_find(struct tmem_hashbucket
*hb
, struct tmem_oid
*oidp
,
77 struct rb_node
**parent
, struct rb_node
***link
)
79 struct rb_node
*_parent
= NULL
, **rbnode
;
80 struct tmem_obj
*obj
= NULL
;
82 rbnode
= &hb
->obj_rb_root
.rb_node
;
84 BUG_ON(RB_EMPTY_NODE(*rbnode
));
86 obj
= rb_entry(*rbnode
, struct tmem_obj
,
88 switch (tmem_oid_compare(oidp
, &obj
->oid
)) {
92 rbnode
= &(*rbnode
)->rb_left
;
95 rbnode
= &(*rbnode
)->rb_right
;
111 /* searches for object==oid in pool, returns locked object if found */
112 static struct tmem_obj
*tmem_obj_find(struct tmem_hashbucket
*hb
,
113 struct tmem_oid
*oidp
)
115 return __tmem_obj_find(hb
, oidp
, NULL
, NULL
);
118 static void tmem_pampd_destroy_all_in_obj(struct tmem_obj
*);
120 /* free an object that has no more pampds in it */
121 static void tmem_obj_free(struct tmem_obj
*obj
, struct tmem_hashbucket
*hb
)
123 struct tmem_pool
*pool
;
126 ASSERT_SENTINEL(obj
, OBJ
);
127 BUG_ON(obj
->pampd_count
> 0);
129 BUG_ON(pool
== NULL
);
130 if (obj
->objnode_tree_root
!= NULL
) /* may be "stump" with no leaves */
131 tmem_pampd_destroy_all_in_obj(obj
);
132 BUG_ON(obj
->objnode_tree_root
!= NULL
);
133 BUG_ON((long)obj
->objnode_count
!= 0);
134 atomic_dec(&pool
->obj_count
);
135 BUG_ON(atomic_read(&pool
->obj_count
) < 0);
136 INVERT_SENTINEL(obj
, OBJ
);
138 tmem_oid_set_invalid(&obj
->oid
);
139 rb_erase(&obj
->rb_tree_node
, &hb
->obj_rb_root
);
143 * initialize, and insert an tmem_object_root (called only if find failed)
145 static void tmem_obj_init(struct tmem_obj
*obj
, struct tmem_hashbucket
*hb
,
146 struct tmem_pool
*pool
,
147 struct tmem_oid
*oidp
)
149 struct rb_root
*root
= &hb
->obj_rb_root
;
150 struct rb_node
**new = NULL
, *parent
= NULL
;
152 BUG_ON(pool
== NULL
);
153 atomic_inc(&pool
->obj_count
);
154 obj
->objnode_tree_height
= 0;
155 obj
->objnode_tree_root
= NULL
;
158 obj
->objnode_count
= 0;
159 obj
->pampd_count
= 0;
160 (*tmem_pamops
.new_obj
)(obj
);
161 SET_SENTINEL(obj
, OBJ
);
163 if (__tmem_obj_find(hb
, oidp
, &parent
, &new))
166 rb_link_node(&obj
->rb_tree_node
, parent
, new);
167 rb_insert_color(&obj
->rb_tree_node
, root
);
171 * Tmem is managed as a set of tmem_pools with certain attributes, such as
172 * "ephemeral" vs "persistent". These attributes apply to all tmem_objs
173 * and all pampds that belong to a tmem_pool. A tmem_pool is created
174 * or deleted relatively rarely (for example, when a filesystem is
175 * mounted or unmounted.
178 /* flush all data from a pool and, optionally, free it */
179 static void tmem_pool_flush(struct tmem_pool
*pool
, bool destroy
)
181 struct rb_node
*rbnode
;
182 struct tmem_obj
*obj
;
183 struct tmem_hashbucket
*hb
= &pool
->hashbucket
[0];
186 BUG_ON(pool
== NULL
);
187 for (i
= 0; i
< TMEM_HASH_BUCKETS
; i
++, hb
++) {
188 spin_lock(&hb
->lock
);
189 rbnode
= rb_first(&hb
->obj_rb_root
);
190 while (rbnode
!= NULL
) {
191 obj
= rb_entry(rbnode
, struct tmem_obj
, rb_tree_node
);
192 rbnode
= rb_next(rbnode
);
193 tmem_pampd_destroy_all_in_obj(obj
);
194 tmem_obj_free(obj
, hb
);
195 (*tmem_hostops
.obj_free
)(obj
, pool
);
197 spin_unlock(&hb
->lock
);
200 list_del(&pool
->pool_list
);
204 * A tmem_obj contains a radix-tree-like tree in which the intermediate
205 * nodes are called tmem_objnodes. (The kernel lib/radix-tree.c implementation
206 * is very specialized and tuned for specific uses and is not particularly
207 * suited for use from this code, though some code from the core algorithms has
208 * been reused, thus the copyright notices below). Each tmem_objnode contains
209 * a set of pointers which point to either a set of intermediate tmem_objnodes
210 * or a set of of pampds.
212 * Portions Copyright (C) 2001 Momchil Velikov
213 * Portions Copyright (C) 2001 Christoph Hellwig
214 * Portions Copyright (C) 2005 SGI, Christoph Lameter <clameter@sgi.com>
217 struct tmem_objnode_tree_path
{
218 struct tmem_objnode
*objnode
;
222 /* objnode height_to_maxindex translation */
223 static unsigned long tmem_objnode_tree_h2max
[OBJNODE_TREE_MAX_PATH
+ 1];
225 static void tmem_objnode_tree_init(void)
227 unsigned int ht
, tmp
;
229 for (ht
= 0; ht
< ARRAY_SIZE(tmem_objnode_tree_h2max
); ht
++) {
230 tmp
= ht
* OBJNODE_TREE_MAP_SHIFT
;
231 if (tmp
>= OBJNODE_TREE_INDEX_BITS
)
232 tmem_objnode_tree_h2max
[ht
] = ~0UL;
234 tmem_objnode_tree_h2max
[ht
] =
235 (~0UL >> (OBJNODE_TREE_INDEX_BITS
- tmp
- 1)) >> 1;
239 static struct tmem_objnode
*tmem_objnode_alloc(struct tmem_obj
*obj
)
241 struct tmem_objnode
*objnode
;
243 ASSERT_SENTINEL(obj
, OBJ
);
244 BUG_ON(obj
->pool
== NULL
);
245 ASSERT_SENTINEL(obj
->pool
, POOL
);
246 objnode
= (*tmem_hostops
.objnode_alloc
)(obj
->pool
);
247 if (unlikely(objnode
== NULL
))
250 SET_SENTINEL(objnode
, OBJNODE
);
251 memset(&objnode
->slots
, 0, sizeof(objnode
->slots
));
252 objnode
->slots_in_use
= 0;
253 obj
->objnode_count
++;
258 static void tmem_objnode_free(struct tmem_objnode
*objnode
)
260 struct tmem_pool
*pool
;
263 BUG_ON(objnode
== NULL
);
264 for (i
= 0; i
< OBJNODE_TREE_MAP_SIZE
; i
++)
265 BUG_ON(objnode
->slots
[i
] != NULL
);
266 ASSERT_SENTINEL(objnode
, OBJNODE
);
267 INVERT_SENTINEL(objnode
, OBJNODE
);
268 BUG_ON(objnode
->obj
== NULL
);
269 ASSERT_SENTINEL(objnode
->obj
, OBJ
);
270 pool
= objnode
->obj
->pool
;
271 BUG_ON(pool
== NULL
);
272 ASSERT_SENTINEL(pool
, POOL
);
273 objnode
->obj
->objnode_count
--;
275 (*tmem_hostops
.objnode_free
)(objnode
, pool
);
279 * lookup index in object and return associated pampd (or NULL if not found)
281 static void **__tmem_pampd_lookup_in_obj(struct tmem_obj
*obj
, uint32_t index
)
283 unsigned int height
, shift
;
284 struct tmem_objnode
**slot
= NULL
;
287 ASSERT_SENTINEL(obj
, OBJ
);
288 BUG_ON(obj
->pool
== NULL
);
289 ASSERT_SENTINEL(obj
->pool
, POOL
);
291 height
= obj
->objnode_tree_height
;
292 if (index
> tmem_objnode_tree_h2max
[obj
->objnode_tree_height
])
294 if (height
== 0 && obj
->objnode_tree_root
) {
295 slot
= &obj
->objnode_tree_root
;
298 shift
= (height
-1) * OBJNODE_TREE_MAP_SHIFT
;
299 slot
= &obj
->objnode_tree_root
;
303 slot
= (struct tmem_objnode
**)
305 ((index
>> shift
) & OBJNODE_TREE_MAP_MASK
));
306 shift
-= OBJNODE_TREE_MAP_SHIFT
;
310 return slot
!= NULL
? (void **)slot
: NULL
;
313 static void *tmem_pampd_lookup_in_obj(struct tmem_obj
*obj
, uint32_t index
)
315 struct tmem_objnode
**slot
;
317 slot
= (struct tmem_objnode
**)__tmem_pampd_lookup_in_obj(obj
, index
);
318 return slot
!= NULL
? *slot
: NULL
;
321 static void *tmem_pampd_replace_in_obj(struct tmem_obj
*obj
, uint32_t index
,
324 struct tmem_objnode
**slot
;
327 slot
= (struct tmem_objnode
**)__tmem_pampd_lookup_in_obj(obj
, index
);
328 if ((slot
!= NULL
) && (*slot
!= NULL
)) {
329 void *old_pampd
= *(void **)slot
;
330 *(void **)slot
= new_pampd
;
331 (*tmem_pamops
.free
)(old_pampd
, obj
->pool
, NULL
, 0);
337 static int tmem_pampd_add_to_obj(struct tmem_obj
*obj
, uint32_t index
,
341 struct tmem_objnode
*objnode
= NULL
, *newnode
, *slot
;
342 unsigned int height
, shift
;
345 /* if necessary, extend the tree to be higher */
346 if (index
> tmem_objnode_tree_h2max
[obj
->objnode_tree_height
]) {
347 height
= obj
->objnode_tree_height
+ 1;
348 if (index
> tmem_objnode_tree_h2max
[height
])
349 while (index
> tmem_objnode_tree_h2max
[height
])
351 if (obj
->objnode_tree_root
== NULL
) {
352 obj
->objnode_tree_height
= height
;
356 newnode
= tmem_objnode_alloc(obj
);
361 newnode
->slots
[0] = obj
->objnode_tree_root
;
362 newnode
->slots_in_use
= 1;
363 obj
->objnode_tree_root
= newnode
;
364 obj
->objnode_tree_height
++;
365 } while (height
> obj
->objnode_tree_height
);
368 slot
= obj
->objnode_tree_root
;
369 height
= obj
->objnode_tree_height
;
370 shift
= (height
-1) * OBJNODE_TREE_MAP_SHIFT
;
373 /* add a child objnode. */
374 slot
= tmem_objnode_alloc(obj
);
381 objnode
->slots
[offset
] = slot
;
382 objnode
->slots_in_use
++;
384 obj
->objnode_tree_root
= slot
;
386 /* go down a level */
387 offset
= (index
>> shift
) & OBJNODE_TREE_MAP_MASK
;
389 slot
= objnode
->slots
[offset
];
390 shift
-= OBJNODE_TREE_MAP_SHIFT
;
393 BUG_ON(slot
!= NULL
);
395 objnode
->slots_in_use
++;
396 objnode
->slots
[offset
] = pampd
;
398 obj
->objnode_tree_root
= pampd
;
404 static void *tmem_pampd_delete_from_obj(struct tmem_obj
*obj
, uint32_t index
)
406 struct tmem_objnode_tree_path path
[OBJNODE_TREE_MAX_PATH
+ 1];
407 struct tmem_objnode_tree_path
*pathp
= path
;
408 struct tmem_objnode
*slot
= NULL
;
409 unsigned int height
, shift
;
413 ASSERT_SENTINEL(obj
, OBJ
);
414 BUG_ON(obj
->pool
== NULL
);
415 ASSERT_SENTINEL(obj
->pool
, POOL
);
416 height
= obj
->objnode_tree_height
;
417 if (index
> tmem_objnode_tree_h2max
[height
])
419 slot
= obj
->objnode_tree_root
;
420 if (height
== 0 && obj
->objnode_tree_root
) {
421 obj
->objnode_tree_root
= NULL
;
424 shift
= (height
- 1) * OBJNODE_TREE_MAP_SHIFT
;
425 pathp
->objnode
= NULL
;
430 offset
= (index
>> shift
) & OBJNODE_TREE_MAP_MASK
;
431 pathp
->offset
= offset
;
432 pathp
->objnode
= slot
;
433 slot
= slot
->slots
[offset
];
434 shift
-= OBJNODE_TREE_MAP_SHIFT
;
436 } while (height
> 0);
439 while (pathp
->objnode
) {
440 pathp
->objnode
->slots
[pathp
->offset
] = NULL
;
441 pathp
->objnode
->slots_in_use
--;
442 if (pathp
->objnode
->slots_in_use
) {
443 if (pathp
->objnode
== obj
->objnode_tree_root
) {
444 while (obj
->objnode_tree_height
> 0 &&
445 obj
->objnode_tree_root
->slots_in_use
== 1 &&
446 obj
->objnode_tree_root
->slots
[0]) {
447 struct tmem_objnode
*to_free
=
448 obj
->objnode_tree_root
;
450 obj
->objnode_tree_root
=
452 obj
->objnode_tree_height
--;
453 to_free
->slots
[0] = NULL
;
454 to_free
->slots_in_use
= 0;
455 tmem_objnode_free(to_free
);
460 tmem_objnode_free(pathp
->objnode
); /* 0 slots used, free it */
463 obj
->objnode_tree_height
= 0;
464 obj
->objnode_tree_root
= NULL
;
469 BUG_ON(obj
->pampd_count
< 0);
473 /* recursively walk the objnode_tree destroying pampds and objnodes */
474 static void tmem_objnode_node_destroy(struct tmem_obj
*obj
,
475 struct tmem_objnode
*objnode
,
482 for (i
= 0; i
< OBJNODE_TREE_MAP_SIZE
; i
++) {
483 if (objnode
->slots
[i
]) {
486 (*tmem_pamops
.free
)(objnode
->slots
[i
],
488 objnode
->slots
[i
] = NULL
;
491 tmem_objnode_node_destroy(obj
, objnode
->slots
[i
], ht
-1);
492 tmem_objnode_free(objnode
->slots
[i
]);
493 objnode
->slots
[i
] = NULL
;
498 static void tmem_pampd_destroy_all_in_obj(struct tmem_obj
*obj
)
500 if (obj
->objnode_tree_root
== NULL
)
502 if (obj
->objnode_tree_height
== 0) {
504 (*tmem_pamops
.free
)(obj
->objnode_tree_root
, obj
->pool
, NULL
, 0);
506 tmem_objnode_node_destroy(obj
, obj
->objnode_tree_root
,
507 obj
->objnode_tree_height
);
508 tmem_objnode_free(obj
->objnode_tree_root
);
509 obj
->objnode_tree_height
= 0;
511 obj
->objnode_tree_root
= NULL
;
512 (*tmem_pamops
.free_obj
)(obj
->pool
, obj
);
516 * Tmem is operated on by a set of well-defined actions:
517 * "put", "get", "flush", "flush_object", "new pool" and "destroy pool".
518 * (The tmem ABI allows for subpages and exchanges but these operations
519 * are not included in this implementation.)
521 * These "tmem core" operations are implemented in the following functions.
525 * "Put" a page, e.g. copy a page from the kernel into newly allocated
526 * PAM space (if such space is available). Tmem_put is complicated by
527 * a corner case: What if a page with matching handle already exists in
528 * tmem? To guarantee coherency, one of two actions is necessary: Either
529 * the data for the page must be overwritten, or the page must be
530 * "flushed" so that the data is not accessible to a subsequent "get".
531 * Since these "duplicate puts" are relatively rare, this implementation
532 * always flushes for simplicity.
534 int tmem_put(struct tmem_pool
*pool
, struct tmem_oid
*oidp
, uint32_t index
,
535 char *data
, size_t size
, bool raw
, bool ephemeral
)
537 struct tmem_obj
*obj
= NULL
, *objfound
= NULL
, *objnew
= NULL
;
538 void *pampd
= NULL
, *pampd_del
= NULL
;
540 struct tmem_hashbucket
*hb
;
542 hb
= &pool
->hashbucket
[tmem_oid_hash(oidp
)];
543 spin_lock(&hb
->lock
);
544 obj
= objfound
= tmem_obj_find(hb
, oidp
);
546 pampd
= tmem_pampd_lookup_in_obj(objfound
, index
);
548 /* if found, is a dup put, flush the old one */
549 pampd_del
= tmem_pampd_delete_from_obj(obj
, index
);
550 BUG_ON(pampd_del
!= pampd
);
551 (*tmem_pamops
.free
)(pampd
, pool
, oidp
, index
);
552 if (obj
->pampd_count
== 0) {
559 obj
= objnew
= (*tmem_hostops
.obj_alloc
)(pool
);
560 if (unlikely(obj
== NULL
)) {
564 tmem_obj_init(obj
, hb
, pool
, oidp
);
567 BUG_ON(((objnew
!= obj
) && (objfound
!= obj
)) || (objnew
== objfound
));
568 pampd
= (*tmem_pamops
.create
)(data
, size
, raw
, ephemeral
,
569 obj
->pool
, &obj
->oid
, index
);
570 if (unlikely(pampd
== NULL
))
572 ret
= tmem_pampd_add_to_obj(obj
, index
, pampd
);
573 if (unlikely(ret
== -ENOMEM
))
574 /* may have partially built objnode tree ("stump") */
575 goto delete_and_free
;
579 (void)tmem_pampd_delete_from_obj(obj
, index
);
582 (*tmem_pamops
.free
)(pampd
, pool
, NULL
, 0);
584 tmem_obj_free(objnew
, hb
);
585 (*tmem_hostops
.obj_free
)(objnew
, pool
);
588 spin_unlock(&hb
->lock
);
593 * "Get" a page, e.g. if one can be found, copy the tmem page with the
594 * matching handle from PAM space to the kernel. By tmem definition,
595 * when a "get" is successful on an ephemeral page, the page is "flushed",
596 * and when a "get" is successful on a persistent page, the page is retained
597 * in tmem. Note that to preserve
598 * coherency, "get" can never be skipped if tmem contains the data.
599 * That is, if a get is done with a certain handle and fails, any
600 * subsequent "get" must also fail (unless of course there is a
601 * "put" done with the same handle).
604 int tmem_get(struct tmem_pool
*pool
, struct tmem_oid
*oidp
, uint32_t index
,
605 char *data
, size_t *size
, bool raw
, int get_and_free
)
607 struct tmem_obj
*obj
;
609 bool ephemeral
= is_ephemeral(pool
);
611 struct tmem_hashbucket
*hb
;
612 bool free
= (get_and_free
== 1) || ((get_and_free
== 0) && ephemeral
);
613 bool lock_held
= false;
615 hb
= &pool
->hashbucket
[tmem_oid_hash(oidp
)];
616 spin_lock(&hb
->lock
);
618 obj
= tmem_obj_find(hb
, oidp
);
622 pampd
= tmem_pampd_delete_from_obj(obj
, index
);
624 pampd
= tmem_pampd_lookup_in_obj(obj
, index
);
628 if (obj
->pampd_count
== 0) {
629 tmem_obj_free(obj
, hb
);
630 (*tmem_hostops
.obj_free
)(obj
, pool
);
634 if (tmem_pamops
.is_remote(pampd
)) {
636 spin_unlock(&hb
->lock
);
639 ret
= (*tmem_pamops
.get_data_and_free
)(
640 data
, size
, raw
, pampd
, pool
, oidp
, index
);
642 ret
= (*tmem_pamops
.get_data
)(
643 data
, size
, raw
, pampd
, pool
, oidp
, index
);
649 spin_unlock(&hb
->lock
);
654 * If a page in tmem matches the handle, "flush" this page from tmem such
655 * that any subsequent "get" does not succeed (unless, of course, there
656 * was another "put" with the same handle).
658 int tmem_flush_page(struct tmem_pool
*pool
,
659 struct tmem_oid
*oidp
, uint32_t index
)
661 struct tmem_obj
*obj
;
664 struct tmem_hashbucket
*hb
;
666 hb
= &pool
->hashbucket
[tmem_oid_hash(oidp
)];
667 spin_lock(&hb
->lock
);
668 obj
= tmem_obj_find(hb
, oidp
);
671 pampd
= tmem_pampd_delete_from_obj(obj
, index
);
674 (*tmem_pamops
.free
)(pampd
, pool
, oidp
, index
);
675 if (obj
->pampd_count
== 0) {
676 tmem_obj_free(obj
, hb
);
677 (*tmem_hostops
.obj_free
)(obj
, pool
);
682 spin_unlock(&hb
->lock
);
687 * If a page in tmem matches the handle, replace the page so that any
688 * subsequent "get" gets the new page. Returns 0 if
689 * there was a page to replace, else returns -1.
691 int tmem_replace(struct tmem_pool
*pool
, struct tmem_oid
*oidp
,
692 uint32_t index
, void *new_pampd
)
694 struct tmem_obj
*obj
;
696 struct tmem_hashbucket
*hb
;
698 hb
= &pool
->hashbucket
[tmem_oid_hash(oidp
)];
699 spin_lock(&hb
->lock
);
700 obj
= tmem_obj_find(hb
, oidp
);
703 new_pampd
= tmem_pampd_replace_in_obj(obj
, index
, new_pampd
);
704 ret
= (*tmem_pamops
.replace_in_obj
)(new_pampd
, obj
);
706 spin_unlock(&hb
->lock
);
711 * "Flush" all pages in tmem matching this oid.
713 int tmem_flush_object(struct tmem_pool
*pool
, struct tmem_oid
*oidp
)
715 struct tmem_obj
*obj
;
716 struct tmem_hashbucket
*hb
;
719 hb
= &pool
->hashbucket
[tmem_oid_hash(oidp
)];
720 spin_lock(&hb
->lock
);
721 obj
= tmem_obj_find(hb
, oidp
);
724 tmem_pampd_destroy_all_in_obj(obj
);
725 tmem_obj_free(obj
, hb
);
726 (*tmem_hostops
.obj_free
)(obj
, pool
);
730 spin_unlock(&hb
->lock
);
735 * "Flush" all pages (and tmem_objs) from this tmem_pool and disable
736 * all subsequent access to this tmem_pool.
738 int tmem_destroy_pool(struct tmem_pool
*pool
)
744 tmem_pool_flush(pool
, 1);
750 static LIST_HEAD(tmem_global_pool_list
);
753 * Create a new tmem_pool with the provided flag and return
754 * a pool id provided by the tmem host implementation.
756 void tmem_new_pool(struct tmem_pool
*pool
, uint32_t flags
)
758 int persistent
= flags
& TMEM_POOL_PERSIST
;
759 int shared
= flags
& TMEM_POOL_SHARED
;
760 struct tmem_hashbucket
*hb
= &pool
->hashbucket
[0];
763 for (i
= 0; i
< TMEM_HASH_BUCKETS
; i
++, hb
++) {
764 hb
->obj_rb_root
= RB_ROOT
;
765 spin_lock_init(&hb
->lock
);
767 INIT_LIST_HEAD(&pool
->pool_list
);
768 atomic_set(&pool
->obj_count
, 0);
769 SET_SENTINEL(pool
, POOL
);
770 list_add_tail(&pool
->pool_list
, &tmem_global_pool_list
);
771 pool
->persistent
= persistent
;
772 pool
->shared
= shared
;