]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - mm/zbud.c
Bluetooth: btusb: Add support of IMC Networks PID 0x3568
[mirror_ubuntu-jammy-kernel.git] / mm / zbud.c
CommitLineData
09c434b8 1// SPDX-License-Identifier: GPL-2.0-only
4e2e2770
SJ
2/*
3 * zbud.c
4 *
5 * Copyright (C) 2013, Seth Jennings, IBM
6 *
7 * Concepts based on zcache internal zbud allocator by Dan Magenheimer.
8 *
9 * zbud is an special purpose allocator for storing compressed pages. Contrary
10 * to what its name may suggest, zbud is not a buddy allocator, but rather an
11 * allocator that "buddies" two compressed pages together in a single memory
12 * page.
13 *
14 * While this design limits storage density, it has simple and deterministic
15 * reclaim properties that make it preferable to a higher density approach when
16 * reclaim will be used.
17 *
18 * zbud works by storing compressed pages, or "zpages", together in pairs in a
19 * single memory page called a "zbud page". The first buddy is "left
eee87e17 20 * justified" at the beginning of the zbud page, and the last buddy is "right
4e2e2770
SJ
21 * justified" at the end of the zbud page. The benefit is that if either
22 * buddy is freed, the freed buddy space, coalesced with whatever slack space
23 * that existed between the buddies, results in the largest possible free region
24 * within the zbud page.
25 *
26 * zbud also provides an attractive lower bound on density. The ratio of zpages
27 * to zbud pages can not be less than 1. This ensures that zbud can never "do
28 * harm" by using more pages to store zpages than the uncompressed zpages would
29 * have used on their own.
30 *
31 * zbud pages are divided into "chunks". The size of the chunks is fixed at
32 * compile time and determined by NCHUNKS_ORDER below. Dividing zbud pages
33 * into chunks allows organizing unbuddied zbud pages into a manageable number
34 * of unbuddied lists according to the number of free chunks available in the
35 * zbud page.
36 *
37 * The zbud API differs from that of conventional allocators in that the
38 * allocation function, zbud_alloc(), returns an opaque handle to the user,
39 * not a dereferenceable pointer. The user must map the handle using
40 * zbud_map() in order to get a usable pointer by which to access the
41 * allocation data and unmap the handle with zbud_unmap() when operations
42 * on the allocation data are complete.
43 */
44
45#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
46
47#include <linux/atomic.h>
48#include <linux/list.h>
49#include <linux/mm.h>
50#include <linux/module.h>
51#include <linux/preempt.h>
52#include <linux/slab.h>
53#include <linux/spinlock.h>
c795779d 54#include <linux/zpool.h>
4e2e2770
SJ
55
56/*****************
57 * Structures
58*****************/
59/*
60 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
61 * adjusting internal fragmentation. It also determines the number of
62 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
f203c3b3
CY
63 * allocation granularity will be in chunks of size PAGE_SIZE/64. As one chunk
64 * in allocated page is occupied by zbud header, NCHUNKS will be calculated to
65 * 63 which shows the max number of free chunks in zbud page, also there will be
66 * 63 freelists per pool.
4e2e2770
SJ
67 */
68#define NCHUNKS_ORDER 6
69
70#define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER)
71#define CHUNK_SIZE (1 << CHUNK_SHIFT)
4e2e2770 72#define ZHDR_SIZE_ALIGNED CHUNK_SIZE
f203c3b3 73#define NCHUNKS ((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT)
4e2e2770 74
2a03085c
ML
75struct zbud_pool;
76
77struct zbud_ops {
78 int (*evict)(struct zbud_pool *pool, unsigned long handle);
79};
80
4e2e2770
SJ
81/**
82 * struct zbud_pool - stores metadata for each zbud pool
83 * @lock: protects all pool fields and first|last_chunk fields of any
84 * zbud page in the pool
85 * @unbuddied: array of lists tracking zbud pages that only contain one buddy;
86 * the lists each zbud page is added to depends on the size of
87 * its free region.
88 * @buddied: list tracking the zbud pages that contain two buddies;
89 * these zbud pages are full
90 * @lru: list tracking the zbud pages in LRU order by most recently
91 * added buddy.
92 * @pages_nr: number of zbud pages in the pool.
93 * @ops: pointer to a structure of user defined operations specified at
94 * pool creation time.
a29a7506
MG
95 * @zpool: zpool driver
96 * @zpool_ops: zpool operations structure with an evict callback
4e2e2770
SJ
97 *
98 * This structure is allocated at pool creation time and maintains metadata
99 * pertaining to a particular zbud pool.
100 */
101struct zbud_pool {
102 spinlock_t lock;
f356aeac
ML
103 union {
104 /*
105 * Reuse unbuddied[0] as buddied on the ground that
106 * unbuddied[0] is unused.
107 */
108 struct list_head buddied;
109 struct list_head unbuddied[NCHUNKS];
110 };
4e2e2770
SJ
111 struct list_head lru;
112 u64 pages_nr;
c83db4f4 113 const struct zbud_ops *ops;
479305fd 114 struct zpool *zpool;
78672779 115 const struct zpool_ops *zpool_ops;
4e2e2770
SJ
116};
117
118/*
119 * struct zbud_header - zbud page metadata occupying the first chunk of each
120 * zbud page.
121 * @buddy: links the zbud page into the unbuddied/buddied lists in the pool
122 * @lru: links the zbud page into the lru list in the pool
123 * @first_chunks: the size of the first buddy in chunks, 0 if free
124 * @last_chunks: the size of the last buddy in chunks, 0 if free
125 */
126struct zbud_header {
127 struct list_head buddy;
128 struct list_head lru;
129 unsigned int first_chunks;
130 unsigned int last_chunks;
131 bool under_reclaim;
132};
133
134/*****************
135 * Helpers
136*****************/
137/* Just to make the code easier to read */
138enum buddy {
139 FIRST,
140 LAST
141};
142
143/* Converts an allocation size in bytes to size in zbud chunks */
99eef8e9 144static int size_to_chunks(size_t size)
4e2e2770
SJ
145{
146 return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
147}
148
149#define for_each_unbuddied_list(_iter, _begin) \
150 for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
151
152/* Initializes the zbud header of a newly allocated zbud page */
153static struct zbud_header *init_zbud_page(struct page *page)
154{
155 struct zbud_header *zhdr = page_address(page);
156 zhdr->first_chunks = 0;
157 zhdr->last_chunks = 0;
158 INIT_LIST_HEAD(&zhdr->buddy);
159 INIT_LIST_HEAD(&zhdr->lru);
fa1f68cc 160 zhdr->under_reclaim = false;
4e2e2770
SJ
161 return zhdr;
162}
163
164/* Resets the struct page fields and frees the page */
165static void free_zbud_page(struct zbud_header *zhdr)
166{
167 __free_page(virt_to_page(zhdr));
168}
169
170/*
171 * Encodes the handle of a particular buddy within a zbud page
172 * Pool lock should be held as this function accesses first|last_chunks
173 */
174static unsigned long encode_handle(struct zbud_header *zhdr, enum buddy bud)
175{
176 unsigned long handle;
177
178 /*
179 * For now, the encoded handle is actually just the pointer to the data
180 * but this might not always be the case. A little information hiding.
181 * Add CHUNK_SIZE to the handle if it is the first allocation to jump
182 * over the zbud header in the first chunk.
183 */
184 handle = (unsigned long)zhdr;
185 if (bud == FIRST)
186 /* skip over zbud header */
187 handle += ZHDR_SIZE_ALIGNED;
188 else /* bud == LAST */
189 handle += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
190 return handle;
191}
192
193/* Returns the zbud page where a given handle is stored */
194static struct zbud_header *handle_to_zbud_header(unsigned long handle)
195{
196 return (struct zbud_header *)(handle & PAGE_MASK);
197}
198
199/* Returns the number of free chunks in a zbud page */
200static int num_free_chunks(struct zbud_header *zhdr)
201{
202 /*
203 * Rather than branch for different situations, just use the fact that
f203c3b3 204 * free buddies have a length of zero to simplify everything.
4e2e2770 205 */
f203c3b3 206 return NCHUNKS - zhdr->first_chunks - zhdr->last_chunks;
4e2e2770
SJ
207}
208
209/*****************
210 * API Functions
211*****************/
212/**
213 * zbud_create_pool() - create a new zbud pool
214 * @gfp: gfp flags when allocating the zbud pool structure
215 * @ops: user-defined operations for the zbud pool
216 *
217 * Return: pointer to the new zbud pool or NULL if the metadata allocation
218 * failed.
219 */
2a03085c 220static struct zbud_pool *zbud_create_pool(gfp_t gfp, const struct zbud_ops *ops)
4e2e2770
SJ
221{
222 struct zbud_pool *pool;
223 int i;
224
479305fd 225 pool = kzalloc(sizeof(struct zbud_pool), gfp);
4e2e2770
SJ
226 if (!pool)
227 return NULL;
228 spin_lock_init(&pool->lock);
229 for_each_unbuddied_list(i, 0)
230 INIT_LIST_HEAD(&pool->unbuddied[i]);
231 INIT_LIST_HEAD(&pool->buddied);
232 INIT_LIST_HEAD(&pool->lru);
233 pool->pages_nr = 0;
234 pool->ops = ops;
235 return pool;
236}
237
238/**
239 * zbud_destroy_pool() - destroys an existing zbud pool
240 * @pool: the zbud pool to be destroyed
241 *
242 * The pool should be emptied before this function is called.
243 */
2a03085c 244static void zbud_destroy_pool(struct zbud_pool *pool)
4e2e2770
SJ
245{
246 kfree(pool);
247}
248
249/**
250 * zbud_alloc() - allocates a region of a given size
251 * @pool: zbud pool from which to allocate
252 * @size: size in bytes of the desired allocation
253 * @gfp: gfp flags used if the pool needs to grow
254 * @handle: handle of the new allocation
255 *
256 * This function will attempt to find a free region in the pool large enough to
257 * satisfy the allocation request. A search of the unbuddied lists is
258 * performed first. If no suitable free region is found, then a new page is
259 * allocated and added to the pool to satisfy the request.
260 *
261 * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used
262 * as zbud pool pages.
263 *
eee87e17 264 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
4e2e2770
SJ
265 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
266 * a new page.
267 */
2a03085c 268static int zbud_alloc(struct zbud_pool *pool, size_t size, gfp_t gfp,
4e2e2770
SJ
269 unsigned long *handle)
270{
271 int chunks, i, freechunks;
272 struct zbud_header *zhdr = NULL;
273 enum buddy bud;
274 struct page *page;
275
50417c55 276 if (!size || (gfp & __GFP_HIGHMEM))
4e2e2770 277 return -EINVAL;
9d8c5b52 278 if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
4e2e2770
SJ
279 return -ENOSPC;
280 chunks = size_to_chunks(size);
281 spin_lock(&pool->lock);
282
283 /* First, try to find an unbuddied zbud page. */
4e2e2770
SJ
284 for_each_unbuddied_list(i, chunks) {
285 if (!list_empty(&pool->unbuddied[i])) {
286 zhdr = list_first_entry(&pool->unbuddied[i],
287 struct zbud_header, buddy);
288 list_del(&zhdr->buddy);
289 if (zhdr->first_chunks == 0)
290 bud = FIRST;
291 else
292 bud = LAST;
293 goto found;
294 }
295 }
296
297 /* Couldn't find unbuddied zbud page, create new one */
298 spin_unlock(&pool->lock);
299 page = alloc_page(gfp);
300 if (!page)
301 return -ENOMEM;
302 spin_lock(&pool->lock);
303 pool->pages_nr++;
304 zhdr = init_zbud_page(page);
305 bud = FIRST;
306
307found:
308 if (bud == FIRST)
309 zhdr->first_chunks = chunks;
310 else
311 zhdr->last_chunks = chunks;
312
313 if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0) {
314 /* Add to unbuddied list */
315 freechunks = num_free_chunks(zhdr);
316 list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
317 } else {
318 /* Add to buddied list */
319 list_add(&zhdr->buddy, &pool->buddied);
320 }
321
322 /* Add/move zbud page to beginning of LRU */
323 if (!list_empty(&zhdr->lru))
324 list_del(&zhdr->lru);
325 list_add(&zhdr->lru, &pool->lru);
326
327 *handle = encode_handle(zhdr, bud);
328 spin_unlock(&pool->lock);
329
330 return 0;
331}
332
333/**
334 * zbud_free() - frees the allocation associated with the given handle
335 * @pool: pool in which the allocation resided
336 * @handle: handle associated with the allocation returned by zbud_alloc()
337 *
338 * In the case that the zbud page in which the allocation resides is under
339 * reclaim, as indicated by the PG_reclaim flag being set, this function
340 * only sets the first|last_chunks to 0. The page is actually freed
341 * once both buddies are evicted (see zbud_reclaim_page() below).
342 */
2a03085c 343static void zbud_free(struct zbud_pool *pool, unsigned long handle)
4e2e2770
SJ
344{
345 struct zbud_header *zhdr;
346 int freechunks;
347
348 spin_lock(&pool->lock);
349 zhdr = handle_to_zbud_header(handle);
350
351 /* If first buddy, handle will be page aligned */
352 if ((handle - ZHDR_SIZE_ALIGNED) & ~PAGE_MASK)
353 zhdr->last_chunks = 0;
354 else
355 zhdr->first_chunks = 0;
356
357 if (zhdr->under_reclaim) {
358 /* zbud page is under reclaim, reclaim will free */
359 spin_unlock(&pool->lock);
360 return;
361 }
362
363 /* Remove from existing buddy list */
364 list_del(&zhdr->buddy);
365
366 if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
367 /* zbud page is empty, free */
368 list_del(&zhdr->lru);
369 free_zbud_page(zhdr);
370 pool->pages_nr--;
371 } else {
372 /* Add to unbuddied list */
373 freechunks = num_free_chunks(zhdr);
374 list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
375 }
376
377 spin_unlock(&pool->lock);
378}
379
4e2e2770
SJ
380/**
381 * zbud_reclaim_page() - evicts allocations from a pool page and frees it
382 * @pool: pool from which a page will attempt to be evicted
f144c390 383 * @retries: number of pages on the LRU list for which eviction will
4e2e2770
SJ
384 * be attempted before failing
385 *
386 * zbud reclaim is different from normal system reclaim in that the reclaim is
387 * done from the bottom, up. This is because only the bottom layer, zbud, has
388 * information on how the allocations are organized within each zbud page. This
389 * has the potential to create interesting locking situations between zbud and
390 * the user, however.
391 *
392 * To avoid these, this is how zbud_reclaim_page() should be called:
f144c390 393 *
4e2e2770
SJ
394 * The user detects a page should be reclaimed and calls zbud_reclaim_page().
395 * zbud_reclaim_page() will remove a zbud page from the pool LRU list and call
396 * the user-defined eviction handler with the pool and handle as arguments.
397 *
398 * If the handle can not be evicted, the eviction handler should return
399 * non-zero. zbud_reclaim_page() will add the zbud page back to the
400 * appropriate list and try the next zbud page on the LRU up to
401 * a user defined number of retries.
402 *
403 * If the handle is successfully evicted, the eviction handler should
404 * return 0 _and_ should have called zbud_free() on the handle. zbud_free()
405 * contains logic to delay freeing the page if the page is under reclaim,
406 * as indicated by the setting of the PG_reclaim flag on the underlying page.
407 *
408 * If all buddies in the zbud page are successfully evicted, then the
409 * zbud page can be freed.
410 *
411 * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are
412 * no pages to evict or an eviction handler is not registered, -EAGAIN if
413 * the retry limit was hit.
414 */
2a03085c 415static int zbud_reclaim_page(struct zbud_pool *pool, unsigned int retries)
4e2e2770
SJ
416{
417 int i, ret, freechunks;
418 struct zbud_header *zhdr;
419 unsigned long first_handle = 0, last_handle = 0;
420
421 spin_lock(&pool->lock);
422 if (!pool->ops || !pool->ops->evict || list_empty(&pool->lru) ||
423 retries == 0) {
424 spin_unlock(&pool->lock);
425 return -EINVAL;
426 }
427 for (i = 0; i < retries; i++) {
f58fb5e7 428 zhdr = list_last_entry(&pool->lru, struct zbud_header, lru);
4e2e2770
SJ
429 list_del(&zhdr->lru);
430 list_del(&zhdr->buddy);
431 /* Protect zbud page against free */
432 zhdr->under_reclaim = true;
433 /*
434 * We need encode the handles before unlocking, since we can
435 * race with free that will set (first|last)_chunks to 0
436 */
437 first_handle = 0;
438 last_handle = 0;
439 if (zhdr->first_chunks)
440 first_handle = encode_handle(zhdr, FIRST);
441 if (zhdr->last_chunks)
442 last_handle = encode_handle(zhdr, LAST);
443 spin_unlock(&pool->lock);
444
445 /* Issue the eviction callback(s) */
446 if (first_handle) {
447 ret = pool->ops->evict(pool, first_handle);
448 if (ret)
449 goto next;
450 }
451 if (last_handle) {
452 ret = pool->ops->evict(pool, last_handle);
453 if (ret)
454 goto next;
455 }
456next:
457 spin_lock(&pool->lock);
458 zhdr->under_reclaim = false;
459 if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
460 /*
461 * Both buddies are now free, free the zbud page and
462 * return success.
463 */
464 free_zbud_page(zhdr);
465 pool->pages_nr--;
466 spin_unlock(&pool->lock);
467 return 0;
468 } else if (zhdr->first_chunks == 0 ||
469 zhdr->last_chunks == 0) {
470 /* add to unbuddied list */
471 freechunks = num_free_chunks(zhdr);
472 list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
473 } else {
474 /* add to buddied list */
475 list_add(&zhdr->buddy, &pool->buddied);
476 }
477
478 /* add to beginning of LRU */
479 list_add(&zhdr->lru, &pool->lru);
480 }
481 spin_unlock(&pool->lock);
482 return -EAGAIN;
483}
484
485/**
486 * zbud_map() - maps the allocation associated with the given handle
487 * @pool: pool in which the allocation resides
488 * @handle: handle associated with the allocation to be mapped
489 *
490 * While trivial for zbud, the mapping functions for others allocators
491 * implementing this allocation API could have more complex information encoded
492 * in the handle and could create temporary mappings to make the data
493 * accessible to the user.
494 *
495 * Returns: a pointer to the mapped allocation
496 */
2a03085c 497static void *zbud_map(struct zbud_pool *pool, unsigned long handle)
4e2e2770
SJ
498{
499 return (void *)(handle);
500}
501
502/**
503 * zbud_unmap() - maps the allocation associated with the given handle
504 * @pool: pool in which the allocation resides
505 * @handle: handle associated with the allocation to be unmapped
506 */
2a03085c 507static void zbud_unmap(struct zbud_pool *pool, unsigned long handle)
4e2e2770
SJ
508{
509}
510
511/**
512 * zbud_get_pool_size() - gets the zbud pool size in pages
513 * @pool: pool whose size is being queried
514 *
515 * Returns: size in pages of the given pool. The pool lock need not be
516 * taken to access pages_nr.
517 */
2a03085c 518static u64 zbud_get_pool_size(struct zbud_pool *pool)
4e2e2770
SJ
519{
520 return pool->pages_nr;
521}
522
2a03085c
ML
523/*****************
524 * zpool
525 ****************/
526
527static int zbud_zpool_evict(struct zbud_pool *pool, unsigned long handle)
528{
529 if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict)
530 return pool->zpool_ops->evict(pool->zpool, handle);
531 else
532 return -ENOENT;
533}
534
535static const struct zbud_ops zbud_zpool_ops = {
536 .evict = zbud_zpool_evict
537};
538
539static void *zbud_zpool_create(const char *name, gfp_t gfp,
540 const struct zpool_ops *zpool_ops,
541 struct zpool *zpool)
542{
543 struct zbud_pool *pool;
544
545 pool = zbud_create_pool(gfp, zpool_ops ? &zbud_zpool_ops : NULL);
546 if (pool) {
547 pool->zpool = zpool;
548 pool->zpool_ops = zpool_ops;
549 }
550 return pool;
551}
552
553static void zbud_zpool_destroy(void *pool)
554{
555 zbud_destroy_pool(pool);
556}
557
558static int zbud_zpool_malloc(void *pool, size_t size, gfp_t gfp,
559 unsigned long *handle)
560{
561 return zbud_alloc(pool, size, gfp, handle);
562}
563static void zbud_zpool_free(void *pool, unsigned long handle)
564{
565 zbud_free(pool, handle);
566}
567
568static int zbud_zpool_shrink(void *pool, unsigned int pages,
569 unsigned int *reclaimed)
570{
571 unsigned int total = 0;
572 int ret = -EINVAL;
573
574 while (total < pages) {
575 ret = zbud_reclaim_page(pool, 8);
576 if (ret < 0)
577 break;
578 total++;
579 }
580
581 if (reclaimed)
582 *reclaimed = total;
583
584 return ret;
585}
586
587static void *zbud_zpool_map(void *pool, unsigned long handle,
588 enum zpool_mapmode mm)
589{
590 return zbud_map(pool, handle);
591}
592static void zbud_zpool_unmap(void *pool, unsigned long handle)
593{
594 zbud_unmap(pool, handle);
595}
596
597static u64 zbud_zpool_total_size(void *pool)
598{
599 return zbud_get_pool_size(pool) * PAGE_SIZE;
600}
601
602static struct zpool_driver zbud_zpool_driver = {
603 .type = "zbud",
604 .sleep_mapped = true,
605 .owner = THIS_MODULE,
606 .create = zbud_zpool_create,
607 .destroy = zbud_zpool_destroy,
608 .malloc = zbud_zpool_malloc,
609 .free = zbud_zpool_free,
610 .shrink = zbud_zpool_shrink,
611 .map = zbud_zpool_map,
612 .unmap = zbud_zpool_unmap,
613 .total_size = zbud_zpool_total_size,
614};
615
616MODULE_ALIAS("zpool-zbud");
617
4e2e2770
SJ
618static int __init init_zbud(void)
619{
620 /* Make sure the zbud header will fit in one chunk */
621 BUILD_BUG_ON(sizeof(struct zbud_header) > ZHDR_SIZE_ALIGNED);
622 pr_info("loaded\n");
c795779d 623
c795779d 624 zpool_register_driver(&zbud_zpool_driver);
c795779d 625
4e2e2770
SJ
626 return 0;
627}
628
629static void __exit exit_zbud(void)
630{
c795779d 631 zpool_unregister_driver(&zbud_zpool_driver);
4e2e2770
SJ
632 pr_info("unloaded\n");
633}
634
635module_init(init_zbud);
636module_exit(exit_zbud);
637
638MODULE_LICENSE("GPL");
68386da8 639MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
4e2e2770 640MODULE_DESCRIPTION("Buddy Allocator for Compressed Pages");