Linux Audio

Check our new training course

Loading...
v6.8
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * z3fold.c
   4 *
   5 * Author: Vitaly Wool <vitaly.wool@konsulko.com>
   6 * Copyright (C) 2016, Sony Mobile Communications Inc.
   7 *
   8 * This implementation is based on zbud written by Seth Jennings.
   9 *
  10 * z3fold is an special purpose allocator for storing compressed pages. It
  11 * can store up to three compressed pages per page which improves the
  12 * compression ratio of zbud while retaining its main concepts (e. g. always
  13 * storing an integral number of objects per page) and simplicity.
  14 * It still has simple and deterministic reclaim properties that make it
  15 * preferable to a higher density approach (with no requirement on integral
  16 * number of object per page) when reclaim is used.
  17 *
  18 * As in zbud, pages are divided into "chunks".  The size of the chunks is
  19 * fixed at compile time and is determined by NCHUNKS_ORDER below.
  20 *
  21 * z3fold doesn't export any API and is meant to be used via zpool API.
  22 */
  23
  24#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  25
  26#include <linux/atomic.h>
  27#include <linux/sched.h>
  28#include <linux/cpumask.h>
  29#include <linux/list.h>
  30#include <linux/mm.h>
  31#include <linux/module.h>
  32#include <linux/page-flags.h>
  33#include <linux/migrate.h>
  34#include <linux/node.h>
  35#include <linux/compaction.h>
  36#include <linux/percpu.h>
 
 
 
  37#include <linux/preempt.h>
  38#include <linux/workqueue.h>
  39#include <linux/slab.h>
  40#include <linux/spinlock.h>
  41#include <linux/zpool.h>
 
  42#include <linux/kmemleak.h>
  43
  44/*
  45 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
  46 * adjusting internal fragmentation.  It also determines the number of
  47 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
  48 * allocation granularity will be in chunks of size PAGE_SIZE/64. Some chunks
  49 * in the beginning of an allocated page are occupied by z3fold header, so
  50 * NCHUNKS will be calculated to 63 (or 62 in case CONFIG_DEBUG_SPINLOCK=y),
  51 * which shows the max number of free chunks in z3fold page, also there will
  52 * be 63, or 62, respectively, freelists per pool.
  53 */
  54#define NCHUNKS_ORDER	6
  55
  56#define CHUNK_SHIFT	(PAGE_SHIFT - NCHUNKS_ORDER)
  57#define CHUNK_SIZE	(1 << CHUNK_SHIFT)
  58#define ZHDR_SIZE_ALIGNED round_up(sizeof(struct z3fold_header), CHUNK_SIZE)
  59#define ZHDR_CHUNKS	(ZHDR_SIZE_ALIGNED >> CHUNK_SHIFT)
  60#define TOTAL_CHUNKS	(PAGE_SIZE >> CHUNK_SHIFT)
  61#define NCHUNKS		(TOTAL_CHUNKS - ZHDR_CHUNKS)
  62
  63#define BUDDY_MASK	(0x3)
  64#define BUDDY_SHIFT	2
  65#define SLOTS_ALIGN	(0x40)
  66
  67/*****************
  68 * Structures
  69*****************/
  70struct z3fold_pool;
 
 
 
  71
  72enum buddy {
  73	HEADLESS = 0,
  74	FIRST,
  75	MIDDLE,
  76	LAST,
  77	BUDDIES_MAX = LAST
  78};
  79
  80struct z3fold_buddy_slots {
  81	/*
  82	 * we are using BUDDY_MASK in handle_to_buddy etc. so there should
  83	 * be enough slots to hold all possible variants
  84	 */
  85	unsigned long slot[BUDDY_MASK + 1];
  86	unsigned long pool; /* back link */
  87	rwlock_t lock;
  88};
  89#define HANDLE_FLAG_MASK	(0x03)
  90
  91/*
  92 * struct z3fold_header - z3fold page metadata occupying first chunks of each
  93 *			z3fold page, except for HEADLESS pages
  94 * @buddy:		links the z3fold page into the relevant list in the
  95 *			pool
  96 * @page_lock:		per-page lock
  97 * @refcount:		reference count for the z3fold page
  98 * @work:		work_struct for page layout optimization
  99 * @slots:		pointer to the structure holding buddy slots
 100 * @pool:		pointer to the containing pool
 101 * @cpu:		CPU which this page "belongs" to
 102 * @first_chunks:	the size of the first buddy in chunks, 0 if free
 103 * @middle_chunks:	the size of the middle buddy in chunks, 0 if free
 104 * @last_chunks:	the size of the last buddy in chunks, 0 if free
 105 * @first_num:		the starting number (for the first handle)
 106 * @mapped_count:	the number of objects currently mapped
 107 */
 108struct z3fold_header {
 109	struct list_head buddy;
 110	spinlock_t page_lock;
 111	struct kref refcount;
 112	struct work_struct work;
 113	struct z3fold_buddy_slots *slots;
 114	struct z3fold_pool *pool;
 115	short cpu;
 116	unsigned short first_chunks;
 117	unsigned short middle_chunks;
 118	unsigned short last_chunks;
 119	unsigned short start_middle;
 120	unsigned short first_num:2;
 121	unsigned short mapped_count:2;
 122	unsigned short foreign_handles:2;
 123};
 124
 125/**
 126 * struct z3fold_pool - stores metadata for each z3fold pool
 127 * @name:	pool name
 128 * @lock:	protects pool unbuddied lists
 129 * @stale_lock:	protects pool stale page list
 130 * @unbuddied:	per-cpu array of lists tracking z3fold pages that contain 2-
 131 *		buddies; the list each z3fold page is added to depends on
 132 *		the size of its free region.
 
 
 133 * @stale:	list of pages marked for freeing
 134 * @pages_nr:	number of z3fold pages in the pool.
 135 * @c_handle:	cache for z3fold_buddy_slots allocation
 
 
 136 * @compact_wq:	workqueue for page layout background optimization
 137 * @release_wq:	workqueue for safe page release
 138 * @work:	work_struct for safe page release
 
 139 *
 140 * This structure is allocated at pool creation time and maintains metadata
 141 * pertaining to a particular z3fold pool.
 142 */
 143struct z3fold_pool {
 144	const char *name;
 145	spinlock_t lock;
 146	spinlock_t stale_lock;
 147	struct list_head *unbuddied;
 
 148	struct list_head stale;
 149	atomic64_t pages_nr;
 150	struct kmem_cache *c_handle;
 
 
 
 151	struct workqueue_struct *compact_wq;
 152	struct workqueue_struct *release_wq;
 153	struct work_struct work;
 
 154};
 155
 156/*
 157 * Internal z3fold page flags
 158 */
 159enum z3fold_page_flags {
 160	PAGE_HEADLESS = 0,
 161	MIDDLE_CHUNK_MAPPED,
 162	NEEDS_COMPACTING,
 163	PAGE_STALE,
 164	PAGE_CLAIMED, /* by either reclaim or free */
 165	PAGE_MIGRATED, /* page is migrated and soon to be released */
 166};
 167
 168/*
 169 * handle flags, go under HANDLE_FLAG_MASK
 170 */
 171enum z3fold_handle_flags {
 172	HANDLES_NOFREE = 0,
 173};
 174
 175/*
 176 * Forward declarations
 177 */
 178static struct z3fold_header *__z3fold_alloc(struct z3fold_pool *, size_t, bool);
 179static void compact_page_work(struct work_struct *w);
 180
 181/*****************
 182 * Helpers
 183*****************/
 184
 185/* Converts an allocation size in bytes to size in z3fold chunks */
 186static int size_to_chunks(size_t size)
 187{
 188	return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
 189}
 190
 191#define for_each_unbuddied_list(_iter, _begin) \
 192	for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
 193
 194static inline struct z3fold_buddy_slots *alloc_slots(struct z3fold_pool *pool,
 195							gfp_t gfp)
 196{
 197	struct z3fold_buddy_slots *slots = kmem_cache_zalloc(pool->c_handle,
 198							     gfp);
 
 
 199
 200	if (slots) {
 201		/* It will be freed separately in free_handle(). */
 202		kmemleak_not_leak(slots);
 
 203		slots->pool = (unsigned long)pool;
 204		rwlock_init(&slots->lock);
 205	}
 206
 207	return slots;
 208}
 209
 210static inline struct z3fold_pool *slots_to_pool(struct z3fold_buddy_slots *s)
 211{
 212	return (struct z3fold_pool *)(s->pool & ~HANDLE_FLAG_MASK);
 213}
 214
 215static inline struct z3fold_buddy_slots *handle_to_slots(unsigned long handle)
 216{
 217	return (struct z3fold_buddy_slots *)(handle & ~(SLOTS_ALIGN - 1));
 218}
 219
 220/* Lock a z3fold page */
 221static inline void z3fold_page_lock(struct z3fold_header *zhdr)
 222{
 223	spin_lock(&zhdr->page_lock);
 224}
 225
 226/* Try to lock a z3fold page */
 227static inline int z3fold_page_trylock(struct z3fold_header *zhdr)
 228{
 229	return spin_trylock(&zhdr->page_lock);
 230}
 231
 232/* Unlock a z3fold page */
 233static inline void z3fold_page_unlock(struct z3fold_header *zhdr)
 234{
 235	spin_unlock(&zhdr->page_lock);
 236}
 237
 238/* return locked z3fold page if it's not headless */
 239static inline struct z3fold_header *get_z3fold_header(unsigned long handle)
 
 240{
 241	struct z3fold_buddy_slots *slots;
 242	struct z3fold_header *zhdr;
 243	int locked = 0;
 244
 245	if (!(handle & (1 << PAGE_HEADLESS))) {
 246		slots = handle_to_slots(handle);
 247		do {
 248			unsigned long addr;
 249
 250			read_lock(&slots->lock);
 251			addr = *(unsigned long *)handle;
 252			zhdr = (struct z3fold_header *)(addr & PAGE_MASK);
 253			locked = z3fold_page_trylock(zhdr);
 
 254			read_unlock(&slots->lock);
 255			if (locked) {
 256				struct page *page = virt_to_page(zhdr);
 257
 258				if (!test_bit(PAGE_MIGRATED, &page->private))
 259					break;
 260				z3fold_page_unlock(zhdr);
 261			}
 262			cpu_relax();
 263		} while (true);
 264	} else {
 265		zhdr = (struct z3fold_header *)(handle & PAGE_MASK);
 266	}
 267
 268	return zhdr;
 269}
 270
 
 
 
 
 
 
 
 
 
 
 
 
 271static inline void put_z3fold_header(struct z3fold_header *zhdr)
 272{
 273	struct page *page = virt_to_page(zhdr);
 274
 275	if (!test_bit(PAGE_HEADLESS, &page->private))
 276		z3fold_page_unlock(zhdr);
 277}
 278
 279static inline void free_handle(unsigned long handle, struct z3fold_header *zhdr)
 280{
 281	struct z3fold_buddy_slots *slots;
 
 282	int i;
 283	bool is_free;
 284
 
 
 
 285	if (WARN_ON(*(unsigned long *)handle == 0))
 286		return;
 287
 
 288	slots = handle_to_slots(handle);
 289	write_lock(&slots->lock);
 290	*(unsigned long *)handle = 0;
 291
 292	if (test_bit(HANDLES_NOFREE, &slots->pool)) {
 293		write_unlock(&slots->lock);
 294		return; /* simple case, nothing else to do */
 295	}
 296
 297	if (zhdr->slots != slots)
 298		zhdr->foreign_handles--;
 299
 300	is_free = true;
 
 
 
 
 301	for (i = 0; i <= BUDDY_MASK; i++) {
 302		if (slots->slot[i]) {
 303			is_free = false;
 304			break;
 305		}
 306	}
 307	write_unlock(&slots->lock);
 308
 309	if (is_free) {
 310		struct z3fold_pool *pool = slots_to_pool(slots);
 311
 312		if (zhdr->slots == slots)
 313			zhdr->slots = NULL;
 314		kmem_cache_free(pool->c_handle, slots);
 315	}
 316}
 317
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 318/* Initializes the z3fold header of a newly allocated z3fold page */
 319static struct z3fold_header *init_z3fold_page(struct page *page, bool headless,
 320					struct z3fold_pool *pool, gfp_t gfp)
 321{
 322	struct z3fold_header *zhdr = page_address(page);
 323	struct z3fold_buddy_slots *slots;
 324
 
 325	clear_bit(PAGE_HEADLESS, &page->private);
 326	clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
 327	clear_bit(NEEDS_COMPACTING, &page->private);
 328	clear_bit(PAGE_STALE, &page->private);
 329	clear_bit(PAGE_CLAIMED, &page->private);
 330	clear_bit(PAGE_MIGRATED, &page->private);
 331	if (headless)
 332		return zhdr;
 333
 334	slots = alloc_slots(pool, gfp);
 335	if (!slots)
 336		return NULL;
 337
 338	memset(zhdr, 0, sizeof(*zhdr));
 339	spin_lock_init(&zhdr->page_lock);
 340	kref_init(&zhdr->refcount);
 
 
 
 
 
 341	zhdr->cpu = -1;
 
 
 342	zhdr->slots = slots;
 343	zhdr->pool = pool;
 344	INIT_LIST_HEAD(&zhdr->buddy);
 345	INIT_WORK(&zhdr->work, compact_page_work);
 346	return zhdr;
 347}
 348
 349/* Resets the struct page fields and frees the page */
 350static void free_z3fold_page(struct page *page, bool headless)
 351{
 352	if (!headless) {
 353		lock_page(page);
 354		__ClearPageMovable(page);
 355		unlock_page(page);
 356	}
 
 357	__free_page(page);
 358}
 359
 360/* Helper function to build the index */
 361static inline int __idx(struct z3fold_header *zhdr, enum buddy bud)
 362{
 363	return (bud + zhdr->first_num) & BUDDY_MASK;
 364}
 365
 366/*
 367 * Encodes the handle of a particular buddy within a z3fold page
 368 * Pool lock should be held as this function accesses first_num
 369 */
 370static unsigned long __encode_handle(struct z3fold_header *zhdr,
 371				struct z3fold_buddy_slots *slots,
 372				enum buddy bud)
 373{
 374	unsigned long h = (unsigned long)zhdr;
 375	int idx = 0;
 376
 377	/*
 378	 * For a headless page, its handle is its pointer with the extra
 379	 * PAGE_HEADLESS bit set
 380	 */
 381	if (bud == HEADLESS)
 382		return h | (1 << PAGE_HEADLESS);
 383
 384	/* otherwise, return pointer to encoded handle */
 385	idx = __idx(zhdr, bud);
 386	h += idx;
 387	if (bud == LAST)
 388		h |= (zhdr->last_chunks << BUDDY_SHIFT);
 389
 390	write_lock(&slots->lock);
 391	slots->slot[idx] = h;
 392	write_unlock(&slots->lock);
 393	return (unsigned long)&slots->slot[idx];
 394}
 395
 396static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud)
 397{
 398	return __encode_handle(zhdr, zhdr->slots, bud);
 399}
 400
 401/* only for LAST bud, returns zero otherwise */
 402static unsigned short handle_to_chunks(unsigned long handle)
 403{
 404	struct z3fold_buddy_slots *slots = handle_to_slots(handle);
 405	unsigned long addr;
 406
 407	read_lock(&slots->lock);
 408	addr = *(unsigned long *)handle;
 409	read_unlock(&slots->lock);
 410	return (addr & ~PAGE_MASK) >> BUDDY_SHIFT;
 411}
 412
 413/*
 414 * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle
 415 *  but that doesn't matter. because the masking will result in the
 416 *  correct buddy number.
 417 */
 418static enum buddy handle_to_buddy(unsigned long handle)
 419{
 420	struct z3fold_header *zhdr;
 421	struct z3fold_buddy_slots *slots = handle_to_slots(handle);
 422	unsigned long addr;
 423
 424	read_lock(&slots->lock);
 425	WARN_ON(handle & (1 << PAGE_HEADLESS));
 426	addr = *(unsigned long *)handle;
 427	read_unlock(&slots->lock);
 428	zhdr = (struct z3fold_header *)(addr & PAGE_MASK);
 429	return (addr - zhdr->first_num) & BUDDY_MASK;
 430}
 431
 432static inline struct z3fold_pool *zhdr_to_pool(struct z3fold_header *zhdr)
 433{
 434	return zhdr->pool;
 435}
 436
 437static void __release_z3fold_page(struct z3fold_header *zhdr, bool locked)
 438{
 439	struct page *page = virt_to_page(zhdr);
 440	struct z3fold_pool *pool = zhdr_to_pool(zhdr);
 
 
 441
 442	WARN_ON(!list_empty(&zhdr->buddy));
 443	set_bit(PAGE_STALE, &page->private);
 444	clear_bit(NEEDS_COMPACTING, &page->private);
 445	spin_lock(&pool->lock);
 
 
 446	spin_unlock(&pool->lock);
 447
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 448	if (locked)
 449		z3fold_page_unlock(zhdr);
 450
 451	spin_lock(&pool->stale_lock);
 452	list_add(&zhdr->buddy, &pool->stale);
 453	queue_work(pool->release_wq, &pool->work);
 454	spin_unlock(&pool->stale_lock);
 
 455
 456	atomic64_dec(&pool->pages_nr);
 
 
 
 
 
 457}
 458
 459static void release_z3fold_page_locked(struct kref *ref)
 460{
 461	struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
 462						refcount);
 463	WARN_ON(z3fold_page_trylock(zhdr));
 464	__release_z3fold_page(zhdr, true);
 465}
 466
 467static void release_z3fold_page_locked_list(struct kref *ref)
 468{
 469	struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
 470					       refcount);
 471	struct z3fold_pool *pool = zhdr_to_pool(zhdr);
 472
 473	spin_lock(&pool->lock);
 474	list_del_init(&zhdr->buddy);
 475	spin_unlock(&pool->lock);
 476
 477	WARN_ON(z3fold_page_trylock(zhdr));
 478	__release_z3fold_page(zhdr, true);
 479}
 480
 481static inline int put_z3fold_locked(struct z3fold_header *zhdr)
 482{
 483	return kref_put(&zhdr->refcount, release_z3fold_page_locked);
 484}
 485
 486static inline int put_z3fold_locked_list(struct z3fold_header *zhdr)
 487{
 488	return kref_put(&zhdr->refcount, release_z3fold_page_locked_list);
 489}
 490
 491static void free_pages_work(struct work_struct *w)
 492{
 493	struct z3fold_pool *pool = container_of(w, struct z3fold_pool, work);
 494
 495	spin_lock(&pool->stale_lock);
 496	while (!list_empty(&pool->stale)) {
 497		struct z3fold_header *zhdr = list_first_entry(&pool->stale,
 498						struct z3fold_header, buddy);
 499		struct page *page = virt_to_page(zhdr);
 500
 501		list_del(&zhdr->buddy);
 502		if (WARN_ON(!test_bit(PAGE_STALE, &page->private)))
 503			continue;
 504		spin_unlock(&pool->stale_lock);
 505		cancel_work_sync(&zhdr->work);
 506		free_z3fold_page(page, false);
 507		cond_resched();
 508		spin_lock(&pool->stale_lock);
 509	}
 510	spin_unlock(&pool->stale_lock);
 511}
 512
 513/*
 514 * Returns the number of free chunks in a z3fold page.
 515 * NB: can't be used with HEADLESS pages.
 516 */
 517static int num_free_chunks(struct z3fold_header *zhdr)
 518{
 519	int nfree;
 520	/*
 521	 * If there is a middle object, pick up the bigger free space
 522	 * either before or after it. Otherwise just subtract the number
 523	 * of chunks occupied by the first and the last objects.
 524	 */
 525	if (zhdr->middle_chunks != 0) {
 526		int nfree_before = zhdr->first_chunks ?
 527			0 : zhdr->start_middle - ZHDR_CHUNKS;
 528		int nfree_after = zhdr->last_chunks ?
 529			0 : TOTAL_CHUNKS -
 530				(zhdr->start_middle + zhdr->middle_chunks);
 531		nfree = max(nfree_before, nfree_after);
 532	} else
 533		nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks;
 534	return nfree;
 535}
 536
 537/* Add to the appropriate unbuddied list */
 538static inline void add_to_unbuddied(struct z3fold_pool *pool,
 539				struct z3fold_header *zhdr)
 540{
 541	if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 ||
 542			zhdr->middle_chunks == 0) {
 543		struct list_head *unbuddied;
 544		int freechunks = num_free_chunks(zhdr);
 545
 546		migrate_disable();
 547		unbuddied = this_cpu_ptr(pool->unbuddied);
 548		spin_lock(&pool->lock);
 549		list_add(&zhdr->buddy, &unbuddied[freechunks]);
 550		spin_unlock(&pool->lock);
 551		zhdr->cpu = smp_processor_id();
 552		migrate_enable();
 553	}
 554}
 555
 556static inline enum buddy get_free_buddy(struct z3fold_header *zhdr, int chunks)
 557{
 558	enum buddy bud = HEADLESS;
 559
 560	if (zhdr->middle_chunks) {
 561		if (!zhdr->first_chunks &&
 562		    chunks <= zhdr->start_middle - ZHDR_CHUNKS)
 563			bud = FIRST;
 564		else if (!zhdr->last_chunks)
 565			bud = LAST;
 566	} else {
 567		if (!zhdr->first_chunks)
 568			bud = FIRST;
 569		else if (!zhdr->last_chunks)
 570			bud = LAST;
 571		else
 572			bud = MIDDLE;
 573	}
 574
 575	return bud;
 576}
 577
 578static inline void *mchunk_memmove(struct z3fold_header *zhdr,
 579				unsigned short dst_chunk)
 580{
 581	void *beg = zhdr;
 582	return memmove(beg + (dst_chunk << CHUNK_SHIFT),
 583		       beg + (zhdr->start_middle << CHUNK_SHIFT),
 584		       zhdr->middle_chunks << CHUNK_SHIFT);
 585}
 586
 587static inline bool buddy_single(struct z3fold_header *zhdr)
 588{
 589	return !((zhdr->first_chunks && zhdr->middle_chunks) ||
 590			(zhdr->first_chunks && zhdr->last_chunks) ||
 591			(zhdr->middle_chunks && zhdr->last_chunks));
 592}
 593
 594static struct z3fold_header *compact_single_buddy(struct z3fold_header *zhdr)
 595{
 596	struct z3fold_pool *pool = zhdr_to_pool(zhdr);
 597	void *p = zhdr;
 598	unsigned long old_handle = 0;
 599	size_t sz = 0;
 600	struct z3fold_header *new_zhdr = NULL;
 601	int first_idx = __idx(zhdr, FIRST);
 602	int middle_idx = __idx(zhdr, MIDDLE);
 603	int last_idx = __idx(zhdr, LAST);
 604	unsigned short *moved_chunks = NULL;
 605
 606	/*
 607	 * No need to protect slots here -- all the slots are "local" and
 608	 * the page lock is already taken
 609	 */
 610	if (zhdr->first_chunks && zhdr->slots->slot[first_idx]) {
 611		p += ZHDR_SIZE_ALIGNED;
 612		sz = zhdr->first_chunks << CHUNK_SHIFT;
 613		old_handle = (unsigned long)&zhdr->slots->slot[first_idx];
 614		moved_chunks = &zhdr->first_chunks;
 615	} else if (zhdr->middle_chunks && zhdr->slots->slot[middle_idx]) {
 616		p += zhdr->start_middle << CHUNK_SHIFT;
 617		sz = zhdr->middle_chunks << CHUNK_SHIFT;
 618		old_handle = (unsigned long)&zhdr->slots->slot[middle_idx];
 619		moved_chunks = &zhdr->middle_chunks;
 620	} else if (zhdr->last_chunks && zhdr->slots->slot[last_idx]) {
 621		p += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
 622		sz = zhdr->last_chunks << CHUNK_SHIFT;
 623		old_handle = (unsigned long)&zhdr->slots->slot[last_idx];
 624		moved_chunks = &zhdr->last_chunks;
 625	}
 626
 627	if (sz > 0) {
 628		enum buddy new_bud = HEADLESS;
 629		short chunks = size_to_chunks(sz);
 630		void *q;
 631
 632		new_zhdr = __z3fold_alloc(pool, sz, false);
 633		if (!new_zhdr)
 634			return NULL;
 635
 636		if (WARN_ON(new_zhdr == zhdr))
 637			goto out_fail;
 638
 639		new_bud = get_free_buddy(new_zhdr, chunks);
 
 
 
 
 
 
 
 
 
 
 
 640		q = new_zhdr;
 641		switch (new_bud) {
 642		case FIRST:
 643			new_zhdr->first_chunks = chunks;
 644			q += ZHDR_SIZE_ALIGNED;
 645			break;
 646		case MIDDLE:
 647			new_zhdr->middle_chunks = chunks;
 648			new_zhdr->start_middle =
 649				new_zhdr->first_chunks + ZHDR_CHUNKS;
 650			q += new_zhdr->start_middle << CHUNK_SHIFT;
 651			break;
 652		case LAST:
 653			new_zhdr->last_chunks = chunks;
 654			q += PAGE_SIZE - (new_zhdr->last_chunks << CHUNK_SHIFT);
 655			break;
 656		default:
 657			goto out_fail;
 658		}
 659		new_zhdr->foreign_handles++;
 660		memcpy(q, p, sz);
 661		write_lock(&zhdr->slots->lock);
 662		*(unsigned long *)old_handle = (unsigned long)new_zhdr +
 663			__idx(new_zhdr, new_bud);
 664		if (new_bud == LAST)
 665			*(unsigned long *)old_handle |=
 666					(new_zhdr->last_chunks << BUDDY_SHIFT);
 667		write_unlock(&zhdr->slots->lock);
 668		add_to_unbuddied(pool, new_zhdr);
 669		z3fold_page_unlock(new_zhdr);
 670
 671		*moved_chunks = 0;
 672	}
 673
 674	return new_zhdr;
 675
 676out_fail:
 677	if (new_zhdr && !put_z3fold_locked(new_zhdr)) {
 678		add_to_unbuddied(pool, new_zhdr);
 679		z3fold_page_unlock(new_zhdr);
 
 
 
 
 680	}
 681	return NULL;
 682
 683}
 684
 685#define BIG_CHUNK_GAP	3
 686/* Has to be called with lock held */
 687static int z3fold_compact_page(struct z3fold_header *zhdr)
 688{
 689	struct page *page = virt_to_page(zhdr);
 690
 691	if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private))
 692		return 0; /* can't move middle chunk, it's used */
 693
 694	if (unlikely(PageIsolated(page)))
 695		return 0;
 696
 697	if (zhdr->middle_chunks == 0)
 698		return 0; /* nothing to compact */
 699
 700	if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
 701		/* move to the beginning */
 702		mchunk_memmove(zhdr, ZHDR_CHUNKS);
 703		zhdr->first_chunks = zhdr->middle_chunks;
 704		zhdr->middle_chunks = 0;
 705		zhdr->start_middle = 0;
 706		zhdr->first_num++;
 707		return 1;
 708	}
 709
 710	/*
 711	 * moving data is expensive, so let's only do that if
 712	 * there's substantial gain (at least BIG_CHUNK_GAP chunks)
 713	 */
 714	if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 &&
 715	    zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >=
 716			BIG_CHUNK_GAP) {
 717		mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS);
 718		zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
 719		return 1;
 720	} else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 &&
 721		   TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle
 722					+ zhdr->middle_chunks) >=
 723			BIG_CHUNK_GAP) {
 724		unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks -
 725			zhdr->middle_chunks;
 726		mchunk_memmove(zhdr, new_start);
 727		zhdr->start_middle = new_start;
 728		return 1;
 729	}
 730
 731	return 0;
 732}
 733
 734static void do_compact_page(struct z3fold_header *zhdr, bool locked)
 735{
 736	struct z3fold_pool *pool = zhdr_to_pool(zhdr);
 737	struct page *page;
 738
 739	page = virt_to_page(zhdr);
 740	if (locked)
 741		WARN_ON(z3fold_page_trylock(zhdr));
 742	else
 743		z3fold_page_lock(zhdr);
 744	if (WARN_ON(!test_and_clear_bit(NEEDS_COMPACTING, &page->private))) {
 745		z3fold_page_unlock(zhdr);
 746		return;
 747	}
 748	spin_lock(&pool->lock);
 749	list_del_init(&zhdr->buddy);
 750	spin_unlock(&pool->lock);
 751
 752	if (put_z3fold_locked(zhdr))
 
 753		return;
 
 754
 755	if (test_bit(PAGE_STALE, &page->private) ||
 756	    test_and_set_bit(PAGE_CLAIMED, &page->private)) {
 
 757		z3fold_page_unlock(zhdr);
 758		return;
 759	}
 760
 761	if (!zhdr->foreign_handles && buddy_single(zhdr) &&
 762	    zhdr->mapped_count == 0 && compact_single_buddy(zhdr)) {
 763		if (!put_z3fold_locked(zhdr)) {
 764			clear_bit(PAGE_CLAIMED, &page->private);
 
 765			z3fold_page_unlock(zhdr);
 766		}
 767		return;
 768	}
 769
 770	z3fold_compact_page(zhdr);
 771	add_to_unbuddied(pool, zhdr);
 772	clear_bit(PAGE_CLAIMED, &page->private);
 773	z3fold_page_unlock(zhdr);
 774}
 775
 776static void compact_page_work(struct work_struct *w)
 777{
 778	struct z3fold_header *zhdr = container_of(w, struct z3fold_header,
 779						work);
 780
 781	do_compact_page(zhdr, false);
 782}
 783
 784/* returns _locked_ z3fold page header or NULL */
 785static inline struct z3fold_header *__z3fold_alloc(struct z3fold_pool *pool,
 786						size_t size, bool can_sleep)
 787{
 788	struct z3fold_header *zhdr = NULL;
 789	struct page *page;
 790	struct list_head *unbuddied;
 791	int chunks = size_to_chunks(size), i;
 792
 793lookup:
 794	migrate_disable();
 795	/* First, try to find an unbuddied z3fold page. */
 796	unbuddied = this_cpu_ptr(pool->unbuddied);
 797	for_each_unbuddied_list(i, chunks) {
 798		struct list_head *l = &unbuddied[i];
 799
 800		zhdr = list_first_entry_or_null(READ_ONCE(l),
 801					struct z3fold_header, buddy);
 802
 803		if (!zhdr)
 804			continue;
 805
 806		/* Re-check under lock. */
 807		spin_lock(&pool->lock);
 
 808		if (unlikely(zhdr != list_first_entry(READ_ONCE(l),
 809						struct z3fold_header, buddy)) ||
 810		    !z3fold_page_trylock(zhdr)) {
 811			spin_unlock(&pool->lock);
 812			zhdr = NULL;
 813			migrate_enable();
 814			if (can_sleep)
 815				cond_resched();
 816			goto lookup;
 817		}
 818		list_del_init(&zhdr->buddy);
 819		zhdr->cpu = -1;
 820		spin_unlock(&pool->lock);
 821
 822		page = virt_to_page(zhdr);
 823		if (test_bit(NEEDS_COMPACTING, &page->private) ||
 824		    test_bit(PAGE_CLAIMED, &page->private)) {
 825			z3fold_page_unlock(zhdr);
 826			zhdr = NULL;
 827			migrate_enable();
 828			if (can_sleep)
 829				cond_resched();
 830			goto lookup;
 831		}
 832
 833		/*
 834		 * this page could not be removed from its unbuddied
 835		 * list while pool lock was held, and then we've taken
 836		 * page lock so kref_put could not be called before
 837		 * we got here, so it's safe to just call kref_get()
 838		 */
 839		kref_get(&zhdr->refcount);
 840		break;
 841	}
 842	migrate_enable();
 843
 844	if (!zhdr) {
 845		int cpu;
 846
 847		/* look for _exact_ match on other cpus' lists */
 848		for_each_online_cpu(cpu) {
 849			struct list_head *l;
 850
 851			unbuddied = per_cpu_ptr(pool->unbuddied, cpu);
 852			spin_lock(&pool->lock);
 853			l = &unbuddied[chunks];
 854
 855			zhdr = list_first_entry_or_null(READ_ONCE(l),
 856						struct z3fold_header, buddy);
 857
 858			if (!zhdr || !z3fold_page_trylock(zhdr)) {
 859				spin_unlock(&pool->lock);
 860				zhdr = NULL;
 861				continue;
 862			}
 863			list_del_init(&zhdr->buddy);
 864			zhdr->cpu = -1;
 865			spin_unlock(&pool->lock);
 866
 867			page = virt_to_page(zhdr);
 868			if (test_bit(NEEDS_COMPACTING, &page->private) ||
 869			    test_bit(PAGE_CLAIMED, &page->private)) {
 870				z3fold_page_unlock(zhdr);
 871				zhdr = NULL;
 872				if (can_sleep)
 873					cond_resched();
 874				continue;
 875			}
 876			kref_get(&zhdr->refcount);
 877			break;
 878		}
 879	}
 880
 881	if (zhdr && !zhdr->slots) {
 882		zhdr->slots = alloc_slots(pool, GFP_ATOMIC);
 883		if (!zhdr->slots)
 884			goto out_fail;
 885	}
 886	return zhdr;
 887
 888out_fail:
 889	if (!put_z3fold_locked(zhdr)) {
 890		add_to_unbuddied(pool, zhdr);
 891		z3fold_page_unlock(zhdr);
 892	}
 893	return NULL;
 894}
 895
 896/*
 897 * API Functions
 898 */
 899
 900/**
 901 * z3fold_create_pool() - create a new z3fold pool
 902 * @name:	pool name
 903 * @gfp:	gfp flags when allocating the z3fold pool structure
 
 904 *
 905 * Return: pointer to the new z3fold pool or NULL if the metadata allocation
 906 * failed.
 907 */
 908static struct z3fold_pool *z3fold_create_pool(const char *name, gfp_t gfp)
 
 909{
 910	struct z3fold_pool *pool = NULL;
 911	int i, cpu;
 912
 913	pool = kzalloc(sizeof(struct z3fold_pool), gfp);
 914	if (!pool)
 915		goto out;
 916	pool->c_handle = kmem_cache_create("z3fold_handle",
 917				sizeof(struct z3fold_buddy_slots),
 918				SLOTS_ALIGN, 0, NULL);
 919	if (!pool->c_handle)
 920		goto out_c;
 921	spin_lock_init(&pool->lock);
 922	spin_lock_init(&pool->stale_lock);
 923	pool->unbuddied = __alloc_percpu(sizeof(struct list_head) * NCHUNKS,
 924					 __alignof__(struct list_head));
 925	if (!pool->unbuddied)
 926		goto out_pool;
 927	for_each_possible_cpu(cpu) {
 928		struct list_head *unbuddied =
 929				per_cpu_ptr(pool->unbuddied, cpu);
 930		for_each_unbuddied_list(i, 0)
 931			INIT_LIST_HEAD(&unbuddied[i]);
 932	}
 
 933	INIT_LIST_HEAD(&pool->stale);
 934	atomic64_set(&pool->pages_nr, 0);
 935	pool->name = name;
 936	pool->compact_wq = create_singlethread_workqueue(pool->name);
 937	if (!pool->compact_wq)
 938		goto out_unbuddied;
 939	pool->release_wq = create_singlethread_workqueue(pool->name);
 940	if (!pool->release_wq)
 941		goto out_wq;
 
 
 942	INIT_WORK(&pool->work, free_pages_work);
 
 943	return pool;
 944
 
 
 945out_wq:
 946	destroy_workqueue(pool->compact_wq);
 947out_unbuddied:
 948	free_percpu(pool->unbuddied);
 949out_pool:
 950	kmem_cache_destroy(pool->c_handle);
 951out_c:
 952	kfree(pool);
 953out:
 954	return NULL;
 955}
 956
 957/**
 958 * z3fold_destroy_pool() - destroys an existing z3fold pool
 959 * @pool:	the z3fold pool to be destroyed
 960 *
 961 * The pool should be emptied before this function is called.
 962 */
 963static void z3fold_destroy_pool(struct z3fold_pool *pool)
 964{
 965	kmem_cache_destroy(pool->c_handle);
 966
 967	/*
 968	 * We need to destroy pool->compact_wq before pool->release_wq,
 969	 * as any pending work on pool->compact_wq will call
 970	 * queue_work(pool->release_wq, &pool->work).
 971	 *
 972	 * There are still outstanding pages until both workqueues are drained,
 973	 * so we cannot unregister migration until then.
 974	 */
 975
 976	destroy_workqueue(pool->compact_wq);
 977	destroy_workqueue(pool->release_wq);
 978	free_percpu(pool->unbuddied);
 979	kfree(pool);
 980}
 981
 982static const struct movable_operations z3fold_mops;
 983
 984/**
 985 * z3fold_alloc() - allocates a region of a given size
 986 * @pool:	z3fold pool from which to allocate
 987 * @size:	size in bytes of the desired allocation
 988 * @gfp:	gfp flags used if the pool needs to grow
 989 * @handle:	handle of the new allocation
 990 *
 991 * This function will attempt to find a free region in the pool large enough to
 992 * satisfy the allocation request.  A search of the unbuddied lists is
 993 * performed first. If no suitable free region is found, then a new page is
 994 * allocated and added to the pool to satisfy the request.
 995 *
 
 
 
 996 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
 997 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
 998 * a new page.
 999 */
1000static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
1001			unsigned long *handle)
1002{
1003	int chunks = size_to_chunks(size);
1004	struct z3fold_header *zhdr = NULL;
1005	struct page *page = NULL;
1006	enum buddy bud;
1007	bool can_sleep = gfpflags_allow_blocking(gfp);
1008
1009	if (!size || (gfp & __GFP_HIGHMEM))
1010		return -EINVAL;
1011
1012	if (size > PAGE_SIZE)
1013		return -ENOSPC;
1014
1015	if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
1016		bud = HEADLESS;
1017	else {
1018retry:
1019		zhdr = __z3fold_alloc(pool, size, can_sleep);
1020		if (zhdr) {
1021			bud = get_free_buddy(zhdr, chunks);
1022			if (bud == HEADLESS) {
1023				if (!put_z3fold_locked(zhdr))
 
 
 
 
 
 
 
 
 
 
 
 
1024					z3fold_page_unlock(zhdr);
1025				pr_err("No free chunks in unbuddied\n");
1026				WARN_ON(1);
1027				goto retry;
1028			}
1029			page = virt_to_page(zhdr);
1030			goto found;
1031		}
1032		bud = FIRST;
1033	}
1034
1035	page = alloc_page(gfp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1036	if (!page)
1037		return -ENOMEM;
1038
1039	zhdr = init_z3fold_page(page, bud == HEADLESS, pool, gfp);
1040	if (!zhdr) {
1041		__free_page(page);
1042		return -ENOMEM;
1043	}
1044	atomic64_inc(&pool->pages_nr);
1045
1046	if (bud == HEADLESS) {
1047		set_bit(PAGE_HEADLESS, &page->private);
1048		goto headless;
1049	}
1050	if (can_sleep) {
1051		lock_page(page);
1052		__SetPageMovable(page, &z3fold_mops);
1053		unlock_page(page);
1054	} else {
1055		WARN_ON(!trylock_page(page));
1056		__SetPageMovable(page, &z3fold_mops);
1057		unlock_page(page);
 
1058	}
1059	z3fold_page_lock(zhdr);
1060
1061found:
1062	if (bud == FIRST)
1063		zhdr->first_chunks = chunks;
1064	else if (bud == LAST)
1065		zhdr->last_chunks = chunks;
1066	else {
1067		zhdr->middle_chunks = chunks;
1068		zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
1069	}
1070	add_to_unbuddied(pool, zhdr);
1071
1072headless:
1073	spin_lock(&pool->lock);
 
 
 
 
 
 
1074	*handle = encode_handle(zhdr, bud);
1075	spin_unlock(&pool->lock);
1076	if (bud != HEADLESS)
1077		z3fold_page_unlock(zhdr);
1078
1079	return 0;
1080}
1081
1082/**
1083 * z3fold_free() - frees the allocation associated with the given handle
1084 * @pool:	pool in which the allocation resided
1085 * @handle:	handle associated with the allocation returned by z3fold_alloc()
1086 *
1087 * In the case that the z3fold page in which the allocation resides is under
1088 * reclaim, as indicated by the PAGE_CLAIMED flag being set, this function
1089 * only sets the first|middle|last_chunks to 0.  The page is actually freed
1090 * once all buddies are evicted (see z3fold_reclaim_page() below).
1091 */
1092static void z3fold_free(struct z3fold_pool *pool, unsigned long handle)
1093{
1094	struct z3fold_header *zhdr;
1095	struct page *page;
1096	enum buddy bud;
1097	bool page_claimed;
1098
1099	zhdr = get_z3fold_header(handle);
1100	page = virt_to_page(zhdr);
1101	page_claimed = test_and_set_bit(PAGE_CLAIMED, &page->private);
1102
1103	if (test_bit(PAGE_HEADLESS, &page->private)) {
1104		/* if a headless page is under reclaim, just leave.
1105		 * NB: we use test_and_set_bit for a reason: if the bit
1106		 * has not been set before, we release this page
1107		 * immediately so we don't care about its value any more.
1108		 */
1109		if (!page_claimed) {
 
 
 
1110			put_z3fold_header(zhdr);
1111			free_z3fold_page(page, true);
1112			atomic64_dec(&pool->pages_nr);
1113		}
1114		return;
1115	}
1116
1117	/* Non-headless case */
1118	bud = handle_to_buddy(handle);
1119
1120	switch (bud) {
1121	case FIRST:
1122		zhdr->first_chunks = 0;
1123		break;
1124	case MIDDLE:
1125		zhdr->middle_chunks = 0;
1126		break;
1127	case LAST:
1128		zhdr->last_chunks = 0;
1129		break;
1130	default:
1131		pr_err("%s: unknown bud %d\n", __func__, bud);
1132		WARN_ON(1);
1133		put_z3fold_header(zhdr);
 
1134		return;
1135	}
1136
1137	if (!page_claimed)
1138		free_handle(handle, zhdr);
1139	if (put_z3fold_locked_list(zhdr))
 
1140		return;
 
1141	if (page_claimed) {
1142		/* the page has not been claimed by us */
1143		put_z3fold_header(zhdr);
1144		return;
1145	}
1146	if (test_and_set_bit(NEEDS_COMPACTING, &page->private)) {
1147		clear_bit(PAGE_CLAIMED, &page->private);
1148		put_z3fold_header(zhdr);
 
1149		return;
1150	}
1151	if (zhdr->cpu < 0 || !cpu_online(zhdr->cpu)) {
 
 
 
1152		zhdr->cpu = -1;
1153		kref_get(&zhdr->refcount);
1154		clear_bit(PAGE_CLAIMED, &page->private);
1155		do_compact_page(zhdr, true);
1156		return;
1157	}
1158	kref_get(&zhdr->refcount);
1159	clear_bit(PAGE_CLAIMED, &page->private);
1160	queue_work_on(zhdr->cpu, pool->compact_wq, &zhdr->work);
1161	put_z3fold_header(zhdr);
1162}
1163
1164/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1165 * z3fold_map() - maps the allocation associated with the given handle
1166 * @pool:	pool in which the allocation resides
1167 * @handle:	handle associated with the allocation to be mapped
1168 *
1169 * Extracts the buddy number from handle and constructs the pointer to the
1170 * correct starting chunk within the page.
1171 *
1172 * Returns: a pointer to the mapped allocation
1173 */
1174static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle)
1175{
1176	struct z3fold_header *zhdr;
1177	struct page *page;
1178	void *addr;
1179	enum buddy buddy;
1180
1181	zhdr = get_z3fold_header(handle);
1182	addr = zhdr;
1183	page = virt_to_page(zhdr);
1184
1185	if (test_bit(PAGE_HEADLESS, &page->private))
1186		goto out;
1187
1188	buddy = handle_to_buddy(handle);
1189	switch (buddy) {
1190	case FIRST:
1191		addr += ZHDR_SIZE_ALIGNED;
1192		break;
1193	case MIDDLE:
1194		addr += zhdr->start_middle << CHUNK_SHIFT;
1195		set_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1196		break;
1197	case LAST:
1198		addr += PAGE_SIZE - (handle_to_chunks(handle) << CHUNK_SHIFT);
1199		break;
1200	default:
1201		pr_err("unknown buddy id %d\n", buddy);
1202		WARN_ON(1);
1203		addr = NULL;
1204		break;
1205	}
1206
1207	if (addr)
1208		zhdr->mapped_count++;
1209out:
1210	put_z3fold_header(zhdr);
1211	return addr;
1212}
1213
1214/**
1215 * z3fold_unmap() - unmaps the allocation associated with the given handle
1216 * @pool:	pool in which the allocation resides
1217 * @handle:	handle associated with the allocation to be unmapped
1218 */
1219static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle)
1220{
1221	struct z3fold_header *zhdr;
1222	struct page *page;
1223	enum buddy buddy;
1224
1225	zhdr = get_z3fold_header(handle);
1226	page = virt_to_page(zhdr);
1227
1228	if (test_bit(PAGE_HEADLESS, &page->private))
1229		return;
1230
1231	buddy = handle_to_buddy(handle);
1232	if (buddy == MIDDLE)
1233		clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1234	zhdr->mapped_count--;
1235	put_z3fold_header(zhdr);
1236}
1237
1238/**
1239 * z3fold_get_pool_size() - gets the z3fold pool size in pages
1240 * @pool:	pool whose size is being queried
1241 *
1242 * Returns: size in pages of the given pool.
1243 */
1244static u64 z3fold_get_pool_size(struct z3fold_pool *pool)
1245{
1246	return atomic64_read(&pool->pages_nr);
1247}
1248
1249static bool z3fold_page_isolate(struct page *page, isolate_mode_t mode)
1250{
1251	struct z3fold_header *zhdr;
1252	struct z3fold_pool *pool;
1253
 
1254	VM_BUG_ON_PAGE(PageIsolated(page), page);
1255
1256	if (test_bit(PAGE_HEADLESS, &page->private))
 
1257		return false;
1258
1259	zhdr = page_address(page);
1260	z3fold_page_lock(zhdr);
1261	if (test_bit(NEEDS_COMPACTING, &page->private) ||
1262	    test_bit(PAGE_STALE, &page->private))
1263		goto out;
1264
1265	if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0)
1266		goto out;
1267
1268	if (test_and_set_bit(PAGE_CLAIMED, &page->private))
1269		goto out;
1270	pool = zhdr_to_pool(zhdr);
1271	spin_lock(&pool->lock);
1272	if (!list_empty(&zhdr->buddy))
1273		list_del_init(&zhdr->buddy);
 
 
1274	spin_unlock(&pool->lock);
1275
1276	kref_get(&zhdr->refcount);
1277	z3fold_page_unlock(zhdr);
1278	return true;
1279
1280out:
1281	z3fold_page_unlock(zhdr);
1282	return false;
1283}
1284
1285static int z3fold_page_migrate(struct page *newpage, struct page *page,
1286		enum migrate_mode mode)
1287{
1288	struct z3fold_header *zhdr, *new_zhdr;
1289	struct z3fold_pool *pool;
 
1290
 
1291	VM_BUG_ON_PAGE(!PageIsolated(page), page);
1292	VM_BUG_ON_PAGE(!test_bit(PAGE_CLAIMED, &page->private), page);
1293	VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
1294
1295	zhdr = page_address(page);
1296	pool = zhdr_to_pool(zhdr);
1297
1298	if (!z3fold_page_trylock(zhdr))
1299		return -EAGAIN;
 
1300	if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0) {
1301		clear_bit(PAGE_CLAIMED, &page->private);
1302		z3fold_page_unlock(zhdr);
1303		return -EBUSY;
1304	}
1305	if (work_pending(&zhdr->work)) {
1306		z3fold_page_unlock(zhdr);
1307		return -EAGAIN;
1308	}
1309	new_zhdr = page_address(newpage);
1310	memcpy(new_zhdr, zhdr, PAGE_SIZE);
1311	newpage->private = page->private;
1312	set_bit(PAGE_MIGRATED, &page->private);
1313	z3fold_page_unlock(zhdr);
1314	spin_lock_init(&new_zhdr->page_lock);
1315	INIT_WORK(&new_zhdr->work, compact_page_work);
1316	/*
1317	 * z3fold_page_isolate() ensures that new_zhdr->buddy is empty,
1318	 * so we only have to reinitialize it.
1319	 */
1320	INIT_LIST_HEAD(&new_zhdr->buddy);
 
1321	__ClearPageMovable(page);
 
1322
1323	get_page(newpage);
1324	z3fold_page_lock(new_zhdr);
1325	if (new_zhdr->first_chunks)
1326		encode_handle(new_zhdr, FIRST);
1327	if (new_zhdr->last_chunks)
1328		encode_handle(new_zhdr, LAST);
1329	if (new_zhdr->middle_chunks)
1330		encode_handle(new_zhdr, MIDDLE);
1331	set_bit(NEEDS_COMPACTING, &newpage->private);
1332	new_zhdr->cpu = smp_processor_id();
1333	__SetPageMovable(newpage, &z3fold_mops);
 
 
 
1334	z3fold_page_unlock(new_zhdr);
1335
1336	queue_work_on(new_zhdr->cpu, pool->compact_wq, &new_zhdr->work);
1337
1338	/* PAGE_CLAIMED and PAGE_MIGRATED are cleared now. */
1339	page->private = 0;
1340	put_page(page);
1341	return 0;
1342}
1343
1344static void z3fold_page_putback(struct page *page)
1345{
1346	struct z3fold_header *zhdr;
1347	struct z3fold_pool *pool;
1348
1349	zhdr = page_address(page);
1350	pool = zhdr_to_pool(zhdr);
1351
1352	z3fold_page_lock(zhdr);
1353	if (!list_empty(&zhdr->buddy))
1354		list_del_init(&zhdr->buddy);
1355	INIT_LIST_HEAD(&page->lru);
1356	if (put_z3fold_locked(zhdr))
 
1357		return;
1358	if (list_empty(&zhdr->buddy))
1359		add_to_unbuddied(pool, zhdr);
1360	clear_bit(PAGE_CLAIMED, &page->private);
 
1361	z3fold_page_unlock(zhdr);
1362}
1363
1364static const struct movable_operations z3fold_mops = {
1365	.isolate_page = z3fold_page_isolate,
1366	.migrate_page = z3fold_page_migrate,
1367	.putback_page = z3fold_page_putback,
1368};
1369
1370/*****************
1371 * zpool
1372 ****************/
1373
1374static void *z3fold_zpool_create(const char *name, gfp_t gfp)
1375{
1376	return z3fold_create_pool(name, gfp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1377}
1378
1379static void z3fold_zpool_destroy(void *pool)
1380{
1381	z3fold_destroy_pool(pool);
1382}
1383
1384static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp,
1385			unsigned long *handle)
1386{
1387	return z3fold_alloc(pool, size, gfp, handle);
1388}
1389static void z3fold_zpool_free(void *pool, unsigned long handle)
1390{
1391	z3fold_free(pool, handle);
1392}
1393
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1394static void *z3fold_zpool_map(void *pool, unsigned long handle,
1395			enum zpool_mapmode mm)
1396{
1397	return z3fold_map(pool, handle);
1398}
1399static void z3fold_zpool_unmap(void *pool, unsigned long handle)
1400{
1401	z3fold_unmap(pool, handle);
1402}
1403
1404static u64 z3fold_zpool_total_size(void *pool)
1405{
1406	return z3fold_get_pool_size(pool) * PAGE_SIZE;
1407}
1408
1409static struct zpool_driver z3fold_zpool_driver = {
1410	.type =		"z3fold",
1411	.sleep_mapped = true,
1412	.owner =	THIS_MODULE,
1413	.create =	z3fold_zpool_create,
1414	.destroy =	z3fold_zpool_destroy,
1415	.malloc =	z3fold_zpool_malloc,
1416	.free =		z3fold_zpool_free,
 
1417	.map =		z3fold_zpool_map,
1418	.unmap =	z3fold_zpool_unmap,
1419	.total_size =	z3fold_zpool_total_size,
1420};
1421
1422MODULE_ALIAS("zpool-z3fold");
1423
1424static int __init init_z3fold(void)
1425{
1426	/*
1427	 * Make sure the z3fold header is not larger than the page size and
1428	 * there has remaining spaces for its buddy.
1429	 */
1430	BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE - CHUNK_SIZE);
 
 
 
1431	zpool_register_driver(&z3fold_zpool_driver);
1432
1433	return 0;
1434}
1435
1436static void __exit exit_z3fold(void)
1437{
 
1438	zpool_unregister_driver(&z3fold_zpool_driver);
1439}
1440
1441module_init(init_z3fold);
1442module_exit(exit_z3fold);
1443
1444MODULE_LICENSE("GPL");
1445MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>");
1446MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages");
v5.9
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * z3fold.c
   4 *
   5 * Author: Vitaly Wool <vitaly.wool@konsulko.com>
   6 * Copyright (C) 2016, Sony Mobile Communications Inc.
   7 *
   8 * This implementation is based on zbud written by Seth Jennings.
   9 *
  10 * z3fold is an special purpose allocator for storing compressed pages. It
  11 * can store up to three compressed pages per page which improves the
  12 * compression ratio of zbud while retaining its main concepts (e. g. always
  13 * storing an integral number of objects per page) and simplicity.
  14 * It still has simple and deterministic reclaim properties that make it
  15 * preferable to a higher density approach (with no requirement on integral
  16 * number of object per page) when reclaim is used.
  17 *
  18 * As in zbud, pages are divided into "chunks".  The size of the chunks is
  19 * fixed at compile time and is determined by NCHUNKS_ORDER below.
  20 *
  21 * z3fold doesn't export any API and is meant to be used via zpool API.
  22 */
  23
  24#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  25
  26#include <linux/atomic.h>
  27#include <linux/sched.h>
  28#include <linux/cpumask.h>
  29#include <linux/list.h>
  30#include <linux/mm.h>
  31#include <linux/module.h>
  32#include <linux/page-flags.h>
  33#include <linux/migrate.h>
  34#include <linux/node.h>
  35#include <linux/compaction.h>
  36#include <linux/percpu.h>
  37#include <linux/mount.h>
  38#include <linux/pseudo_fs.h>
  39#include <linux/fs.h>
  40#include <linux/preempt.h>
  41#include <linux/workqueue.h>
  42#include <linux/slab.h>
  43#include <linux/spinlock.h>
  44#include <linux/zpool.h>
  45#include <linux/magic.h>
  46#include <linux/kmemleak.h>
  47
  48/*
  49 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
  50 * adjusting internal fragmentation.  It also determines the number of
  51 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
  52 * allocation granularity will be in chunks of size PAGE_SIZE/64. Some chunks
  53 * in the beginning of an allocated page are occupied by z3fold header, so
  54 * NCHUNKS will be calculated to 63 (or 62 in case CONFIG_DEBUG_SPINLOCK=y),
  55 * which shows the max number of free chunks in z3fold page, also there will
  56 * be 63, or 62, respectively, freelists per pool.
  57 */
  58#define NCHUNKS_ORDER	6
  59
  60#define CHUNK_SHIFT	(PAGE_SHIFT - NCHUNKS_ORDER)
  61#define CHUNK_SIZE	(1 << CHUNK_SHIFT)
  62#define ZHDR_SIZE_ALIGNED round_up(sizeof(struct z3fold_header), CHUNK_SIZE)
  63#define ZHDR_CHUNKS	(ZHDR_SIZE_ALIGNED >> CHUNK_SHIFT)
  64#define TOTAL_CHUNKS	(PAGE_SIZE >> CHUNK_SHIFT)
  65#define NCHUNKS		((PAGE_SIZE - ZHDR_SIZE_ALIGNED) >> CHUNK_SHIFT)
  66
  67#define BUDDY_MASK	(0x3)
  68#define BUDDY_SHIFT	2
  69#define SLOTS_ALIGN	(0x40)
  70
  71/*****************
  72 * Structures
  73*****************/
  74struct z3fold_pool;
  75struct z3fold_ops {
  76	int (*evict)(struct z3fold_pool *pool, unsigned long handle);
  77};
  78
  79enum buddy {
  80	HEADLESS = 0,
  81	FIRST,
  82	MIDDLE,
  83	LAST,
  84	BUDDIES_MAX = LAST
  85};
  86
  87struct z3fold_buddy_slots {
  88	/*
  89	 * we are using BUDDY_MASK in handle_to_buddy etc. so there should
  90	 * be enough slots to hold all possible variants
  91	 */
  92	unsigned long slot[BUDDY_MASK + 1];
  93	unsigned long pool; /* back link + flags */
  94	rwlock_t lock;
  95};
  96#define HANDLE_FLAG_MASK	(0x03)
  97
  98/*
  99 * struct z3fold_header - z3fold page metadata occupying first chunks of each
 100 *			z3fold page, except for HEADLESS pages
 101 * @buddy:		links the z3fold page into the relevant list in the
 102 *			pool
 103 * @page_lock:		per-page lock
 104 * @refcount:		reference count for the z3fold page
 105 * @work:		work_struct for page layout optimization
 106 * @slots:		pointer to the structure holding buddy slots
 107 * @pool:		pointer to the containing pool
 108 * @cpu:		CPU which this page "belongs" to
 109 * @first_chunks:	the size of the first buddy in chunks, 0 if free
 110 * @middle_chunks:	the size of the middle buddy in chunks, 0 if free
 111 * @last_chunks:	the size of the last buddy in chunks, 0 if free
 112 * @first_num:		the starting number (for the first handle)
 113 * @mapped_count:	the number of objects currently mapped
 114 */
 115struct z3fold_header {
 116	struct list_head buddy;
 117	spinlock_t page_lock;
 118	struct kref refcount;
 119	struct work_struct work;
 120	struct z3fold_buddy_slots *slots;
 121	struct z3fold_pool *pool;
 122	short cpu;
 123	unsigned short first_chunks;
 124	unsigned short middle_chunks;
 125	unsigned short last_chunks;
 126	unsigned short start_middle;
 127	unsigned short first_num:2;
 128	unsigned short mapped_count:2;
 129	unsigned short foreign_handles:2;
 130};
 131
 132/**
 133 * struct z3fold_pool - stores metadata for each z3fold pool
 134 * @name:	pool name
 135 * @lock:	protects pool unbuddied/lru lists
 136 * @stale_lock:	protects pool stale page list
 137 * @unbuddied:	per-cpu array of lists tracking z3fold pages that contain 2-
 138 *		buddies; the list each z3fold page is added to depends on
 139 *		the size of its free region.
 140 * @lru:	list tracking the z3fold pages in LRU order by most recently
 141 *		added buddy.
 142 * @stale:	list of pages marked for freeing
 143 * @pages_nr:	number of z3fold pages in the pool.
 144 * @c_handle:	cache for z3fold_buddy_slots allocation
 145 * @ops:	pointer to a structure of user defined operations specified at
 146 *		pool creation time.
 147 * @compact_wq:	workqueue for page layout background optimization
 148 * @release_wq:	workqueue for safe page release
 149 * @work:	work_struct for safe page release
 150 * @inode:	inode for z3fold pseudo filesystem
 151 *
 152 * This structure is allocated at pool creation time and maintains metadata
 153 * pertaining to a particular z3fold pool.
 154 */
 155struct z3fold_pool {
 156	const char *name;
 157	spinlock_t lock;
 158	spinlock_t stale_lock;
 159	struct list_head *unbuddied;
 160	struct list_head lru;
 161	struct list_head stale;
 162	atomic64_t pages_nr;
 163	struct kmem_cache *c_handle;
 164	const struct z3fold_ops *ops;
 165	struct zpool *zpool;
 166	const struct zpool_ops *zpool_ops;
 167	struct workqueue_struct *compact_wq;
 168	struct workqueue_struct *release_wq;
 169	struct work_struct work;
 170	struct inode *inode;
 171};
 172
 173/*
 174 * Internal z3fold page flags
 175 */
 176enum z3fold_page_flags {
 177	PAGE_HEADLESS = 0,
 178	MIDDLE_CHUNK_MAPPED,
 179	NEEDS_COMPACTING,
 180	PAGE_STALE,
 181	PAGE_CLAIMED, /* by either reclaim or free */
 
 182};
 183
 184/*
 185 * handle flags, go under HANDLE_FLAG_MASK
 186 */
 187enum z3fold_handle_flags {
 188	HANDLES_ORPHANED = 0,
 189};
 190
 191/*
 192 * Forward declarations
 193 */
 194static struct z3fold_header *__z3fold_alloc(struct z3fold_pool *, size_t, bool);
 195static void compact_page_work(struct work_struct *w);
 196
 197/*****************
 198 * Helpers
 199*****************/
 200
 201/* Converts an allocation size in bytes to size in z3fold chunks */
 202static int size_to_chunks(size_t size)
 203{
 204	return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
 205}
 206
 207#define for_each_unbuddied_list(_iter, _begin) \
 208	for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
 209
 210static inline struct z3fold_buddy_slots *alloc_slots(struct z3fold_pool *pool,
 211							gfp_t gfp)
 212{
 213	struct z3fold_buddy_slots *slots;
 214
 215	slots = kmem_cache_alloc(pool->c_handle,
 216				 (gfp & ~(__GFP_HIGHMEM | __GFP_MOVABLE)));
 217
 218	if (slots) {
 219		/* It will be freed separately in free_handle(). */
 220		kmemleak_not_leak(slots);
 221		memset(slots->slot, 0, sizeof(slots->slot));
 222		slots->pool = (unsigned long)pool;
 223		rwlock_init(&slots->lock);
 224	}
 225
 226	return slots;
 227}
 228
 229static inline struct z3fold_pool *slots_to_pool(struct z3fold_buddy_slots *s)
 230{
 231	return (struct z3fold_pool *)(s->pool & ~HANDLE_FLAG_MASK);
 232}
 233
 234static inline struct z3fold_buddy_slots *handle_to_slots(unsigned long handle)
 235{
 236	return (struct z3fold_buddy_slots *)(handle & ~(SLOTS_ALIGN - 1));
 237}
 238
 239/* Lock a z3fold page */
 240static inline void z3fold_page_lock(struct z3fold_header *zhdr)
 241{
 242	spin_lock(&zhdr->page_lock);
 243}
 244
 245/* Try to lock a z3fold page */
 246static inline int z3fold_page_trylock(struct z3fold_header *zhdr)
 247{
 248	return spin_trylock(&zhdr->page_lock);
 249}
 250
 251/* Unlock a z3fold page */
 252static inline void z3fold_page_unlock(struct z3fold_header *zhdr)
 253{
 254	spin_unlock(&zhdr->page_lock);
 255}
 256
 257
 258static inline struct z3fold_header *__get_z3fold_header(unsigned long handle,
 259							bool lock)
 260{
 261	struct z3fold_buddy_slots *slots;
 262	struct z3fold_header *zhdr;
 263	int locked = 0;
 264
 265	if (!(handle & (1 << PAGE_HEADLESS))) {
 266		slots = handle_to_slots(handle);
 267		do {
 268			unsigned long addr;
 269
 270			read_lock(&slots->lock);
 271			addr = *(unsigned long *)handle;
 272			zhdr = (struct z3fold_header *)(addr & PAGE_MASK);
 273			if (lock)
 274				locked = z3fold_page_trylock(zhdr);
 275			read_unlock(&slots->lock);
 276			if (locked)
 277				break;
 
 
 
 
 
 278			cpu_relax();
 279		} while (lock);
 280	} else {
 281		zhdr = (struct z3fold_header *)(handle & PAGE_MASK);
 282	}
 283
 284	return zhdr;
 285}
 286
 287/* Returns the z3fold page where a given handle is stored */
 288static inline struct z3fold_header *handle_to_z3fold_header(unsigned long h)
 289{
 290	return __get_z3fold_header(h, false);
 291}
 292
 293/* return locked z3fold page if it's not headless */
 294static inline struct z3fold_header *get_z3fold_header(unsigned long h)
 295{
 296	return __get_z3fold_header(h, true);
 297}
 298
 299static inline void put_z3fold_header(struct z3fold_header *zhdr)
 300{
 301	struct page *page = virt_to_page(zhdr);
 302
 303	if (!test_bit(PAGE_HEADLESS, &page->private))
 304		z3fold_page_unlock(zhdr);
 305}
 306
 307static inline void free_handle(unsigned long handle)
 308{
 309	struct z3fold_buddy_slots *slots;
 310	struct z3fold_header *zhdr;
 311	int i;
 312	bool is_free;
 313
 314	if (handle & (1 << PAGE_HEADLESS))
 315		return;
 316
 317	if (WARN_ON(*(unsigned long *)handle == 0))
 318		return;
 319
 320	zhdr = handle_to_z3fold_header(handle);
 321	slots = handle_to_slots(handle);
 322	write_lock(&slots->lock);
 323	*(unsigned long *)handle = 0;
 324	if (zhdr->slots == slots) {
 
 325		write_unlock(&slots->lock);
 326		return; /* simple case, nothing else to do */
 327	}
 328
 329	/* we are freeing a foreign handle if we are here */
 330	zhdr->foreign_handles--;
 
 331	is_free = true;
 332	if (!test_bit(HANDLES_ORPHANED, &slots->pool)) {
 333		write_unlock(&slots->lock);
 334		return;
 335	}
 336	for (i = 0; i <= BUDDY_MASK; i++) {
 337		if (slots->slot[i]) {
 338			is_free = false;
 339			break;
 340		}
 341	}
 342	write_unlock(&slots->lock);
 343
 344	if (is_free) {
 345		struct z3fold_pool *pool = slots_to_pool(slots);
 346
 
 
 347		kmem_cache_free(pool->c_handle, slots);
 348	}
 349}
 350
 351static int z3fold_init_fs_context(struct fs_context *fc)
 352{
 353	return init_pseudo(fc, Z3FOLD_MAGIC) ? 0 : -ENOMEM;
 354}
 355
 356static struct file_system_type z3fold_fs = {
 357	.name		= "z3fold",
 358	.init_fs_context = z3fold_init_fs_context,
 359	.kill_sb	= kill_anon_super,
 360};
 361
 362static struct vfsmount *z3fold_mnt;
 363static int z3fold_mount(void)
 364{
 365	int ret = 0;
 366
 367	z3fold_mnt = kern_mount(&z3fold_fs);
 368	if (IS_ERR(z3fold_mnt))
 369		ret = PTR_ERR(z3fold_mnt);
 370
 371	return ret;
 372}
 373
 374static void z3fold_unmount(void)
 375{
 376	kern_unmount(z3fold_mnt);
 377}
 378
 379static const struct address_space_operations z3fold_aops;
 380static int z3fold_register_migration(struct z3fold_pool *pool)
 381{
 382	pool->inode = alloc_anon_inode(z3fold_mnt->mnt_sb);
 383	if (IS_ERR(pool->inode)) {
 384		pool->inode = NULL;
 385		return 1;
 386	}
 387
 388	pool->inode->i_mapping->private_data = pool;
 389	pool->inode->i_mapping->a_ops = &z3fold_aops;
 390	return 0;
 391}
 392
 393static void z3fold_unregister_migration(struct z3fold_pool *pool)
 394{
 395	if (pool->inode)
 396		iput(pool->inode);
 397 }
 398
 399/* Initializes the z3fold header of a newly allocated z3fold page */
 400static struct z3fold_header *init_z3fold_page(struct page *page, bool headless,
 401					struct z3fold_pool *pool, gfp_t gfp)
 402{
 403	struct z3fold_header *zhdr = page_address(page);
 404	struct z3fold_buddy_slots *slots;
 405
 406	INIT_LIST_HEAD(&page->lru);
 407	clear_bit(PAGE_HEADLESS, &page->private);
 408	clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
 409	clear_bit(NEEDS_COMPACTING, &page->private);
 410	clear_bit(PAGE_STALE, &page->private);
 411	clear_bit(PAGE_CLAIMED, &page->private);
 
 412	if (headless)
 413		return zhdr;
 414
 415	slots = alloc_slots(pool, gfp);
 416	if (!slots)
 417		return NULL;
 418
 
 419	spin_lock_init(&zhdr->page_lock);
 420	kref_init(&zhdr->refcount);
 421	zhdr->first_chunks = 0;
 422	zhdr->middle_chunks = 0;
 423	zhdr->last_chunks = 0;
 424	zhdr->first_num = 0;
 425	zhdr->start_middle = 0;
 426	zhdr->cpu = -1;
 427	zhdr->foreign_handles = 0;
 428	zhdr->mapped_count = 0;
 429	zhdr->slots = slots;
 430	zhdr->pool = pool;
 431	INIT_LIST_HEAD(&zhdr->buddy);
 432	INIT_WORK(&zhdr->work, compact_page_work);
 433	return zhdr;
 434}
 435
 436/* Resets the struct page fields and frees the page */
 437static void free_z3fold_page(struct page *page, bool headless)
 438{
 439	if (!headless) {
 440		lock_page(page);
 441		__ClearPageMovable(page);
 442		unlock_page(page);
 443	}
 444	ClearPagePrivate(page);
 445	__free_page(page);
 446}
 447
 448/* Helper function to build the index */
 449static inline int __idx(struct z3fold_header *zhdr, enum buddy bud)
 450{
 451	return (bud + zhdr->first_num) & BUDDY_MASK;
 452}
 453
 454/*
 455 * Encodes the handle of a particular buddy within a z3fold page
 456 * Pool lock should be held as this function accesses first_num
 457 */
 458static unsigned long __encode_handle(struct z3fold_header *zhdr,
 459				struct z3fold_buddy_slots *slots,
 460				enum buddy bud)
 461{
 462	unsigned long h = (unsigned long)zhdr;
 463	int idx = 0;
 464
 465	/*
 466	 * For a headless page, its handle is its pointer with the extra
 467	 * PAGE_HEADLESS bit set
 468	 */
 469	if (bud == HEADLESS)
 470		return h | (1 << PAGE_HEADLESS);
 471
 472	/* otherwise, return pointer to encoded handle */
 473	idx = __idx(zhdr, bud);
 474	h += idx;
 475	if (bud == LAST)
 476		h |= (zhdr->last_chunks << BUDDY_SHIFT);
 477
 478	write_lock(&slots->lock);
 479	slots->slot[idx] = h;
 480	write_unlock(&slots->lock);
 481	return (unsigned long)&slots->slot[idx];
 482}
 483
 484static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud)
 485{
 486	return __encode_handle(zhdr, zhdr->slots, bud);
 487}
 488
 489/* only for LAST bud, returns zero otherwise */
 490static unsigned short handle_to_chunks(unsigned long handle)
 491{
 492	struct z3fold_buddy_slots *slots = handle_to_slots(handle);
 493	unsigned long addr;
 494
 495	read_lock(&slots->lock);
 496	addr = *(unsigned long *)handle;
 497	read_unlock(&slots->lock);
 498	return (addr & ~PAGE_MASK) >> BUDDY_SHIFT;
 499}
 500
 501/*
 502 * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle
 503 *  but that doesn't matter. because the masking will result in the
 504 *  correct buddy number.
 505 */
 506static enum buddy handle_to_buddy(unsigned long handle)
 507{
 508	struct z3fold_header *zhdr;
 509	struct z3fold_buddy_slots *slots = handle_to_slots(handle);
 510	unsigned long addr;
 511
 512	read_lock(&slots->lock);
 513	WARN_ON(handle & (1 << PAGE_HEADLESS));
 514	addr = *(unsigned long *)handle;
 515	read_unlock(&slots->lock);
 516	zhdr = (struct z3fold_header *)(addr & PAGE_MASK);
 517	return (addr - zhdr->first_num) & BUDDY_MASK;
 518}
 519
 520static inline struct z3fold_pool *zhdr_to_pool(struct z3fold_header *zhdr)
 521{
 522	return zhdr->pool;
 523}
 524
 525static void __release_z3fold_page(struct z3fold_header *zhdr, bool locked)
 526{
 527	struct page *page = virt_to_page(zhdr);
 528	struct z3fold_pool *pool = zhdr_to_pool(zhdr);
 529	bool is_free = true;
 530	int i;
 531
 532	WARN_ON(!list_empty(&zhdr->buddy));
 533	set_bit(PAGE_STALE, &page->private);
 534	clear_bit(NEEDS_COMPACTING, &page->private);
 535	spin_lock(&pool->lock);
 536	if (!list_empty(&page->lru))
 537		list_del_init(&page->lru);
 538	spin_unlock(&pool->lock);
 539
 540	/* If there are no foreign handles, free the handles array */
 541	read_lock(&zhdr->slots->lock);
 542	for (i = 0; i <= BUDDY_MASK; i++) {
 543		if (zhdr->slots->slot[i]) {
 544			is_free = false;
 545			break;
 546		}
 547	}
 548	if (!is_free)
 549		set_bit(HANDLES_ORPHANED, &zhdr->slots->pool);
 550	read_unlock(&zhdr->slots->lock);
 551
 552	if (is_free)
 553		kmem_cache_free(pool->c_handle, zhdr->slots);
 554
 555	if (locked)
 556		z3fold_page_unlock(zhdr);
 557
 558	spin_lock(&pool->stale_lock);
 559	list_add(&zhdr->buddy, &pool->stale);
 560	queue_work(pool->release_wq, &pool->work);
 561	spin_unlock(&pool->stale_lock);
 562}
 563
 564static void __attribute__((__unused__))
 565			release_z3fold_page(struct kref *ref)
 566{
 567	struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
 568						refcount);
 569	__release_z3fold_page(zhdr, false);
 570}
 571
 572static void release_z3fold_page_locked(struct kref *ref)
 573{
 574	struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
 575						refcount);
 576	WARN_ON(z3fold_page_trylock(zhdr));
 577	__release_z3fold_page(zhdr, true);
 578}
 579
 580static void release_z3fold_page_locked_list(struct kref *ref)
 581{
 582	struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
 583					       refcount);
 584	struct z3fold_pool *pool = zhdr_to_pool(zhdr);
 585
 586	spin_lock(&pool->lock);
 587	list_del_init(&zhdr->buddy);
 588	spin_unlock(&pool->lock);
 589
 590	WARN_ON(z3fold_page_trylock(zhdr));
 591	__release_z3fold_page(zhdr, true);
 592}
 593
 
 
 
 
 
 
 
 
 
 
 594static void free_pages_work(struct work_struct *w)
 595{
 596	struct z3fold_pool *pool = container_of(w, struct z3fold_pool, work);
 597
 598	spin_lock(&pool->stale_lock);
 599	while (!list_empty(&pool->stale)) {
 600		struct z3fold_header *zhdr = list_first_entry(&pool->stale,
 601						struct z3fold_header, buddy);
 602		struct page *page = virt_to_page(zhdr);
 603
 604		list_del(&zhdr->buddy);
 605		if (WARN_ON(!test_bit(PAGE_STALE, &page->private)))
 606			continue;
 607		spin_unlock(&pool->stale_lock);
 608		cancel_work_sync(&zhdr->work);
 609		free_z3fold_page(page, false);
 610		cond_resched();
 611		spin_lock(&pool->stale_lock);
 612	}
 613	spin_unlock(&pool->stale_lock);
 614}
 615
 616/*
 617 * Returns the number of free chunks in a z3fold page.
 618 * NB: can't be used with HEADLESS pages.
 619 */
 620static int num_free_chunks(struct z3fold_header *zhdr)
 621{
 622	int nfree;
 623	/*
 624	 * If there is a middle object, pick up the bigger free space
 625	 * either before or after it. Otherwise just subtract the number
 626	 * of chunks occupied by the first and the last objects.
 627	 */
 628	if (zhdr->middle_chunks != 0) {
 629		int nfree_before = zhdr->first_chunks ?
 630			0 : zhdr->start_middle - ZHDR_CHUNKS;
 631		int nfree_after = zhdr->last_chunks ?
 632			0 : TOTAL_CHUNKS -
 633				(zhdr->start_middle + zhdr->middle_chunks);
 634		nfree = max(nfree_before, nfree_after);
 635	} else
 636		nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks;
 637	return nfree;
 638}
 639
 640/* Add to the appropriate unbuddied list */
 641static inline void add_to_unbuddied(struct z3fold_pool *pool,
 642				struct z3fold_header *zhdr)
 643{
 644	if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 ||
 645			zhdr->middle_chunks == 0) {
 646		struct list_head *unbuddied = get_cpu_ptr(pool->unbuddied);
 
 647
 648		int freechunks = num_free_chunks(zhdr);
 
 649		spin_lock(&pool->lock);
 650		list_add(&zhdr->buddy, &unbuddied[freechunks]);
 651		spin_unlock(&pool->lock);
 652		zhdr->cpu = smp_processor_id();
 653		put_cpu_ptr(pool->unbuddied);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 654	}
 
 
 655}
 656
 657static inline void *mchunk_memmove(struct z3fold_header *zhdr,
 658				unsigned short dst_chunk)
 659{
 660	void *beg = zhdr;
 661	return memmove(beg + (dst_chunk << CHUNK_SHIFT),
 662		       beg + (zhdr->start_middle << CHUNK_SHIFT),
 663		       zhdr->middle_chunks << CHUNK_SHIFT);
 664}
 665
 666static inline bool buddy_single(struct z3fold_header *zhdr)
 667{
 668	return !((zhdr->first_chunks && zhdr->middle_chunks) ||
 669			(zhdr->first_chunks && zhdr->last_chunks) ||
 670			(zhdr->middle_chunks && zhdr->last_chunks));
 671}
 672
 673static struct z3fold_header *compact_single_buddy(struct z3fold_header *zhdr)
 674{
 675	struct z3fold_pool *pool = zhdr_to_pool(zhdr);
 676	void *p = zhdr;
 677	unsigned long old_handle = 0;
 678	size_t sz = 0;
 679	struct z3fold_header *new_zhdr = NULL;
 680	int first_idx = __idx(zhdr, FIRST);
 681	int middle_idx = __idx(zhdr, MIDDLE);
 682	int last_idx = __idx(zhdr, LAST);
 683	unsigned short *moved_chunks = NULL;
 684
 685	/*
 686	 * No need to protect slots here -- all the slots are "local" and
 687	 * the page lock is already taken
 688	 */
 689	if (zhdr->first_chunks && zhdr->slots->slot[first_idx]) {
 690		p += ZHDR_SIZE_ALIGNED;
 691		sz = zhdr->first_chunks << CHUNK_SHIFT;
 692		old_handle = (unsigned long)&zhdr->slots->slot[first_idx];
 693		moved_chunks = &zhdr->first_chunks;
 694	} else if (zhdr->middle_chunks && zhdr->slots->slot[middle_idx]) {
 695		p += zhdr->start_middle << CHUNK_SHIFT;
 696		sz = zhdr->middle_chunks << CHUNK_SHIFT;
 697		old_handle = (unsigned long)&zhdr->slots->slot[middle_idx];
 698		moved_chunks = &zhdr->middle_chunks;
 699	} else if (zhdr->last_chunks && zhdr->slots->slot[last_idx]) {
 700		p += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
 701		sz = zhdr->last_chunks << CHUNK_SHIFT;
 702		old_handle = (unsigned long)&zhdr->slots->slot[last_idx];
 703		moved_chunks = &zhdr->last_chunks;
 704	}
 705
 706	if (sz > 0) {
 707		enum buddy new_bud = HEADLESS;
 708		short chunks = size_to_chunks(sz);
 709		void *q;
 710
 711		new_zhdr = __z3fold_alloc(pool, sz, false);
 712		if (!new_zhdr)
 713			return NULL;
 714
 715		if (WARN_ON(new_zhdr == zhdr))
 716			goto out_fail;
 717
 718		if (new_zhdr->first_chunks == 0) {
 719			if (new_zhdr->middle_chunks != 0 &&
 720					chunks >= new_zhdr->start_middle) {
 721				new_bud = LAST;
 722			} else {
 723				new_bud = FIRST;
 724			}
 725		} else if (new_zhdr->last_chunks == 0) {
 726			new_bud = LAST;
 727		} else if (new_zhdr->middle_chunks == 0) {
 728			new_bud = MIDDLE;
 729		}
 730		q = new_zhdr;
 731		switch (new_bud) {
 732		case FIRST:
 733			new_zhdr->first_chunks = chunks;
 734			q += ZHDR_SIZE_ALIGNED;
 735			break;
 736		case MIDDLE:
 737			new_zhdr->middle_chunks = chunks;
 738			new_zhdr->start_middle =
 739				new_zhdr->first_chunks + ZHDR_CHUNKS;
 740			q += new_zhdr->start_middle << CHUNK_SHIFT;
 741			break;
 742		case LAST:
 743			new_zhdr->last_chunks = chunks;
 744			q += PAGE_SIZE - (new_zhdr->last_chunks << CHUNK_SHIFT);
 745			break;
 746		default:
 747			goto out_fail;
 748		}
 749		new_zhdr->foreign_handles++;
 750		memcpy(q, p, sz);
 751		write_lock(&zhdr->slots->lock);
 752		*(unsigned long *)old_handle = (unsigned long)new_zhdr +
 753			__idx(new_zhdr, new_bud);
 754		if (new_bud == LAST)
 755			*(unsigned long *)old_handle |=
 756					(new_zhdr->last_chunks << BUDDY_SHIFT);
 757		write_unlock(&zhdr->slots->lock);
 758		add_to_unbuddied(pool, new_zhdr);
 759		z3fold_page_unlock(new_zhdr);
 760
 761		*moved_chunks = 0;
 762	}
 763
 764	return new_zhdr;
 765
 766out_fail:
 767	if (new_zhdr) {
 768		if (kref_put(&new_zhdr->refcount, release_z3fold_page_locked))
 769			atomic64_dec(&pool->pages_nr);
 770		else {
 771			add_to_unbuddied(pool, new_zhdr);
 772			z3fold_page_unlock(new_zhdr);
 773		}
 774	}
 775	return NULL;
 776
 777}
 778
 779#define BIG_CHUNK_GAP	3
 780/* Has to be called with lock held */
 781static int z3fold_compact_page(struct z3fold_header *zhdr)
 782{
 783	struct page *page = virt_to_page(zhdr);
 784
 785	if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private))
 786		return 0; /* can't move middle chunk, it's used */
 787
 788	if (unlikely(PageIsolated(page)))
 789		return 0;
 790
 791	if (zhdr->middle_chunks == 0)
 792		return 0; /* nothing to compact */
 793
 794	if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
 795		/* move to the beginning */
 796		mchunk_memmove(zhdr, ZHDR_CHUNKS);
 797		zhdr->first_chunks = zhdr->middle_chunks;
 798		zhdr->middle_chunks = 0;
 799		zhdr->start_middle = 0;
 800		zhdr->first_num++;
 801		return 1;
 802	}
 803
 804	/*
 805	 * moving data is expensive, so let's only do that if
 806	 * there's substantial gain (at least BIG_CHUNK_GAP chunks)
 807	 */
 808	if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 &&
 809	    zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >=
 810			BIG_CHUNK_GAP) {
 811		mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS);
 812		zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
 813		return 1;
 814	} else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 &&
 815		   TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle
 816					+ zhdr->middle_chunks) >=
 817			BIG_CHUNK_GAP) {
 818		unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks -
 819			zhdr->middle_chunks;
 820		mchunk_memmove(zhdr, new_start);
 821		zhdr->start_middle = new_start;
 822		return 1;
 823	}
 824
 825	return 0;
 826}
 827
 828static void do_compact_page(struct z3fold_header *zhdr, bool locked)
 829{
 830	struct z3fold_pool *pool = zhdr_to_pool(zhdr);
 831	struct page *page;
 832
 833	page = virt_to_page(zhdr);
 834	if (locked)
 835		WARN_ON(z3fold_page_trylock(zhdr));
 836	else
 837		z3fold_page_lock(zhdr);
 838	if (WARN_ON(!test_and_clear_bit(NEEDS_COMPACTING, &page->private))) {
 839		z3fold_page_unlock(zhdr);
 840		return;
 841	}
 842	spin_lock(&pool->lock);
 843	list_del_init(&zhdr->buddy);
 844	spin_unlock(&pool->lock);
 845
 846	if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) {
 847		atomic64_dec(&pool->pages_nr);
 848		return;
 849	}
 850
 851	if (unlikely(PageIsolated(page) ||
 852		     test_bit(PAGE_CLAIMED, &page->private) ||
 853		     test_bit(PAGE_STALE, &page->private))) {
 854		z3fold_page_unlock(zhdr);
 855		return;
 856	}
 857
 858	if (!zhdr->foreign_handles && buddy_single(zhdr) &&
 859	    zhdr->mapped_count == 0 && compact_single_buddy(zhdr)) {
 860		if (kref_put(&zhdr->refcount, release_z3fold_page_locked))
 861			atomic64_dec(&pool->pages_nr);
 862		else
 863			z3fold_page_unlock(zhdr);
 
 864		return;
 865	}
 866
 867	z3fold_compact_page(zhdr);
 868	add_to_unbuddied(pool, zhdr);
 
 869	z3fold_page_unlock(zhdr);
 870}
 871
 872static void compact_page_work(struct work_struct *w)
 873{
 874	struct z3fold_header *zhdr = container_of(w, struct z3fold_header,
 875						work);
 876
 877	do_compact_page(zhdr, false);
 878}
 879
 880/* returns _locked_ z3fold page header or NULL */
 881static inline struct z3fold_header *__z3fold_alloc(struct z3fold_pool *pool,
 882						size_t size, bool can_sleep)
 883{
 884	struct z3fold_header *zhdr = NULL;
 885	struct page *page;
 886	struct list_head *unbuddied;
 887	int chunks = size_to_chunks(size), i;
 888
 889lookup:
 
 890	/* First, try to find an unbuddied z3fold page. */
 891	unbuddied = get_cpu_ptr(pool->unbuddied);
 892	for_each_unbuddied_list(i, chunks) {
 893		struct list_head *l = &unbuddied[i];
 894
 895		zhdr = list_first_entry_or_null(READ_ONCE(l),
 896					struct z3fold_header, buddy);
 897
 898		if (!zhdr)
 899			continue;
 900
 901		/* Re-check under lock. */
 902		spin_lock(&pool->lock);
 903		l = &unbuddied[i];
 904		if (unlikely(zhdr != list_first_entry(READ_ONCE(l),
 905						struct z3fold_header, buddy)) ||
 906		    !z3fold_page_trylock(zhdr)) {
 907			spin_unlock(&pool->lock);
 908			zhdr = NULL;
 909			put_cpu_ptr(pool->unbuddied);
 910			if (can_sleep)
 911				cond_resched();
 912			goto lookup;
 913		}
 914		list_del_init(&zhdr->buddy);
 915		zhdr->cpu = -1;
 916		spin_unlock(&pool->lock);
 917
 918		page = virt_to_page(zhdr);
 919		if (test_bit(NEEDS_COMPACTING, &page->private) ||
 920		    test_bit(PAGE_CLAIMED, &page->private)) {
 921			z3fold_page_unlock(zhdr);
 922			zhdr = NULL;
 923			put_cpu_ptr(pool->unbuddied);
 924			if (can_sleep)
 925				cond_resched();
 926			goto lookup;
 927		}
 928
 929		/*
 930		 * this page could not be removed from its unbuddied
 931		 * list while pool lock was held, and then we've taken
 932		 * page lock so kref_put could not be called before
 933		 * we got here, so it's safe to just call kref_get()
 934		 */
 935		kref_get(&zhdr->refcount);
 936		break;
 937	}
 938	put_cpu_ptr(pool->unbuddied);
 939
 940	if (!zhdr) {
 941		int cpu;
 942
 943		/* look for _exact_ match on other cpus' lists */
 944		for_each_online_cpu(cpu) {
 945			struct list_head *l;
 946
 947			unbuddied = per_cpu_ptr(pool->unbuddied, cpu);
 948			spin_lock(&pool->lock);
 949			l = &unbuddied[chunks];
 950
 951			zhdr = list_first_entry_or_null(READ_ONCE(l),
 952						struct z3fold_header, buddy);
 953
 954			if (!zhdr || !z3fold_page_trylock(zhdr)) {
 955				spin_unlock(&pool->lock);
 956				zhdr = NULL;
 957				continue;
 958			}
 959			list_del_init(&zhdr->buddy);
 960			zhdr->cpu = -1;
 961			spin_unlock(&pool->lock);
 962
 963			page = virt_to_page(zhdr);
 964			if (test_bit(NEEDS_COMPACTING, &page->private) ||
 965			    test_bit(PAGE_CLAIMED, &page->private)) {
 966				z3fold_page_unlock(zhdr);
 967				zhdr = NULL;
 968				if (can_sleep)
 969					cond_resched();
 970				continue;
 971			}
 972			kref_get(&zhdr->refcount);
 973			break;
 974		}
 975	}
 976
 
 
 
 
 
 977	return zhdr;
 
 
 
 
 
 
 
 978}
 979
 980/*
 981 * API Functions
 982 */
 983
 984/**
 985 * z3fold_create_pool() - create a new z3fold pool
 986 * @name:	pool name
 987 * @gfp:	gfp flags when allocating the z3fold pool structure
 988 * @ops:	user-defined operations for the z3fold pool
 989 *
 990 * Return: pointer to the new z3fold pool or NULL if the metadata allocation
 991 * failed.
 992 */
 993static struct z3fold_pool *z3fold_create_pool(const char *name, gfp_t gfp,
 994		const struct z3fold_ops *ops)
 995{
 996	struct z3fold_pool *pool = NULL;
 997	int i, cpu;
 998
 999	pool = kzalloc(sizeof(struct z3fold_pool), gfp);
1000	if (!pool)
1001		goto out;
1002	pool->c_handle = kmem_cache_create("z3fold_handle",
1003				sizeof(struct z3fold_buddy_slots),
1004				SLOTS_ALIGN, 0, NULL);
1005	if (!pool->c_handle)
1006		goto out_c;
1007	spin_lock_init(&pool->lock);
1008	spin_lock_init(&pool->stale_lock);
1009	pool->unbuddied = __alloc_percpu(sizeof(struct list_head)*NCHUNKS, 2);
 
1010	if (!pool->unbuddied)
1011		goto out_pool;
1012	for_each_possible_cpu(cpu) {
1013		struct list_head *unbuddied =
1014				per_cpu_ptr(pool->unbuddied, cpu);
1015		for_each_unbuddied_list(i, 0)
1016			INIT_LIST_HEAD(&unbuddied[i]);
1017	}
1018	INIT_LIST_HEAD(&pool->lru);
1019	INIT_LIST_HEAD(&pool->stale);
1020	atomic64_set(&pool->pages_nr, 0);
1021	pool->name = name;
1022	pool->compact_wq = create_singlethread_workqueue(pool->name);
1023	if (!pool->compact_wq)
1024		goto out_unbuddied;
1025	pool->release_wq = create_singlethread_workqueue(pool->name);
1026	if (!pool->release_wq)
1027		goto out_wq;
1028	if (z3fold_register_migration(pool))
1029		goto out_rwq;
1030	INIT_WORK(&pool->work, free_pages_work);
1031	pool->ops = ops;
1032	return pool;
1033
1034out_rwq:
1035	destroy_workqueue(pool->release_wq);
1036out_wq:
1037	destroy_workqueue(pool->compact_wq);
1038out_unbuddied:
1039	free_percpu(pool->unbuddied);
1040out_pool:
1041	kmem_cache_destroy(pool->c_handle);
1042out_c:
1043	kfree(pool);
1044out:
1045	return NULL;
1046}
1047
1048/**
1049 * z3fold_destroy_pool() - destroys an existing z3fold pool
1050 * @pool:	the z3fold pool to be destroyed
1051 *
1052 * The pool should be emptied before this function is called.
1053 */
1054static void z3fold_destroy_pool(struct z3fold_pool *pool)
1055{
1056	kmem_cache_destroy(pool->c_handle);
1057
1058	/*
1059	 * We need to destroy pool->compact_wq before pool->release_wq,
1060	 * as any pending work on pool->compact_wq will call
1061	 * queue_work(pool->release_wq, &pool->work).
1062	 *
1063	 * There are still outstanding pages until both workqueues are drained,
1064	 * so we cannot unregister migration until then.
1065	 */
1066
1067	destroy_workqueue(pool->compact_wq);
1068	destroy_workqueue(pool->release_wq);
1069	z3fold_unregister_migration(pool);
1070	kfree(pool);
1071}
1072
 
 
1073/**
1074 * z3fold_alloc() - allocates a region of a given size
1075 * @pool:	z3fold pool from which to allocate
1076 * @size:	size in bytes of the desired allocation
1077 * @gfp:	gfp flags used if the pool needs to grow
1078 * @handle:	handle of the new allocation
1079 *
1080 * This function will attempt to find a free region in the pool large enough to
1081 * satisfy the allocation request.  A search of the unbuddied lists is
1082 * performed first. If no suitable free region is found, then a new page is
1083 * allocated and added to the pool to satisfy the request.
1084 *
1085 * gfp should not set __GFP_HIGHMEM as highmem pages cannot be used
1086 * as z3fold pool pages.
1087 *
1088 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
1089 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
1090 * a new page.
1091 */
1092static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
1093			unsigned long *handle)
1094{
1095	int chunks = size_to_chunks(size);
1096	struct z3fold_header *zhdr = NULL;
1097	struct page *page = NULL;
1098	enum buddy bud;
1099	bool can_sleep = gfpflags_allow_blocking(gfp);
1100
1101	if (!size)
1102		return -EINVAL;
1103
1104	if (size > PAGE_SIZE)
1105		return -ENOSPC;
1106
1107	if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
1108		bud = HEADLESS;
1109	else {
1110retry:
1111		zhdr = __z3fold_alloc(pool, size, can_sleep);
1112		if (zhdr) {
1113			if (zhdr->first_chunks == 0) {
1114				if (zhdr->middle_chunks != 0 &&
1115				    chunks >= zhdr->start_middle)
1116					bud = LAST;
1117				else
1118					bud = FIRST;
1119			} else if (zhdr->last_chunks == 0)
1120				bud = LAST;
1121			else if (zhdr->middle_chunks == 0)
1122				bud = MIDDLE;
1123			else {
1124				if (kref_put(&zhdr->refcount,
1125					     release_z3fold_page_locked))
1126					atomic64_dec(&pool->pages_nr);
1127				else
1128					z3fold_page_unlock(zhdr);
1129				pr_err("No free chunks in unbuddied\n");
1130				WARN_ON(1);
1131				goto retry;
1132			}
1133			page = virt_to_page(zhdr);
1134			goto found;
1135		}
1136		bud = FIRST;
1137	}
1138
1139	page = NULL;
1140	if (can_sleep) {
1141		spin_lock(&pool->stale_lock);
1142		zhdr = list_first_entry_or_null(&pool->stale,
1143						struct z3fold_header, buddy);
1144		/*
1145		 * Before allocating a page, let's see if we can take one from
1146		 * the stale pages list. cancel_work_sync() can sleep so we
1147		 * limit this case to the contexts where we can sleep
1148		 */
1149		if (zhdr) {
1150			list_del(&zhdr->buddy);
1151			spin_unlock(&pool->stale_lock);
1152			cancel_work_sync(&zhdr->work);
1153			page = virt_to_page(zhdr);
1154		} else {
1155			spin_unlock(&pool->stale_lock);
1156		}
1157	}
1158	if (!page)
1159		page = alloc_page(gfp);
1160
1161	if (!page)
1162		return -ENOMEM;
1163
1164	zhdr = init_z3fold_page(page, bud == HEADLESS, pool, gfp);
1165	if (!zhdr) {
1166		__free_page(page);
1167		return -ENOMEM;
1168	}
1169	atomic64_inc(&pool->pages_nr);
1170
1171	if (bud == HEADLESS) {
1172		set_bit(PAGE_HEADLESS, &page->private);
1173		goto headless;
1174	}
1175	if (can_sleep) {
1176		lock_page(page);
1177		__SetPageMovable(page, pool->inode->i_mapping);
1178		unlock_page(page);
1179	} else {
1180		if (trylock_page(page)) {
1181			__SetPageMovable(page, pool->inode->i_mapping);
1182			unlock_page(page);
1183		}
1184	}
1185	z3fold_page_lock(zhdr);
1186
1187found:
1188	if (bud == FIRST)
1189		zhdr->first_chunks = chunks;
1190	else if (bud == LAST)
1191		zhdr->last_chunks = chunks;
1192	else {
1193		zhdr->middle_chunks = chunks;
1194		zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
1195	}
1196	add_to_unbuddied(pool, zhdr);
1197
1198headless:
1199	spin_lock(&pool->lock);
1200	/* Add/move z3fold page to beginning of LRU */
1201	if (!list_empty(&page->lru))
1202		list_del(&page->lru);
1203
1204	list_add(&page->lru, &pool->lru);
1205
1206	*handle = encode_handle(zhdr, bud);
1207	spin_unlock(&pool->lock);
1208	if (bud != HEADLESS)
1209		z3fold_page_unlock(zhdr);
1210
1211	return 0;
1212}
1213
1214/**
1215 * z3fold_free() - frees the allocation associated with the given handle
1216 * @pool:	pool in which the allocation resided
1217 * @handle:	handle associated with the allocation returned by z3fold_alloc()
1218 *
1219 * In the case that the z3fold page in which the allocation resides is under
1220 * reclaim, as indicated by the PG_reclaim flag being set, this function
1221 * only sets the first|last_chunks to 0.  The page is actually freed
1222 * once both buddies are evicted (see z3fold_reclaim_page() below).
1223 */
1224static void z3fold_free(struct z3fold_pool *pool, unsigned long handle)
1225{
1226	struct z3fold_header *zhdr;
1227	struct page *page;
1228	enum buddy bud;
1229	bool page_claimed;
1230
1231	zhdr = get_z3fold_header(handle);
1232	page = virt_to_page(zhdr);
1233	page_claimed = test_and_set_bit(PAGE_CLAIMED, &page->private);
1234
1235	if (test_bit(PAGE_HEADLESS, &page->private)) {
1236		/* if a headless page is under reclaim, just leave.
1237		 * NB: we use test_and_set_bit for a reason: if the bit
1238		 * has not been set before, we release this page
1239		 * immediately so we don't care about its value any more.
1240		 */
1241		if (!page_claimed) {
1242			spin_lock(&pool->lock);
1243			list_del(&page->lru);
1244			spin_unlock(&pool->lock);
1245			put_z3fold_header(zhdr);
1246			free_z3fold_page(page, true);
1247			atomic64_dec(&pool->pages_nr);
1248		}
1249		return;
1250	}
1251
1252	/* Non-headless case */
1253	bud = handle_to_buddy(handle);
1254
1255	switch (bud) {
1256	case FIRST:
1257		zhdr->first_chunks = 0;
1258		break;
1259	case MIDDLE:
1260		zhdr->middle_chunks = 0;
1261		break;
1262	case LAST:
1263		zhdr->last_chunks = 0;
1264		break;
1265	default:
1266		pr_err("%s: unknown bud %d\n", __func__, bud);
1267		WARN_ON(1);
1268		put_z3fold_header(zhdr);
1269		clear_bit(PAGE_CLAIMED, &page->private);
1270		return;
1271	}
1272
1273	if (!page_claimed)
1274		free_handle(handle);
1275	if (kref_put(&zhdr->refcount, release_z3fold_page_locked_list)) {
1276		atomic64_dec(&pool->pages_nr);
1277		return;
1278	}
1279	if (page_claimed) {
1280		/* the page has not been claimed by us */
1281		z3fold_page_unlock(zhdr);
1282		return;
1283	}
1284	if (unlikely(PageIsolated(page)) ||
1285	    test_and_set_bit(NEEDS_COMPACTING, &page->private)) {
1286		put_z3fold_header(zhdr);
1287		clear_bit(PAGE_CLAIMED, &page->private);
1288		return;
1289	}
1290	if (zhdr->cpu < 0 || !cpu_online(zhdr->cpu)) {
1291		spin_lock(&pool->lock);
1292		list_del_init(&zhdr->buddy);
1293		spin_unlock(&pool->lock);
1294		zhdr->cpu = -1;
1295		kref_get(&zhdr->refcount);
1296		clear_bit(PAGE_CLAIMED, &page->private);
1297		do_compact_page(zhdr, true);
1298		return;
1299	}
1300	kref_get(&zhdr->refcount);
1301	clear_bit(PAGE_CLAIMED, &page->private);
1302	queue_work_on(zhdr->cpu, pool->compact_wq, &zhdr->work);
1303	put_z3fold_header(zhdr);
1304}
1305
1306/**
1307 * z3fold_reclaim_page() - evicts allocations from a pool page and frees it
1308 * @pool:	pool from which a page will attempt to be evicted
1309 * @retries:	number of pages on the LRU list for which eviction will
1310 *		be attempted before failing
1311 *
1312 * z3fold reclaim is different from normal system reclaim in that it is done
1313 * from the bottom, up. This is because only the bottom layer, z3fold, has
1314 * information on how the allocations are organized within each z3fold page.
1315 * This has the potential to create interesting locking situations between
1316 * z3fold and the user, however.
1317 *
1318 * To avoid these, this is how z3fold_reclaim_page() should be called:
1319 *
1320 * The user detects a page should be reclaimed and calls z3fold_reclaim_page().
1321 * z3fold_reclaim_page() will remove a z3fold page from the pool LRU list and
1322 * call the user-defined eviction handler with the pool and handle as
1323 * arguments.
1324 *
1325 * If the handle can not be evicted, the eviction handler should return
1326 * non-zero. z3fold_reclaim_page() will add the z3fold page back to the
1327 * appropriate list and try the next z3fold page on the LRU up to
1328 * a user defined number of retries.
1329 *
1330 * If the handle is successfully evicted, the eviction handler should
1331 * return 0 _and_ should have called z3fold_free() on the handle. z3fold_free()
1332 * contains logic to delay freeing the page if the page is under reclaim,
1333 * as indicated by the setting of the PG_reclaim flag on the underlying page.
1334 *
1335 * If all buddies in the z3fold page are successfully evicted, then the
1336 * z3fold page can be freed.
1337 *
1338 * Returns: 0 if page is successfully freed, otherwise -EINVAL if there are
1339 * no pages to evict or an eviction handler is not registered, -EAGAIN if
1340 * the retry limit was hit.
1341 */
1342static int z3fold_reclaim_page(struct z3fold_pool *pool, unsigned int retries)
1343{
1344	int i, ret = -1;
1345	struct z3fold_header *zhdr = NULL;
1346	struct page *page = NULL;
1347	struct list_head *pos;
1348	unsigned long first_handle = 0, middle_handle = 0, last_handle = 0;
1349
1350	spin_lock(&pool->lock);
1351	if (!pool->ops || !pool->ops->evict || retries == 0) {
1352		spin_unlock(&pool->lock);
1353		return -EINVAL;
1354	}
1355	for (i = 0; i < retries; i++) {
1356		if (list_empty(&pool->lru)) {
1357			spin_unlock(&pool->lock);
1358			return -EINVAL;
1359		}
1360		list_for_each_prev(pos, &pool->lru) {
1361			page = list_entry(pos, struct page, lru);
1362
1363			/* this bit could have been set by free, in which case
1364			 * we pass over to the next page in the pool.
1365			 */
1366			if (test_and_set_bit(PAGE_CLAIMED, &page->private)) {
1367				page = NULL;
1368				continue;
1369			}
1370
1371			if (unlikely(PageIsolated(page))) {
1372				clear_bit(PAGE_CLAIMED, &page->private);
1373				page = NULL;
1374				continue;
1375			}
1376			zhdr = page_address(page);
1377			if (test_bit(PAGE_HEADLESS, &page->private))
1378				break;
1379
1380			if (!z3fold_page_trylock(zhdr)) {
1381				clear_bit(PAGE_CLAIMED, &page->private);
1382				zhdr = NULL;
1383				continue; /* can't evict at this point */
1384			}
1385			if (zhdr->foreign_handles) {
1386				clear_bit(PAGE_CLAIMED, &page->private);
1387				z3fold_page_unlock(zhdr);
1388				zhdr = NULL;
1389				continue; /* can't evict such page */
1390			}
1391			kref_get(&zhdr->refcount);
1392			list_del_init(&zhdr->buddy);
1393			zhdr->cpu = -1;
1394			break;
1395		}
1396
1397		if (!zhdr)
1398			break;
1399
1400		list_del_init(&page->lru);
1401		spin_unlock(&pool->lock);
1402
1403		if (!test_bit(PAGE_HEADLESS, &page->private)) {
1404			/*
1405			 * We need encode the handles before unlocking, and
1406			 * use our local slots structure because z3fold_free
1407			 * can zero out zhdr->slots and we can't do much
1408			 * about that
1409			 */
1410			first_handle = 0;
1411			last_handle = 0;
1412			middle_handle = 0;
1413			if (zhdr->first_chunks)
1414				first_handle = encode_handle(zhdr, FIRST);
1415			if (zhdr->middle_chunks)
1416				middle_handle = encode_handle(zhdr, MIDDLE);
1417			if (zhdr->last_chunks)
1418				last_handle = encode_handle(zhdr, LAST);
1419			/*
1420			 * it's safe to unlock here because we hold a
1421			 * reference to this page
1422			 */
1423			z3fold_page_unlock(zhdr);
1424		} else {
1425			first_handle = encode_handle(zhdr, HEADLESS);
1426			last_handle = middle_handle = 0;
1427		}
1428		/* Issue the eviction callback(s) */
1429		if (middle_handle) {
1430			ret = pool->ops->evict(pool, middle_handle);
1431			if (ret)
1432				goto next;
1433			free_handle(middle_handle);
1434		}
1435		if (first_handle) {
1436			ret = pool->ops->evict(pool, first_handle);
1437			if (ret)
1438				goto next;
1439			free_handle(first_handle);
1440		}
1441		if (last_handle) {
1442			ret = pool->ops->evict(pool, last_handle);
1443			if (ret)
1444				goto next;
1445			free_handle(last_handle);
1446		}
1447next:
1448		if (test_bit(PAGE_HEADLESS, &page->private)) {
1449			if (ret == 0) {
1450				free_z3fold_page(page, true);
1451				atomic64_dec(&pool->pages_nr);
1452				return 0;
1453			}
1454			spin_lock(&pool->lock);
1455			list_add(&page->lru, &pool->lru);
1456			spin_unlock(&pool->lock);
1457			clear_bit(PAGE_CLAIMED, &page->private);
1458		} else {
1459			z3fold_page_lock(zhdr);
1460			if (kref_put(&zhdr->refcount,
1461					release_z3fold_page_locked)) {
1462				atomic64_dec(&pool->pages_nr);
1463				return 0;
1464			}
1465			/*
1466			 * if we are here, the page is still not completely
1467			 * free. Take the global pool lock then to be able
1468			 * to add it back to the lru list
1469			 */
1470			spin_lock(&pool->lock);
1471			list_add(&page->lru, &pool->lru);
1472			spin_unlock(&pool->lock);
1473			z3fold_page_unlock(zhdr);
1474			clear_bit(PAGE_CLAIMED, &page->private);
1475		}
1476
1477		/* We started off locked to we need to lock the pool back */
1478		spin_lock(&pool->lock);
1479	}
1480	spin_unlock(&pool->lock);
1481	return -EAGAIN;
1482}
1483
1484/**
1485 * z3fold_map() - maps the allocation associated with the given handle
1486 * @pool:	pool in which the allocation resides
1487 * @handle:	handle associated with the allocation to be mapped
1488 *
1489 * Extracts the buddy number from handle and constructs the pointer to the
1490 * correct starting chunk within the page.
1491 *
1492 * Returns: a pointer to the mapped allocation
1493 */
1494static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle)
1495{
1496	struct z3fold_header *zhdr;
1497	struct page *page;
1498	void *addr;
1499	enum buddy buddy;
1500
1501	zhdr = get_z3fold_header(handle);
1502	addr = zhdr;
1503	page = virt_to_page(zhdr);
1504
1505	if (test_bit(PAGE_HEADLESS, &page->private))
1506		goto out;
1507
1508	buddy = handle_to_buddy(handle);
1509	switch (buddy) {
1510	case FIRST:
1511		addr += ZHDR_SIZE_ALIGNED;
1512		break;
1513	case MIDDLE:
1514		addr += zhdr->start_middle << CHUNK_SHIFT;
1515		set_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1516		break;
1517	case LAST:
1518		addr += PAGE_SIZE - (handle_to_chunks(handle) << CHUNK_SHIFT);
1519		break;
1520	default:
1521		pr_err("unknown buddy id %d\n", buddy);
1522		WARN_ON(1);
1523		addr = NULL;
1524		break;
1525	}
1526
1527	if (addr)
1528		zhdr->mapped_count++;
1529out:
1530	put_z3fold_header(zhdr);
1531	return addr;
1532}
1533
1534/**
1535 * z3fold_unmap() - unmaps the allocation associated with the given handle
1536 * @pool:	pool in which the allocation resides
1537 * @handle:	handle associated with the allocation to be unmapped
1538 */
1539static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle)
1540{
1541	struct z3fold_header *zhdr;
1542	struct page *page;
1543	enum buddy buddy;
1544
1545	zhdr = get_z3fold_header(handle);
1546	page = virt_to_page(zhdr);
1547
1548	if (test_bit(PAGE_HEADLESS, &page->private))
1549		return;
1550
1551	buddy = handle_to_buddy(handle);
1552	if (buddy == MIDDLE)
1553		clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1554	zhdr->mapped_count--;
1555	put_z3fold_header(zhdr);
1556}
1557
1558/**
1559 * z3fold_get_pool_size() - gets the z3fold pool size in pages
1560 * @pool:	pool whose size is being queried
1561 *
1562 * Returns: size in pages of the given pool.
1563 */
1564static u64 z3fold_get_pool_size(struct z3fold_pool *pool)
1565{
1566	return atomic64_read(&pool->pages_nr);
1567}
1568
1569static bool z3fold_page_isolate(struct page *page, isolate_mode_t mode)
1570{
1571	struct z3fold_header *zhdr;
1572	struct z3fold_pool *pool;
1573
1574	VM_BUG_ON_PAGE(!PageMovable(page), page);
1575	VM_BUG_ON_PAGE(PageIsolated(page), page);
1576
1577	if (test_bit(PAGE_HEADLESS, &page->private) ||
1578	    test_bit(PAGE_CLAIMED, &page->private))
1579		return false;
1580
1581	zhdr = page_address(page);
1582	z3fold_page_lock(zhdr);
1583	if (test_bit(NEEDS_COMPACTING, &page->private) ||
1584	    test_bit(PAGE_STALE, &page->private))
1585		goto out;
1586
1587	if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0)
1588		goto out;
1589
 
 
1590	pool = zhdr_to_pool(zhdr);
1591	spin_lock(&pool->lock);
1592	if (!list_empty(&zhdr->buddy))
1593		list_del_init(&zhdr->buddy);
1594	if (!list_empty(&page->lru))
1595		list_del_init(&page->lru);
1596	spin_unlock(&pool->lock);
1597
1598	kref_get(&zhdr->refcount);
1599	z3fold_page_unlock(zhdr);
1600	return true;
1601
1602out:
1603	z3fold_page_unlock(zhdr);
1604	return false;
1605}
1606
1607static int z3fold_page_migrate(struct address_space *mapping, struct page *newpage,
1608			       struct page *page, enum migrate_mode mode)
1609{
1610	struct z3fold_header *zhdr, *new_zhdr;
1611	struct z3fold_pool *pool;
1612	struct address_space *new_mapping;
1613
1614	VM_BUG_ON_PAGE(!PageMovable(page), page);
1615	VM_BUG_ON_PAGE(!PageIsolated(page), page);
 
1616	VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
1617
1618	zhdr = page_address(page);
1619	pool = zhdr_to_pool(zhdr);
1620
1621	if (!z3fold_page_trylock(zhdr)) {
1622		return -EAGAIN;
1623	}
1624	if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0) {
 
1625		z3fold_page_unlock(zhdr);
1626		return -EBUSY;
1627	}
1628	if (work_pending(&zhdr->work)) {
1629		z3fold_page_unlock(zhdr);
1630		return -EAGAIN;
1631	}
1632	new_zhdr = page_address(newpage);
1633	memcpy(new_zhdr, zhdr, PAGE_SIZE);
1634	newpage->private = page->private;
1635	page->private = 0;
1636	z3fold_page_unlock(zhdr);
1637	spin_lock_init(&new_zhdr->page_lock);
1638	INIT_WORK(&new_zhdr->work, compact_page_work);
1639	/*
1640	 * z3fold_page_isolate() ensures that new_zhdr->buddy is empty,
1641	 * so we only have to reinitialize it.
1642	 */
1643	INIT_LIST_HEAD(&new_zhdr->buddy);
1644	new_mapping = page_mapping(page);
1645	__ClearPageMovable(page);
1646	ClearPagePrivate(page);
1647
1648	get_page(newpage);
1649	z3fold_page_lock(new_zhdr);
1650	if (new_zhdr->first_chunks)
1651		encode_handle(new_zhdr, FIRST);
1652	if (new_zhdr->last_chunks)
1653		encode_handle(new_zhdr, LAST);
1654	if (new_zhdr->middle_chunks)
1655		encode_handle(new_zhdr, MIDDLE);
1656	set_bit(NEEDS_COMPACTING, &newpage->private);
1657	new_zhdr->cpu = smp_processor_id();
1658	spin_lock(&pool->lock);
1659	list_add(&newpage->lru, &pool->lru);
1660	spin_unlock(&pool->lock);
1661	__SetPageMovable(newpage, new_mapping);
1662	z3fold_page_unlock(new_zhdr);
1663
1664	queue_work_on(new_zhdr->cpu, pool->compact_wq, &new_zhdr->work);
1665
1666	page_mapcount_reset(page);
 
1667	put_page(page);
1668	return 0;
1669}
1670
1671static void z3fold_page_putback(struct page *page)
1672{
1673	struct z3fold_header *zhdr;
1674	struct z3fold_pool *pool;
1675
1676	zhdr = page_address(page);
1677	pool = zhdr_to_pool(zhdr);
1678
1679	z3fold_page_lock(zhdr);
1680	if (!list_empty(&zhdr->buddy))
1681		list_del_init(&zhdr->buddy);
1682	INIT_LIST_HEAD(&page->lru);
1683	if (kref_put(&zhdr->refcount, release_z3fold_page_locked)) {
1684		atomic64_dec(&pool->pages_nr);
1685		return;
1686	}
1687	spin_lock(&pool->lock);
1688	list_add(&page->lru, &pool->lru);
1689	spin_unlock(&pool->lock);
1690	z3fold_page_unlock(zhdr);
1691}
1692
1693static const struct address_space_operations z3fold_aops = {
1694	.isolate_page = z3fold_page_isolate,
1695	.migratepage = z3fold_page_migrate,
1696	.putback_page = z3fold_page_putback,
1697};
1698
1699/*****************
1700 * zpool
1701 ****************/
1702
1703static int z3fold_zpool_evict(struct z3fold_pool *pool, unsigned long handle)
1704{
1705	if (pool->zpool && pool->zpool_ops && pool->zpool_ops->evict)
1706		return pool->zpool_ops->evict(pool->zpool, handle);
1707	else
1708		return -ENOENT;
1709}
1710
1711static const struct z3fold_ops z3fold_zpool_ops = {
1712	.evict =	z3fold_zpool_evict
1713};
1714
1715static void *z3fold_zpool_create(const char *name, gfp_t gfp,
1716			       const struct zpool_ops *zpool_ops,
1717			       struct zpool *zpool)
1718{
1719	struct z3fold_pool *pool;
1720
1721	pool = z3fold_create_pool(name, gfp,
1722				zpool_ops ? &z3fold_zpool_ops : NULL);
1723	if (pool) {
1724		pool->zpool = zpool;
1725		pool->zpool_ops = zpool_ops;
1726	}
1727	return pool;
1728}
1729
1730static void z3fold_zpool_destroy(void *pool)
1731{
1732	z3fold_destroy_pool(pool);
1733}
1734
1735static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp,
1736			unsigned long *handle)
1737{
1738	return z3fold_alloc(pool, size, gfp, handle);
1739}
1740static void z3fold_zpool_free(void *pool, unsigned long handle)
1741{
1742	z3fold_free(pool, handle);
1743}
1744
1745static int z3fold_zpool_shrink(void *pool, unsigned int pages,
1746			unsigned int *reclaimed)
1747{
1748	unsigned int total = 0;
1749	int ret = -EINVAL;
1750
1751	while (total < pages) {
1752		ret = z3fold_reclaim_page(pool, 8);
1753		if (ret < 0)
1754			break;
1755		total++;
1756	}
1757
1758	if (reclaimed)
1759		*reclaimed = total;
1760
1761	return ret;
1762}
1763
1764static void *z3fold_zpool_map(void *pool, unsigned long handle,
1765			enum zpool_mapmode mm)
1766{
1767	return z3fold_map(pool, handle);
1768}
1769static void z3fold_zpool_unmap(void *pool, unsigned long handle)
1770{
1771	z3fold_unmap(pool, handle);
1772}
1773
1774static u64 z3fold_zpool_total_size(void *pool)
1775{
1776	return z3fold_get_pool_size(pool) * PAGE_SIZE;
1777}
1778
1779static struct zpool_driver z3fold_zpool_driver = {
1780	.type =		"z3fold",
 
1781	.owner =	THIS_MODULE,
1782	.create =	z3fold_zpool_create,
1783	.destroy =	z3fold_zpool_destroy,
1784	.malloc =	z3fold_zpool_malloc,
1785	.free =		z3fold_zpool_free,
1786	.shrink =	z3fold_zpool_shrink,
1787	.map =		z3fold_zpool_map,
1788	.unmap =	z3fold_zpool_unmap,
1789	.total_size =	z3fold_zpool_total_size,
1790};
1791
1792MODULE_ALIAS("zpool-z3fold");
1793
1794static int __init init_z3fold(void)
1795{
1796	int ret;
1797
1798	/* Make sure the z3fold header is not larger than the page size */
1799	BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE);
1800	ret = z3fold_mount();
1801	if (ret)
1802		return ret;
1803
1804	zpool_register_driver(&z3fold_zpool_driver);
1805
1806	return 0;
1807}
1808
1809static void __exit exit_z3fold(void)
1810{
1811	z3fold_unmount();
1812	zpool_unregister_driver(&z3fold_zpool_driver);
1813}
1814
1815module_init(init_z3fold);
1816module_exit(exit_z3fold);
1817
1818MODULE_LICENSE("GPL");
1819MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>");
1820MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages");