]>
Commit | Line | Data |
---|---|---|
3b20eb23 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
f8de50eb | 2 | /* |
a15a519e | 3 | * Copyright © 2006-2009, Intel Corporation. |
f8de50eb | 4 | * |
98bcef56 | 5 | * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com> |
f8de50eb KA |
6 | */ |
7 | ||
38717946 | 8 | #include <linux/iova.h> |
15bbdec3 | 9 | #include <linux/module.h> |
85b45456 | 10 | #include <linux/slab.h> |
9257b4a2 OP |
11 | #include <linux/smp.h> |
12 | #include <linux/bitops.h> | |
aaffaa8a | 13 | #include <linux/cpu.h> |
9257b4a2 | 14 | |
bb68b2fb RM |
15 | /* The anchor node sits above the top of the usable address space */ |
16 | #define IOVA_ANCHOR ~0UL | |
17 | ||
9257b4a2 OP |
18 | static bool iova_rcache_insert(struct iova_domain *iovad, |
19 | unsigned long pfn, | |
20 | unsigned long size); | |
21 | static unsigned long iova_rcache_get(struct iova_domain *iovad, | |
22 | unsigned long size, | |
23 | unsigned long limit_pfn); | |
24 | static void init_iova_rcaches(struct iova_domain *iovad); | |
149448b3 | 25 | static void free_cpu_cached_iovas(unsigned int cpu, struct iova_domain *iovad); |
9257b4a2 | 26 | static void free_iova_rcaches(struct iova_domain *iovad); |
19282101 | 27 | static void fq_destroy_all_entries(struct iova_domain *iovad); |
e99e88a9 | 28 | static void fq_flush_timeout(struct timer_list *t); |
f598a497 JG |
29 | |
30 | static int iova_cpuhp_dead(unsigned int cpu, struct hlist_node *node) | |
31 | { | |
32 | struct iova_domain *iovad; | |
33 | ||
34 | iovad = hlist_entry_safe(node, struct iova_domain, cpuhp_dead); | |
35 | ||
36 | free_cpu_cached_iovas(cpu, iovad); | |
37 | return 0; | |
38 | } | |
39 | ||
6fa3525b | 40 | static void free_global_cached_iovas(struct iova_domain *iovad); |
85b45456 | 41 | |
7ae31cec RM |
42 | static struct iova *to_iova(struct rb_node *node) |
43 | { | |
44 | return rb_entry(node, struct iova, node); | |
45 | } | |
46 | ||
f8de50eb | 47 | void |
0fb5fe87 | 48 | init_iova_domain(struct iova_domain *iovad, unsigned long granule, |
aa3ac946 | 49 | unsigned long start_pfn) |
f8de50eb | 50 | { |
0fb5fe87 RM |
51 | /* |
52 | * IOVA granularity will normally be equal to the smallest | |
53 | * supported IOMMU page size; both *must* be capable of | |
54 | * representing individual CPU pages exactly. | |
55 | */ | |
56 | BUG_ON((granule > PAGE_SIZE) || !is_power_of_2(granule)); | |
57 | ||
f8de50eb KA |
58 | spin_lock_init(&iovad->iova_rbtree_lock); |
59 | iovad->rbroot = RB_ROOT; | |
973f5fbe RM |
60 | iovad->cached_node = &iovad->anchor.node; |
61 | iovad->cached32_node = &iovad->anchor.node; | |
0fb5fe87 | 62 | iovad->granule = granule; |
1b722500 | 63 | iovad->start_pfn = start_pfn; |
aa3ac946 | 64 | iovad->dma_32bit_pfn = 1UL << (32 - iova_shift(iovad)); |
bee60e94 | 65 | iovad->max32_alloc_size = iovad->dma_32bit_pfn; |
42f87e71 JR |
66 | iovad->flush_cb = NULL; |
67 | iovad->fq = NULL; | |
bb68b2fb RM |
68 | iovad->anchor.pfn_lo = iovad->anchor.pfn_hi = IOVA_ANCHOR; |
69 | rb_link_node(&iovad->anchor.node, NULL, &iovad->rbroot.rb_node); | |
70 | rb_insert_color(&iovad->anchor.node, &iovad->rbroot); | |
f598a497 | 71 | cpuhp_state_add_instance_nocalls(CPUHP_IOMMU_IOVA_DEAD, &iovad->cpuhp_dead); |
9257b4a2 | 72 | init_iova_rcaches(iovad); |
f8de50eb | 73 | } |
9b41760b | 74 | EXPORT_SYMBOL_GPL(init_iova_domain); |
f8de50eb | 75 | |
9cc0aaeb | 76 | static bool has_iova_flush_queue(struct iova_domain *iovad) |
effa4678 DS |
77 | { |
78 | return !!iovad->fq; | |
79 | } | |
80 | ||
42f87e71 JR |
81 | static void free_iova_flush_queue(struct iova_domain *iovad) |
82 | { | |
effa4678 | 83 | if (!has_iova_flush_queue(iovad)) |
42f87e71 JR |
84 | return; |
85 | ||
9a005a80 JR |
86 | if (timer_pending(&iovad->fq_timer)) |
87 | del_timer(&iovad->fq_timer); | |
88 | ||
19282101 | 89 | fq_destroy_all_entries(iovad); |
9a005a80 | 90 | |
42f87e71 JR |
91 | free_percpu(iovad->fq); |
92 | ||
93 | iovad->fq = NULL; | |
94 | iovad->flush_cb = NULL; | |
95 | iovad->entry_dtor = NULL; | |
96 | } | |
97 | ||
98 | int init_iova_flush_queue(struct iova_domain *iovad, | |
99 | iova_flush_cb flush_cb, iova_entry_dtor entry_dtor) | |
100 | { | |
effa4678 | 101 | struct iova_fq __percpu *queue; |
42f87e71 JR |
102 | int cpu; |
103 | ||
fb418dab JR |
104 | atomic64_set(&iovad->fq_flush_start_cnt, 0); |
105 | atomic64_set(&iovad->fq_flush_finish_cnt, 0); | |
106 | ||
effa4678 DS |
107 | queue = alloc_percpu(struct iova_fq); |
108 | if (!queue) | |
42f87e71 JR |
109 | return -ENOMEM; |
110 | ||
111 | iovad->flush_cb = flush_cb; | |
112 | iovad->entry_dtor = entry_dtor; | |
113 | ||
114 | for_each_possible_cpu(cpu) { | |
115 | struct iova_fq *fq; | |
116 | ||
effa4678 | 117 | fq = per_cpu_ptr(queue, cpu); |
42f87e71 JR |
118 | fq->head = 0; |
119 | fq->tail = 0; | |
8109c2a2 JR |
120 | |
121 | spin_lock_init(&fq->lock); | |
42f87e71 | 122 | } |
effa4678 | 123 | |
effa4678 | 124 | iovad->fq = queue; |
42f87e71 | 125 | |
e99e88a9 | 126 | timer_setup(&iovad->fq_timer, fq_flush_timeout, 0); |
9a005a80 JR |
127 | atomic_set(&iovad->fq_timer_on, 0); |
128 | ||
42f87e71 JR |
129 | return 0; |
130 | } | |
42f87e71 | 131 | |
f8de50eb | 132 | static struct rb_node * |
973f5fbe | 133 | __get_cached_rbnode(struct iova_domain *iovad, unsigned long limit_pfn) |
f8de50eb | 134 | { |
973f5fbe RM |
135 | if (limit_pfn <= iovad->dma_32bit_pfn) |
136 | return iovad->cached32_node; | |
e60aa7b5 | 137 | |
973f5fbe | 138 | return iovad->cached_node; |
f8de50eb KA |
139 | } |
140 | ||
141 | static void | |
e60aa7b5 | 142 | __cached_rbnode_insert_update(struct iova_domain *iovad, struct iova *new) |
f8de50eb | 143 | { |
e60aa7b5 RM |
144 | if (new->pfn_hi < iovad->dma_32bit_pfn) |
145 | iovad->cached32_node = &new->node; | |
146 | else | |
147 | iovad->cached_node = &new->node; | |
f8de50eb KA |
148 | } |
149 | ||
150 | static void | |
151 | __cached_rbnode_delete_update(struct iova_domain *iovad, struct iova *free) | |
152 | { | |
153 | struct iova *cached_iova; | |
f8de50eb | 154 | |
7ae31cec | 155 | cached_iova = to_iova(iovad->cached32_node); |
9eed17d3 CW |
156 | if (free == cached_iova || |
157 | (free->pfn_hi < iovad->dma_32bit_pfn && | |
158 | free->pfn_lo >= cached_iova->pfn_lo)) { | |
e60aa7b5 | 159 | iovad->cached32_node = rb_next(&free->node); |
bee60e94 GK |
160 | iovad->max32_alloc_size = iovad->dma_32bit_pfn; |
161 | } | |
e60aa7b5 | 162 | |
7ae31cec | 163 | cached_iova = to_iova(iovad->cached_node); |
973f5fbe | 164 | if (free->pfn_lo >= cached_iova->pfn_lo) |
e60aa7b5 | 165 | iovad->cached_node = rb_next(&free->node); |
f8de50eb KA |
166 | } |
167 | ||
371d7955 RM |
168 | static struct rb_node *iova_find_limit(struct iova_domain *iovad, unsigned long limit_pfn) |
169 | { | |
170 | struct rb_node *node, *next; | |
171 | /* | |
172 | * Ideally what we'd like to judge here is whether limit_pfn is close | |
173 | * enough to the highest-allocated IOVA that starting the allocation | |
174 | * walk from the anchor node will be quicker than this initial work to | |
175 | * find an exact starting point (especially if that ends up being the | |
176 | * anchor node anyway). This is an incredibly crude approximation which | |
177 | * only really helps the most likely case, but is at least trivially easy. | |
178 | */ | |
179 | if (limit_pfn > iovad->dma_32bit_pfn) | |
180 | return &iovad->anchor.node; | |
181 | ||
182 | node = iovad->rbroot.rb_node; | |
183 | while (to_iova(node)->pfn_hi < limit_pfn) | |
184 | node = node->rb_right; | |
185 | ||
186 | search_left: | |
187 | while (node->rb_left && to_iova(node->rb_left)->pfn_lo >= limit_pfn) | |
188 | node = node->rb_left; | |
189 | ||
190 | if (!node->rb_left) | |
191 | return node; | |
192 | ||
193 | next = node->rb_left; | |
194 | while (next->rb_right) { | |
195 | next = next->rb_right; | |
196 | if (to_iova(next)->pfn_lo >= limit_pfn) { | |
197 | node = next; | |
198 | goto search_left; | |
199 | } | |
200 | } | |
201 | ||
202 | return node; | |
203 | } | |
204 | ||
d751751a MS |
205 | /* Insert the iova into domain rbtree by holding writer lock */ |
206 | static void | |
207 | iova_insert_rbtree(struct rb_root *root, struct iova *iova, | |
208 | struct rb_node *start) | |
209 | { | |
210 | struct rb_node **new, *parent = NULL; | |
211 | ||
212 | new = (start) ? &start : &(root->rb_node); | |
213 | /* Figure out where to put new node */ | |
214 | while (*new) { | |
7ae31cec | 215 | struct iova *this = to_iova(*new); |
d751751a MS |
216 | |
217 | parent = *new; | |
218 | ||
219 | if (iova->pfn_lo < this->pfn_lo) | |
220 | new = &((*new)->rb_left); | |
221 | else if (iova->pfn_lo > this->pfn_lo) | |
222 | new = &((*new)->rb_right); | |
223 | else { | |
224 | WARN_ON(1); /* this should not happen */ | |
225 | return; | |
226 | } | |
227 | } | |
228 | /* Add new node and rebalance tree. */ | |
229 | rb_link_node(&iova->node, parent, new); | |
230 | rb_insert_color(&iova->node, root); | |
231 | } | |
232 | ||
ddf02886 | 233 | static int __alloc_and_insert_iova_range(struct iova_domain *iovad, |
234 | unsigned long size, unsigned long limit_pfn, | |
235 | struct iova *new, bool size_aligned) | |
f8de50eb | 236 | { |
973f5fbe RM |
237 | struct rb_node *curr, *prev; |
238 | struct iova *curr_iova; | |
f8de50eb | 239 | unsigned long flags; |
4e89dce7 | 240 | unsigned long new_pfn, retry_pfn; |
086c83ac | 241 | unsigned long align_mask = ~0UL; |
4e89dce7 | 242 | unsigned long high_pfn = limit_pfn, low_pfn = iovad->start_pfn; |
086c83ac ZL |
243 | |
244 | if (size_aligned) | |
245 | align_mask <<= fls_long(size - 1); | |
f8de50eb KA |
246 | |
247 | /* Walk the tree backwards */ | |
248 | spin_lock_irqsave(&iovad->iova_rbtree_lock, flags); | |
bee60e94 GK |
249 | if (limit_pfn <= iovad->dma_32bit_pfn && |
250 | size >= iovad->max32_alloc_size) | |
251 | goto iova32_full; | |
252 | ||
973f5fbe | 253 | curr = __get_cached_rbnode(iovad, limit_pfn); |
7ae31cec | 254 | curr_iova = to_iova(curr); |
4e89dce7 VJ |
255 | retry_pfn = curr_iova->pfn_hi + 1; |
256 | ||
257 | retry: | |
973f5fbe | 258 | do { |
4e89dce7 VJ |
259 | high_pfn = min(high_pfn, curr_iova->pfn_lo); |
260 | new_pfn = (high_pfn - size) & align_mask; | |
ddf02886 | 261 | prev = curr; |
f8de50eb | 262 | curr = rb_prev(curr); |
7ae31cec | 263 | curr_iova = to_iova(curr); |
4e89dce7 VJ |
264 | } while (curr && new_pfn <= curr_iova->pfn_hi && new_pfn >= low_pfn); |
265 | ||
266 | if (high_pfn < size || new_pfn < low_pfn) { | |
267 | if (low_pfn == iovad->start_pfn && retry_pfn < limit_pfn) { | |
268 | high_pfn = limit_pfn; | |
269 | low_pfn = retry_pfn; | |
371d7955 | 270 | curr = iova_find_limit(iovad, limit_pfn); |
7ae31cec | 271 | curr_iova = to_iova(curr); |
4e89dce7 VJ |
272 | goto retry; |
273 | } | |
80ef4464 | 274 | iovad->max32_alloc_size = size; |
bee60e94 | 275 | goto iova32_full; |
80ef4464 | 276 | } |
f76aec76 KA |
277 | |
278 | /* pfn_lo will point to size aligned address if size_aligned is set */ | |
086c83ac | 279 | new->pfn_lo = new_pfn; |
f76aec76 | 280 | new->pfn_hi = new->pfn_lo + size - 1; |
f8de50eb | 281 | |
d751751a MS |
282 | /* If we have 'prev', it's a valid place to start the insertion. */ |
283 | iova_insert_rbtree(&iovad->rbroot, new, prev); | |
e60aa7b5 | 284 | __cached_rbnode_insert_update(iovad, new); |
ddf02886 | 285 | |
f8de50eb KA |
286 | spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags); |
287 | return 0; | |
bee60e94 GK |
288 | |
289 | iova32_full: | |
bee60e94 GK |
290 | spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags); |
291 | return -ENOMEM; | |
f8de50eb KA |
292 | } |
293 | ||
ae1ff3d6 SA |
294 | static struct kmem_cache *iova_cache; |
295 | static unsigned int iova_cache_users; | |
296 | static DEFINE_MUTEX(iova_cache_mutex); | |
297 | ||
51b70b81 | 298 | static struct iova *alloc_iova_mem(void) |
ae1ff3d6 | 299 | { |
944c9175 | 300 | return kmem_cache_zalloc(iova_cache, GFP_ATOMIC | __GFP_NOWARN); |
ae1ff3d6 | 301 | } |
ae1ff3d6 | 302 | |
176cfc18 | 303 | static void free_iova_mem(struct iova *iova) |
ae1ff3d6 | 304 | { |
bb68b2fb RM |
305 | if (iova->pfn_lo != IOVA_ANCHOR) |
306 | kmem_cache_free(iova_cache, iova); | |
ae1ff3d6 | 307 | } |
ae1ff3d6 SA |
308 | |
309 | int iova_cache_get(void) | |
310 | { | |
311 | mutex_lock(&iova_cache_mutex); | |
312 | if (!iova_cache_users) { | |
f598a497 JG |
313 | int ret; |
314 | ||
315 | ret = cpuhp_setup_state_multi(CPUHP_IOMMU_IOVA_DEAD, "iommu/iova:dead", NULL, | |
316 | iova_cpuhp_dead); | |
317 | if (ret) { | |
318 | mutex_unlock(&iova_cache_mutex); | |
319 | pr_err("Couldn't register cpuhp handler\n"); | |
320 | return ret; | |
321 | } | |
322 | ||
ae1ff3d6 SA |
323 | iova_cache = kmem_cache_create( |
324 | "iommu_iova", sizeof(struct iova), 0, | |
325 | SLAB_HWCACHE_ALIGN, NULL); | |
326 | if (!iova_cache) { | |
f598a497 | 327 | cpuhp_remove_multi_state(CPUHP_IOMMU_IOVA_DEAD); |
ae1ff3d6 | 328 | mutex_unlock(&iova_cache_mutex); |
3a0ce12e | 329 | pr_err("Couldn't create iova cache\n"); |
ae1ff3d6 SA |
330 | return -ENOMEM; |
331 | } | |
332 | } | |
333 | ||
334 | iova_cache_users++; | |
335 | mutex_unlock(&iova_cache_mutex); | |
336 | ||
337 | return 0; | |
338 | } | |
9b41760b | 339 | EXPORT_SYMBOL_GPL(iova_cache_get); |
ae1ff3d6 SA |
340 | |
341 | void iova_cache_put(void) | |
342 | { | |
343 | mutex_lock(&iova_cache_mutex); | |
344 | if (WARN_ON(!iova_cache_users)) { | |
345 | mutex_unlock(&iova_cache_mutex); | |
346 | return; | |
347 | } | |
348 | iova_cache_users--; | |
f598a497 JG |
349 | if (!iova_cache_users) { |
350 | cpuhp_remove_multi_state(CPUHP_IOMMU_IOVA_DEAD); | |
ae1ff3d6 | 351 | kmem_cache_destroy(iova_cache); |
f598a497 | 352 | } |
ae1ff3d6 SA |
353 | mutex_unlock(&iova_cache_mutex); |
354 | } | |
9b41760b | 355 | EXPORT_SYMBOL_GPL(iova_cache_put); |
ae1ff3d6 | 356 | |
f8de50eb KA |
357 | /** |
358 | * alloc_iova - allocates an iova | |
07db0409 MI |
359 | * @iovad: - iova domain in question |
360 | * @size: - size of page frames to allocate | |
361 | * @limit_pfn: - max limit address | |
362 | * @size_aligned: - set if size_aligned address range is required | |
1b722500 RM |
363 | * This function allocates an iova in the range iovad->start_pfn to limit_pfn, |
364 | * searching top-down from limit_pfn to iovad->start_pfn. If the size_aligned | |
f76aec76 KA |
365 | * flag is set then the allocated address iova->pfn_lo will be naturally |
366 | * aligned on roundup_power_of_two(size). | |
f8de50eb KA |
367 | */ |
368 | struct iova * | |
369 | alloc_iova(struct iova_domain *iovad, unsigned long size, | |
f76aec76 KA |
370 | unsigned long limit_pfn, |
371 | bool size_aligned) | |
f8de50eb | 372 | { |
f8de50eb KA |
373 | struct iova *new_iova; |
374 | int ret; | |
375 | ||
376 | new_iova = alloc_iova_mem(); | |
377 | if (!new_iova) | |
378 | return NULL; | |
379 | ||
757c370f | 380 | ret = __alloc_and_insert_iova_range(iovad, size, limit_pfn + 1, |
ddf02886 | 381 | new_iova, size_aligned); |
f8de50eb KA |
382 | |
383 | if (ret) { | |
f8de50eb KA |
384 | free_iova_mem(new_iova); |
385 | return NULL; | |
386 | } | |
387 | ||
f8de50eb KA |
388 | return new_iova; |
389 | } | |
9b41760b | 390 | EXPORT_SYMBOL_GPL(alloc_iova); |
f8de50eb | 391 | |
9257b4a2 OP |
392 | static struct iova * |
393 | private_find_iova(struct iova_domain *iovad, unsigned long pfn) | |
f8de50eb | 394 | { |
9257b4a2 OP |
395 | struct rb_node *node = iovad->rbroot.rb_node; |
396 | ||
397 | assert_spin_locked(&iovad->iova_rbtree_lock); | |
f8de50eb | 398 | |
f8de50eb | 399 | while (node) { |
7ae31cec | 400 | struct iova *iova = to_iova(node); |
f8de50eb | 401 | |
f8de50eb KA |
402 | if (pfn < iova->pfn_lo) |
403 | node = node->rb_left; | |
2070f940 | 404 | else if (pfn > iova->pfn_hi) |
f8de50eb | 405 | node = node->rb_right; |
2070f940 ZL |
406 | else |
407 | return iova; /* pfn falls within iova's range */ | |
f8de50eb KA |
408 | } |
409 | ||
f8de50eb KA |
410 | return NULL; |
411 | } | |
9257b4a2 | 412 | |
7978724f | 413 | static void remove_iova(struct iova_domain *iovad, struct iova *iova) |
9257b4a2 OP |
414 | { |
415 | assert_spin_locked(&iovad->iova_rbtree_lock); | |
416 | __cached_rbnode_delete_update(iovad, iova); | |
417 | rb_erase(&iova->node, &iovad->rbroot); | |
9257b4a2 OP |
418 | } |
419 | ||
420 | /** | |
421 | * find_iova - finds an iova for a given pfn | |
422 | * @iovad: - iova domain in question. | |
423 | * @pfn: - page frame number | |
424 | * This function finds and returns an iova belonging to the | |
6775ae90 | 425 | * given domain which matches the given pfn. |
9257b4a2 OP |
426 | */ |
427 | struct iova *find_iova(struct iova_domain *iovad, unsigned long pfn) | |
428 | { | |
429 | unsigned long flags; | |
430 | struct iova *iova; | |
431 | ||
432 | /* Take the lock so that no other thread is manipulating the rbtree */ | |
433 | spin_lock_irqsave(&iovad->iova_rbtree_lock, flags); | |
434 | iova = private_find_iova(iovad, pfn); | |
435 | spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags); | |
436 | return iova; | |
437 | } | |
9b41760b | 438 | EXPORT_SYMBOL_GPL(find_iova); |
f8de50eb KA |
439 | |
440 | /** | |
441 | * __free_iova - frees the given iova | |
442 | * @iovad: iova domain in question. | |
443 | * @iova: iova in question. | |
444 | * Frees the given iova belonging to the giving domain | |
445 | */ | |
446 | void | |
447 | __free_iova(struct iova_domain *iovad, struct iova *iova) | |
448 | { | |
449 | unsigned long flags; | |
450 | ||
451 | spin_lock_irqsave(&iovad->iova_rbtree_lock, flags); | |
7978724f | 452 | remove_iova(iovad, iova); |
f8de50eb | 453 | spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags); |
7978724f | 454 | free_iova_mem(iova); |
f8de50eb | 455 | } |
9b41760b | 456 | EXPORT_SYMBOL_GPL(__free_iova); |
f8de50eb KA |
457 | |
458 | /** | |
459 | * free_iova - finds and frees the iova for a given pfn | |
460 | * @iovad: - iova domain in question. | |
461 | * @pfn: - pfn that is allocated previously | |
462 | * This functions finds an iova for a given pfn and then | |
463 | * frees the iova from that domain. | |
464 | */ | |
465 | void | |
466 | free_iova(struct iova_domain *iovad, unsigned long pfn) | |
467 | { | |
3a651b3a CW |
468 | unsigned long flags; |
469 | struct iova *iova; | |
733cac2a | 470 | |
3a651b3a CW |
471 | spin_lock_irqsave(&iovad->iova_rbtree_lock, flags); |
472 | iova = private_find_iova(iovad, pfn); | |
7978724f XC |
473 | if (!iova) { |
474 | spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags); | |
475 | return; | |
476 | } | |
477 | remove_iova(iovad, iova); | |
3a651b3a | 478 | spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags); |
7978724f | 479 | free_iova_mem(iova); |
f8de50eb | 480 | } |
9b41760b | 481 | EXPORT_SYMBOL_GPL(free_iova); |
f8de50eb | 482 | |
9257b4a2 OP |
483 | /** |
484 | * alloc_iova_fast - allocates an iova from rcache | |
485 | * @iovad: - iova domain in question | |
486 | * @size: - size of page frames to allocate | |
487 | * @limit_pfn: - max limit address | |
538d5b33 | 488 | * @flush_rcache: - set to flush rcache on regular allocation failure |
9257b4a2 | 489 | * This function tries to satisfy an iova allocation from the rcache, |
538d5b33 TN |
490 | * and falls back to regular allocation on failure. If regular allocation |
491 | * fails too and the flush_rcache flag is set then the rcache will be flushed. | |
9257b4a2 OP |
492 | */ |
493 | unsigned long | |
494 | alloc_iova_fast(struct iova_domain *iovad, unsigned long size, | |
538d5b33 | 495 | unsigned long limit_pfn, bool flush_rcache) |
9257b4a2 | 496 | { |
9257b4a2 OP |
497 | unsigned long iova_pfn; |
498 | struct iova *new_iova; | |
499 | ||
b826ee9a | 500 | iova_pfn = iova_rcache_get(iovad, size, limit_pfn + 1); |
9257b4a2 OP |
501 | if (iova_pfn) |
502 | return iova_pfn; | |
503 | ||
504 | retry: | |
505 | new_iova = alloc_iova(iovad, size, limit_pfn, true); | |
506 | if (!new_iova) { | |
507 | unsigned int cpu; | |
508 | ||
538d5b33 | 509 | if (!flush_rcache) |
9257b4a2 OP |
510 | return 0; |
511 | ||
512 | /* Try replenishing IOVAs by flushing rcache. */ | |
538d5b33 | 513 | flush_rcache = false; |
9257b4a2 OP |
514 | for_each_online_cpu(cpu) |
515 | free_cpu_cached_iovas(cpu, iovad); | |
6fa3525b | 516 | free_global_cached_iovas(iovad); |
9257b4a2 OP |
517 | goto retry; |
518 | } | |
519 | ||
520 | return new_iova->pfn_lo; | |
521 | } | |
a93a9626 | 522 | EXPORT_SYMBOL_GPL(alloc_iova_fast); |
9257b4a2 OP |
523 | |
524 | /** | |
525 | * free_iova_fast - free iova pfn range into rcache | |
526 | * @iovad: - iova domain in question. | |
527 | * @pfn: - pfn that is allocated previously | |
528 | * @size: - # of pages in range | |
529 | * This functions frees an iova range by trying to put it into the rcache, | |
530 | * falling back to regular iova deallocation via free_iova() if this fails. | |
531 | */ | |
532 | void | |
533 | free_iova_fast(struct iova_domain *iovad, unsigned long pfn, unsigned long size) | |
534 | { | |
535 | if (iova_rcache_insert(iovad, pfn, size)) | |
536 | return; | |
537 | ||
538 | free_iova(iovad, pfn); | |
539 | } | |
a93a9626 | 540 | EXPORT_SYMBOL_GPL(free_iova_fast); |
9257b4a2 | 541 | |
19282101 JR |
542 | #define fq_ring_for_each(i, fq) \ |
543 | for ((i) = (fq)->head; (i) != (fq)->tail; (i) = ((i) + 1) % IOVA_FQ_SIZE) | |
544 | ||
545 | static inline bool fq_full(struct iova_fq *fq) | |
546 | { | |
8109c2a2 | 547 | assert_spin_locked(&fq->lock); |
19282101 JR |
548 | return (((fq->tail + 1) % IOVA_FQ_SIZE) == fq->head); |
549 | } | |
550 | ||
551 | static inline unsigned fq_ring_add(struct iova_fq *fq) | |
552 | { | |
553 | unsigned idx = fq->tail; | |
554 | ||
8109c2a2 JR |
555 | assert_spin_locked(&fq->lock); |
556 | ||
19282101 JR |
557 | fq->tail = (idx + 1) % IOVA_FQ_SIZE; |
558 | ||
559 | return idx; | |
560 | } | |
561 | ||
562 | static void fq_ring_free(struct iova_domain *iovad, struct iova_fq *fq) | |
563 | { | |
fb418dab | 564 | u64 counter = atomic64_read(&iovad->fq_flush_finish_cnt); |
19282101 JR |
565 | unsigned idx; |
566 | ||
8109c2a2 JR |
567 | assert_spin_locked(&fq->lock); |
568 | ||
19282101 JR |
569 | fq_ring_for_each(idx, fq) { |
570 | ||
fb418dab JR |
571 | if (fq->entries[idx].counter >= counter) |
572 | break; | |
573 | ||
19282101 JR |
574 | if (iovad->entry_dtor) |
575 | iovad->entry_dtor(fq->entries[idx].data); | |
576 | ||
577 | free_iova_fast(iovad, | |
578 | fq->entries[idx].iova_pfn, | |
579 | fq->entries[idx].pages); | |
fb418dab JR |
580 | |
581 | fq->head = (fq->head + 1) % IOVA_FQ_SIZE; | |
19282101 | 582 | } |
fb418dab | 583 | } |
19282101 | 584 | |
fb418dab JR |
585 | static void iova_domain_flush(struct iova_domain *iovad) |
586 | { | |
587 | atomic64_inc(&iovad->fq_flush_start_cnt); | |
588 | iovad->flush_cb(iovad); | |
589 | atomic64_inc(&iovad->fq_flush_finish_cnt); | |
19282101 JR |
590 | } |
591 | ||
592 | static void fq_destroy_all_entries(struct iova_domain *iovad) | |
593 | { | |
594 | int cpu; | |
595 | ||
596 | /* | |
597 | * This code runs when the iova_domain is being detroyed, so don't | |
598 | * bother to free iovas, just call the entry_dtor on all remaining | |
599 | * entries. | |
600 | */ | |
601 | if (!iovad->entry_dtor) | |
602 | return; | |
603 | ||
604 | for_each_possible_cpu(cpu) { | |
605 | struct iova_fq *fq = per_cpu_ptr(iovad->fq, cpu); | |
606 | int idx; | |
607 | ||
608 | fq_ring_for_each(idx, fq) | |
609 | iovad->entry_dtor(fq->entries[idx].data); | |
610 | } | |
611 | } | |
612 | ||
e99e88a9 | 613 | static void fq_flush_timeout(struct timer_list *t) |
9a005a80 | 614 | { |
e99e88a9 | 615 | struct iova_domain *iovad = from_timer(iovad, t, fq_timer); |
9a005a80 JR |
616 | int cpu; |
617 | ||
618 | atomic_set(&iovad->fq_timer_on, 0); | |
619 | iova_domain_flush(iovad); | |
620 | ||
621 | for_each_possible_cpu(cpu) { | |
622 | unsigned long flags; | |
623 | struct iova_fq *fq; | |
624 | ||
625 | fq = per_cpu_ptr(iovad->fq, cpu); | |
626 | spin_lock_irqsave(&fq->lock, flags); | |
627 | fq_ring_free(iovad, fq); | |
628 | spin_unlock_irqrestore(&fq->lock, flags); | |
629 | } | |
630 | } | |
631 | ||
19282101 JR |
632 | void queue_iova(struct iova_domain *iovad, |
633 | unsigned long pfn, unsigned long pages, | |
634 | unsigned long data) | |
635 | { | |
452e69b5 | 636 | struct iova_fq *fq; |
8109c2a2 | 637 | unsigned long flags; |
19282101 JR |
638 | unsigned idx; |
639 | ||
7a7c5bad RM |
640 | /* |
641 | * Order against the IOMMU driver's pagetable update from unmapping | |
642 | * @pte, to guarantee that iova_domain_flush() observes that if called | |
452e69b5 RM |
643 | * from a different CPU before we release the lock below. Full barrier |
644 | * so it also pairs with iommu_dma_init_fq() to avoid seeing partially | |
645 | * written fq state here. | |
7a7c5bad | 646 | */ |
452e69b5 | 647 | smp_mb(); |
7a7c5bad | 648 | |
452e69b5 | 649 | fq = raw_cpu_ptr(iovad->fq); |
8109c2a2 JR |
650 | spin_lock_irqsave(&fq->lock, flags); |
651 | ||
fb418dab JR |
652 | /* |
653 | * First remove all entries from the flush queue that have already been | |
654 | * flushed out on another CPU. This makes the fq_full() check below less | |
655 | * likely to be true. | |
656 | */ | |
657 | fq_ring_free(iovad, fq); | |
658 | ||
19282101 | 659 | if (fq_full(fq)) { |
fb418dab | 660 | iova_domain_flush(iovad); |
19282101 JR |
661 | fq_ring_free(iovad, fq); |
662 | } | |
663 | ||
664 | idx = fq_ring_add(fq); | |
665 | ||
666 | fq->entries[idx].iova_pfn = pfn; | |
667 | fq->entries[idx].pages = pages; | |
668 | fq->entries[idx].data = data; | |
fb418dab | 669 | fq->entries[idx].counter = atomic64_read(&iovad->fq_flush_start_cnt); |
19282101 | 670 | |
8109c2a2 | 671 | spin_unlock_irqrestore(&fq->lock, flags); |
9a005a80 | 672 | |
0d87308c ED |
673 | /* Avoid false sharing as much as possible. */ |
674 | if (!atomic_read(&iovad->fq_timer_on) && | |
ba328f82 | 675 | !atomic_xchg(&iovad->fq_timer_on, 1)) |
9a005a80 JR |
676 | mod_timer(&iovad->fq_timer, |
677 | jiffies + msecs_to_jiffies(IOVA_FQ_TIMEOUT)); | |
19282101 | 678 | } |
19282101 | 679 | |
f8de50eb | 680 | /** |
6775ae90 | 681 | * put_iova_domain - destroys the iova domain |
f8de50eb KA |
682 | * @iovad: - iova domain in question. |
683 | * All the iova's in that domain are destroyed. | |
684 | */ | |
685 | void put_iova_domain(struct iova_domain *iovad) | |
686 | { | |
7595dc58 | 687 | struct iova *iova, *tmp; |
f8de50eb | 688 | |
f598a497 JG |
689 | cpuhp_state_remove_instance_nocalls(CPUHP_IOMMU_IOVA_DEAD, |
690 | &iovad->cpuhp_dead); | |
691 | ||
42f87e71 | 692 | free_iova_flush_queue(iovad); |
9257b4a2 | 693 | free_iova_rcaches(iovad); |
7595dc58 | 694 | rbtree_postorder_for_each_entry_safe(iova, tmp, &iovad->rbroot, node) |
f8de50eb | 695 | free_iova_mem(iova); |
f8de50eb | 696 | } |
9b41760b | 697 | EXPORT_SYMBOL_GPL(put_iova_domain); |
f8de50eb KA |
698 | |
699 | static int | |
700 | __is_range_overlap(struct rb_node *node, | |
701 | unsigned long pfn_lo, unsigned long pfn_hi) | |
702 | { | |
7ae31cec | 703 | struct iova *iova = to_iova(node); |
f8de50eb KA |
704 | |
705 | if ((pfn_lo <= iova->pfn_hi) && (pfn_hi >= iova->pfn_lo)) | |
706 | return 1; | |
707 | return 0; | |
708 | } | |
709 | ||
75f05569 JL |
710 | static inline struct iova * |
711 | alloc_and_init_iova(unsigned long pfn_lo, unsigned long pfn_hi) | |
712 | { | |
713 | struct iova *iova; | |
714 | ||
715 | iova = alloc_iova_mem(); | |
716 | if (iova) { | |
717 | iova->pfn_lo = pfn_lo; | |
718 | iova->pfn_hi = pfn_hi; | |
719 | } | |
720 | ||
721 | return iova; | |
722 | } | |
723 | ||
f8de50eb KA |
724 | static struct iova * |
725 | __insert_new_range(struct iova_domain *iovad, | |
726 | unsigned long pfn_lo, unsigned long pfn_hi) | |
727 | { | |
728 | struct iova *iova; | |
729 | ||
75f05569 JL |
730 | iova = alloc_and_init_iova(pfn_lo, pfn_hi); |
731 | if (iova) | |
d751751a | 732 | iova_insert_rbtree(&iovad->rbroot, iova, NULL); |
f8de50eb | 733 | |
f8de50eb KA |
734 | return iova; |
735 | } | |
736 | ||
737 | static void | |
738 | __adjust_overlap_range(struct iova *iova, | |
739 | unsigned long *pfn_lo, unsigned long *pfn_hi) | |
740 | { | |
741 | if (*pfn_lo < iova->pfn_lo) | |
742 | iova->pfn_lo = *pfn_lo; | |
743 | if (*pfn_hi > iova->pfn_hi) | |
744 | *pfn_lo = iova->pfn_hi + 1; | |
745 | } | |
746 | ||
747 | /** | |
748 | * reserve_iova - reserves an iova in the given range | |
749 | * @iovad: - iova domain pointer | |
750 | * @pfn_lo: - lower page frame address | |
751 | * @pfn_hi:- higher pfn adderss | |
752 | * This function allocates reserves the address range from pfn_lo to pfn_hi so | |
753 | * that this address is not dished out as part of alloc_iova. | |
754 | */ | |
755 | struct iova * | |
756 | reserve_iova(struct iova_domain *iovad, | |
757 | unsigned long pfn_lo, unsigned long pfn_hi) | |
758 | { | |
759 | struct rb_node *node; | |
760 | unsigned long flags; | |
761 | struct iova *iova; | |
762 | unsigned int overlap = 0; | |
763 | ||
bb68b2fb RM |
764 | /* Don't allow nonsensical pfns */ |
765 | if (WARN_ON((pfn_hi | pfn_lo) > (ULLONG_MAX >> iova_shift(iovad)))) | |
766 | return NULL; | |
767 | ||
3d39cecc | 768 | spin_lock_irqsave(&iovad->iova_rbtree_lock, flags); |
f8de50eb KA |
769 | for (node = rb_first(&iovad->rbroot); node; node = rb_next(node)) { |
770 | if (__is_range_overlap(node, pfn_lo, pfn_hi)) { | |
7ae31cec | 771 | iova = to_iova(node); |
f8de50eb KA |
772 | __adjust_overlap_range(iova, &pfn_lo, &pfn_hi); |
773 | if ((pfn_lo >= iova->pfn_lo) && | |
774 | (pfn_hi <= iova->pfn_hi)) | |
775 | goto finish; | |
776 | overlap = 1; | |
777 | ||
778 | } else if (overlap) | |
779 | break; | |
780 | } | |
781 | ||
25985edc | 782 | /* We are here either because this is the first reserver node |
f8de50eb KA |
783 | * or need to insert remaining non overlap addr range |
784 | */ | |
785 | iova = __insert_new_range(iovad, pfn_lo, pfn_hi); | |
786 | finish: | |
787 | ||
3d39cecc | 788 | spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags); |
f8de50eb KA |
789 | return iova; |
790 | } | |
9b41760b | 791 | EXPORT_SYMBOL_GPL(reserve_iova); |
f8de50eb | 792 | |
9257b4a2 OP |
793 | /* |
794 | * Magazine caches for IOVA ranges. For an introduction to magazines, | |
795 | * see the USENIX 2001 paper "Magazines and Vmem: Extending the Slab | |
796 | * Allocator to Many CPUs and Arbitrary Resources" by Bonwick and Adams. | |
797 | * For simplicity, we use a static magazine size and don't implement the | |
798 | * dynamic size tuning described in the paper. | |
799 | */ | |
800 | ||
801 | #define IOVA_MAG_SIZE 128 | |
802 | ||
803 | struct iova_magazine { | |
804 | unsigned long size; | |
805 | unsigned long pfns[IOVA_MAG_SIZE]; | |
806 | }; | |
807 | ||
808 | struct iova_cpu_rcache { | |
809 | spinlock_t lock; | |
810 | struct iova_magazine *loaded; | |
811 | struct iova_magazine *prev; | |
812 | }; | |
813 | ||
814 | static struct iova_magazine *iova_magazine_alloc(gfp_t flags) | |
815 | { | |
816 | return kzalloc(sizeof(struct iova_magazine), flags); | |
817 | } | |
818 | ||
819 | static void iova_magazine_free(struct iova_magazine *mag) | |
820 | { | |
821 | kfree(mag); | |
822 | } | |
823 | ||
824 | static void | |
825 | iova_magazine_free_pfns(struct iova_magazine *mag, struct iova_domain *iovad) | |
826 | { | |
827 | unsigned long flags; | |
828 | int i; | |
829 | ||
830 | if (!mag) | |
831 | return; | |
832 | ||
833 | spin_lock_irqsave(&iovad->iova_rbtree_lock, flags); | |
834 | ||
835 | for (i = 0 ; i < mag->size; ++i) { | |
836 | struct iova *iova = private_find_iova(iovad, mag->pfns[i]); | |
837 | ||
d3e3d2be RM |
838 | if (WARN_ON(!iova)) |
839 | continue; | |
840 | ||
7978724f XC |
841 | remove_iova(iovad, iova); |
842 | free_iova_mem(iova); | |
9257b4a2 OP |
843 | } |
844 | ||
845 | spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags); | |
846 | ||
847 | mag->size = 0; | |
848 | } | |
849 | ||
850 | static bool iova_magazine_full(struct iova_magazine *mag) | |
851 | { | |
852 | return (mag && mag->size == IOVA_MAG_SIZE); | |
853 | } | |
854 | ||
855 | static bool iova_magazine_empty(struct iova_magazine *mag) | |
856 | { | |
857 | return (!mag || mag->size == 0); | |
858 | } | |
859 | ||
860 | static unsigned long iova_magazine_pop(struct iova_magazine *mag, | |
861 | unsigned long limit_pfn) | |
862 | { | |
e8b19840 RM |
863 | int i; |
864 | unsigned long pfn; | |
865 | ||
9257b4a2 OP |
866 | BUG_ON(iova_magazine_empty(mag)); |
867 | ||
e8b19840 RM |
868 | /* Only fall back to the rbtree if we have no suitable pfns at all */ |
869 | for (i = mag->size - 1; mag->pfns[i] > limit_pfn; i--) | |
870 | if (i == 0) | |
871 | return 0; | |
872 | ||
873 | /* Swap it to pop it */ | |
874 | pfn = mag->pfns[i]; | |
875 | mag->pfns[i] = mag->pfns[--mag->size]; | |
9257b4a2 | 876 | |
e8b19840 | 877 | return pfn; |
9257b4a2 OP |
878 | } |
879 | ||
880 | static void iova_magazine_push(struct iova_magazine *mag, unsigned long pfn) | |
881 | { | |
882 | BUG_ON(iova_magazine_full(mag)); | |
883 | ||
884 | mag->pfns[mag->size++] = pfn; | |
885 | } | |
886 | ||
887 | static void init_iova_rcaches(struct iova_domain *iovad) | |
888 | { | |
889 | struct iova_cpu_rcache *cpu_rcache; | |
890 | struct iova_rcache *rcache; | |
891 | unsigned int cpu; | |
892 | int i; | |
893 | ||
894 | for (i = 0; i < IOVA_RANGE_CACHE_MAX_SIZE; ++i) { | |
895 | rcache = &iovad->rcaches[i]; | |
896 | spin_lock_init(&rcache->lock); | |
897 | rcache->depot_size = 0; | |
898 | rcache->cpu_rcaches = __alloc_percpu(sizeof(*cpu_rcache), cache_line_size()); | |
899 | if (WARN_ON(!rcache->cpu_rcaches)) | |
900 | continue; | |
901 | for_each_possible_cpu(cpu) { | |
902 | cpu_rcache = per_cpu_ptr(rcache->cpu_rcaches, cpu); | |
903 | spin_lock_init(&cpu_rcache->lock); | |
904 | cpu_rcache->loaded = iova_magazine_alloc(GFP_KERNEL); | |
905 | cpu_rcache->prev = iova_magazine_alloc(GFP_KERNEL); | |
906 | } | |
907 | } | |
908 | } | |
909 | ||
910 | /* | |
911 | * Try inserting IOVA range starting with 'iova_pfn' into 'rcache', and | |
912 | * return true on success. Can fail if rcache is full and we can't free | |
913 | * space, and free_iova() (our only caller) will then return the IOVA | |
914 | * range to the rbtree instead. | |
915 | */ | |
916 | static bool __iova_rcache_insert(struct iova_domain *iovad, | |
917 | struct iova_rcache *rcache, | |
918 | unsigned long iova_pfn) | |
919 | { | |
920 | struct iova_magazine *mag_to_free = NULL; | |
921 | struct iova_cpu_rcache *cpu_rcache; | |
922 | bool can_insert = false; | |
923 | unsigned long flags; | |
924 | ||
aaffaa8a | 925 | cpu_rcache = raw_cpu_ptr(rcache->cpu_rcaches); |
9257b4a2 OP |
926 | spin_lock_irqsave(&cpu_rcache->lock, flags); |
927 | ||
928 | if (!iova_magazine_full(cpu_rcache->loaded)) { | |
929 | can_insert = true; | |
930 | } else if (!iova_magazine_full(cpu_rcache->prev)) { | |
931 | swap(cpu_rcache->prev, cpu_rcache->loaded); | |
932 | can_insert = true; | |
933 | } else { | |
934 | struct iova_magazine *new_mag = iova_magazine_alloc(GFP_ATOMIC); | |
935 | ||
936 | if (new_mag) { | |
937 | spin_lock(&rcache->lock); | |
938 | if (rcache->depot_size < MAX_GLOBAL_MAGS) { | |
939 | rcache->depot[rcache->depot_size++] = | |
940 | cpu_rcache->loaded; | |
941 | } else { | |
942 | mag_to_free = cpu_rcache->loaded; | |
943 | } | |
944 | spin_unlock(&rcache->lock); | |
945 | ||
946 | cpu_rcache->loaded = new_mag; | |
947 | can_insert = true; | |
948 | } | |
949 | } | |
950 | ||
951 | if (can_insert) | |
952 | iova_magazine_push(cpu_rcache->loaded, iova_pfn); | |
953 | ||
954 | spin_unlock_irqrestore(&cpu_rcache->lock, flags); | |
955 | ||
956 | if (mag_to_free) { | |
957 | iova_magazine_free_pfns(mag_to_free, iovad); | |
958 | iova_magazine_free(mag_to_free); | |
959 | } | |
960 | ||
961 | return can_insert; | |
962 | } | |
963 | ||
964 | static bool iova_rcache_insert(struct iova_domain *iovad, unsigned long pfn, | |
965 | unsigned long size) | |
966 | { | |
967 | unsigned int log_size = order_base_2(size); | |
968 | ||
969 | if (log_size >= IOVA_RANGE_CACHE_MAX_SIZE) | |
970 | return false; | |
971 | ||
972 | return __iova_rcache_insert(iovad, &iovad->rcaches[log_size], pfn); | |
973 | } | |
974 | ||
975 | /* | |
976 | * Caller wants to allocate a new IOVA range from 'rcache'. If we can | |
977 | * satisfy the request, return a matching non-NULL range and remove | |
978 | * it from the 'rcache'. | |
979 | */ | |
980 | static unsigned long __iova_rcache_get(struct iova_rcache *rcache, | |
981 | unsigned long limit_pfn) | |
982 | { | |
983 | struct iova_cpu_rcache *cpu_rcache; | |
984 | unsigned long iova_pfn = 0; | |
985 | bool has_pfn = false; | |
986 | unsigned long flags; | |
987 | ||
aaffaa8a | 988 | cpu_rcache = raw_cpu_ptr(rcache->cpu_rcaches); |
9257b4a2 OP |
989 | spin_lock_irqsave(&cpu_rcache->lock, flags); |
990 | ||
991 | if (!iova_magazine_empty(cpu_rcache->loaded)) { | |
992 | has_pfn = true; | |
993 | } else if (!iova_magazine_empty(cpu_rcache->prev)) { | |
994 | swap(cpu_rcache->prev, cpu_rcache->loaded); | |
995 | has_pfn = true; | |
996 | } else { | |
997 | spin_lock(&rcache->lock); | |
998 | if (rcache->depot_size > 0) { | |
999 | iova_magazine_free(cpu_rcache->loaded); | |
1000 | cpu_rcache->loaded = rcache->depot[--rcache->depot_size]; | |
1001 | has_pfn = true; | |
1002 | } | |
1003 | spin_unlock(&rcache->lock); | |
1004 | } | |
1005 | ||
1006 | if (has_pfn) | |
1007 | iova_pfn = iova_magazine_pop(cpu_rcache->loaded, limit_pfn); | |
1008 | ||
1009 | spin_unlock_irqrestore(&cpu_rcache->lock, flags); | |
1010 | ||
1011 | return iova_pfn; | |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * Try to satisfy IOVA allocation range from rcache. Fail if requested | |
1016 | * size is too big or the DMA limit we are given isn't satisfied by the | |
1017 | * top element in the magazine. | |
1018 | */ | |
1019 | static unsigned long iova_rcache_get(struct iova_domain *iovad, | |
1020 | unsigned long size, | |
1021 | unsigned long limit_pfn) | |
1022 | { | |
1023 | unsigned int log_size = order_base_2(size); | |
1024 | ||
1025 | if (log_size >= IOVA_RANGE_CACHE_MAX_SIZE) | |
1026 | return 0; | |
1027 | ||
b826ee9a | 1028 | return __iova_rcache_get(&iovad->rcaches[log_size], limit_pfn - size); |
9257b4a2 OP |
1029 | } |
1030 | ||
9257b4a2 OP |
1031 | /* |
1032 | * free rcache data structures. | |
1033 | */ | |
1034 | static void free_iova_rcaches(struct iova_domain *iovad) | |
1035 | { | |
1036 | struct iova_rcache *rcache; | |
7595dc58 | 1037 | struct iova_cpu_rcache *cpu_rcache; |
9257b4a2 OP |
1038 | unsigned int cpu; |
1039 | int i, j; | |
1040 | ||
1041 | for (i = 0; i < IOVA_RANGE_CACHE_MAX_SIZE; ++i) { | |
1042 | rcache = &iovad->rcaches[i]; | |
7595dc58 RM |
1043 | for_each_possible_cpu(cpu) { |
1044 | cpu_rcache = per_cpu_ptr(rcache->cpu_rcaches, cpu); | |
1045 | iova_magazine_free(cpu_rcache->loaded); | |
1046 | iova_magazine_free(cpu_rcache->prev); | |
1047 | } | |
9257b4a2 | 1048 | free_percpu(rcache->cpu_rcaches); |
7595dc58 | 1049 | for (j = 0; j < rcache->depot_size; ++j) |
9257b4a2 | 1050 | iova_magazine_free(rcache->depot[j]); |
9257b4a2 OP |
1051 | } |
1052 | } | |
1053 | ||
1054 | /* | |
1055 | * free all the IOVA ranges cached by a cpu (used when cpu is unplugged) | |
1056 | */ | |
149448b3 | 1057 | static void free_cpu_cached_iovas(unsigned int cpu, struct iova_domain *iovad) |
9257b4a2 OP |
1058 | { |
1059 | struct iova_cpu_rcache *cpu_rcache; | |
1060 | struct iova_rcache *rcache; | |
1061 | unsigned long flags; | |
1062 | int i; | |
1063 | ||
1064 | for (i = 0; i < IOVA_RANGE_CACHE_MAX_SIZE; ++i) { | |
1065 | rcache = &iovad->rcaches[i]; | |
1066 | cpu_rcache = per_cpu_ptr(rcache->cpu_rcaches, cpu); | |
1067 | spin_lock_irqsave(&cpu_rcache->lock, flags); | |
1068 | iova_magazine_free_pfns(cpu_rcache->loaded, iovad); | |
1069 | iova_magazine_free_pfns(cpu_rcache->prev, iovad); | |
1070 | spin_unlock_irqrestore(&cpu_rcache->lock, flags); | |
1071 | } | |
1072 | } | |
1073 | ||
6fa3525b VJ |
1074 | /* |
1075 | * free all the IOVA ranges of global cache | |
1076 | */ | |
1077 | static void free_global_cached_iovas(struct iova_domain *iovad) | |
1078 | { | |
1079 | struct iova_rcache *rcache; | |
1080 | unsigned long flags; | |
1081 | int i, j; | |
1082 | ||
1083 | for (i = 0; i < IOVA_RANGE_CACHE_MAX_SIZE; ++i) { | |
1084 | rcache = &iovad->rcaches[i]; | |
1085 | spin_lock_irqsave(&rcache->lock, flags); | |
1086 | for (j = 0; j < rcache->depot_size; ++j) { | |
1087 | iova_magazine_free_pfns(rcache->depot[j], iovad); | |
1088 | iova_magazine_free(rcache->depot[j]); | |
1089 | } | |
1090 | rcache->depot_size = 0; | |
1091 | spin_unlock_irqrestore(&rcache->lock, flags); | |
1092 | } | |
1093 | } | |
15bbdec3 SA |
1094 | MODULE_AUTHOR("Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>"); |
1095 | MODULE_LICENSE("GPL"); |