2 * Compressed RAM block device
4 * Copyright (C) 2008, 2009, 2010 Nitin Gupta
5 * 2012, 2013 Minchan Kim
7 * This code is released using a dual license strategy: BSD/GPL
8 * You can choose the licence that better fits your requirements.
10 * Released under the terms of 3-clause BSD License
11 * Released under the terms of GNU General Public License Version 2.0
15 #define KMSG_COMPONENT "zram"
16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
18 #include <linux/module.h>
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/bitops.h>
22 #include <linux/blkdev.h>
23 #include <linux/buffer_head.h>
24 #include <linux/device.h>
25 #include <linux/highmem.h>
26 #include <linux/slab.h>
27 #include <linux/backing-dev.h>
28 #include <linux/string.h>
29 #include <linux/vmalloc.h>
30 #include <linux/err.h>
31 #include <linux/idr.h>
32 #include <linux/sysfs.h>
33 #include <linux/debugfs.h>
34 #include <linux/cpuhotplug.h>
35 #include <linux/part_stat.h>
39 static DEFINE_IDR(zram_index_idr);
40 /* idr index must be protected */
41 static DEFINE_MUTEX(zram_index_mutex);
43 static int zram_major;
44 static const char *default_compressor = CONFIG_ZRAM_DEF_COMP;
46 /* Module params (documentation at end) */
47 static unsigned int num_devices = 1;
49 * Pages that compress to sizes equals or greater than this are stored
50 * uncompressed in memory.
52 static size_t huge_class_size;
54 static const struct block_device_operations zram_devops;
56 static void zram_free_page(struct zram *zram, size_t index);
57 static int zram_read_page(struct zram *zram, struct page *page, u32 index,
60 static int zram_slot_trylock(struct zram *zram, u32 index)
62 return bit_spin_trylock(ZRAM_LOCK, &zram->table[index].flags);
65 static void zram_slot_lock(struct zram *zram, u32 index)
67 bit_spin_lock(ZRAM_LOCK, &zram->table[index].flags);
70 static void zram_slot_unlock(struct zram *zram, u32 index)
72 bit_spin_unlock(ZRAM_LOCK, &zram->table[index].flags);
75 static inline bool init_done(struct zram *zram)
77 return zram->disksize;
80 static inline struct zram *dev_to_zram(struct device *dev)
82 return (struct zram *)dev_to_disk(dev)->private_data;
85 static unsigned long zram_get_handle(struct zram *zram, u32 index)
87 return zram->table[index].handle;
90 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
92 zram->table[index].handle = handle;
95 /* flag operations require table entry bit_spin_lock() being held */
96 static bool zram_test_flag(struct zram *zram, u32 index,
97 enum zram_pageflags flag)
99 return zram->table[index].flags & BIT(flag);
102 static void zram_set_flag(struct zram *zram, u32 index,
103 enum zram_pageflags flag)
105 zram->table[index].flags |= BIT(flag);
108 static void zram_clear_flag(struct zram *zram, u32 index,
109 enum zram_pageflags flag)
111 zram->table[index].flags &= ~BIT(flag);
114 static inline void zram_set_element(struct zram *zram, u32 index,
115 unsigned long element)
117 zram->table[index].element = element;
120 static unsigned long zram_get_element(struct zram *zram, u32 index)
122 return zram->table[index].element;
125 static size_t zram_get_obj_size(struct zram *zram, u32 index)
127 return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
130 static void zram_set_obj_size(struct zram *zram,
131 u32 index, size_t size)
133 unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
135 zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
138 static inline bool zram_allocated(struct zram *zram, u32 index)
140 return zram_get_obj_size(zram, index) ||
141 zram_test_flag(zram, index, ZRAM_SAME) ||
142 zram_test_flag(zram, index, ZRAM_WB);
145 #if PAGE_SIZE != 4096
146 static inline bool is_partial_io(struct bio_vec *bvec)
148 return bvec->bv_len != PAGE_SIZE;
150 #define ZRAM_PARTIAL_IO 1
152 static inline bool is_partial_io(struct bio_vec *bvec)
158 static inline void zram_set_priority(struct zram *zram, u32 index, u32 prio)
160 prio &= ZRAM_COMP_PRIORITY_MASK;
162 * Clear previous priority value first, in case if we recompress
163 * further an already recompressed page
165 zram->table[index].flags &= ~(ZRAM_COMP_PRIORITY_MASK <<
166 ZRAM_COMP_PRIORITY_BIT1);
167 zram->table[index].flags |= (prio << ZRAM_COMP_PRIORITY_BIT1);
170 static inline u32 zram_get_priority(struct zram *zram, u32 index)
172 u32 prio = zram->table[index].flags >> ZRAM_COMP_PRIORITY_BIT1;
174 return prio & ZRAM_COMP_PRIORITY_MASK;
177 static inline void update_used_max(struct zram *zram,
178 const unsigned long pages)
180 unsigned long cur_max = atomic_long_read(&zram->stats.max_used_pages);
183 if (cur_max >= pages)
185 } while (!atomic_long_try_cmpxchg(&zram->stats.max_used_pages,
189 static inline void zram_fill_page(void *ptr, unsigned long len,
192 WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
193 memset_l(ptr, value, len / sizeof(unsigned long));
196 static bool page_same_filled(void *ptr, unsigned long *element)
200 unsigned int pos, last_pos = PAGE_SIZE / sizeof(*page) - 1;
202 page = (unsigned long *)ptr;
205 if (val != page[last_pos])
208 for (pos = 1; pos < last_pos; pos++) {
209 if (val != page[pos])
218 static ssize_t initstate_show(struct device *dev,
219 struct device_attribute *attr, char *buf)
222 struct zram *zram = dev_to_zram(dev);
224 down_read(&zram->init_lock);
225 val = init_done(zram);
226 up_read(&zram->init_lock);
228 return scnprintf(buf, PAGE_SIZE, "%u\n", val);
231 static ssize_t disksize_show(struct device *dev,
232 struct device_attribute *attr, char *buf)
234 struct zram *zram = dev_to_zram(dev);
236 return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
239 static ssize_t mem_limit_store(struct device *dev,
240 struct device_attribute *attr, const char *buf, size_t len)
244 struct zram *zram = dev_to_zram(dev);
246 limit = memparse(buf, &tmp);
247 if (buf == tmp) /* no chars parsed, invalid input */
250 down_write(&zram->init_lock);
251 zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
252 up_write(&zram->init_lock);
257 static ssize_t mem_used_max_store(struct device *dev,
258 struct device_attribute *attr, const char *buf, size_t len)
262 struct zram *zram = dev_to_zram(dev);
264 err = kstrtoul(buf, 10, &val);
268 down_read(&zram->init_lock);
269 if (init_done(zram)) {
270 atomic_long_set(&zram->stats.max_used_pages,
271 zs_get_total_pages(zram->mem_pool));
273 up_read(&zram->init_lock);
279 * Mark all pages which are older than or equal to cutoff as IDLE.
280 * Callers should hold the zram init lock in read mode
282 static void mark_idle(struct zram *zram, ktime_t cutoff)
285 unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
288 for (index = 0; index < nr_pages; index++) {
290 * Do not mark ZRAM_UNDER_WB slot as ZRAM_IDLE to close race.
291 * See the comment in writeback_store.
293 zram_slot_lock(zram, index);
294 if (zram_allocated(zram, index) &&
295 !zram_test_flag(zram, index, ZRAM_UNDER_WB)) {
296 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
297 is_idle = !cutoff || ktime_after(cutoff, zram->table[index].ac_time);
300 zram_set_flag(zram, index, ZRAM_IDLE);
302 zram_slot_unlock(zram, index);
306 static ssize_t idle_store(struct device *dev,
307 struct device_attribute *attr, const char *buf, size_t len)
309 struct zram *zram = dev_to_zram(dev);
310 ktime_t cutoff_time = 0;
311 ssize_t rv = -EINVAL;
313 if (!sysfs_streq(buf, "all")) {
315 * If it did not parse as 'all' try to treat it as an integer
316 * when we have memory tracking enabled.
320 if (IS_ENABLED(CONFIG_ZRAM_MEMORY_TRACKING) && !kstrtoull(buf, 0, &age_sec))
321 cutoff_time = ktime_sub(ktime_get_boottime(),
322 ns_to_ktime(age_sec * NSEC_PER_SEC));
327 down_read(&zram->init_lock);
328 if (!init_done(zram))
332 * A cutoff_time of 0 marks everything as idle, this is the
335 mark_idle(zram, cutoff_time);
339 up_read(&zram->init_lock);
344 #ifdef CONFIG_ZRAM_WRITEBACK
345 static ssize_t writeback_limit_enable_store(struct device *dev,
346 struct device_attribute *attr, const char *buf, size_t len)
348 struct zram *zram = dev_to_zram(dev);
350 ssize_t ret = -EINVAL;
352 if (kstrtoull(buf, 10, &val))
355 down_read(&zram->init_lock);
356 spin_lock(&zram->wb_limit_lock);
357 zram->wb_limit_enable = val;
358 spin_unlock(&zram->wb_limit_lock);
359 up_read(&zram->init_lock);
365 static ssize_t writeback_limit_enable_show(struct device *dev,
366 struct device_attribute *attr, char *buf)
369 struct zram *zram = dev_to_zram(dev);
371 down_read(&zram->init_lock);
372 spin_lock(&zram->wb_limit_lock);
373 val = zram->wb_limit_enable;
374 spin_unlock(&zram->wb_limit_lock);
375 up_read(&zram->init_lock);
377 return scnprintf(buf, PAGE_SIZE, "%d\n", val);
380 static ssize_t writeback_limit_store(struct device *dev,
381 struct device_attribute *attr, const char *buf, size_t len)
383 struct zram *zram = dev_to_zram(dev);
385 ssize_t ret = -EINVAL;
387 if (kstrtoull(buf, 10, &val))
390 down_read(&zram->init_lock);
391 spin_lock(&zram->wb_limit_lock);
392 zram->bd_wb_limit = val;
393 spin_unlock(&zram->wb_limit_lock);
394 up_read(&zram->init_lock);
400 static ssize_t writeback_limit_show(struct device *dev,
401 struct device_attribute *attr, char *buf)
404 struct zram *zram = dev_to_zram(dev);
406 down_read(&zram->init_lock);
407 spin_lock(&zram->wb_limit_lock);
408 val = zram->bd_wb_limit;
409 spin_unlock(&zram->wb_limit_lock);
410 up_read(&zram->init_lock);
412 return scnprintf(buf, PAGE_SIZE, "%llu\n", val);
415 static void reset_bdev(struct zram *zram)
417 struct block_device *bdev;
419 if (!zram->backing_dev)
423 blkdev_put(bdev, zram);
424 /* hope filp_close flush all of IO */
425 filp_close(zram->backing_dev, NULL);
426 zram->backing_dev = NULL;
428 zram->disk->fops = &zram_devops;
429 kvfree(zram->bitmap);
433 static ssize_t backing_dev_show(struct device *dev,
434 struct device_attribute *attr, char *buf)
437 struct zram *zram = dev_to_zram(dev);
441 down_read(&zram->init_lock);
442 file = zram->backing_dev;
444 memcpy(buf, "none\n", 5);
445 up_read(&zram->init_lock);
449 p = file_path(file, buf, PAGE_SIZE - 1);
456 memmove(buf, p, ret);
459 up_read(&zram->init_lock);
463 static ssize_t backing_dev_store(struct device *dev,
464 struct device_attribute *attr, const char *buf, size_t len)
468 struct file *backing_dev = NULL;
470 struct address_space *mapping;
471 unsigned int bitmap_sz;
472 unsigned long nr_pages, *bitmap = NULL;
473 struct block_device *bdev = NULL;
475 struct zram *zram = dev_to_zram(dev);
477 file_name = kmalloc(PATH_MAX, GFP_KERNEL);
481 down_write(&zram->init_lock);
482 if (init_done(zram)) {
483 pr_info("Can't setup backing device for initialized device\n");
488 strscpy(file_name, buf, PATH_MAX);
489 /* ignore trailing newline */
490 sz = strlen(file_name);
491 if (sz > 0 && file_name[sz - 1] == '\n')
492 file_name[sz - 1] = 0x00;
494 backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
495 if (IS_ERR(backing_dev)) {
496 err = PTR_ERR(backing_dev);
501 mapping = backing_dev->f_mapping;
502 inode = mapping->host;
504 /* Support only block device in this moment */
505 if (!S_ISBLK(inode->i_mode)) {
510 bdev = blkdev_get_by_dev(inode->i_rdev, BLK_OPEN_READ | BLK_OPEN_WRITE,
518 nr_pages = i_size_read(inode) >> PAGE_SHIFT;
519 bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
520 bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
529 zram->backing_dev = backing_dev;
530 zram->bitmap = bitmap;
531 zram->nr_pages = nr_pages;
532 up_write(&zram->init_lock);
534 pr_info("setup backing device %s\n", file_name);
542 blkdev_put(bdev, zram);
545 filp_close(backing_dev, NULL);
547 up_write(&zram->init_lock);
554 static unsigned long alloc_block_bdev(struct zram *zram)
556 unsigned long blk_idx = 1;
558 /* skip 0 bit to confuse zram.handle = 0 */
559 blk_idx = find_next_zero_bit(zram->bitmap, zram->nr_pages, blk_idx);
560 if (blk_idx == zram->nr_pages)
563 if (test_and_set_bit(blk_idx, zram->bitmap))
566 atomic64_inc(&zram->stats.bd_count);
570 static void free_block_bdev(struct zram *zram, unsigned long blk_idx)
574 was_set = test_and_clear_bit(blk_idx, zram->bitmap);
575 WARN_ON_ONCE(!was_set);
576 atomic64_dec(&zram->stats.bd_count);
579 static void read_from_bdev_async(struct zram *zram, struct page *page,
580 unsigned long entry, struct bio *parent)
584 bio = bio_alloc(zram->bdev, 1, parent->bi_opf, GFP_NOIO);
585 bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
586 __bio_add_page(bio, page, PAGE_SIZE, 0);
587 bio_chain(bio, parent);
591 #define PAGE_WB_SIG "page_index="
593 #define PAGE_WRITEBACK 0
594 #define HUGE_WRITEBACK (1<<0)
595 #define IDLE_WRITEBACK (1<<1)
596 #define INCOMPRESSIBLE_WRITEBACK (1<<2)
598 static ssize_t writeback_store(struct device *dev,
599 struct device_attribute *attr, const char *buf, size_t len)
601 struct zram *zram = dev_to_zram(dev);
602 unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
603 unsigned long index = 0;
605 struct bio_vec bio_vec;
609 unsigned long blk_idx = 0;
611 if (sysfs_streq(buf, "idle"))
612 mode = IDLE_WRITEBACK;
613 else if (sysfs_streq(buf, "huge"))
614 mode = HUGE_WRITEBACK;
615 else if (sysfs_streq(buf, "huge_idle"))
616 mode = IDLE_WRITEBACK | HUGE_WRITEBACK;
617 else if (sysfs_streq(buf, "incompressible"))
618 mode = INCOMPRESSIBLE_WRITEBACK;
620 if (strncmp(buf, PAGE_WB_SIG, sizeof(PAGE_WB_SIG) - 1))
623 if (kstrtol(buf + sizeof(PAGE_WB_SIG) - 1, 10, &index) ||
628 mode = PAGE_WRITEBACK;
631 down_read(&zram->init_lock);
632 if (!init_done(zram)) {
634 goto release_init_lock;
637 if (!zram->backing_dev) {
639 goto release_init_lock;
642 page = alloc_page(GFP_KERNEL);
645 goto release_init_lock;
648 for (; nr_pages != 0; index++, nr_pages--) {
649 spin_lock(&zram->wb_limit_lock);
650 if (zram->wb_limit_enable && !zram->bd_wb_limit) {
651 spin_unlock(&zram->wb_limit_lock);
655 spin_unlock(&zram->wb_limit_lock);
658 blk_idx = alloc_block_bdev(zram);
665 zram_slot_lock(zram, index);
666 if (!zram_allocated(zram, index))
669 if (zram_test_flag(zram, index, ZRAM_WB) ||
670 zram_test_flag(zram, index, ZRAM_SAME) ||
671 zram_test_flag(zram, index, ZRAM_UNDER_WB))
674 if (mode & IDLE_WRITEBACK &&
675 !zram_test_flag(zram, index, ZRAM_IDLE))
677 if (mode & HUGE_WRITEBACK &&
678 !zram_test_flag(zram, index, ZRAM_HUGE))
680 if (mode & INCOMPRESSIBLE_WRITEBACK &&
681 !zram_test_flag(zram, index, ZRAM_INCOMPRESSIBLE))
685 * Clearing ZRAM_UNDER_WB is duty of caller.
686 * IOW, zram_free_page never clear it.
688 zram_set_flag(zram, index, ZRAM_UNDER_WB);
689 /* Need for hugepage writeback racing */
690 zram_set_flag(zram, index, ZRAM_IDLE);
691 zram_slot_unlock(zram, index);
692 if (zram_read_page(zram, page, index, NULL)) {
693 zram_slot_lock(zram, index);
694 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
695 zram_clear_flag(zram, index, ZRAM_IDLE);
696 zram_slot_unlock(zram, index);
700 bio_init(&bio, zram->bdev, &bio_vec, 1,
701 REQ_OP_WRITE | REQ_SYNC);
702 bio.bi_iter.bi_sector = blk_idx * (PAGE_SIZE >> 9);
703 __bio_add_page(&bio, page, PAGE_SIZE, 0);
706 * XXX: A single page IO would be inefficient for write
707 * but it would be not bad as starter.
709 err = submit_bio_wait(&bio);
711 zram_slot_lock(zram, index);
712 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
713 zram_clear_flag(zram, index, ZRAM_IDLE);
714 zram_slot_unlock(zram, index);
716 * BIO errors are not fatal, we continue and simply
717 * attempt to writeback the remaining objects (pages).
718 * At the same time we need to signal user-space that
719 * some writes (at least one, but also could be all of
720 * them) were not successful and we do so by returning
721 * the most recent BIO error.
727 atomic64_inc(&zram->stats.bd_writes);
729 * We released zram_slot_lock so need to check if the slot was
730 * changed. If there is freeing for the slot, we can catch it
731 * easily by zram_allocated.
732 * A subtle case is the slot is freed/reallocated/marked as
733 * ZRAM_IDLE again. To close the race, idle_store doesn't
734 * mark ZRAM_IDLE once it found the slot was ZRAM_UNDER_WB.
735 * Thus, we could close the race by checking ZRAM_IDLE bit.
737 zram_slot_lock(zram, index);
738 if (!zram_allocated(zram, index) ||
739 !zram_test_flag(zram, index, ZRAM_IDLE)) {
740 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
741 zram_clear_flag(zram, index, ZRAM_IDLE);
745 zram_free_page(zram, index);
746 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
747 zram_set_flag(zram, index, ZRAM_WB);
748 zram_set_element(zram, index, blk_idx);
750 atomic64_inc(&zram->stats.pages_stored);
751 spin_lock(&zram->wb_limit_lock);
752 if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
753 zram->bd_wb_limit -= 1UL << (PAGE_SHIFT - 12);
754 spin_unlock(&zram->wb_limit_lock);
756 zram_slot_unlock(zram, index);
760 free_block_bdev(zram, blk_idx);
763 up_read(&zram->init_lock);
769 struct work_struct work;
776 static void zram_sync_read(struct work_struct *work)
778 struct zram_work *zw = container_of(work, struct zram_work, work);
782 bio_init(&bio, zw->zram->bdev, &bv, 1, REQ_OP_READ);
783 bio.bi_iter.bi_sector = zw->entry * (PAGE_SIZE >> 9);
784 __bio_add_page(&bio, zw->page, PAGE_SIZE, 0);
785 zw->error = submit_bio_wait(&bio);
789 * Block layer want one ->submit_bio to be active at a time, so if we use
790 * chained IO with parent IO in same context, it's a deadlock. To avoid that,
791 * use a worker thread context.
793 static int read_from_bdev_sync(struct zram *zram, struct page *page,
796 struct zram_work work;
802 INIT_WORK_ONSTACK(&work.work, zram_sync_read);
803 queue_work(system_unbound_wq, &work.work);
804 flush_work(&work.work);
805 destroy_work_on_stack(&work.work);
810 static int read_from_bdev(struct zram *zram, struct page *page,
811 unsigned long entry, struct bio *parent)
813 atomic64_inc(&zram->stats.bd_reads);
815 if (WARN_ON_ONCE(!IS_ENABLED(ZRAM_PARTIAL_IO)))
817 return read_from_bdev_sync(zram, page, entry);
819 read_from_bdev_async(zram, page, entry, parent);
823 static inline void reset_bdev(struct zram *zram) {};
824 static int read_from_bdev(struct zram *zram, struct page *page,
825 unsigned long entry, struct bio *parent)
830 static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
833 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
835 static struct dentry *zram_debugfs_root;
837 static void zram_debugfs_create(void)
839 zram_debugfs_root = debugfs_create_dir("zram", NULL);
842 static void zram_debugfs_destroy(void)
844 debugfs_remove_recursive(zram_debugfs_root);
847 static void zram_accessed(struct zram *zram, u32 index)
849 zram_clear_flag(zram, index, ZRAM_IDLE);
850 zram->table[index].ac_time = ktime_get_boottime();
853 static ssize_t read_block_state(struct file *file, char __user *buf,
854 size_t count, loff_t *ppos)
857 ssize_t index, written = 0;
858 struct zram *zram = file->private_data;
859 unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
860 struct timespec64 ts;
862 kbuf = kvmalloc(count, GFP_KERNEL);
866 down_read(&zram->init_lock);
867 if (!init_done(zram)) {
868 up_read(&zram->init_lock);
873 for (index = *ppos; index < nr_pages; index++) {
876 zram_slot_lock(zram, index);
877 if (!zram_allocated(zram, index))
880 ts = ktime_to_timespec64(zram->table[index].ac_time);
881 copied = snprintf(kbuf + written, count,
882 "%12zd %12lld.%06lu %c%c%c%c%c%c\n",
883 index, (s64)ts.tv_sec,
884 ts.tv_nsec / NSEC_PER_USEC,
885 zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
886 zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
887 zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.',
888 zram_test_flag(zram, index, ZRAM_IDLE) ? 'i' : '.',
889 zram_get_priority(zram, index) ? 'r' : '.',
890 zram_test_flag(zram, index,
891 ZRAM_INCOMPRESSIBLE) ? 'n' : '.');
893 if (count <= copied) {
894 zram_slot_unlock(zram, index);
900 zram_slot_unlock(zram, index);
904 up_read(&zram->init_lock);
905 if (copy_to_user(buf, kbuf, written))
912 static const struct file_operations proc_zram_block_state_op = {
914 .read = read_block_state,
915 .llseek = default_llseek,
918 static void zram_debugfs_register(struct zram *zram)
920 if (!zram_debugfs_root)
923 zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
925 debugfs_create_file("block_state", 0400, zram->debugfs_dir,
926 zram, &proc_zram_block_state_op);
929 static void zram_debugfs_unregister(struct zram *zram)
931 debugfs_remove_recursive(zram->debugfs_dir);
934 static void zram_debugfs_create(void) {};
935 static void zram_debugfs_destroy(void) {};
936 static void zram_accessed(struct zram *zram, u32 index)
938 zram_clear_flag(zram, index, ZRAM_IDLE);
940 static void zram_debugfs_register(struct zram *zram) {};
941 static void zram_debugfs_unregister(struct zram *zram) {};
945 * We switched to per-cpu streams and this attr is not needed anymore.
946 * However, we will keep it around for some time, because:
947 * a) we may revert per-cpu streams in the future
948 * b) it's visible to user space and we need to follow our 2 years
949 * retirement rule; but we already have a number of 'soon to be
950 * altered' attrs, so max_comp_streams need to wait for the next
953 static ssize_t max_comp_streams_show(struct device *dev,
954 struct device_attribute *attr, char *buf)
956 return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
959 static ssize_t max_comp_streams_store(struct device *dev,
960 struct device_attribute *attr, const char *buf, size_t len)
965 static void comp_algorithm_set(struct zram *zram, u32 prio, const char *alg)
967 /* Do not free statically defined compression algorithms */
968 if (zram->comp_algs[prio] != default_compressor)
969 kfree(zram->comp_algs[prio]);
971 zram->comp_algs[prio] = alg;
974 static ssize_t __comp_algorithm_show(struct zram *zram, u32 prio, char *buf)
978 down_read(&zram->init_lock);
979 sz = zcomp_available_show(zram->comp_algs[prio], buf);
980 up_read(&zram->init_lock);
985 static int __comp_algorithm_store(struct zram *zram, u32 prio, const char *buf)
991 if (sz >= CRYPTO_MAX_ALG_NAME)
994 compressor = kstrdup(buf, GFP_KERNEL);
998 /* ignore trailing newline */
999 if (sz > 0 && compressor[sz - 1] == '\n')
1000 compressor[sz - 1] = 0x00;
1002 if (!zcomp_available_algorithm(compressor)) {
1007 down_write(&zram->init_lock);
1008 if (init_done(zram)) {
1009 up_write(&zram->init_lock);
1011 pr_info("Can't change algorithm for initialized device\n");
1015 comp_algorithm_set(zram, prio, compressor);
1016 up_write(&zram->init_lock);
1020 static ssize_t comp_algorithm_show(struct device *dev,
1021 struct device_attribute *attr,
1024 struct zram *zram = dev_to_zram(dev);
1026 return __comp_algorithm_show(zram, ZRAM_PRIMARY_COMP, buf);
1029 static ssize_t comp_algorithm_store(struct device *dev,
1030 struct device_attribute *attr,
1034 struct zram *zram = dev_to_zram(dev);
1037 ret = __comp_algorithm_store(zram, ZRAM_PRIMARY_COMP, buf);
1038 return ret ? ret : len;
1041 #ifdef CONFIG_ZRAM_MULTI_COMP
1042 static ssize_t recomp_algorithm_show(struct device *dev,
1043 struct device_attribute *attr,
1046 struct zram *zram = dev_to_zram(dev);
1050 for (prio = ZRAM_SECONDARY_COMP; prio < ZRAM_MAX_COMPS; prio++) {
1051 if (!zram->comp_algs[prio])
1054 sz += scnprintf(buf + sz, PAGE_SIZE - sz - 2, "#%d: ", prio);
1055 sz += __comp_algorithm_show(zram, prio, buf + sz);
1061 static ssize_t recomp_algorithm_store(struct device *dev,
1062 struct device_attribute *attr,
1066 struct zram *zram = dev_to_zram(dev);
1067 int prio = ZRAM_SECONDARY_COMP;
1068 char *args, *param, *val;
1072 args = skip_spaces(buf);
1074 args = next_arg(args, ¶m, &val);
1079 if (!strcmp(param, "algo")) {
1084 if (!strcmp(param, "priority")) {
1085 ret = kstrtoint(val, 10, &prio);
1095 if (prio < ZRAM_SECONDARY_COMP || prio >= ZRAM_MAX_COMPS)
1098 ret = __comp_algorithm_store(zram, prio, alg);
1099 return ret ? ret : len;
1103 static ssize_t compact_store(struct device *dev,
1104 struct device_attribute *attr, const char *buf, size_t len)
1106 struct zram *zram = dev_to_zram(dev);
1108 down_read(&zram->init_lock);
1109 if (!init_done(zram)) {
1110 up_read(&zram->init_lock);
1114 zs_compact(zram->mem_pool);
1115 up_read(&zram->init_lock);
1120 static ssize_t io_stat_show(struct device *dev,
1121 struct device_attribute *attr, char *buf)
1123 struct zram *zram = dev_to_zram(dev);
1126 down_read(&zram->init_lock);
1127 ret = scnprintf(buf, PAGE_SIZE,
1128 "%8llu %8llu 0 %8llu\n",
1129 (u64)atomic64_read(&zram->stats.failed_reads),
1130 (u64)atomic64_read(&zram->stats.failed_writes),
1131 (u64)atomic64_read(&zram->stats.notify_free));
1132 up_read(&zram->init_lock);
1137 static ssize_t mm_stat_show(struct device *dev,
1138 struct device_attribute *attr, char *buf)
1140 struct zram *zram = dev_to_zram(dev);
1141 struct zs_pool_stats pool_stats;
1142 u64 orig_size, mem_used = 0;
1146 memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
1148 down_read(&zram->init_lock);
1149 if (init_done(zram)) {
1150 mem_used = zs_get_total_pages(zram->mem_pool);
1151 zs_pool_stats(zram->mem_pool, &pool_stats);
1154 orig_size = atomic64_read(&zram->stats.pages_stored);
1155 max_used = atomic_long_read(&zram->stats.max_used_pages);
1157 ret = scnprintf(buf, PAGE_SIZE,
1158 "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu %8llu\n",
1159 orig_size << PAGE_SHIFT,
1160 (u64)atomic64_read(&zram->stats.compr_data_size),
1161 mem_used << PAGE_SHIFT,
1162 zram->limit_pages << PAGE_SHIFT,
1163 max_used << PAGE_SHIFT,
1164 (u64)atomic64_read(&zram->stats.same_pages),
1165 atomic_long_read(&pool_stats.pages_compacted),
1166 (u64)atomic64_read(&zram->stats.huge_pages),
1167 (u64)atomic64_read(&zram->stats.huge_pages_since));
1168 up_read(&zram->init_lock);
1173 #ifdef CONFIG_ZRAM_WRITEBACK
1174 #define FOUR_K(x) ((x) * (1 << (PAGE_SHIFT - 12)))
1175 static ssize_t bd_stat_show(struct device *dev,
1176 struct device_attribute *attr, char *buf)
1178 struct zram *zram = dev_to_zram(dev);
1181 down_read(&zram->init_lock);
1182 ret = scnprintf(buf, PAGE_SIZE,
1183 "%8llu %8llu %8llu\n",
1184 FOUR_K((u64)atomic64_read(&zram->stats.bd_count)),
1185 FOUR_K((u64)atomic64_read(&zram->stats.bd_reads)),
1186 FOUR_K((u64)atomic64_read(&zram->stats.bd_writes)));
1187 up_read(&zram->init_lock);
1193 static ssize_t debug_stat_show(struct device *dev,
1194 struct device_attribute *attr, char *buf)
1197 struct zram *zram = dev_to_zram(dev);
1200 down_read(&zram->init_lock);
1201 ret = scnprintf(buf, PAGE_SIZE,
1202 "version: %d\n%8llu %8llu\n",
1204 (u64)atomic64_read(&zram->stats.writestall),
1205 (u64)atomic64_read(&zram->stats.miss_free));
1206 up_read(&zram->init_lock);
1211 static DEVICE_ATTR_RO(io_stat);
1212 static DEVICE_ATTR_RO(mm_stat);
1213 #ifdef CONFIG_ZRAM_WRITEBACK
1214 static DEVICE_ATTR_RO(bd_stat);
1216 static DEVICE_ATTR_RO(debug_stat);
1218 static void zram_meta_free(struct zram *zram, u64 disksize)
1220 size_t num_pages = disksize >> PAGE_SHIFT;
1223 /* Free all pages that are still in this zram device */
1224 for (index = 0; index < num_pages; index++)
1225 zram_free_page(zram, index);
1227 zs_destroy_pool(zram->mem_pool);
1231 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
1235 num_pages = disksize >> PAGE_SHIFT;
1236 zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
1240 zram->mem_pool = zs_create_pool(zram->disk->disk_name);
1241 if (!zram->mem_pool) {
1246 if (!huge_class_size)
1247 huge_class_size = zs_huge_class_size(zram->mem_pool);
1252 * To protect concurrent access to the same index entry,
1253 * caller should hold this table index entry's bit_spinlock to
1254 * indicate this index entry is accessing.
1256 static void zram_free_page(struct zram *zram, size_t index)
1258 unsigned long handle;
1260 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
1261 zram->table[index].ac_time = 0;
1263 if (zram_test_flag(zram, index, ZRAM_IDLE))
1264 zram_clear_flag(zram, index, ZRAM_IDLE);
1266 if (zram_test_flag(zram, index, ZRAM_HUGE)) {
1267 zram_clear_flag(zram, index, ZRAM_HUGE);
1268 atomic64_dec(&zram->stats.huge_pages);
1271 if (zram_test_flag(zram, index, ZRAM_INCOMPRESSIBLE))
1272 zram_clear_flag(zram, index, ZRAM_INCOMPRESSIBLE);
1274 zram_set_priority(zram, index, 0);
1276 if (zram_test_flag(zram, index, ZRAM_WB)) {
1277 zram_clear_flag(zram, index, ZRAM_WB);
1278 free_block_bdev(zram, zram_get_element(zram, index));
1283 * No memory is allocated for same element filled pages.
1284 * Simply clear same page flag.
1286 if (zram_test_flag(zram, index, ZRAM_SAME)) {
1287 zram_clear_flag(zram, index, ZRAM_SAME);
1288 atomic64_dec(&zram->stats.same_pages);
1292 handle = zram_get_handle(zram, index);
1296 zs_free(zram->mem_pool, handle);
1298 atomic64_sub(zram_get_obj_size(zram, index),
1299 &zram->stats.compr_data_size);
1301 atomic64_dec(&zram->stats.pages_stored);
1302 zram_set_handle(zram, index, 0);
1303 zram_set_obj_size(zram, index, 0);
1304 WARN_ON_ONCE(zram->table[index].flags &
1305 ~(1UL << ZRAM_LOCK | 1UL << ZRAM_UNDER_WB));
1309 * Reads (decompresses if needed) a page from zspool (zsmalloc).
1310 * Corresponding ZRAM slot should be locked.
1312 static int zram_read_from_zspool(struct zram *zram, struct page *page,
1315 struct zcomp_strm *zstrm;
1316 unsigned long handle;
1322 handle = zram_get_handle(zram, index);
1323 if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1324 unsigned long value;
1327 value = handle ? zram_get_element(zram, index) : 0;
1328 mem = kmap_atomic(page);
1329 zram_fill_page(mem, PAGE_SIZE, value);
1334 size = zram_get_obj_size(zram, index);
1336 if (size != PAGE_SIZE) {
1337 prio = zram_get_priority(zram, index);
1338 zstrm = zcomp_stream_get(zram->comps[prio]);
1341 src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1342 if (size == PAGE_SIZE) {
1343 dst = kmap_atomic(page);
1344 memcpy(dst, src, PAGE_SIZE);
1348 dst = kmap_atomic(page);
1349 ret = zcomp_decompress(zstrm, src, size, dst);
1351 zcomp_stream_put(zram->comps[prio]);
1353 zs_unmap_object(zram->mem_pool, handle);
1357 static int zram_read_page(struct zram *zram, struct page *page, u32 index,
1362 zram_slot_lock(zram, index);
1363 if (!zram_test_flag(zram, index, ZRAM_WB)) {
1364 /* Slot should be locked through out the function call */
1365 ret = zram_read_from_zspool(zram, page, index);
1366 zram_slot_unlock(zram, index);
1369 * The slot should be unlocked before reading from the backing
1372 zram_slot_unlock(zram, index);
1374 ret = read_from_bdev(zram, page, zram_get_element(zram, index),
1378 /* Should NEVER happen. Return bio error if it does. */
1379 if (WARN_ON(ret < 0))
1380 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1386 * Use a temporary buffer to decompress the page, as the decompressor
1387 * always expects a full page for the output.
1389 static int zram_bvec_read_partial(struct zram *zram, struct bio_vec *bvec,
1390 u32 index, int offset)
1392 struct page *page = alloc_page(GFP_NOIO);
1397 ret = zram_read_page(zram, page, index, NULL);
1399 memcpy_to_bvec(bvec, page_address(page) + offset);
1404 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1405 u32 index, int offset, struct bio *bio)
1407 if (is_partial_io(bvec))
1408 return zram_bvec_read_partial(zram, bvec, index, offset);
1409 return zram_read_page(zram, bvec->bv_page, index, bio);
1412 static int zram_write_page(struct zram *zram, struct page *page, u32 index)
1415 unsigned long alloced_pages;
1416 unsigned long handle = -ENOMEM;
1417 unsigned int comp_len = 0;
1418 void *src, *dst, *mem;
1419 struct zcomp_strm *zstrm;
1420 unsigned long element = 0;
1421 enum zram_pageflags flags = 0;
1423 mem = kmap_atomic(page);
1424 if (page_same_filled(mem, &element)) {
1426 /* Free memory associated with this sector now. */
1428 atomic64_inc(&zram->stats.same_pages);
1434 zstrm = zcomp_stream_get(zram->comps[ZRAM_PRIMARY_COMP]);
1435 src = kmap_atomic(page);
1436 ret = zcomp_compress(zstrm, src, &comp_len);
1439 if (unlikely(ret)) {
1440 zcomp_stream_put(zram->comps[ZRAM_PRIMARY_COMP]);
1441 pr_err("Compression failed! err=%d\n", ret);
1442 zs_free(zram->mem_pool, handle);
1446 if (comp_len >= huge_class_size)
1447 comp_len = PAGE_SIZE;
1449 * handle allocation has 2 paths:
1450 * a) fast path is executed with preemption disabled (for
1451 * per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1452 * since we can't sleep;
1453 * b) slow path enables preemption and attempts to allocate
1454 * the page with __GFP_DIRECT_RECLAIM bit set. we have to
1455 * put per-cpu compression stream and, thus, to re-do
1456 * the compression once handle is allocated.
1458 * if we have a 'non-null' handle here then we are coming
1459 * from the slow path and handle has already been allocated.
1461 if (IS_ERR_VALUE(handle))
1462 handle = zs_malloc(zram->mem_pool, comp_len,
1463 __GFP_KSWAPD_RECLAIM |
1467 if (IS_ERR_VALUE(handle)) {
1468 zcomp_stream_put(zram->comps[ZRAM_PRIMARY_COMP]);
1469 atomic64_inc(&zram->stats.writestall);
1470 handle = zs_malloc(zram->mem_pool, comp_len,
1471 GFP_NOIO | __GFP_HIGHMEM |
1473 if (IS_ERR_VALUE(handle))
1474 return PTR_ERR((void *)handle);
1476 if (comp_len != PAGE_SIZE)
1477 goto compress_again;
1479 * If the page is not compressible, you need to acquire the
1480 * lock and execute the code below. The zcomp_stream_get()
1481 * call is needed to disable the cpu hotplug and grab the
1482 * zstrm buffer back. It is necessary that the dereferencing
1483 * of the zstrm variable below occurs correctly.
1485 zstrm = zcomp_stream_get(zram->comps[ZRAM_PRIMARY_COMP]);
1488 alloced_pages = zs_get_total_pages(zram->mem_pool);
1489 update_used_max(zram, alloced_pages);
1491 if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1492 zcomp_stream_put(zram->comps[ZRAM_PRIMARY_COMP]);
1493 zs_free(zram->mem_pool, handle);
1497 dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1499 src = zstrm->buffer;
1500 if (comp_len == PAGE_SIZE)
1501 src = kmap_atomic(page);
1502 memcpy(dst, src, comp_len);
1503 if (comp_len == PAGE_SIZE)
1506 zcomp_stream_put(zram->comps[ZRAM_PRIMARY_COMP]);
1507 zs_unmap_object(zram->mem_pool, handle);
1508 atomic64_add(comp_len, &zram->stats.compr_data_size);
1511 * Free memory associated with this sector
1512 * before overwriting unused sectors.
1514 zram_slot_lock(zram, index);
1515 zram_free_page(zram, index);
1517 if (comp_len == PAGE_SIZE) {
1518 zram_set_flag(zram, index, ZRAM_HUGE);
1519 atomic64_inc(&zram->stats.huge_pages);
1520 atomic64_inc(&zram->stats.huge_pages_since);
1524 zram_set_flag(zram, index, flags);
1525 zram_set_element(zram, index, element);
1527 zram_set_handle(zram, index, handle);
1528 zram_set_obj_size(zram, index, comp_len);
1530 zram_slot_unlock(zram, index);
1533 atomic64_inc(&zram->stats.pages_stored);
1538 * This is a partial IO. Read the full page before writing the changes.
1540 static int zram_bvec_write_partial(struct zram *zram, struct bio_vec *bvec,
1541 u32 index, int offset, struct bio *bio)
1543 struct page *page = alloc_page(GFP_NOIO);
1549 ret = zram_read_page(zram, page, index, bio);
1551 memcpy_from_bvec(page_address(page) + offset, bvec);
1552 ret = zram_write_page(zram, page, index);
1558 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1559 u32 index, int offset, struct bio *bio)
1561 if (is_partial_io(bvec))
1562 return zram_bvec_write_partial(zram, bvec, index, offset, bio);
1563 return zram_write_page(zram, bvec->bv_page, index);
1566 #ifdef CONFIG_ZRAM_MULTI_COMP
1568 * This function will decompress (unless it's ZRAM_HUGE) the page and then
1569 * attempt to compress it using provided compression algorithm priority
1570 * (which is potentially more effective).
1572 * Corresponding ZRAM slot should be locked.
1574 static int zram_recompress(struct zram *zram, u32 index, struct page *page,
1575 u32 threshold, u32 prio, u32 prio_max)
1577 struct zcomp_strm *zstrm = NULL;
1578 unsigned long handle_old;
1579 unsigned long handle_new;
1580 unsigned int comp_len_old;
1581 unsigned int comp_len_new;
1582 unsigned int class_index_old;
1583 unsigned int class_index_new;
1584 u32 num_recomps = 0;
1588 handle_old = zram_get_handle(zram, index);
1592 comp_len_old = zram_get_obj_size(zram, index);
1594 * Do not recompress objects that are already "small enough".
1596 if (comp_len_old < threshold)
1599 ret = zram_read_from_zspool(zram, page, index);
1603 class_index_old = zs_lookup_class_index(zram->mem_pool, comp_len_old);
1605 * Iterate the secondary comp algorithms list (in order of priority)
1606 * and try to recompress the page.
1608 for (; prio < prio_max; prio++) {
1609 if (!zram->comps[prio])
1613 * Skip if the object is already re-compressed with a higher
1614 * priority algorithm (or same algorithm).
1616 if (prio <= zram_get_priority(zram, index))
1620 zstrm = zcomp_stream_get(zram->comps[prio]);
1621 src = kmap_atomic(page);
1622 ret = zcomp_compress(zstrm, src, &comp_len_new);
1626 zcomp_stream_put(zram->comps[prio]);
1630 class_index_new = zs_lookup_class_index(zram->mem_pool,
1633 /* Continue until we make progress */
1634 if (class_index_new >= class_index_old ||
1635 (threshold && comp_len_new >= threshold)) {
1636 zcomp_stream_put(zram->comps[prio]);
1640 /* Recompression was successful so break out */
1645 * We did not try to recompress, e.g. when we have only one
1646 * secondary algorithm and the page is already recompressed
1647 * using that algorithm
1652 if (class_index_new >= class_index_old) {
1654 * Secondary algorithms failed to re-compress the page
1655 * in a way that would save memory, mark the object as
1656 * incompressible so that we will not try to compress
1659 * We need to make sure that all secondary algorithms have
1660 * failed, so we test if the number of recompressions matches
1661 * the number of active secondary algorithms.
1663 if (num_recomps == zram->num_active_comps - 1)
1664 zram_set_flag(zram, index, ZRAM_INCOMPRESSIBLE);
1668 /* Successful recompression but above threshold */
1669 if (threshold && comp_len_new >= threshold)
1673 * No direct reclaim (slow path) for handle allocation and no
1674 * re-compression attempt (unlike in zram_write_bvec()) since
1675 * we already have stored that object in zsmalloc. If we cannot
1676 * alloc memory for recompressed object then we bail out and
1677 * simply keep the old (existing) object in zsmalloc.
1679 handle_new = zs_malloc(zram->mem_pool, comp_len_new,
1680 __GFP_KSWAPD_RECLAIM |
1684 if (IS_ERR_VALUE(handle_new)) {
1685 zcomp_stream_put(zram->comps[prio]);
1686 return PTR_ERR((void *)handle_new);
1689 dst = zs_map_object(zram->mem_pool, handle_new, ZS_MM_WO);
1690 memcpy(dst, zstrm->buffer, comp_len_new);
1691 zcomp_stream_put(zram->comps[prio]);
1693 zs_unmap_object(zram->mem_pool, handle_new);
1695 zram_free_page(zram, index);
1696 zram_set_handle(zram, index, handle_new);
1697 zram_set_obj_size(zram, index, comp_len_new);
1698 zram_set_priority(zram, index, prio);
1700 atomic64_add(comp_len_new, &zram->stats.compr_data_size);
1701 atomic64_inc(&zram->stats.pages_stored);
1706 #define RECOMPRESS_IDLE (1 << 0)
1707 #define RECOMPRESS_HUGE (1 << 1)
1709 static ssize_t recompress_store(struct device *dev,
1710 struct device_attribute *attr,
1711 const char *buf, size_t len)
1713 u32 prio = ZRAM_SECONDARY_COMP, prio_max = ZRAM_MAX_COMPS;
1714 struct zram *zram = dev_to_zram(dev);
1715 unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
1716 char *args, *param, *val, *algo = NULL;
1717 u32 mode = 0, threshold = 0;
1718 unsigned long index;
1722 args = skip_spaces(buf);
1724 args = next_arg(args, ¶m, &val);
1729 if (!strcmp(param, "type")) {
1730 if (!strcmp(val, "idle"))
1731 mode = RECOMPRESS_IDLE;
1732 if (!strcmp(val, "huge"))
1733 mode = RECOMPRESS_HUGE;
1734 if (!strcmp(val, "huge_idle"))
1735 mode = RECOMPRESS_IDLE | RECOMPRESS_HUGE;
1739 if (!strcmp(param, "threshold")) {
1741 * We will re-compress only idle objects equal or
1742 * greater in size than watermark.
1744 ret = kstrtouint(val, 10, &threshold);
1750 if (!strcmp(param, "algo")) {
1756 if (threshold >= huge_class_size)
1759 down_read(&zram->init_lock);
1760 if (!init_done(zram)) {
1762 goto release_init_lock;
1768 for (; prio < ZRAM_MAX_COMPS; prio++) {
1769 if (!zram->comp_algs[prio])
1772 if (!strcmp(zram->comp_algs[prio], algo)) {
1773 prio_max = min(prio + 1, ZRAM_MAX_COMPS);
1781 goto release_init_lock;
1785 page = alloc_page(GFP_KERNEL);
1788 goto release_init_lock;
1792 for (index = 0; index < nr_pages; index++) {
1795 zram_slot_lock(zram, index);
1797 if (!zram_allocated(zram, index))
1800 if (mode & RECOMPRESS_IDLE &&
1801 !zram_test_flag(zram, index, ZRAM_IDLE))
1804 if (mode & RECOMPRESS_HUGE &&
1805 !zram_test_flag(zram, index, ZRAM_HUGE))
1808 if (zram_test_flag(zram, index, ZRAM_WB) ||
1809 zram_test_flag(zram, index, ZRAM_UNDER_WB) ||
1810 zram_test_flag(zram, index, ZRAM_SAME) ||
1811 zram_test_flag(zram, index, ZRAM_INCOMPRESSIBLE))
1814 err = zram_recompress(zram, index, page, threshold,
1817 zram_slot_unlock(zram, index);
1829 up_read(&zram->init_lock);
1834 static void zram_bio_discard(struct zram *zram, struct bio *bio)
1836 size_t n = bio->bi_iter.bi_size;
1837 u32 index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1838 u32 offset = (bio->bi_iter.bi_sector & (SECTORS_PER_PAGE - 1)) <<
1842 * zram manages data in physical block size units. Because logical block
1843 * size isn't identical with physical block size on some arch, we
1844 * could get a discard request pointing to a specific offset within a
1845 * certain physical block. Although we can handle this request by
1846 * reading that physiclal block and decompressing and partially zeroing
1847 * and re-compressing and then re-storing it, this isn't reasonable
1848 * because our intent with a discard request is to save memory. So
1849 * skipping this logical block is appropriate here.
1852 if (n <= (PAGE_SIZE - offset))
1855 n -= (PAGE_SIZE - offset);
1859 while (n >= PAGE_SIZE) {
1860 zram_slot_lock(zram, index);
1861 zram_free_page(zram, index);
1862 zram_slot_unlock(zram, index);
1863 atomic64_inc(&zram->stats.notify_free);
1871 static void zram_bio_read(struct zram *zram, struct bio *bio)
1873 struct bvec_iter iter;
1875 unsigned long start_time;
1877 start_time = bio_start_io_acct(bio);
1878 bio_for_each_segment(bv, bio, iter) {
1879 u32 index = iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1880 u32 offset = (iter.bi_sector & (SECTORS_PER_PAGE - 1)) <<
1883 if (zram_bvec_read(zram, &bv, index, offset, bio) < 0) {
1884 atomic64_inc(&zram->stats.failed_reads);
1885 bio->bi_status = BLK_STS_IOERR;
1888 flush_dcache_page(bv.bv_page);
1890 zram_slot_lock(zram, index);
1891 zram_accessed(zram, index);
1892 zram_slot_unlock(zram, index);
1894 bio_end_io_acct(bio, start_time);
1898 static void zram_bio_write(struct zram *zram, struct bio *bio)
1900 struct bvec_iter iter;
1902 unsigned long start_time;
1904 start_time = bio_start_io_acct(bio);
1905 bio_for_each_segment(bv, bio, iter) {
1906 u32 index = iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1907 u32 offset = (iter.bi_sector & (SECTORS_PER_PAGE - 1)) <<
1910 if (zram_bvec_write(zram, &bv, index, offset, bio) < 0) {
1911 atomic64_inc(&zram->stats.failed_writes);
1912 bio->bi_status = BLK_STS_IOERR;
1916 zram_slot_lock(zram, index);
1917 zram_accessed(zram, index);
1918 zram_slot_unlock(zram, index);
1920 bio_end_io_acct(bio, start_time);
1925 * Handler function for all zram I/O requests.
1927 static void zram_submit_bio(struct bio *bio)
1929 struct zram *zram = bio->bi_bdev->bd_disk->private_data;
1931 switch (bio_op(bio)) {
1933 zram_bio_read(zram, bio);
1936 zram_bio_write(zram, bio);
1938 case REQ_OP_DISCARD:
1939 case REQ_OP_WRITE_ZEROES:
1940 zram_bio_discard(zram, bio);
1948 static void zram_slot_free_notify(struct block_device *bdev,
1949 unsigned long index)
1953 zram = bdev->bd_disk->private_data;
1955 atomic64_inc(&zram->stats.notify_free);
1956 if (!zram_slot_trylock(zram, index)) {
1957 atomic64_inc(&zram->stats.miss_free);
1961 zram_free_page(zram, index);
1962 zram_slot_unlock(zram, index);
1965 static void zram_destroy_comps(struct zram *zram)
1969 for (prio = 0; prio < ZRAM_MAX_COMPS; prio++) {
1970 struct zcomp *comp = zram->comps[prio];
1972 zram->comps[prio] = NULL;
1975 zcomp_destroy(comp);
1976 zram->num_active_comps--;
1980 static void zram_reset_device(struct zram *zram)
1982 down_write(&zram->init_lock);
1984 zram->limit_pages = 0;
1986 if (!init_done(zram)) {
1987 up_write(&zram->init_lock);
1991 set_capacity_and_notify(zram->disk, 0);
1992 part_stat_set_all(zram->disk->part0, 0);
1994 /* I/O operation under all of CPU are done so let's free */
1995 zram_meta_free(zram, zram->disksize);
1997 zram_destroy_comps(zram);
1998 memset(&zram->stats, 0, sizeof(zram->stats));
2001 comp_algorithm_set(zram, ZRAM_PRIMARY_COMP, default_compressor);
2002 up_write(&zram->init_lock);
2005 static ssize_t disksize_store(struct device *dev,
2006 struct device_attribute *attr, const char *buf, size_t len)
2010 struct zram *zram = dev_to_zram(dev);
2014 disksize = memparse(buf, NULL);
2018 down_write(&zram->init_lock);
2019 if (init_done(zram)) {
2020 pr_info("Cannot change disksize for initialized device\n");
2025 disksize = PAGE_ALIGN(disksize);
2026 if (!zram_meta_alloc(zram, disksize)) {
2031 for (prio = 0; prio < ZRAM_MAX_COMPS; prio++) {
2032 if (!zram->comp_algs[prio])
2035 comp = zcomp_create(zram->comp_algs[prio]);
2037 pr_err("Cannot initialise %s compressing backend\n",
2038 zram->comp_algs[prio]);
2039 err = PTR_ERR(comp);
2040 goto out_free_comps;
2043 zram->comps[prio] = comp;
2044 zram->num_active_comps++;
2046 zram->disksize = disksize;
2047 set_capacity_and_notify(zram->disk, zram->disksize >> SECTOR_SHIFT);
2048 up_write(&zram->init_lock);
2053 zram_destroy_comps(zram);
2054 zram_meta_free(zram, disksize);
2056 up_write(&zram->init_lock);
2060 static ssize_t reset_store(struct device *dev,
2061 struct device_attribute *attr, const char *buf, size_t len)
2064 unsigned short do_reset;
2066 struct gendisk *disk;
2068 ret = kstrtou16(buf, 10, &do_reset);
2075 zram = dev_to_zram(dev);
2078 mutex_lock(&disk->open_mutex);
2079 /* Do not reset an active device or claimed device */
2080 if (disk_openers(disk) || zram->claim) {
2081 mutex_unlock(&disk->open_mutex);
2085 /* From now on, anyone can't open /dev/zram[0-9] */
2087 mutex_unlock(&disk->open_mutex);
2089 /* Make sure all the pending I/O are finished */
2090 sync_blockdev(disk->part0);
2091 zram_reset_device(zram);
2093 mutex_lock(&disk->open_mutex);
2094 zram->claim = false;
2095 mutex_unlock(&disk->open_mutex);
2100 static int zram_open(struct gendisk *disk, blk_mode_t mode)
2102 struct zram *zram = disk->private_data;
2104 WARN_ON(!mutex_is_locked(&disk->open_mutex));
2106 /* zram was claimed to reset so open request fails */
2112 static const struct block_device_operations zram_devops = {
2114 .submit_bio = zram_submit_bio,
2115 .swap_slot_free_notify = zram_slot_free_notify,
2116 .owner = THIS_MODULE
2119 static DEVICE_ATTR_WO(compact);
2120 static DEVICE_ATTR_RW(disksize);
2121 static DEVICE_ATTR_RO(initstate);
2122 static DEVICE_ATTR_WO(reset);
2123 static DEVICE_ATTR_WO(mem_limit);
2124 static DEVICE_ATTR_WO(mem_used_max);
2125 static DEVICE_ATTR_WO(idle);
2126 static DEVICE_ATTR_RW(max_comp_streams);
2127 static DEVICE_ATTR_RW(comp_algorithm);
2128 #ifdef CONFIG_ZRAM_WRITEBACK
2129 static DEVICE_ATTR_RW(backing_dev);
2130 static DEVICE_ATTR_WO(writeback);
2131 static DEVICE_ATTR_RW(writeback_limit);
2132 static DEVICE_ATTR_RW(writeback_limit_enable);
2134 #ifdef CONFIG_ZRAM_MULTI_COMP
2135 static DEVICE_ATTR_RW(recomp_algorithm);
2136 static DEVICE_ATTR_WO(recompress);
2139 static struct attribute *zram_disk_attrs[] = {
2140 &dev_attr_disksize.attr,
2141 &dev_attr_initstate.attr,
2142 &dev_attr_reset.attr,
2143 &dev_attr_compact.attr,
2144 &dev_attr_mem_limit.attr,
2145 &dev_attr_mem_used_max.attr,
2146 &dev_attr_idle.attr,
2147 &dev_attr_max_comp_streams.attr,
2148 &dev_attr_comp_algorithm.attr,
2149 #ifdef CONFIG_ZRAM_WRITEBACK
2150 &dev_attr_backing_dev.attr,
2151 &dev_attr_writeback.attr,
2152 &dev_attr_writeback_limit.attr,
2153 &dev_attr_writeback_limit_enable.attr,
2155 &dev_attr_io_stat.attr,
2156 &dev_attr_mm_stat.attr,
2157 #ifdef CONFIG_ZRAM_WRITEBACK
2158 &dev_attr_bd_stat.attr,
2160 &dev_attr_debug_stat.attr,
2161 #ifdef CONFIG_ZRAM_MULTI_COMP
2162 &dev_attr_recomp_algorithm.attr,
2163 &dev_attr_recompress.attr,
2168 ATTRIBUTE_GROUPS(zram_disk);
2171 * Allocate and initialize new zram device. the function returns
2172 * '>= 0' device_id upon success, and negative value otherwise.
2174 static int zram_add(void)
2179 zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
2183 ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
2188 init_rwsem(&zram->init_lock);
2189 #ifdef CONFIG_ZRAM_WRITEBACK
2190 spin_lock_init(&zram->wb_limit_lock);
2193 /* gendisk structure */
2194 zram->disk = blk_alloc_disk(NUMA_NO_NODE);
2196 pr_err("Error allocating disk structure for device %d\n",
2202 zram->disk->major = zram_major;
2203 zram->disk->first_minor = device_id;
2204 zram->disk->minors = 1;
2205 zram->disk->flags |= GENHD_FL_NO_PART;
2206 zram->disk->fops = &zram_devops;
2207 zram->disk->private_data = zram;
2208 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
2210 /* Actual capacity set using sysfs (/sys/block/zram<id>/disksize */
2211 set_capacity(zram->disk, 0);
2212 /* zram devices sort of resembles non-rotational disks */
2213 blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
2214 blk_queue_flag_set(QUEUE_FLAG_SYNCHRONOUS, zram->disk->queue);
2217 * To ensure that we always get PAGE_SIZE aligned
2218 * and n*PAGE_SIZED sized I/O requests.
2220 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
2221 blk_queue_logical_block_size(zram->disk->queue,
2222 ZRAM_LOGICAL_BLOCK_SIZE);
2223 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
2224 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
2225 zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
2226 blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
2229 * zram_bio_discard() will clear all logical blocks if logical block
2230 * size is identical with physical block size(PAGE_SIZE). But if it is
2231 * different, we will skip discarding some parts of logical blocks in
2232 * the part of the request range which isn't aligned to physical block
2233 * size. So we can't ensure that all discarded logical blocks are
2236 if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
2237 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
2239 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, zram->disk->queue);
2240 ret = device_add_disk(NULL, zram->disk, zram_disk_groups);
2242 goto out_cleanup_disk;
2244 comp_algorithm_set(zram, ZRAM_PRIMARY_COMP, default_compressor);
2246 zram_debugfs_register(zram);
2247 pr_info("Added device: %s\n", zram->disk->disk_name);
2251 put_disk(zram->disk);
2253 idr_remove(&zram_index_idr, device_id);
2259 static int zram_remove(struct zram *zram)
2263 mutex_lock(&zram->disk->open_mutex);
2264 if (disk_openers(zram->disk)) {
2265 mutex_unlock(&zram->disk->open_mutex);
2269 claimed = zram->claim;
2272 mutex_unlock(&zram->disk->open_mutex);
2274 zram_debugfs_unregister(zram);
2278 * If we were claimed by reset_store(), del_gendisk() will
2279 * wait until reset_store() is done, so nothing need to do.
2283 /* Make sure all the pending I/O are finished */
2284 sync_blockdev(zram->disk->part0);
2285 zram_reset_device(zram);
2288 pr_info("Removed device: %s\n", zram->disk->disk_name);
2290 del_gendisk(zram->disk);
2292 /* del_gendisk drains pending reset_store */
2293 WARN_ON_ONCE(claimed && zram->claim);
2296 * disksize_store() may be called in between zram_reset_device()
2297 * and del_gendisk(), so run the last reset to avoid leaking
2298 * anything allocated with disksize_store()
2300 zram_reset_device(zram);
2302 put_disk(zram->disk);
2307 /* zram-control sysfs attributes */
2310 * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
2311 * sense that reading from this file does alter the state of your system -- it
2312 * creates a new un-initialized zram device and returns back this device's
2313 * device_id (or an error code if it fails to create a new device).
2315 static ssize_t hot_add_show(const struct class *class,
2316 const struct class_attribute *attr,
2321 mutex_lock(&zram_index_mutex);
2323 mutex_unlock(&zram_index_mutex);
2327 return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
2329 /* This attribute must be set to 0400, so CLASS_ATTR_RO() can not be used */
2330 static struct class_attribute class_attr_hot_add =
2331 __ATTR(hot_add, 0400, hot_add_show, NULL);
2333 static ssize_t hot_remove_store(const struct class *class,
2334 const struct class_attribute *attr,
2341 /* dev_id is gendisk->first_minor, which is `int' */
2342 ret = kstrtoint(buf, 10, &dev_id);
2348 mutex_lock(&zram_index_mutex);
2350 zram = idr_find(&zram_index_idr, dev_id);
2352 ret = zram_remove(zram);
2354 idr_remove(&zram_index_idr, dev_id);
2359 mutex_unlock(&zram_index_mutex);
2360 return ret ? ret : count;
2362 static CLASS_ATTR_WO(hot_remove);
2364 static struct attribute *zram_control_class_attrs[] = {
2365 &class_attr_hot_add.attr,
2366 &class_attr_hot_remove.attr,
2369 ATTRIBUTE_GROUPS(zram_control_class);
2371 static struct class zram_control_class = {
2372 .name = "zram-control",
2373 .class_groups = zram_control_class_groups,
2376 static int zram_remove_cb(int id, void *ptr, void *data)
2378 WARN_ON_ONCE(zram_remove(ptr));
2382 static void destroy_devices(void)
2384 class_unregister(&zram_control_class);
2385 idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
2386 zram_debugfs_destroy();
2387 idr_destroy(&zram_index_idr);
2388 unregister_blkdev(zram_major, "zram");
2389 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2392 static int __init zram_init(void)
2396 BUILD_BUG_ON(__NR_ZRAM_PAGEFLAGS > BITS_PER_LONG);
2398 ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
2399 zcomp_cpu_up_prepare, zcomp_cpu_dead);
2403 ret = class_register(&zram_control_class);
2405 pr_err("Unable to register zram-control class\n");
2406 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2410 zram_debugfs_create();
2411 zram_major = register_blkdev(0, "zram");
2412 if (zram_major <= 0) {
2413 pr_err("Unable to get major number\n");
2414 class_unregister(&zram_control_class);
2415 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2419 while (num_devices != 0) {
2420 mutex_lock(&zram_index_mutex);
2422 mutex_unlock(&zram_index_mutex);
2435 static void __exit zram_exit(void)
2440 module_init(zram_init);
2441 module_exit(zram_exit);
2443 module_param(num_devices, uint, 0);
2444 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
2446 MODULE_LICENSE("Dual BSD/GPL");
2447 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
2448 MODULE_DESCRIPTION("Compressed RAM Block Device");