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)
40 #define PBLK_MAX_REQ_ADDRS (64)
41 #define PBLK_MAX_REQ_ADDRS_PW (6)
43 #define PBLK_NR_CLOSE_JOBS (4)
45 #define PBLK_CACHE_NAME_LEN (DISK_NAME_LEN + 16)
47 #define PBLK_COMMAND_TIMEOUT_MS 30000
49 /* Max 512 LUNs per device */
50 #define PBLK_MAX_LUNS_BITMAP (4)
52 #define NR_PHY_IN_LOG (PBLK_EXPOSED_PAGE_SIZE / PBLK_SECTOR)
54 #define pblk_for_each_lun(pblk, rlun, i) \
55 for ((i) = 0, rlun = &(pblk)->luns[0]; \
56 (i) < (pblk)->nr_luns; (i)++, rlun = &(pblk)->luns[(i)])
58 /* Static pool sizes */
59 #define PBLK_GEN_WS_POOL_SIZE (2)
63 PBLK_WRITE = WRITE,/* Write from write buffer */
64 PBLK_WRITE_INT, /* Internal write - no write buffer */
70 PBLK_IOTYPE_USER = 1 << 0,
71 PBLK_IOTYPE_GC = 1 << 1,
73 /* Write buffer flags */
74 PBLK_FLUSH_ENTRY = 1 << 2,
75 PBLK_WRITTEN_DATA = 1 << 3,
76 PBLK_SUBMITTED_ENTRY = 1 << 4,
77 PBLK_WRITABLE_ENTRY = 1 << 5,
81 PBLK_BLK_ST_OPEN = 0x1,
82 PBLK_BLK_ST_CLOSED = 0x2,
85 struct pblk_sec_meta {
90 /* The number of GC lists and the rate-limiter states go together. This way the
91 * rate-limiter can dictate how much GC is needed based on resource utilization.
93 #define PBLK_GC_NR_LISTS 3
101 #define pblk_dma_meta_size (sizeof(struct pblk_sec_meta) * PBLK_MAX_REQ_ADDRS)
103 /* write buffer completion context */
105 struct list_head list; /* Head for out-of-order completion */
107 unsigned long *lun_bitmap; /* Luns used on current request */
109 unsigned int nr_valid;
110 unsigned int nr_padded;
113 /* generic context */
121 struct completion wait;
125 /* Recovery context */
126 struct pblk_rec_ctx {
129 struct list_head failed;
130 struct work_struct ws_rec;
135 struct bio_list bios; /* Original bios - used for completion
136 * in REQ_FUA, REQ_FLUSH case
138 u64 lba; /* Logic addr. associated with entry */
139 struct ppa_addr ppa; /* Physic addr. associated with entry */
140 int flags; /* Write context flags */
143 struct pblk_rb_entry {
144 struct ppa_addr cacheline; /* Cacheline for this entry */
145 void *data; /* Pointer to data on this entry */
146 struct pblk_w_ctx w_ctx; /* Context for this entry */
147 struct list_head index; /* List head to enable indexes */
150 #define EMPTY_ENTRY (~0U)
152 struct pblk_rb_pages {
155 struct list_head list;
159 struct pblk_rb_entry *entries; /* Ring buffer entries */
160 unsigned int mem; /* Write offset - points to next
161 * writable entry in memory
163 unsigned int subm; /* Read offset - points to last entry
164 * that has been submitted to the media
167 unsigned int sync; /* Synced - backpointer that signals
168 * the last submitted entry that has
169 * been successfully persisted to media
171 unsigned int sync_point; /* Sync point - last entry that must be
172 * flushed to the media. Used with
173 * REQ_FLUSH and REQ_FUA
175 unsigned int l2p_update; /* l2p update point - next entry for
176 * which l2p mapping will be updated to
177 * contain a device ppa address (instead
180 unsigned int nr_entries; /* Number of entries in write buffer -
181 * must be a power of two
183 unsigned int seg_size; /* Size of the data segments being
184 * stored on each entry. Typically this
188 struct list_head pages; /* List of data pages */
190 spinlock_t w_lock; /* Write lock */
191 spinlock_t s_lock; /* Sync lock */
193 #ifdef CONFIG_NVM_DEBUG
194 atomic_t inflight_sync_point; /* Not served REQ_FLUSH | REQ_FUA */
198 #define PBLK_RECOVERY_SECTORS 16
201 struct ppa_addr bppa;
203 u8 *bb_list; /* Bad block list for LUN. Only used on
204 * bring up. Bad blocks are managed
205 * within lines on run-time.
208 struct semaphore wr_sem;
212 struct pblk_line *line;
214 u64 paddr_list[PBLK_MAX_REQ_ADDRS];
215 u64 lba_list[PBLK_MAX_REQ_ADDRS];
218 struct list_head list;
222 /* These states are not protected by a lock since (i) they are in the
223 * fast path, and (ii) they are not critical.
229 struct task_struct *gc_ts;
230 struct task_struct *gc_writer_ts;
231 struct task_struct *gc_reader_ts;
233 struct workqueue_struct *gc_line_reader_wq;
234 struct workqueue_struct *gc_reader_wq;
236 struct timer_list gc_timer;
238 struct semaphore gc_sem;
239 atomic_t inflight_gc;
242 struct list_head w_list;
243 struct list_head r_list;
251 unsigned int high; /* Upper threshold for rate limiter (free run -
252 * user I/O rate limiter
254 unsigned int low; /* Lower threshold for rate limiter (user I/O
255 * rate limiter - stall)
257 unsigned int high_pw; /* High rounded up as a power of 2 */
259 #define PBLK_USER_HIGH_THRS 8 /* Begin write limit at 12% available blks */
260 #define PBLK_USER_LOW_THRS 10 /* Aggressive GC at 10% available blocks */
262 int rb_windows_pw; /* Number of rate windows in the write buffer
263 * given as a power-of-2. This guarantees that
264 * when user I/O is being rate limited, there
265 * will be reserved enough space for the GC to
266 * place its payload. A window is of
267 * pblk->max_write_pgs size, which in NVMe is
270 int rb_budget; /* Total number of entries available for I/O */
271 int rb_user_max; /* Max buffer entries available for user I/O */
272 int rb_gc_max; /* Max buffer entries available for GC I/O */
273 int rb_gc_rsv; /* Reserved buffer entries for GC I/O */
274 int rb_state; /* Rate-limiter current state */
275 int rb_max_io; /* Maximum size for an I/O giving the config */
277 atomic_t rb_user_cnt; /* User I/O buffer counter */
278 atomic_t rb_gc_cnt; /* GC I/O buffer counter */
279 atomic_t rb_space; /* Space limit in case of reaching capacity */
281 int rsv_blocks; /* Reserved blocks for GC */
286 struct timer_list u_timer;
288 unsigned long long nr_secs;
289 unsigned long total_blocks;
290 atomic_t free_blocks;
293 #define PBLK_LINE_EMPTY (~0U)
297 PBLK_LINETYPE_FREE = 0,
298 PBLK_LINETYPE_LOG = 1,
299 PBLK_LINETYPE_DATA = 2,
302 PBLK_LINESTATE_FREE = 10,
303 PBLK_LINESTATE_OPEN = 11,
304 PBLK_LINESTATE_CLOSED = 12,
305 PBLK_LINESTATE_GC = 13,
306 PBLK_LINESTATE_BAD = 14,
307 PBLK_LINESTATE_CORRUPT = 15,
310 PBLK_LINEGC_NONE = 20,
311 PBLK_LINEGC_EMPTY = 21,
312 PBLK_LINEGC_LOW = 22,
313 PBLK_LINEGC_MID = 23,
314 PBLK_LINEGC_HIGH = 24,
315 PBLK_LINEGC_FULL = 25,
318 #define PBLK_MAGIC 0x70626c6b /*pblk*/
319 #define SMETA_VERSION cpu_to_le16(1)
323 __le32 identifier; /* pblk identifier */
324 __u8 uuid[16]; /* instance uuid */
325 __le16 type; /* line type */
326 __le16 version; /* type version */
327 __le32 id; /* line id for current line */
331 struct line_header header;
333 __le32 crc; /* Full structure including struct crc */
334 /* Previous line metadata */
335 __le32 prev_id; /* Line id for previous line */
337 /* Current line metadata */
338 __le64 seq_nr; /* Sequence number for current line */
341 __le32 window_wr_lun; /* Number of parallel LUNs to write */
349 * Metadata layout in media:
351 * 1. struct line_emeta
352 * 2. bad block bitmap (u64 * window_wr_lun)
353 * Mid sectors (start at lbas_sector):
354 * 3. nr_lbas (u64) forming lba list
355 * Last sectors (start at vsc_sector):
356 * 4. u32 valid sector count (vsc) for all lines (~0U: free line)
359 struct line_header header;
361 __le32 crc; /* Full structure including struct crc */
363 /* Previous line metadata */
364 __le32 prev_id; /* Line id for prev 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 */
372 /* Bookkeeping for recovery */
373 __le32 next_id; /* Line id for next line */
374 __le64 nr_lbas; /* Number of lbas mapped in line */
375 __le64 nr_valid_lbas; /* Number of valid lbas mapped in line */
376 __le64 bb_bitmap[]; /* Updated bad block bitmap for line */
380 struct line_emeta *buf; /* emeta buffer in media format */
381 int mem; /* Write offset - points to next
382 * writable entry in memory
384 atomic_t sync; /* Synced - backpointer that signals the
385 * last entry that has been successfully
388 unsigned int nr_entries; /* Number of emeta entries */
392 struct line_smeta *buf; /* smeta buffer in persistent format */
397 unsigned int id; /* Line number corresponds to the
400 unsigned int seq_nr; /* Unique line sequence number */
402 int state; /* PBLK_LINESTATE_X */
403 int type; /* PBLK_LINETYPE_X */
404 int gc_group; /* PBLK_LINEGC_X */
405 struct list_head list; /* Free, GC lists */
407 unsigned long *lun_bitmap; /* Bitmap for LUNs mapped in line */
409 struct pblk_smeta *smeta; /* Start metadata */
410 struct pblk_emeta *emeta; /* End medatada */
412 int meta_line; /* Metadata line id */
413 int meta_distance; /* Distance between data and metadata */
415 u64 smeta_ssec; /* Sector where smeta starts */
416 u64 emeta_ssec; /* Sector where emeta starts */
418 unsigned int sec_in_line; /* Number of usable secs in line */
420 atomic_t blk_in_line; /* Number of good blocks in line */
421 unsigned long *blk_bitmap; /* Bitmap for valid/invalid blocks */
422 unsigned long *erase_bitmap; /* Bitmap for erased blocks */
424 unsigned long *map_bitmap; /* Bitmap for mapped sectors in line */
425 unsigned long *invalid_bitmap; /* Bitmap for invalid sectors in line */
427 atomic_t left_eblks; /* Blocks left for erasing */
428 atomic_t left_seblks; /* Blocks left for sync erasing */
430 int left_msecs; /* Sectors left for mapping */
431 unsigned int cur_sec; /* Sector map pointer */
432 unsigned int nr_valid_lbas; /* Number of valid lbas in line */
434 __le32 *vsc; /* Valid sector count in line */
436 struct kref ref; /* Write buffer L2P references */
438 spinlock_t lock; /* Necessary for invalid_bitmap only */
441 #define PBLK_DATA_LINES 4
444 PBLK_KMALLOC_META = 1,
445 PBLK_VMALLOC_META = 2,
449 PBLK_EMETA_TYPE_HEADER = 1, /* struct line_emeta first sector */
450 PBLK_EMETA_TYPE_LLBA = 2, /* lba list - type: __le64 */
451 PBLK_EMETA_TYPE_VSC = 3, /* vsc list - type: __le32 */
454 struct pblk_line_mgmt {
455 int nr_lines; /* Total number of full lines */
456 int nr_free_lines; /* Number of full lines in free list */
458 /* Free lists - use free_lock */
459 struct list_head free_list; /* Full lines ready to use */
460 struct list_head corrupt_list; /* Full lines corrupted */
461 struct list_head bad_list; /* Full lines bad */
463 /* GC lists - use gc_lock */
464 struct list_head *gc_lists[PBLK_GC_NR_LISTS];
465 struct list_head gc_high_list; /* Full lines ready to GC, high isc */
466 struct list_head gc_mid_list; /* Full lines ready to GC, mid isc */
467 struct list_head gc_low_list; /* Full lines ready to GC, low isc */
469 struct list_head gc_full_list; /* Full lines ready to GC, no valid */
470 struct list_head gc_empty_list; /* Full lines close, all valid */
472 struct pblk_line *log_line; /* Current FTL log line */
473 struct pblk_line *data_line; /* Current data line */
474 struct pblk_line *log_next; /* Next FTL log line */
475 struct pblk_line *data_next; /* Next data line */
477 struct list_head emeta_list; /* Lines queued to schedule emeta */
479 __le32 *vsc_list; /* Valid sector counts for all lines */
481 /* Metadata allocation type: VMALLOC | KMALLOC */
482 int emeta_alloc_type;
484 /* Pre-allocated metadata for data lines */
485 struct pblk_smeta *sline_meta[PBLK_DATA_LINES];
486 struct pblk_emeta *eline_meta[PBLK_DATA_LINES];
487 unsigned long meta_bitmap;
489 /* Helpers for fast bitmap calculations */
490 unsigned long *bb_template;
491 unsigned long *bb_aux;
493 unsigned long d_seq_nr; /* Data line unique sequence number */
494 unsigned long l_seq_nr; /* Log line unique sequence number */
496 spinlock_t free_lock;
497 spinlock_t close_lock;
501 struct pblk_line_meta {
502 unsigned int smeta_len; /* Total length for smeta */
503 unsigned int smeta_sec; /* Sectors needed for smeta */
505 unsigned int emeta_len[4]; /* Lengths for emeta:
507 * [1]: struct line_emeta length
508 * [2]: L2P portion length
509 * [3]: vsc list length
511 unsigned int emeta_sec[4]; /* Sectors needed for emeta. Same layout
515 unsigned int emeta_bb; /* Boundary for bb that affects emeta */
517 unsigned int vsc_list_len; /* Length for vsc list */
518 unsigned int sec_bitmap_len; /* Length for sector bitmap in line */
519 unsigned int blk_bitmap_len; /* Length for block bitmap in line */
520 unsigned int lun_bitmap_len; /* Length for lun bitmap in line */
522 unsigned int blk_per_line; /* Number of blocks in a full line */
523 unsigned int sec_per_line; /* Number of sectors in a line */
524 unsigned int dsec_per_line; /* Number of data sectors in a line */
525 unsigned int min_blk_line; /* Min. number of good blocks in line */
527 unsigned int mid_thrs; /* Threshold for GC mid list */
528 unsigned int high_thrs; /* Threshold for GC high list */
530 unsigned int meta_distance; /* Distance between data and metadata */
533 struct pblk_addr_format {
549 PBLK_STATE_RUNNING = 0,
550 PBLK_STATE_STOPPING = 1,
551 PBLK_STATE_RECOVERING = 2,
552 PBLK_STATE_STOPPED = 3,
556 struct nvm_tgt_dev *dev;
557 struct gendisk *disk;
561 struct pblk_lun *luns;
563 struct pblk_line *lines; /* Line array */
564 struct pblk_line_mgmt l_mg; /* Line management */
565 struct pblk_line_meta lm; /* Line metadata */
568 struct pblk_addr_format ppaf;
572 int state; /* pblk line state */
574 int min_write_pgs; /* Minimum amount of pages required by controller */
575 int max_write_pgs; /* Maximum amount of pages supported by controller */
576 int pgs_in_buffer; /* Number of pages that need to be held in buffer to
577 * guarantee successful reads.
580 sector_t capacity; /* Device capacity when bad blocks are subtracted */
581 int over_pct; /* Percentage of device used for over-provisioning */
583 /* pblk provisioning values. Used by rate limiter */
588 unsigned char instance_uuid[16];
589 #ifdef CONFIG_NVM_DEBUG
590 /* All debug counters apply to 4kb sector I/Os */
591 atomic_long_t inflight_writes; /* Inflight writes (user and gc) */
592 atomic_long_t padded_writes; /* Sectors padded due to flush/fua */
593 atomic_long_t padded_wb; /* Sectors padded in write buffer */
594 atomic_long_t nr_flush; /* Number of flush/fua I/O */
595 atomic_long_t req_writes; /* Sectors stored on write buffer */
596 atomic_long_t sub_writes; /* Sectors submitted from buffer */
597 atomic_long_t sync_writes; /* Sectors synced to media */
598 atomic_long_t inflight_reads; /* Inflight sector read requests */
599 atomic_long_t cache_reads; /* Read requests that hit the cache */
600 atomic_long_t sync_reads; /* Completed sector read requests */
601 atomic_long_t recov_writes; /* Sectors submitted from recovery */
602 atomic_long_t recov_gc_writes; /* Sectors submitted from write GC */
603 atomic_long_t recov_gc_reads; /* Sectors submitted from read GC */
608 atomic_long_t read_failed;
609 atomic_long_t read_empty;
610 atomic_long_t read_high_ecc;
611 atomic_long_t read_failed_gc;
612 atomic_long_t write_failed;
613 atomic_long_t erase_failed;
615 atomic_t inflight_io; /* General inflight I/O counter */
617 struct task_struct *writer_ts;
619 /* Simple translation map of logical addresses to physical addresses.
620 * The logical addresses is known by the host system, while the physical
621 * addresses are used when writing to the disk block device.
623 unsigned char *trans_map;
624 spinlock_t trans_lock;
626 struct list_head compl_list;
628 mempool_t *page_bio_pool;
629 mempool_t *gen_ws_pool;
631 mempool_t *r_rq_pool;
632 mempool_t *w_rq_pool;
633 mempool_t *e_rq_pool;
635 struct workqueue_struct *close_wq;
636 struct workqueue_struct *bb_wq;
638 struct timer_list wtimer;
643 struct pblk_line_ws {
645 struct pblk_line *line;
647 struct work_struct ws;
650 #define pblk_g_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_g_ctx))
651 #define pblk_w_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_c_ctx))
654 * pblk ring buffer operations
656 int pblk_rb_init(struct pblk_rb *rb, struct pblk_rb_entry *rb_entry_base,
657 unsigned int power_size, unsigned int power_seg_sz);
658 unsigned int pblk_rb_calculate_size(unsigned int nr_entries);
659 void *pblk_rb_entries_ref(struct pblk_rb *rb);
660 int pblk_rb_may_write_user(struct pblk_rb *rb, struct bio *bio,
661 unsigned int nr_entries, unsigned int *pos);
662 int pblk_rb_may_write_gc(struct pblk_rb *rb, unsigned int nr_entries,
664 void pblk_rb_write_entry_user(struct pblk_rb *rb, void *data,
665 struct pblk_w_ctx w_ctx, unsigned int pos);
666 void pblk_rb_write_entry_gc(struct pblk_rb *rb, void *data,
667 struct pblk_w_ctx w_ctx, struct pblk_line *line,
668 u64 paddr, unsigned int pos);
669 struct pblk_w_ctx *pblk_rb_w_ctx(struct pblk_rb *rb, unsigned int pos);
670 void pblk_rb_flush(struct pblk_rb *rb);
672 void pblk_rb_sync_l2p(struct pblk_rb *rb);
673 unsigned int pblk_rb_read_to_bio(struct pblk_rb *rb, struct nvm_rq *rqd,
674 unsigned int pos, unsigned int nr_entries,
676 unsigned int pblk_rb_read_to_bio_list(struct pblk_rb *rb, struct bio *bio,
677 struct list_head *list,
679 int pblk_rb_copy_to_bio(struct pblk_rb *rb, struct bio *bio, sector_t lba,
680 struct ppa_addr ppa, int bio_iter, bool advanced_bio);
681 unsigned int pblk_rb_read_commit(struct pblk_rb *rb, unsigned int entries);
683 unsigned int pblk_rb_sync_init(struct pblk_rb *rb, unsigned long *flags);
684 unsigned int pblk_rb_sync_advance(struct pblk_rb *rb, unsigned int nr_entries);
685 struct pblk_rb_entry *pblk_rb_sync_scan_entry(struct pblk_rb *rb,
686 struct ppa_addr *ppa);
687 void pblk_rb_sync_end(struct pblk_rb *rb, unsigned long *flags);
688 unsigned int pblk_rb_sync_point_count(struct pblk_rb *rb);
690 unsigned int pblk_rb_read_count(struct pblk_rb *rb);
691 unsigned int pblk_rb_sync_count(struct pblk_rb *rb);
692 unsigned int pblk_rb_wrap_pos(struct pblk_rb *rb, unsigned int pos);
694 int pblk_rb_tear_down_check(struct pblk_rb *rb);
695 int pblk_rb_pos_oob(struct pblk_rb *rb, u64 pos);
696 void pblk_rb_data_free(struct pblk_rb *rb);
697 ssize_t pblk_rb_sysfs(struct pblk_rb *rb, char *buf);
702 struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int type);
703 void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int type);
704 void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write);
705 int pblk_setup_w_rec_rq(struct pblk *pblk, struct nvm_rq *rqd,
706 struct pblk_c_ctx *c_ctx);
707 void pblk_wait_for_meta(struct pblk *pblk);
708 struct ppa_addr pblk_get_lba_map(struct pblk *pblk, sector_t lba);
709 void pblk_discard(struct pblk *pblk, struct bio *bio);
710 void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd);
711 void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd);
712 int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd);
713 int pblk_submit_meta_io(struct pblk *pblk, struct pblk_line *meta_line);
714 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
715 unsigned int nr_secs, unsigned int len,
716 int alloc_type, gfp_t gfp_mask);
717 struct pblk_line *pblk_line_get(struct pblk *pblk);
718 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk);
719 void pblk_line_replace_data(struct pblk *pblk);
720 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line);
721 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line);
722 struct pblk_line *pblk_line_get_data(struct pblk *pblk);
723 struct pblk_line *pblk_line_get_erase(struct pblk *pblk);
724 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line);
725 int pblk_line_is_full(struct pblk_line *line);
726 void pblk_line_free(struct pblk *pblk, struct pblk_line *line);
727 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line);
728 void pblk_line_close(struct pblk *pblk, struct pblk_line *line);
729 void pblk_line_close_meta_sync(struct pblk *pblk);
730 void pblk_line_close_ws(struct work_struct *work);
731 void pblk_pipeline_stop(struct pblk *pblk);
732 void pblk_line_mark_bb(struct work_struct *work);
733 void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
734 void (*work)(struct work_struct *), gfp_t gfp_mask,
735 struct workqueue_struct *wq);
736 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line);
737 int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line);
738 int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line,
740 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr erase_ppa);
741 void pblk_line_put(struct kref *ref);
742 struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line);
743 u64 pblk_lookup_page(struct pblk *pblk, struct pblk_line *line);
744 void pblk_dealloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
745 u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
746 u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
747 int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
748 unsigned long secs_to_flush);
749 void pblk_up_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas);
750 void pblk_down_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
751 unsigned long *lun_bitmap);
752 void pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas);
753 void pblk_up_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
754 unsigned long *lun_bitmap);
755 void pblk_end_bio_sync(struct bio *bio);
756 void pblk_end_io_sync(struct nvm_rq *rqd);
757 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
759 void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
761 void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa);
762 void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
764 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa);
765 void pblk_update_map_cache(struct pblk *pblk, sector_t lba,
766 struct ppa_addr ppa);
767 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
768 struct ppa_addr ppa, struct ppa_addr entry_line);
769 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa,
770 struct pblk_line *gc_line, u64 paddr);
771 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
772 u64 *lba_list, int nr_secs);
773 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
774 sector_t blba, int nr_secs);
777 * pblk user I/O write path
779 int pblk_write_to_cache(struct pblk *pblk, struct bio *bio,
780 unsigned long flags);
781 int pblk_write_gc_to_cache(struct pblk *pblk, struct pblk_gc_rq *gc_rq);
786 void pblk_map_erase_rq(struct pblk *pblk, struct nvm_rq *rqd,
787 unsigned int sentry, unsigned long *lun_bitmap,
788 unsigned int valid_secs, struct ppa_addr *erase_ppa);
789 void pblk_map_rq(struct pblk *pblk, struct nvm_rq *rqd, unsigned int sentry,
790 unsigned long *lun_bitmap, unsigned int valid_secs,
796 int pblk_write_ts(void *data);
797 void pblk_write_timer_fn(unsigned long data);
798 void pblk_write_should_kick(struct pblk *pblk);
803 extern struct bio_set *pblk_bio_set;
804 int pblk_submit_read(struct pblk *pblk, struct bio *bio);
805 int pblk_submit_read_gc(struct pblk *pblk, struct pblk_gc_rq *gc_rq);
809 void pblk_submit_rec(struct work_struct *work);
810 struct pblk_line *pblk_recov_l2p(struct pblk *pblk);
811 int pblk_recov_pad(struct pblk *pblk);
812 __le64 *pblk_recov_get_lba_list(struct pblk *pblk, struct line_emeta *emeta);
813 int pblk_recov_setup_rq(struct pblk *pblk, struct pblk_c_ctx *c_ctx,
814 struct pblk_rec_ctx *recovery, u64 *comp_bits,
820 #define PBLK_GC_MAX_READERS 8 /* Max number of outstanding GC reader jobs */
821 #define PBLK_GC_RQ_QD 128 /* Queue depth for inflight GC requests */
822 #define PBLK_GC_L_QD 4 /* Queue depth for inflight GC lines */
823 #define PBLK_GC_RSV_LINE 1 /* Reserved lines for GC */
825 int pblk_gc_init(struct pblk *pblk);
826 void pblk_gc_exit(struct pblk *pblk);
827 void pblk_gc_should_start(struct pblk *pblk);
828 void pblk_gc_should_stop(struct pblk *pblk);
829 void pblk_gc_kick(struct pblk *pblk);
830 void pblk_gc_sysfs_state_show(struct pblk *pblk, int *gc_enabled,
832 int pblk_gc_sysfs_force(struct pblk *pblk, int force);
837 void pblk_rl_init(struct pblk_rl *rl, int budget);
838 void pblk_rl_free(struct pblk_rl *rl);
839 int pblk_rl_high_thrs(struct pblk_rl *rl);
840 int pblk_rl_low_thrs(struct pblk_rl *rl);
841 unsigned long pblk_rl_nr_free_blks(struct pblk_rl *rl);
842 int pblk_rl_user_may_insert(struct pblk_rl *rl, int nr_entries);
843 void pblk_rl_inserted(struct pblk_rl *rl, int nr_entries);
844 void pblk_rl_user_in(struct pblk_rl *rl, int nr_entries);
845 int pblk_rl_gc_may_insert(struct pblk_rl *rl, int nr_entries);
846 void pblk_rl_gc_in(struct pblk_rl *rl, int nr_entries);
847 void pblk_rl_out(struct pblk_rl *rl, int nr_user, int nr_gc);
848 int pblk_rl_sysfs_rate_show(struct pblk_rl *rl);
849 int pblk_rl_max_io(struct pblk_rl *rl);
850 void pblk_rl_free_lines_inc(struct pblk_rl *rl, struct pblk_line *line);
851 void pblk_rl_free_lines_dec(struct pblk_rl *rl, struct pblk_line *line);
852 void pblk_rl_set_space_limit(struct pblk_rl *rl, int entries_left);
853 int pblk_rl_is_limit(struct pblk_rl *rl);
858 int pblk_sysfs_init(struct gendisk *tdisk);
859 void pblk_sysfs_exit(struct gendisk *tdisk);
861 static inline void *pblk_malloc(size_t size, int type, gfp_t flags)
863 if (type == PBLK_KMALLOC_META)
864 return kmalloc(size, flags);
865 return vmalloc(size);
868 static inline void pblk_mfree(void *ptr, int type)
870 if (type == PBLK_KMALLOC_META)
876 static inline struct nvm_rq *nvm_rq_from_c_ctx(void *c_ctx)
878 return c_ctx - sizeof(struct nvm_rq);
881 static inline void *emeta_to_bb(struct line_emeta *emeta)
883 return emeta->bb_bitmap;
886 static inline void *emeta_to_lbas(struct pblk *pblk, struct line_emeta *emeta)
888 return ((void *)emeta + pblk->lm.emeta_len[1]);
891 static inline void *emeta_to_vsc(struct pblk *pblk, struct line_emeta *emeta)
893 return (emeta_to_lbas(pblk, emeta) + pblk->lm.emeta_len[2]);
896 static inline int pblk_line_vsc(struct pblk_line *line)
898 return le32_to_cpu(*line->vsc);
901 #define NVM_MEM_PAGE_WRITE (8)
903 static inline int pblk_pad_distance(struct pblk *pblk)
905 struct nvm_tgt_dev *dev = pblk->dev;
906 struct nvm_geo *geo = &dev->geo;
908 return NVM_MEM_PAGE_WRITE * geo->nr_luns * geo->sec_per_pl;
911 static inline int pblk_dev_ppa_to_line(struct ppa_addr p)
916 static inline int pblk_tgt_ppa_to_line(struct ppa_addr p)
921 static inline int pblk_ppa_to_pos(struct nvm_geo *geo, struct ppa_addr p)
923 return p.g.lun * geo->nr_chnls + p.g.ch;
926 /* A block within a line corresponds to the lun */
927 static inline int pblk_dev_ppa_to_pos(struct nvm_geo *geo, struct ppa_addr p)
929 return p.g.lun * geo->nr_chnls + p.g.ch;
932 static inline struct ppa_addr pblk_ppa32_to_ppa64(struct pblk *pblk, u32 ppa32)
934 struct ppa_addr ppa64;
939 ppa64.ppa = ADDR_EMPTY;
940 } else if (ppa32 & (1U << 31)) {
941 ppa64.c.line = ppa32 & ((~0U) >> 1);
942 ppa64.c.is_cached = 1;
944 ppa64.g.blk = (ppa32 & pblk->ppaf.blk_mask) >>
945 pblk->ppaf.blk_offset;
946 ppa64.g.pg = (ppa32 & pblk->ppaf.pg_mask) >>
947 pblk->ppaf.pg_offset;
948 ppa64.g.lun = (ppa32 & pblk->ppaf.lun_mask) >>
949 pblk->ppaf.lun_offset;
950 ppa64.g.ch = (ppa32 & pblk->ppaf.ch_mask) >>
951 pblk->ppaf.ch_offset;
952 ppa64.g.pl = (ppa32 & pblk->ppaf.pln_mask) >>
953 pblk->ppaf.pln_offset;
954 ppa64.g.sec = (ppa32 & pblk->ppaf.sec_mask) >>
955 pblk->ppaf.sec_offset;
961 static inline struct ppa_addr pblk_trans_map_get(struct pblk *pblk,
966 if (pblk->ppaf_bitsize < 32) {
967 u32 *map = (u32 *)pblk->trans_map;
969 ppa = pblk_ppa32_to_ppa64(pblk, map[lba]);
971 struct ppa_addr *map = (struct ppa_addr *)pblk->trans_map;
979 static inline u32 pblk_ppa64_to_ppa32(struct pblk *pblk, struct ppa_addr ppa64)
983 if (ppa64.ppa == ADDR_EMPTY) {
985 } else if (ppa64.c.is_cached) {
986 ppa32 |= ppa64.c.line;
989 ppa32 |= ppa64.g.blk << pblk->ppaf.blk_offset;
990 ppa32 |= ppa64.g.pg << pblk->ppaf.pg_offset;
991 ppa32 |= ppa64.g.lun << pblk->ppaf.lun_offset;
992 ppa32 |= ppa64.g.ch << pblk->ppaf.ch_offset;
993 ppa32 |= ppa64.g.pl << pblk->ppaf.pln_offset;
994 ppa32 |= ppa64.g.sec << pblk->ppaf.sec_offset;
1000 static inline void pblk_trans_map_set(struct pblk *pblk, sector_t lba,
1001 struct ppa_addr ppa)
1003 if (pblk->ppaf_bitsize < 32) {
1004 u32 *map = (u32 *)pblk->trans_map;
1006 map[lba] = pblk_ppa64_to_ppa32(pblk, ppa);
1008 u64 *map = (u64 *)pblk->trans_map;
1014 static inline u64 pblk_dev_ppa_to_line_addr(struct pblk *pblk,
1020 paddr |= (u64)p.g.pg << pblk->ppaf.pg_offset;
1021 paddr |= (u64)p.g.lun << pblk->ppaf.lun_offset;
1022 paddr |= (u64)p.g.ch << pblk->ppaf.ch_offset;
1023 paddr |= (u64)p.g.pl << pblk->ppaf.pln_offset;
1024 paddr |= (u64)p.g.sec << pblk->ppaf.sec_offset;
1029 static inline int pblk_ppa_empty(struct ppa_addr ppa_addr)
1031 return (ppa_addr.ppa == ADDR_EMPTY);
1034 static inline void pblk_ppa_set_empty(struct ppa_addr *ppa_addr)
1036 ppa_addr->ppa = ADDR_EMPTY;
1039 static inline bool pblk_ppa_comp(struct ppa_addr lppa, struct ppa_addr rppa)
1041 if (lppa.ppa == rppa.ppa)
1047 static inline int pblk_addr_in_cache(struct ppa_addr ppa)
1049 return (ppa.ppa != ADDR_EMPTY && ppa.c.is_cached);
1052 static inline int pblk_addr_to_cacheline(struct ppa_addr ppa)
1057 static inline struct ppa_addr pblk_cacheline_to_addr(int addr)
1067 static inline struct ppa_addr addr_to_gen_ppa(struct pblk *pblk, u64 paddr,
1070 struct ppa_addr ppa;
1073 ppa.g.blk = line_id;
1074 ppa.g.pg = (paddr & pblk->ppaf.pg_mask) >> pblk->ppaf.pg_offset;
1075 ppa.g.lun = (paddr & pblk->ppaf.lun_mask) >> pblk->ppaf.lun_offset;
1076 ppa.g.ch = (paddr & pblk->ppaf.ch_mask) >> pblk->ppaf.ch_offset;
1077 ppa.g.pl = (paddr & pblk->ppaf.pln_mask) >> pblk->ppaf.pln_offset;
1078 ppa.g.sec = (paddr & pblk->ppaf.sec_mask) >> pblk->ppaf.sec_offset;
1083 static inline struct ppa_addr addr_to_pblk_ppa(struct pblk *pblk, u64 paddr,
1086 struct ppa_addr ppa;
1088 ppa = addr_to_gen_ppa(pblk, paddr, line_id);
1093 static inline u32 pblk_calc_meta_header_crc(struct pblk *pblk,
1094 struct line_header *header)
1098 crc = crc32_le(crc, (unsigned char *)header + sizeof(crc),
1099 sizeof(struct line_header) - sizeof(crc));
1104 static inline u32 pblk_calc_smeta_crc(struct pblk *pblk,
1105 struct line_smeta *smeta)
1107 struct pblk_line_meta *lm = &pblk->lm;
1110 crc = crc32_le(crc, (unsigned char *)smeta +
1111 sizeof(struct line_header) + sizeof(crc),
1113 sizeof(struct line_header) - sizeof(crc));
1118 static inline u32 pblk_calc_emeta_crc(struct pblk *pblk,
1119 struct line_emeta *emeta)
1121 struct pblk_line_meta *lm = &pblk->lm;
1124 crc = crc32_le(crc, (unsigned char *)emeta +
1125 sizeof(struct line_header) + sizeof(crc),
1127 sizeof(struct line_header) - sizeof(crc));
1132 static inline int pblk_set_progr_mode(struct pblk *pblk, int type)
1134 struct nvm_tgt_dev *dev = pblk->dev;
1135 struct nvm_geo *geo = &dev->geo;
1138 flags = geo->plane_mode >> 1;
1140 if (type == PBLK_WRITE)
1141 flags |= NVM_IO_SCRAMBLE_ENABLE;
1147 PBLK_READ_RANDOM = 0,
1148 PBLK_READ_SEQUENTIAL = 1,
1151 static inline int pblk_set_read_mode(struct pblk *pblk, int type)
1153 struct nvm_tgt_dev *dev = pblk->dev;
1154 struct nvm_geo *geo = &dev->geo;
1157 flags = NVM_IO_SUSPEND | NVM_IO_SCRAMBLE_ENABLE;
1158 if (type == PBLK_READ_SEQUENTIAL)
1159 flags |= geo->plane_mode >> 1;
1164 static inline int pblk_io_aligned(struct pblk *pblk, int nr_secs)
1166 return !(nr_secs % pblk->min_write_pgs);
1169 #ifdef CONFIG_NVM_DEBUG
1170 static inline void print_ppa(struct ppa_addr *p, char *msg, int error)
1172 if (p->c.is_cached) {
1173 pr_err("ppa: (%s: %x) cache line: %llu\n",
1174 msg, error, (u64)p->c.line);
1176 pr_err("ppa: (%s: %x):ch:%d,lun:%d,blk:%d,pg:%d,pl:%d,sec:%d\n",
1178 p->g.ch, p->g.lun, p->g.blk,
1179 p->g.pg, p->g.pl, p->g.sec);
1183 static inline void pblk_print_failed_rqd(struct pblk *pblk, struct nvm_rq *rqd,
1188 if (rqd->nr_ppas == 1) {
1189 print_ppa(&rqd->ppa_addr, "rqd", error);
1193 while ((bit = find_next_bit((void *)&rqd->ppa_status, rqd->nr_ppas,
1194 bit + 1)) < rqd->nr_ppas) {
1195 print_ppa(&rqd->ppa_list[bit], "rqd", error);
1198 pr_err("error:%d, ppa_status:%llx\n", error, rqd->ppa_status);
1202 static inline int pblk_boundary_ppa_checks(struct nvm_tgt_dev *tgt_dev,
1203 struct ppa_addr *ppas, int nr_ppas)
1205 struct nvm_geo *geo = &tgt_dev->geo;
1206 struct ppa_addr *ppa;
1209 for (i = 0; i < nr_ppas; i++) {
1212 if (!ppa->c.is_cached &&
1213 ppa->g.ch < geo->nr_chnls &&
1214 ppa->g.lun < geo->luns_per_chnl &&
1215 ppa->g.pl < geo->nr_planes &&
1216 ppa->g.blk < geo->blks_per_lun &&
1217 ppa->g.pg < geo->pgs_per_blk &&
1218 ppa->g.sec < geo->sec_per_pg)
1221 #ifdef CONFIG_NVM_DEBUG
1222 print_ppa(ppa, "boundary", i);
1229 static inline int pblk_boundary_paddr_checks(struct pblk *pblk, u64 paddr)
1231 struct pblk_line_meta *lm = &pblk->lm;
1233 if (paddr > lm->sec_per_line)
1239 static inline unsigned int pblk_get_bi_idx(struct bio *bio)
1241 return bio->bi_iter.bi_idx;
1244 static inline sector_t pblk_get_lba(struct bio *bio)
1246 return bio->bi_iter.bi_sector / NR_PHY_IN_LOG;
1249 static inline unsigned int pblk_get_secs(struct bio *bio)
1251 return bio->bi_iter.bi_size / PBLK_EXPOSED_PAGE_SIZE;
1254 static inline sector_t pblk_get_sector(sector_t lba)
1256 return lba * NR_PHY_IN_LOG;
1259 static inline void pblk_setup_uuid(struct pblk *pblk)
1264 memcpy(pblk->instance_uuid, uuid.b, 16);
1266 #endif /* PBLK_H_ */