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