2 * Copyright (C) 2015 IT University of Copenhagen (rrpc.h)
3 * Copyright (C) 2016 CNEX Labs
4 * Initial release: Matias Bjorling <matias@cnexlabs.com>
5 * Write buffering: Javier Gonzalez <javier@cnexlabs.com>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version
9 * 2 as published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * Implementation of a Physical Block-device target for Open-channel SSDs.
23 #include <linux/blkdev.h>
24 #include <linux/blk-mq.h>
25 #include <linux/bio.h>
26 #include <linux/module.h>
27 #include <linux/kthread.h>
28 #include <linux/vmalloc.h>
29 #include <linux/crc32.h>
30 #include <linux/uuid.h>
32 #include <linux/lightnvm.h>
34 /* Run only GC if less than 1/X blocks are free */
35 #define GC_LIMIT_INVERSE 5
36 #define GC_TIME_MSECS 1000
38 #define PBLK_SECTOR (512)
39 #define PBLK_EXPOSED_PAGE_SIZE (4096)
41 #define PBLK_NR_CLOSE_JOBS (4)
43 #define PBLK_CACHE_NAME_LEN (DISK_NAME_LEN + 16)
45 #define PBLK_COMMAND_TIMEOUT_MS 30000
47 /* Max 512 LUNs per device */
48 #define PBLK_MAX_LUNS_BITMAP (4)
50 #define NR_PHY_IN_LOG (PBLK_EXPOSED_PAGE_SIZE / PBLK_SECTOR)
52 /* Static pool sizes */
53 #define PBLK_GEN_WS_POOL_SIZE (2)
55 #define PBLK_DEFAULT_OP (11)
59 PBLK_WRITE = WRITE,/* Write from write buffer */
60 PBLK_WRITE_INT, /* Internal write - no write buffer */
61 PBLK_READ_RECOV, /* Recovery read - errors allowed */
67 PBLK_IOTYPE_USER = 1 << 0,
68 PBLK_IOTYPE_GC = 1 << 1,
70 /* Write buffer flags */
71 PBLK_FLUSH_ENTRY = 1 << 2,
72 PBLK_WRITTEN_DATA = 1 << 3,
73 PBLK_SUBMITTED_ENTRY = 1 << 4,
74 PBLK_WRITABLE_ENTRY = 1 << 5,
78 PBLK_BLK_ST_OPEN = 0x1,
79 PBLK_BLK_ST_CLOSED = 0x2,
83 PBLK_CHUNK_RESET_START,
84 PBLK_CHUNK_RESET_DONE,
85 PBLK_CHUNK_RESET_FAILED,
88 struct pblk_sec_meta {
93 /* The number of GC lists and the rate-limiter states go together. This way the
94 * rate-limiter can dictate how much GC is needed based on resource utilization.
96 #define PBLK_GC_NR_LISTS 4
106 #define pblk_dma_meta_size (sizeof(struct pblk_sec_meta) * NVM_MAX_VLBA)
107 #define pblk_dma_ppa_size (sizeof(u64) * NVM_MAX_VLBA)
109 /* write buffer completion context */
111 struct list_head list; /* Head for out-of-order completion */
113 unsigned long *lun_bitmap; /* Luns used on current request */
115 unsigned int nr_valid;
116 unsigned int nr_padded;
122 unsigned long start_time;
126 /* partial read context */
128 struct bio *orig_bio;
129 DECLARE_BITMAP(bitmap, NVM_MAX_VLBA);
130 unsigned int orig_nr_secs;
131 unsigned int bio_init_idx;
133 dma_addr_t dma_ppa_list;
139 struct completion wait;
143 /* Recovery context */
144 struct pblk_rec_ctx {
147 struct work_struct ws_rec;
152 struct bio_list bios; /* Original bios - used for completion
153 * in REQ_FUA, REQ_FLUSH case
155 u64 lba; /* Logic addr. associated with entry */
156 struct ppa_addr ppa; /* Physic addr. associated with entry */
157 int flags; /* Write context flags */
160 struct pblk_rb_entry {
161 struct ppa_addr cacheline; /* Cacheline for this entry */
162 void *data; /* Pointer to data on this entry */
163 struct pblk_w_ctx w_ctx; /* Context for this entry */
164 struct list_head index; /* List head to enable indexes */
167 #define EMPTY_ENTRY (~0U)
169 struct pblk_rb_pages {
172 struct list_head list;
176 struct pblk_rb_entry *entries; /* Ring buffer entries */
177 unsigned int mem; /* Write offset - points to next
178 * writable entry in memory
180 unsigned int subm; /* Read offset - points to last entry
181 * that has been submitted to the media
184 unsigned int sync; /* Synced - backpointer that signals
185 * the last submitted entry that has
186 * been successfully persisted to media
188 unsigned int flush_point; /* Sync point - last entry that must be
189 * flushed to the media. Used with
190 * REQ_FLUSH and REQ_FUA
192 unsigned int l2p_update; /* l2p update point - next entry for
193 * which l2p mapping will be updated to
194 * contain a device ppa address (instead
197 unsigned int nr_entries; /* Number of entries in write buffer -
198 * must be a power of two
200 unsigned int seg_size; /* Size of the data segments being
201 * stored on each entry. Typically this
205 struct list_head pages; /* List of data pages */
207 spinlock_t w_lock; /* Write lock */
208 spinlock_t s_lock; /* Sync lock */
210 #ifdef CONFIG_NVM_PBLK_DEBUG
211 atomic_t inflight_flush_point; /* Not served REQ_FLUSH | REQ_FUA */
215 #define PBLK_RECOVERY_SECTORS 16
218 struct ppa_addr bppa;
219 struct semaphore wr_sem;
223 struct pblk_line *line;
225 u64 paddr_list[NVM_MAX_VLBA];
226 u64 lba_list[NVM_MAX_VLBA];
229 struct list_head list;
233 /* These states are not protected by a lock since (i) they are in the
234 * fast path, and (ii) they are not critical.
240 struct task_struct *gc_ts;
241 struct task_struct *gc_writer_ts;
242 struct task_struct *gc_reader_ts;
244 struct workqueue_struct *gc_line_reader_wq;
245 struct workqueue_struct *gc_reader_wq;
247 struct timer_list gc_timer;
249 struct semaphore gc_sem;
250 atomic_t read_inflight_gc; /* Number of lines with inflight GC reads */
251 atomic_t pipeline_gc; /* Number of lines in the GC pipeline -
252 * started reads to finished writes
256 struct list_head w_list;
257 struct list_head r_list;
265 unsigned int high; /* Upper threshold for rate limiter (free run -
266 * user I/O rate limiter
268 unsigned int high_pw; /* High rounded up as a power of 2 */
270 #define PBLK_USER_HIGH_THRS 8 /* Begin write limit at 12% available blks */
271 #define PBLK_USER_LOW_THRS 10 /* Aggressive GC at 10% available blocks */
273 int rb_windows_pw; /* Number of rate windows in the write buffer
274 * given as a power-of-2. This guarantees that
275 * when user I/O is being rate limited, there
276 * will be reserved enough space for the GC to
277 * place its payload. A window is of
278 * pblk->max_write_pgs size, which in NVMe is
281 int rb_budget; /* Total number of entries available for I/O */
282 int rb_user_max; /* Max buffer entries available for user I/O */
283 int rb_gc_max; /* Max buffer entries available for GC I/O */
284 int rb_gc_rsv; /* Reserved buffer entries for GC I/O */
285 int rb_state; /* Rate-limiter current state */
286 int rb_max_io; /* Maximum size for an I/O giving the config */
288 atomic_t rb_user_cnt; /* User I/O buffer counter */
289 atomic_t rb_gc_cnt; /* GC I/O buffer counter */
290 atomic_t rb_space; /* Space limit in case of reaching capacity */
292 int rsv_blocks; /* Reserved blocks for GC */
297 atomic_t werr_lines; /* Number of write error lines that needs gc */
299 struct timer_list u_timer;
301 unsigned long long nr_secs;
302 unsigned long total_blocks;
304 atomic_t free_blocks; /* Total number of free blocks (+ OP) */
305 atomic_t free_user_blocks; /* Number of user free blocks (no OP) */
308 #define PBLK_LINE_EMPTY (~0U)
312 PBLK_LINETYPE_FREE = 0,
313 PBLK_LINETYPE_LOG = 1,
314 PBLK_LINETYPE_DATA = 2,
317 PBLK_LINESTATE_NEW = 9,
318 PBLK_LINESTATE_FREE = 10,
319 PBLK_LINESTATE_OPEN = 11,
320 PBLK_LINESTATE_CLOSED = 12,
321 PBLK_LINESTATE_GC = 13,
322 PBLK_LINESTATE_BAD = 14,
323 PBLK_LINESTATE_CORRUPT = 15,
326 PBLK_LINEGC_NONE = 20,
327 PBLK_LINEGC_EMPTY = 21,
328 PBLK_LINEGC_LOW = 22,
329 PBLK_LINEGC_MID = 23,
330 PBLK_LINEGC_HIGH = 24,
331 PBLK_LINEGC_FULL = 25,
332 PBLK_LINEGC_WERR = 26
335 #define PBLK_MAGIC 0x70626c6b /*pblk*/
337 /* emeta/smeta persistent storage format versions:
338 * Changes in major version requires offline migration.
339 * Changes in minor version are handled automatically during
343 #define SMETA_VERSION_MAJOR (0)
344 #define SMETA_VERSION_MINOR (1)
346 #define EMETA_VERSION_MAJOR (0)
347 #define EMETA_VERSION_MINOR (2)
351 __le32 identifier; /* pblk identifier */
352 __u8 uuid[16]; /* instance uuid */
353 __le16 type; /* line type */
354 __u8 version_major; /* version major */
355 __u8 version_minor; /* version minor */
356 __le32 id; /* line id for current line */
360 struct line_header header;
362 __le32 crc; /* Full structure including struct crc */
363 /* Previous line metadata */
364 __le32 prev_id; /* Line id for previous line */
366 /* Current line metadata */
367 __le64 seq_nr; /* Sequence number for current line */
370 __le32 window_wr_lun; /* Number of parallel LUNs to write */
379 * Metadata layout in media:
381 * 1. struct line_emeta
382 * 2. bad block bitmap (u64 * window_wr_lun)
383 * 3. write amplification counters
384 * Mid sectors (start at lbas_sector):
385 * 3. nr_lbas (u64) forming lba list
386 * Last sectors (start at vsc_sector):
387 * 4. u32 valid sector count (vsc) for all lines (~0U: free line)
390 struct line_header header;
392 __le32 crc; /* Full structure including struct crc */
394 /* Previous line metadata */
395 __le32 prev_id; /* Line id for prev line */
397 /* Current line metadata */
398 __le64 seq_nr; /* Sequence number for current line */
401 __le32 window_wr_lun; /* Number of parallel LUNs to write */
403 /* Bookkeeping for recovery */
404 __le32 next_id; /* Line id for next line */
405 __le64 nr_lbas; /* Number of lbas mapped in line */
406 __le64 nr_valid_lbas; /* Number of valid lbas mapped in line */
407 __le64 bb_bitmap[]; /* Updated bad block bitmap for line */
411 /* Write amplification counters stored on media */
413 __le64 user; /* Number of user written sectors */
414 __le64 gc; /* Number of sectors written by GC*/
415 __le64 pad; /* Number of padded sectors */
419 struct line_emeta *buf; /* emeta buffer in media format */
420 int mem; /* Write offset - points to next
421 * writable entry in memory
423 atomic_t sync; /* Synced - backpointer that signals the
424 * last entry that has been successfully
427 unsigned int nr_entries; /* Number of emeta entries */
431 struct line_smeta *buf; /* smeta buffer in persistent format */
434 struct pblk_w_err_gc {
441 unsigned int id; /* Line number corresponds to the
444 unsigned int seq_nr; /* Unique line sequence number */
446 int state; /* PBLK_LINESTATE_X */
447 int type; /* PBLK_LINETYPE_X */
448 int gc_group; /* PBLK_LINEGC_X */
449 struct list_head list; /* Free, GC lists */
451 unsigned long *lun_bitmap; /* Bitmap for LUNs mapped in line */
453 struct nvm_chk_meta *chks; /* Chunks forming line */
455 struct pblk_smeta *smeta; /* Start metadata */
456 struct pblk_emeta *emeta; /* End medatada */
458 int meta_line; /* Metadata line id */
459 int meta_distance; /* Distance between data and metadata */
461 u64 smeta_ssec; /* Sector where smeta starts */
462 u64 emeta_ssec; /* Sector where emeta starts */
464 unsigned int sec_in_line; /* Number of usable secs in line */
466 atomic_t blk_in_line; /* Number of good blocks in line */
467 unsigned long *blk_bitmap; /* Bitmap for valid/invalid blocks */
468 unsigned long *erase_bitmap; /* Bitmap for erased blocks */
470 unsigned long *map_bitmap; /* Bitmap for mapped sectors in line */
471 unsigned long *invalid_bitmap; /* Bitmap for invalid sectors in line */
473 atomic_t left_eblks; /* Blocks left for erasing */
474 atomic_t left_seblks; /* Blocks left for sync erasing */
476 int left_msecs; /* Sectors left for mapping */
477 unsigned int cur_sec; /* Sector map pointer */
478 unsigned int nr_valid_lbas; /* Number of valid lbas in line */
480 __le32 *vsc; /* Valid sector count in line */
482 struct kref ref; /* Write buffer L2P references */
484 struct pblk_w_err_gc *w_err_gc; /* Write error gc recovery metadata */
486 spinlock_t lock; /* Necessary for invalid_bitmap only */
489 #define PBLK_DATA_LINES 4
492 PBLK_KMALLOC_META = 1,
493 PBLK_VMALLOC_META = 2,
497 PBLK_EMETA_TYPE_HEADER = 1, /* struct line_emeta first sector */
498 PBLK_EMETA_TYPE_LLBA = 2, /* lba list - type: __le64 */
499 PBLK_EMETA_TYPE_VSC = 3, /* vsc list - type: __le32 */
502 struct pblk_line_mgmt {
503 int nr_lines; /* Total number of full lines */
504 int nr_free_lines; /* Number of full lines in free list */
506 /* Free lists - use free_lock */
507 struct list_head free_list; /* Full lines ready to use */
508 struct list_head corrupt_list; /* Full lines corrupted */
509 struct list_head bad_list; /* Full lines bad */
511 /* GC lists - use gc_lock */
512 struct list_head *gc_lists[PBLK_GC_NR_LISTS];
513 struct list_head gc_high_list; /* Full lines ready to GC, high isc */
514 struct list_head gc_mid_list; /* Full lines ready to GC, mid isc */
515 struct list_head gc_low_list; /* Full lines ready to GC, low isc */
517 struct list_head gc_werr_list; /* Write err recovery list */
519 struct list_head gc_full_list; /* Full lines ready to GC, no valid */
520 struct list_head gc_empty_list; /* Full lines close, all valid */
522 struct pblk_line *log_line; /* Current FTL log line */
523 struct pblk_line *data_line; /* Current data line */
524 struct pblk_line *log_next; /* Next FTL log line */
525 struct pblk_line *data_next; /* Next data line */
527 struct list_head emeta_list; /* Lines queued to schedule emeta */
529 __le32 *vsc_list; /* Valid sector counts for all lines */
531 /* Metadata allocation type: VMALLOC | KMALLOC */
532 int emeta_alloc_type;
534 /* Pre-allocated metadata for data lines */
535 struct pblk_smeta *sline_meta[PBLK_DATA_LINES];
536 struct pblk_emeta *eline_meta[PBLK_DATA_LINES];
537 unsigned long meta_bitmap;
539 /* Cache and mempool for map/invalid bitmaps */
540 struct kmem_cache *bitmap_cache;
541 mempool_t *bitmap_pool;
543 /* Helpers for fast bitmap calculations */
544 unsigned long *bb_template;
545 unsigned long *bb_aux;
547 unsigned long d_seq_nr; /* Data line unique sequence number */
548 unsigned long l_seq_nr; /* Log line unique sequence number */
550 spinlock_t free_lock;
551 spinlock_t close_lock;
555 struct pblk_line_meta {
556 unsigned int smeta_len; /* Total length for smeta */
557 unsigned int smeta_sec; /* Sectors needed for smeta */
559 unsigned int emeta_len[4]; /* Lengths for emeta:
561 * [1]: struct line_emeta +
562 * bb_bitmap + struct wa_counters
566 unsigned int emeta_sec[4]; /* Sectors needed for emeta. Same layout
570 unsigned int emeta_bb; /* Boundary for bb that affects emeta */
572 unsigned int vsc_list_len; /* Length for vsc list */
573 unsigned int sec_bitmap_len; /* Length for sector bitmap in line */
574 unsigned int blk_bitmap_len; /* Length for block bitmap in line */
575 unsigned int lun_bitmap_len; /* Length for lun bitmap in line */
577 unsigned int blk_per_line; /* Number of blocks in a full line */
578 unsigned int sec_per_line; /* Number of sectors in a line */
579 unsigned int dsec_per_line; /* Number of data sectors in a line */
580 unsigned int min_blk_line; /* Min. number of good blocks in line */
582 unsigned int mid_thrs; /* Threshold for GC mid list */
583 unsigned int high_thrs; /* Threshold for GC high list */
585 unsigned int meta_distance; /* Distance between data and metadata */
589 PBLK_STATE_RUNNING = 0,
590 PBLK_STATE_STOPPING = 1,
591 PBLK_STATE_RECOVERING = 2,
592 PBLK_STATE_STOPPED = 3,
595 /* Internal format to support not power-of-2 device formats */
608 struct nvm_tgt_dev *dev;
609 struct gendisk *disk;
613 struct pblk_lun *luns;
615 struct pblk_line *lines; /* Line array */
616 struct pblk_line_mgmt l_mg; /* Line management */
617 struct pblk_line_meta lm; /* Line metadata */
619 struct nvm_addrf addrf; /* Aligned address format */
620 struct pblk_addrf uaddrf; /* Unaligned address format */
625 int state; /* pblk line state */
627 int min_write_pgs; /* Minimum amount of pages required by controller */
628 int max_write_pgs; /* Maximum amount of pages supported by controller */
630 sector_t capacity; /* Device capacity when bad blocks are subtracted */
632 int op; /* Percentage of device used for over-provisioning */
633 int op_blks; /* Number of blocks used for over-provisioning */
635 /* pblk provisioning values. Used by rate limiter */
640 unsigned char instance_uuid[16];
642 /* Persistent write amplification counters, 4kb sector I/Os */
643 atomic64_t user_wa; /* Sectors written by user */
644 atomic64_t gc_wa; /* Sectors written by GC */
645 atomic64_t pad_wa; /* Padded sectors written */
647 /* Reset values for delta write amplification measurements */
652 /* Counters used for calculating padding distribution */
653 atomic64_t *pad_dist; /* Padding distribution buckets */
654 u64 nr_flush_rst; /* Flushes reset value for pad dist.*/
655 atomic64_t nr_flush; /* Number of flush/fua I/O */
657 #ifdef CONFIG_NVM_PBLK_DEBUG
658 /* Non-persistent debug counters, 4kb sector I/Os */
659 atomic_long_t inflight_writes; /* Inflight writes (user and gc) */
660 atomic_long_t padded_writes; /* Sectors padded due to flush/fua */
661 atomic_long_t padded_wb; /* Sectors padded in write buffer */
662 atomic_long_t req_writes; /* Sectors stored on write buffer */
663 atomic_long_t sub_writes; /* Sectors submitted from buffer */
664 atomic_long_t sync_writes; /* Sectors synced to media */
665 atomic_long_t inflight_reads; /* Inflight sector read requests */
666 atomic_long_t cache_reads; /* Read requests that hit the cache */
667 atomic_long_t sync_reads; /* Completed sector read requests */
668 atomic_long_t recov_writes; /* Sectors submitted from recovery */
669 atomic_long_t recov_gc_writes; /* Sectors submitted from write GC */
670 atomic_long_t recov_gc_reads; /* Sectors submitted from read GC */
675 atomic_long_t read_failed;
676 atomic_long_t read_empty;
677 atomic_long_t read_high_ecc;
678 atomic_long_t read_failed_gc;
679 atomic_long_t write_failed;
680 atomic_long_t erase_failed;
682 atomic_t inflight_io; /* General inflight I/O counter */
684 struct task_struct *writer_ts;
686 /* Simple translation map of logical addresses to physical addresses.
687 * The logical addresses is known by the host system, while the physical
688 * addresses are used when writing to the disk block device.
690 unsigned char *trans_map;
691 spinlock_t trans_lock;
693 struct list_head compl_list;
695 spinlock_t resubmit_lock; /* Resubmit list lock */
696 struct list_head resubmit_list; /* Resubmit list for failed writes*/
698 mempool_t page_bio_pool;
699 mempool_t gen_ws_pool;
705 struct workqueue_struct *close_wq;
706 struct workqueue_struct *bb_wq;
707 struct workqueue_struct *r_end_wq;
709 struct timer_list wtimer;
714 struct pblk_line_ws {
716 struct pblk_line *line;
718 struct work_struct ws;
721 #define pblk_g_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_g_ctx))
722 #define pblk_w_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_c_ctx))
724 #define pblk_err(pblk, fmt, ...) \
725 pr_err("pblk %s: " fmt, pblk->disk->disk_name, ##__VA_ARGS__)
726 #define pblk_info(pblk, fmt, ...) \
727 pr_info("pblk %s: " fmt, pblk->disk->disk_name, ##__VA_ARGS__)
728 #define pblk_warn(pblk, fmt, ...) \
729 pr_warn("pblk %s: " fmt, pblk->disk->disk_name, ##__VA_ARGS__)
730 #define pblk_debug(pblk, fmt, ...) \
731 pr_debug("pblk %s: " fmt, pblk->disk->disk_name, ##__VA_ARGS__)
734 * pblk ring buffer operations
736 int pblk_rb_init(struct pblk_rb *rb, struct pblk_rb_entry *rb_entry_base,
737 unsigned int power_size, unsigned int power_seg_sz);
738 unsigned int pblk_rb_calculate_size(unsigned int nr_entries);
739 void *pblk_rb_entries_ref(struct pblk_rb *rb);
740 int pblk_rb_may_write_user(struct pblk_rb *rb, struct bio *bio,
741 unsigned int nr_entries, unsigned int *pos);
742 int pblk_rb_may_write_gc(struct pblk_rb *rb, unsigned int nr_entries,
744 void pblk_rb_write_entry_user(struct pblk_rb *rb, void *data,
745 struct pblk_w_ctx w_ctx, unsigned int pos);
746 void pblk_rb_write_entry_gc(struct pblk_rb *rb, void *data,
747 struct pblk_w_ctx w_ctx, struct pblk_line *line,
748 u64 paddr, unsigned int pos);
749 struct pblk_w_ctx *pblk_rb_w_ctx(struct pblk_rb *rb, unsigned int pos);
750 void pblk_rb_flush(struct pblk_rb *rb);
752 void pblk_rb_sync_l2p(struct pblk_rb *rb);
753 unsigned int pblk_rb_read_to_bio(struct pblk_rb *rb, struct nvm_rq *rqd,
754 unsigned int pos, unsigned int nr_entries,
756 int pblk_rb_copy_to_bio(struct pblk_rb *rb, struct bio *bio, sector_t lba,
757 struct ppa_addr ppa, int bio_iter, bool advanced_bio);
758 unsigned int pblk_rb_read_commit(struct pblk_rb *rb, unsigned int entries);
760 unsigned int pblk_rb_sync_init(struct pblk_rb *rb, unsigned long *flags);
761 unsigned int pblk_rb_sync_advance(struct pblk_rb *rb, unsigned int nr_entries);
762 struct pblk_rb_entry *pblk_rb_sync_scan_entry(struct pblk_rb *rb,
763 struct ppa_addr *ppa);
764 void pblk_rb_sync_end(struct pblk_rb *rb, unsigned long *flags);
765 unsigned int pblk_rb_flush_point_count(struct pblk_rb *rb);
767 unsigned int pblk_rb_read_count(struct pblk_rb *rb);
768 unsigned int pblk_rb_sync_count(struct pblk_rb *rb);
769 unsigned int pblk_rb_wrap_pos(struct pblk_rb *rb, unsigned int pos);
771 int pblk_rb_tear_down_check(struct pblk_rb *rb);
772 int pblk_rb_pos_oob(struct pblk_rb *rb, u64 pos);
773 void pblk_rb_data_free(struct pblk_rb *rb);
774 ssize_t pblk_rb_sysfs(struct pblk_rb *rb, char *buf);
779 struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int type);
780 void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int type);
781 void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write);
782 int pblk_setup_w_rec_rq(struct pblk *pblk, struct nvm_rq *rqd,
783 struct pblk_c_ctx *c_ctx);
784 void pblk_discard(struct pblk *pblk, struct bio *bio);
785 struct nvm_chk_meta *pblk_get_chunk_meta(struct pblk *pblk);
786 struct nvm_chk_meta *pblk_chunk_get_off(struct pblk *pblk,
787 struct nvm_chk_meta *lp,
788 struct ppa_addr ppa);
789 void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd);
790 void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd);
791 int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd);
792 int pblk_submit_io_sync(struct pblk *pblk, struct nvm_rq *rqd);
793 int pblk_submit_meta_io(struct pblk *pblk, struct pblk_line *meta_line);
794 void pblk_check_chunk_state_update(struct pblk *pblk, struct nvm_rq *rqd);
795 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
796 unsigned int nr_secs, unsigned int len,
797 int alloc_type, gfp_t gfp_mask);
798 struct pblk_line *pblk_line_get(struct pblk *pblk);
799 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk);
800 struct pblk_line *pblk_line_replace_data(struct pblk *pblk);
801 void pblk_ppa_to_line_put(struct pblk *pblk, struct ppa_addr ppa);
802 void pblk_rq_to_line_put(struct pblk *pblk, struct nvm_rq *rqd);
803 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line);
804 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line);
805 struct pblk_line *pblk_line_get_data(struct pblk *pblk);
806 struct pblk_line *pblk_line_get_erase(struct pblk *pblk);
807 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line);
808 int pblk_line_is_full(struct pblk_line *line);
809 void pblk_line_free(struct pblk_line *line);
810 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line);
811 void pblk_line_close(struct pblk *pblk, struct pblk_line *line);
812 void pblk_line_close_ws(struct work_struct *work);
813 void pblk_pipeline_stop(struct pblk *pblk);
814 void __pblk_pipeline_stop(struct pblk *pblk);
815 void __pblk_pipeline_flush(struct pblk *pblk);
816 void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
817 void (*work)(struct work_struct *), gfp_t gfp_mask,
818 struct workqueue_struct *wq);
819 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line);
820 int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line);
821 int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line,
823 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr erase_ppa);
824 void pblk_line_put(struct kref *ref);
825 void pblk_line_put_wq(struct kref *ref);
826 struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line);
827 u64 pblk_lookup_page(struct pblk *pblk, struct pblk_line *line);
828 void pblk_dealloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
829 u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
830 u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
831 int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
832 unsigned long secs_to_flush);
833 void pblk_down_rq(struct pblk *pblk, struct ppa_addr ppa,
834 unsigned long *lun_bitmap);
835 void pblk_down_chunk(struct pblk *pblk, struct ppa_addr ppa);
836 void pblk_up_chunk(struct pblk *pblk, struct ppa_addr ppa);
837 void pblk_up_rq(struct pblk *pblk, unsigned long *lun_bitmap);
838 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
840 void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
842 void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa);
843 void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
845 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa);
846 void pblk_update_map_cache(struct pblk *pblk, sector_t lba,
847 struct ppa_addr ppa);
848 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
849 struct ppa_addr ppa, struct ppa_addr entry_line);
850 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa,
851 struct pblk_line *gc_line, u64 paddr);
852 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
853 u64 *lba_list, int nr_secs);
854 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
855 sector_t blba, int nr_secs);
858 * pblk user I/O write path
860 int pblk_write_to_cache(struct pblk *pblk, struct bio *bio,
861 unsigned long flags);
862 int pblk_write_gc_to_cache(struct pblk *pblk, struct pblk_gc_rq *gc_rq);
867 void pblk_map_erase_rq(struct pblk *pblk, struct nvm_rq *rqd,
868 unsigned int sentry, unsigned long *lun_bitmap,
869 unsigned int valid_secs, struct ppa_addr *erase_ppa);
870 void pblk_map_rq(struct pblk *pblk, struct nvm_rq *rqd, unsigned int sentry,
871 unsigned long *lun_bitmap, unsigned int valid_secs,
877 int pblk_write_ts(void *data);
878 void pblk_write_timer_fn(struct timer_list *t);
879 void pblk_write_should_kick(struct pblk *pblk);
880 void pblk_write_kick(struct pblk *pblk);
885 extern struct bio_set pblk_bio_set;
886 int pblk_submit_read(struct pblk *pblk, struct bio *bio);
887 int pblk_submit_read_gc(struct pblk *pblk, struct pblk_gc_rq *gc_rq);
891 struct pblk_line *pblk_recov_l2p(struct pblk *pblk);
892 int pblk_recov_pad(struct pblk *pblk);
893 int pblk_recov_check_emeta(struct pblk *pblk, struct line_emeta *emeta);
898 #define PBLK_GC_MAX_READERS 8 /* Max number of outstanding GC reader jobs */
899 #define PBLK_GC_RQ_QD 128 /* Queue depth for inflight GC requests */
900 #define PBLK_GC_L_QD 4 /* Queue depth for inflight GC lines */
901 #define PBLK_GC_RSV_LINE 1 /* Reserved lines for GC */
903 int pblk_gc_init(struct pblk *pblk);
904 void pblk_gc_exit(struct pblk *pblk, bool graceful);
905 void pblk_gc_should_start(struct pblk *pblk);
906 void pblk_gc_should_stop(struct pblk *pblk);
907 void pblk_gc_should_kick(struct pblk *pblk);
908 void pblk_gc_free_full_lines(struct pblk *pblk);
909 void pblk_gc_sysfs_state_show(struct pblk *pblk, int *gc_enabled,
911 int pblk_gc_sysfs_force(struct pblk *pblk, int force);
916 void pblk_rl_init(struct pblk_rl *rl, int budget);
917 void pblk_rl_free(struct pblk_rl *rl);
918 void pblk_rl_update_rates(struct pblk_rl *rl);
919 int pblk_rl_high_thrs(struct pblk_rl *rl);
920 unsigned long pblk_rl_nr_free_blks(struct pblk_rl *rl);
921 unsigned long pblk_rl_nr_user_free_blks(struct pblk_rl *rl);
922 int pblk_rl_user_may_insert(struct pblk_rl *rl, int nr_entries);
923 void pblk_rl_inserted(struct pblk_rl *rl, int nr_entries);
924 void pblk_rl_user_in(struct pblk_rl *rl, int nr_entries);
925 int pblk_rl_gc_may_insert(struct pblk_rl *rl, int nr_entries);
926 void pblk_rl_gc_in(struct pblk_rl *rl, int nr_entries);
927 void pblk_rl_out(struct pblk_rl *rl, int nr_user, int nr_gc);
928 int pblk_rl_max_io(struct pblk_rl *rl);
929 void pblk_rl_free_lines_inc(struct pblk_rl *rl, struct pblk_line *line);
930 void pblk_rl_free_lines_dec(struct pblk_rl *rl, struct pblk_line *line,
932 int pblk_rl_is_limit(struct pblk_rl *rl);
934 void pblk_rl_werr_line_in(struct pblk_rl *rl);
935 void pblk_rl_werr_line_out(struct pblk_rl *rl);
940 int pblk_sysfs_init(struct gendisk *tdisk);
941 void pblk_sysfs_exit(struct gendisk *tdisk);
943 static inline void *pblk_malloc(size_t size, int type, gfp_t flags)
945 if (type == PBLK_KMALLOC_META)
946 return kmalloc(size, flags);
947 return vmalloc(size);
950 static inline void pblk_mfree(void *ptr, int type)
952 if (type == PBLK_KMALLOC_META)
958 static inline struct nvm_rq *nvm_rq_from_c_ctx(void *c_ctx)
960 return c_ctx - sizeof(struct nvm_rq);
963 static inline void *emeta_to_bb(struct line_emeta *emeta)
965 return emeta->bb_bitmap;
968 static inline void *emeta_to_wa(struct pblk_line_meta *lm,
969 struct line_emeta *emeta)
971 return emeta->bb_bitmap + lm->blk_bitmap_len;
974 static inline void *emeta_to_lbas(struct pblk *pblk, struct line_emeta *emeta)
976 return ((void *)emeta + pblk->lm.emeta_len[1]);
979 static inline void *emeta_to_vsc(struct pblk *pblk, struct line_emeta *emeta)
981 return (emeta_to_lbas(pblk, emeta) + pblk->lm.emeta_len[2]);
984 static inline int pblk_line_vsc(struct pblk_line *line)
986 return le32_to_cpu(*line->vsc);
989 static inline int pblk_pad_distance(struct pblk *pblk)
991 struct nvm_tgt_dev *dev = pblk->dev;
992 struct nvm_geo *geo = &dev->geo;
994 return geo->mw_cunits * geo->all_luns * geo->ws_opt;
997 static inline int pblk_ppa_to_line_id(struct ppa_addr p)
1002 static inline struct pblk_line *pblk_ppa_to_line(struct pblk *pblk,
1005 return &pblk->lines[pblk_ppa_to_line_id(p)];
1008 static inline int pblk_ppa_to_pos(struct nvm_geo *geo, struct ppa_addr p)
1010 return p.a.lun * geo->num_ch + p.a.ch;
1013 static inline struct ppa_addr addr_to_gen_ppa(struct pblk *pblk, u64 paddr,
1016 struct nvm_tgt_dev *dev = pblk->dev;
1017 struct nvm_geo *geo = &dev->geo;
1018 struct ppa_addr ppa;
1020 if (geo->version == NVM_OCSSD_SPEC_12) {
1021 struct nvm_addrf_12 *ppaf = (struct nvm_addrf_12 *)&pblk->addrf;
1024 ppa.g.blk = line_id;
1025 ppa.g.pg = (paddr & ppaf->pg_mask) >> ppaf->pg_offset;
1026 ppa.g.lun = (paddr & ppaf->lun_mask) >> ppaf->lun_offset;
1027 ppa.g.ch = (paddr & ppaf->ch_mask) >> ppaf->ch_offset;
1028 ppa.g.pl = (paddr & ppaf->pln_mask) >> ppaf->pln_offset;
1029 ppa.g.sec = (paddr & ppaf->sec_mask) >> ppaf->sec_offset;
1031 struct pblk_addrf *uaddrf = &pblk->uaddrf;
1032 int secs, chnls, luns;
1036 ppa.m.chk = line_id;
1038 paddr = div_u64_rem(paddr, uaddrf->sec_stripe, &secs);
1041 paddr = div_u64_rem(paddr, uaddrf->ch_stripe, &chnls);
1044 paddr = div_u64_rem(paddr, uaddrf->lun_stripe, &luns);
1047 ppa.m.sec += uaddrf->sec_stripe * paddr;
1053 static inline struct nvm_chk_meta *pblk_dev_ppa_to_chunk(struct pblk *pblk,
1056 struct nvm_tgt_dev *dev = pblk->dev;
1057 struct nvm_geo *geo = &dev->geo;
1058 struct pblk_line *line = pblk_ppa_to_line(pblk, p);
1059 int pos = pblk_ppa_to_pos(geo, p);
1061 return &line->chks[pos];
1064 static inline u64 pblk_dev_ppa_to_chunk_addr(struct pblk *pblk,
1067 struct nvm_tgt_dev *dev = pblk->dev;
1069 return dev_to_chunk_addr(dev->parent, &pblk->addrf, p);
1072 static inline u64 pblk_dev_ppa_to_line_addr(struct pblk *pblk,
1075 struct nvm_tgt_dev *dev = pblk->dev;
1076 struct nvm_geo *geo = &dev->geo;
1079 if (geo->version == NVM_OCSSD_SPEC_12) {
1080 struct nvm_addrf_12 *ppaf = (struct nvm_addrf_12 *)&pblk->addrf;
1082 paddr = (u64)p.g.ch << ppaf->ch_offset;
1083 paddr |= (u64)p.g.lun << ppaf->lun_offset;
1084 paddr |= (u64)p.g.pg << ppaf->pg_offset;
1085 paddr |= (u64)p.g.pl << ppaf->pln_offset;
1086 paddr |= (u64)p.g.sec << ppaf->sec_offset;
1088 struct pblk_addrf *uaddrf = &pblk->uaddrf;
1092 paddr = (u64)p.m.grp * uaddrf->sec_stripe;
1093 paddr += (u64)p.m.pu * uaddrf->sec_lun_stripe;
1095 secs = div_u64_rem(secs, uaddrf->sec_stripe, &sec_stripe);
1096 paddr += secs * uaddrf->sec_ws_stripe;
1097 paddr += sec_stripe;
1103 static inline struct ppa_addr pblk_ppa32_to_ppa64(struct pblk *pblk, u32 ppa32)
1105 struct nvm_tgt_dev *dev = pblk->dev;
1107 return nvm_ppa32_to_ppa64(dev->parent, &pblk->addrf, ppa32);
1110 static inline u32 pblk_ppa64_to_ppa32(struct pblk *pblk, struct ppa_addr ppa64)
1112 struct nvm_tgt_dev *dev = pblk->dev;
1114 return nvm_ppa64_to_ppa32(dev->parent, &pblk->addrf, ppa64);
1117 static inline struct ppa_addr pblk_trans_map_get(struct pblk *pblk,
1120 struct ppa_addr ppa;
1122 if (pblk->addrf_len < 32) {
1123 u32 *map = (u32 *)pblk->trans_map;
1125 ppa = pblk_ppa32_to_ppa64(pblk, map[lba]);
1127 struct ppa_addr *map = (struct ppa_addr *)pblk->trans_map;
1135 static inline void pblk_trans_map_set(struct pblk *pblk, sector_t lba,
1136 struct ppa_addr ppa)
1138 if (pblk->addrf_len < 32) {
1139 u32 *map = (u32 *)pblk->trans_map;
1141 map[lba] = pblk_ppa64_to_ppa32(pblk, ppa);
1143 u64 *map = (u64 *)pblk->trans_map;
1149 static inline int pblk_ppa_empty(struct ppa_addr ppa_addr)
1151 return (ppa_addr.ppa == ADDR_EMPTY);
1154 static inline void pblk_ppa_set_empty(struct ppa_addr *ppa_addr)
1156 ppa_addr->ppa = ADDR_EMPTY;
1159 static inline bool pblk_ppa_comp(struct ppa_addr lppa, struct ppa_addr rppa)
1161 return (lppa.ppa == rppa.ppa);
1164 static inline int pblk_addr_in_cache(struct ppa_addr ppa)
1166 return (ppa.ppa != ADDR_EMPTY && ppa.c.is_cached);
1169 static inline int pblk_addr_to_cacheline(struct ppa_addr ppa)
1174 static inline struct ppa_addr pblk_cacheline_to_addr(int addr)
1184 static inline u32 pblk_calc_meta_header_crc(struct pblk *pblk,
1185 struct line_header *header)
1189 crc = crc32_le(crc, (unsigned char *)header + sizeof(crc),
1190 sizeof(struct line_header) - sizeof(crc));
1195 static inline u32 pblk_calc_smeta_crc(struct pblk *pblk,
1196 struct line_smeta *smeta)
1198 struct pblk_line_meta *lm = &pblk->lm;
1201 crc = crc32_le(crc, (unsigned char *)smeta +
1202 sizeof(struct line_header) + sizeof(crc),
1204 sizeof(struct line_header) - sizeof(crc));
1209 static inline u32 pblk_calc_emeta_crc(struct pblk *pblk,
1210 struct line_emeta *emeta)
1212 struct pblk_line_meta *lm = &pblk->lm;
1215 crc = crc32_le(crc, (unsigned char *)emeta +
1216 sizeof(struct line_header) + sizeof(crc),
1218 sizeof(struct line_header) - sizeof(crc));
1223 static inline int pblk_io_aligned(struct pblk *pblk, int nr_secs)
1225 return !(nr_secs % pblk->min_write_pgs);
1228 #ifdef CONFIG_NVM_PBLK_DEBUG
1229 static inline void print_ppa(struct pblk *pblk, struct ppa_addr *p,
1230 char *msg, int error)
1232 struct nvm_geo *geo = &pblk->dev->geo;
1234 if (p->c.is_cached) {
1235 pblk_err(pblk, "ppa: (%s: %x) cache line: %llu\n",
1236 msg, error, (u64)p->c.line);
1237 } else if (geo->version == NVM_OCSSD_SPEC_12) {
1238 pblk_err(pblk, "ppa: (%s: %x):ch:%d,lun:%d,blk:%d,pg:%d,pl:%d,sec:%d\n",
1240 p->g.ch, p->g.lun, p->g.blk,
1241 p->g.pg, p->g.pl, p->g.sec);
1243 pblk_err(pblk, "ppa: (%s: %x):ch:%d,lun:%d,chk:%d,sec:%d\n",
1245 p->m.grp, p->m.pu, p->m.chk, p->m.sec);
1249 static inline void pblk_print_failed_rqd(struct pblk *pblk, struct nvm_rq *rqd,
1254 if (rqd->nr_ppas == 1) {
1255 print_ppa(pblk, &rqd->ppa_addr, "rqd", error);
1259 while ((bit = find_next_bit((void *)&rqd->ppa_status, rqd->nr_ppas,
1260 bit + 1)) < rqd->nr_ppas) {
1261 print_ppa(pblk, &rqd->ppa_list[bit], "rqd", error);
1264 pblk_err(pblk, "error:%d, ppa_status:%llx\n", error, rqd->ppa_status);
1267 static inline int pblk_boundary_ppa_checks(struct nvm_tgt_dev *tgt_dev,
1268 struct ppa_addr *ppas, int nr_ppas)
1270 struct nvm_geo *geo = &tgt_dev->geo;
1271 struct ppa_addr *ppa;
1274 for (i = 0; i < nr_ppas; i++) {
1277 if (geo->version == NVM_OCSSD_SPEC_12) {
1278 if (!ppa->c.is_cached &&
1279 ppa->g.ch < geo->num_ch &&
1280 ppa->g.lun < geo->num_lun &&
1281 ppa->g.pl < geo->num_pln &&
1282 ppa->g.blk < geo->num_chk &&
1283 ppa->g.pg < geo->num_pg &&
1284 ppa->g.sec < geo->ws_min)
1287 if (!ppa->c.is_cached &&
1288 ppa->m.grp < geo->num_ch &&
1289 ppa->m.pu < geo->num_lun &&
1290 ppa->m.chk < geo->num_chk &&
1291 ppa->m.sec < geo->clba)
1295 print_ppa(tgt_dev->q->queuedata, ppa, "boundary", i);
1302 static inline int pblk_check_io(struct pblk *pblk, struct nvm_rq *rqd)
1304 struct nvm_tgt_dev *dev = pblk->dev;
1305 struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
1307 if (pblk_boundary_ppa_checks(dev, ppa_list, rqd->nr_ppas)) {
1312 if (rqd->opcode == NVM_OP_PWRITE) {
1313 struct pblk_line *line;
1316 for (i = 0; i < rqd->nr_ppas; i++) {
1317 line = pblk_ppa_to_line(pblk, ppa_list[i]);
1319 spin_lock(&line->lock);
1320 if (line->state != PBLK_LINESTATE_OPEN) {
1321 pblk_err(pblk, "bad ppa: line:%d,state:%d\n",
1322 line->id, line->state);
1324 spin_unlock(&line->lock);
1327 spin_unlock(&line->lock);
1335 static inline int pblk_boundary_paddr_checks(struct pblk *pblk, u64 paddr)
1337 struct pblk_line_meta *lm = &pblk->lm;
1339 if (paddr > lm->sec_per_line)
1345 static inline unsigned int pblk_get_bi_idx(struct bio *bio)
1347 return bio->bi_iter.bi_idx;
1350 static inline sector_t pblk_get_lba(struct bio *bio)
1352 return bio->bi_iter.bi_sector / NR_PHY_IN_LOG;
1355 static inline unsigned int pblk_get_secs(struct bio *bio)
1357 return bio->bi_iter.bi_size / PBLK_EXPOSED_PAGE_SIZE;
1360 static inline void pblk_setup_uuid(struct pblk *pblk)
1365 memcpy(pblk->instance_uuid, uuid.b, 16);
1368 static inline char *pblk_disk_name(struct pblk *pblk)
1370 struct gendisk *disk = pblk->disk;
1372 return disk->disk_name;
1374 #endif /* PBLK_H_ */