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_WS_POOL_SIZE (128)
44 #define PBLK_META_POOL_SIZE (128)
45 #define PBLK_READ_REQ_POOL_SIZE (1024)
47 #define PBLK_NR_CLOSE_JOBS (4)
49 #define PBLK_CACHE_NAME_LEN (DISK_NAME_LEN + 16)
51 #define PBLK_COMMAND_TIMEOUT_MS 30000
53 /* Max 512 LUNs per device */
54 #define PBLK_MAX_LUNS_BITMAP (4)
56 #define NR_PHY_IN_LOG (PBLK_EXPOSED_PAGE_SIZE / PBLK_SECTOR)
58 #define pblk_for_each_lun(pblk, rlun, i) \
59 for ((i) = 0, rlun = &(pblk)->luns[0]; \
60 (i) < (pblk)->nr_luns; (i)++, rlun = &(pblk)->luns[(i)])
62 #define ERASE 2 /* READ = 0, WRITE = 1 */
66 PBLK_IOTYPE_USER = 1 << 0,
67 PBLK_IOTYPE_GC = 1 << 1,
69 /* Write buffer flags */
70 PBLK_FLUSH_ENTRY = 1 << 2,
71 PBLK_WRITTEN_DATA = 1 << 3,
72 PBLK_SUBMITTED_ENTRY = 1 << 4,
73 PBLK_WRITABLE_ENTRY = 1 << 5,
77 PBLK_BLK_ST_OPEN = 0x1,
78 PBLK_BLK_ST_CLOSED = 0x2,
81 struct pblk_sec_meta {
86 /* The number of GC lists and the rate-limiter states go together. This way the
87 * rate-limiter can dictate how much GC is needed based on resource utilization.
89 #define PBLK_GC_NR_LISTS 3
97 #define pblk_dma_meta_size (sizeof(struct pblk_sec_meta) * PBLK_MAX_REQ_ADDRS)
99 /* write buffer completion context */
101 struct list_head list; /* Head for out-of-order completion */
103 unsigned long *lun_bitmap; /* Luns used on current request */
105 unsigned int nr_valid;
106 unsigned int nr_padded;
109 /* generic context */
117 struct completion wait;
121 /* Recovery context */
122 struct pblk_rec_ctx {
125 struct list_head failed;
126 struct work_struct ws_rec;
131 struct bio_list bios; /* Original bios - used for completion
132 * in REQ_FUA, REQ_FLUSH case
134 u64 lba; /* Logic addr. associated with entry */
135 struct ppa_addr ppa; /* Physic addr. associated with entry */
136 int flags; /* Write context flags */
139 struct pblk_rb_entry {
140 struct ppa_addr cacheline; /* Cacheline for this entry */
141 void *data; /* Pointer to data on this entry */
142 struct pblk_w_ctx w_ctx; /* Context for this entry */
143 struct list_head index; /* List head to enable indexes */
146 #define EMPTY_ENTRY (~0U)
148 struct pblk_rb_pages {
151 struct list_head list;
155 struct pblk_rb_entry *entries; /* Ring buffer entries */
156 unsigned int mem; /* Write offset - points to next
157 * writable entry in memory
159 unsigned int subm; /* Read offset - points to last entry
160 * that has been submitted to the media
163 unsigned int sync; /* Synced - backpointer that signals
164 * the last submitted entry that has
165 * been successfully persisted to media
167 unsigned int sync_point; /* Sync point - last entry that must be
168 * flushed to the media. Used with
169 * REQ_FLUSH and REQ_FUA
171 unsigned int l2p_update; /* l2p update point - next entry for
172 * which l2p mapping will be updated to
173 * contain a device ppa address (instead
176 unsigned int nr_entries; /* Number of entries in write buffer -
177 * must be a power of two
179 unsigned int seg_size; /* Size of the data segments being
180 * stored on each entry. Typically this
184 struct list_head pages; /* List of data pages */
186 spinlock_t w_lock; /* Write lock */
187 spinlock_t s_lock; /* Sync lock */
189 #ifdef CONFIG_NVM_DEBUG
190 atomic_t inflight_sync_point; /* Not served REQ_FLUSH | REQ_FUA */
194 #define PBLK_RECOVERY_SECTORS 16
197 struct ppa_addr bppa;
199 u8 *bb_list; /* Bad block list for LUN. Only used on
200 * bring up. Bad blocks are managed
201 * within lines on run-time.
204 struct semaphore wr_sem;
208 struct pblk_line *line;
210 u64 lba_list[PBLK_MAX_REQ_ADDRS];
213 struct list_head list;
217 /* These states are not protected by a lock since (i) they are in the
218 * fast path, and (ii) they are not critical.
224 struct task_struct *gc_ts;
225 struct task_struct *gc_writer_ts;
226 struct task_struct *gc_reader_ts;
228 struct workqueue_struct *gc_line_reader_wq;
229 struct workqueue_struct *gc_reader_wq;
231 struct timer_list gc_timer;
233 struct semaphore gc_sem;
234 atomic_t inflight_gc;
237 struct list_head w_list;
238 struct list_head r_list;
246 unsigned int high; /* Upper threshold for rate limiter (free run -
247 * user I/O rate limiter
249 unsigned int low; /* Lower threshold for rate limiter (user I/O
250 * rate limiter - stall)
252 unsigned int high_pw; /* High rounded up as a power of 2 */
254 #define PBLK_USER_HIGH_THRS 8 /* Begin write limit at 12% available blks */
255 #define PBLK_USER_LOW_THRS 10 /* Aggressive GC at 10% available blocks */
257 int rb_windows_pw; /* Number of rate windows in the write buffer
258 * given as a power-of-2. This guarantees that
259 * when user I/O is being rate limited, there
260 * will be reserved enough space for the GC to
261 * place its payload. A window is of
262 * pblk->max_write_pgs size, which in NVMe is
265 int rb_budget; /* Total number of entries available for I/O */
266 int rb_user_max; /* Max buffer entries available for user I/O */
267 int rb_gc_max; /* Max buffer entries available for GC I/O */
268 int rb_gc_rsv; /* Reserved buffer entries for GC I/O */
269 int rb_state; /* Rate-limiter current state */
270 int rb_max_io; /* Maximum size for an I/O giving the config */
272 atomic_t rb_user_cnt; /* User I/O buffer counter */
273 atomic_t rb_gc_cnt; /* GC I/O buffer counter */
274 atomic_t rb_space; /* Space limit in case of reaching capacity */
276 int rsv_blocks; /* Reserved blocks for GC */
281 struct timer_list u_timer;
283 unsigned long long nr_secs;
284 unsigned long total_blocks;
285 atomic_t free_blocks;
288 #define PBLK_LINE_EMPTY (~0U)
292 PBLK_LINETYPE_FREE = 0,
293 PBLK_LINETYPE_LOG = 1,
294 PBLK_LINETYPE_DATA = 2,
297 PBLK_LINESTATE_FREE = 10,
298 PBLK_LINESTATE_OPEN = 11,
299 PBLK_LINESTATE_CLOSED = 12,
300 PBLK_LINESTATE_GC = 13,
301 PBLK_LINESTATE_BAD = 14,
302 PBLK_LINESTATE_CORRUPT = 15,
305 PBLK_LINEGC_NONE = 20,
306 PBLK_LINEGC_EMPTY = 21,
307 PBLK_LINEGC_LOW = 22,
308 PBLK_LINEGC_MID = 23,
309 PBLK_LINEGC_HIGH = 24,
310 PBLK_LINEGC_FULL = 25,
313 #define PBLK_MAGIC 0x70626c6b /*pblk*/
314 #define SMETA_VERSION cpu_to_le16(1)
318 __le32 identifier; /* pblk identifier */
319 __u8 uuid[16]; /* instance uuid */
320 __le16 type; /* line type */
321 __le16 version; /* type version */
322 __le32 id; /* line id for current line */
326 struct line_header header;
328 __le32 crc; /* Full structure including struct crc */
329 /* Previous line metadata */
330 __le32 prev_id; /* Line id for previous line */
332 /* Current line metadata */
333 __le64 seq_nr; /* Sequence number for current line */
336 __le32 window_wr_lun; /* Number of parallel LUNs to write */
344 * Metadata layout in media:
346 * 1. struct line_emeta
347 * 2. bad block bitmap (u64 * window_wr_lun)
348 * Mid sectors (start at lbas_sector):
349 * 3. nr_lbas (u64) forming lba list
350 * Last sectors (start at vsc_sector):
351 * 4. u32 valid sector count (vsc) for all lines (~0U: free line)
354 struct line_header header;
356 __le32 crc; /* Full structure including struct crc */
358 /* Previous line metadata */
359 __le32 prev_id; /* Line id for prev line */
361 /* Current line metadata */
362 __le64 seq_nr; /* Sequence number for current line */
365 __le32 window_wr_lun; /* Number of parallel LUNs to write */
367 /* Bookkeeping for recovery */
368 __le32 next_id; /* Line id for next line */
369 __le64 nr_lbas; /* Number of lbas mapped in line */
370 __le64 nr_valid_lbas; /* Number of valid lbas mapped in line */
371 __le64 bb_bitmap[]; /* Updated bad block bitmap for line */
375 struct line_emeta *buf; /* emeta buffer in media format */
376 int mem; /* Write offset - points to next
377 * writable entry in memory
379 atomic_t sync; /* Synced - backpointer that signals the
380 * last entry that has been successfully
383 unsigned int nr_entries; /* Number of emeta entries */
387 struct line_smeta *buf; /* smeta buffer in persistent format */
392 unsigned int id; /* Line number corresponds to the
395 unsigned int seq_nr; /* Unique line sequence number */
397 int state; /* PBLK_LINESTATE_X */
398 int type; /* PBLK_LINETYPE_X */
399 int gc_group; /* PBLK_LINEGC_X */
400 struct list_head list; /* Free, GC lists */
402 unsigned long *lun_bitmap; /* Bitmap for LUNs mapped in line */
404 struct pblk_smeta *smeta; /* Start metadata */
405 struct pblk_emeta *emeta; /* End medatada */
407 int meta_line; /* Metadata line id */
408 int meta_distance; /* Distance between data and metadata */
410 u64 smeta_ssec; /* Sector where smeta starts */
411 u64 emeta_ssec; /* Sector where emeta starts */
413 unsigned int sec_in_line; /* Number of usable secs in line */
415 atomic_t blk_in_line; /* Number of good blocks in line */
416 unsigned long *blk_bitmap; /* Bitmap for valid/invalid blocks */
417 unsigned long *erase_bitmap; /* Bitmap for erased blocks */
419 unsigned long *map_bitmap; /* Bitmap for mapped sectors in line */
420 unsigned long *invalid_bitmap; /* Bitmap for invalid sectors in line */
422 atomic_t left_eblks; /* Blocks left for erasing */
423 atomic_t left_seblks; /* Blocks left for sync erasing */
425 int left_msecs; /* Sectors left for mapping */
426 unsigned int cur_sec; /* Sector map pointer */
427 unsigned int nr_valid_lbas; /* Number of valid lbas in line */
429 __le32 *vsc; /* Valid sector count in line */
431 struct kref ref; /* Write buffer L2P references */
433 spinlock_t lock; /* Necessary for invalid_bitmap only */
436 #define PBLK_DATA_LINES 4
439 PBLK_KMALLOC_META = 1,
440 PBLK_VMALLOC_META = 2,
444 PBLK_EMETA_TYPE_HEADER = 1, /* struct line_emeta first sector */
445 PBLK_EMETA_TYPE_LLBA = 2, /* lba list - type: __le64 */
446 PBLK_EMETA_TYPE_VSC = 3, /* vsc list - type: __le32 */
449 struct pblk_line_mgmt {
450 int nr_lines; /* Total number of full lines */
451 int nr_free_lines; /* Number of full lines in free list */
453 /* Free lists - use free_lock */
454 struct list_head free_list; /* Full lines ready to use */
455 struct list_head corrupt_list; /* Full lines corrupted */
456 struct list_head bad_list; /* Full lines bad */
458 /* GC lists - use gc_lock */
459 struct list_head *gc_lists[PBLK_GC_NR_LISTS];
460 struct list_head gc_high_list; /* Full lines ready to GC, high isc */
461 struct list_head gc_mid_list; /* Full lines ready to GC, mid isc */
462 struct list_head gc_low_list; /* Full lines ready to GC, low isc */
464 struct list_head gc_full_list; /* Full lines ready to GC, no valid */
465 struct list_head gc_empty_list; /* Full lines close, all valid */
467 struct pblk_line *log_line; /* Current FTL log line */
468 struct pblk_line *data_line; /* Current data line */
469 struct pblk_line *log_next; /* Next FTL log line */
470 struct pblk_line *data_next; /* Next data line */
472 struct list_head emeta_list; /* Lines queued to schedule emeta */
474 __le32 *vsc_list; /* Valid sector counts for all lines */
476 /* Metadata allocation type: VMALLOC | KMALLOC */
477 int emeta_alloc_type;
479 /* Pre-allocated metadata for data lines */
480 struct pblk_smeta *sline_meta[PBLK_DATA_LINES];
481 struct pblk_emeta *eline_meta[PBLK_DATA_LINES];
482 unsigned long meta_bitmap;
484 /* Helpers for fast bitmap calculations */
485 unsigned long *bb_template;
486 unsigned long *bb_aux;
488 unsigned long d_seq_nr; /* Data line unique sequence number */
489 unsigned long l_seq_nr; /* Log line unique sequence number */
491 spinlock_t free_lock;
492 spinlock_t close_lock;
496 struct pblk_line_meta {
497 unsigned int smeta_len; /* Total length for smeta */
498 unsigned int smeta_sec; /* Sectors needed for smeta */
500 unsigned int emeta_len[4]; /* Lengths for emeta:
502 * [1]: struct line_emeta length
503 * [2]: L2P portion length
504 * [3]: vsc list length
506 unsigned int emeta_sec[4]; /* Sectors needed for emeta. Same layout
510 unsigned int emeta_bb; /* Boundary for bb that affects emeta */
512 unsigned int vsc_list_len; /* Length for vsc list */
513 unsigned int sec_bitmap_len; /* Length for sector bitmap in line */
514 unsigned int blk_bitmap_len; /* Length for block bitmap in line */
515 unsigned int lun_bitmap_len; /* Length for lun bitmap in line */
517 unsigned int blk_per_line; /* Number of blocks in a full line */
518 unsigned int sec_per_line; /* Number of sectors in a line */
519 unsigned int dsec_per_line; /* Number of data sectors in a line */
520 unsigned int min_blk_line; /* Min. number of good blocks in line */
522 unsigned int mid_thrs; /* Threshold for GC mid list */
523 unsigned int high_thrs; /* Threshold for GC high list */
525 unsigned int meta_distance; /* Distance between data and metadata */
528 struct pblk_addr_format {
544 PBLK_STATE_RUNNING = 0,
545 PBLK_STATE_STOPPING = 1,
546 PBLK_STATE_RECOVERING = 2,
547 PBLK_STATE_STOPPED = 3,
551 struct nvm_tgt_dev *dev;
552 struct gendisk *disk;
556 struct pblk_lun *luns;
558 struct pblk_line *lines; /* Line array */
559 struct pblk_line_mgmt l_mg; /* Line management */
560 struct pblk_line_meta lm; /* Line metadata */
563 struct pblk_addr_format ppaf;
567 int state; /* pblk line state */
569 int min_write_pgs; /* Minimum amount of pages required by controller */
570 int max_write_pgs; /* Maximum amount of pages supported by controller */
571 int pgs_in_buffer; /* Number of pages that need to be held in buffer to
572 * guarantee successful reads.
575 sector_t capacity; /* Device capacity when bad blocks are subtracted */
576 int over_pct; /* Percentage of device used for over-provisioning */
578 /* pblk provisioning values. Used by rate limiter */
583 unsigned char instance_uuid[16];
584 #ifdef CONFIG_NVM_DEBUG
585 /* All debug counters apply to 4kb sector I/Os */
586 atomic_long_t inflight_writes; /* Inflight writes (user and gc) */
587 atomic_long_t padded_writes; /* Sectors padded due to flush/fua */
588 atomic_long_t padded_wb; /* Sectors padded in write buffer */
589 atomic_long_t nr_flush; /* Number of flush/fua I/O */
590 atomic_long_t req_writes; /* Sectors stored on write buffer */
591 atomic_long_t sub_writes; /* Sectors submitted from buffer */
592 atomic_long_t sync_writes; /* Sectors synced to media */
593 atomic_long_t inflight_reads; /* Inflight sector read requests */
594 atomic_long_t cache_reads; /* Read requests that hit the cache */
595 atomic_long_t sync_reads; /* Completed sector read requests */
596 atomic_long_t recov_writes; /* Sectors submitted from recovery */
597 atomic_long_t recov_gc_writes; /* Sectors submitted from write GC */
598 atomic_long_t recov_gc_reads; /* Sectors submitted from read GC */
603 atomic_long_t read_failed;
604 atomic_long_t read_empty;
605 atomic_long_t read_high_ecc;
606 atomic_long_t read_failed_gc;
607 atomic_long_t write_failed;
608 atomic_long_t erase_failed;
610 atomic_t inflight_io; /* General inflight I/O counter */
612 struct task_struct *writer_ts;
614 /* Simple translation map of logical addresses to physical addresses.
615 * The logical addresses is known by the host system, while the physical
616 * addresses are used when writing to the disk block device.
618 unsigned char *trans_map;
619 spinlock_t trans_lock;
621 struct list_head compl_list;
623 mempool_t *page_pool;
624 mempool_t *line_ws_pool;
626 mempool_t *g_rq_pool;
627 mempool_t *w_rq_pool;
628 mempool_t *line_meta_pool;
630 struct workqueue_struct *close_wq;
631 struct workqueue_struct *bb_wq;
633 struct timer_list wtimer;
638 struct pblk_line_ws {
640 struct pblk_line *line;
642 struct work_struct ws;
645 #define pblk_g_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_g_ctx))
646 #define pblk_w_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_c_ctx))
649 * pblk ring buffer operations
651 int pblk_rb_init(struct pblk_rb *rb, struct pblk_rb_entry *rb_entry_base,
652 unsigned int power_size, unsigned int power_seg_sz);
653 unsigned int pblk_rb_calculate_size(unsigned int nr_entries);
654 void *pblk_rb_entries_ref(struct pblk_rb *rb);
655 int pblk_rb_may_write_user(struct pblk_rb *rb, struct bio *bio,
656 unsigned int nr_entries, unsigned int *pos);
657 int pblk_rb_may_write_gc(struct pblk_rb *rb, unsigned int nr_entries,
659 void pblk_rb_write_entry_user(struct pblk_rb *rb, void *data,
660 struct pblk_w_ctx w_ctx, unsigned int pos);
661 void pblk_rb_write_entry_gc(struct pblk_rb *rb, void *data,
662 struct pblk_w_ctx w_ctx, struct pblk_line *gc_line,
664 struct pblk_w_ctx *pblk_rb_w_ctx(struct pblk_rb *rb, unsigned int pos);
665 void pblk_rb_flush(struct pblk_rb *rb);
667 void pblk_rb_sync_l2p(struct pblk_rb *rb);
668 unsigned int pblk_rb_read_to_bio(struct pblk_rb *rb, struct nvm_rq *rqd,
669 struct bio *bio, unsigned int pos,
670 unsigned int nr_entries, unsigned int count);
671 unsigned int pblk_rb_read_to_bio_list(struct pblk_rb *rb, struct bio *bio,
672 struct list_head *list,
674 int pblk_rb_copy_to_bio(struct pblk_rb *rb, struct bio *bio, sector_t lba,
675 struct ppa_addr ppa, int bio_iter, bool advanced_bio);
676 unsigned int pblk_rb_read_commit(struct pblk_rb *rb, unsigned int entries);
678 unsigned int pblk_rb_sync_init(struct pblk_rb *rb, unsigned long *flags);
679 unsigned int pblk_rb_sync_advance(struct pblk_rb *rb, unsigned int nr_entries);
680 struct pblk_rb_entry *pblk_rb_sync_scan_entry(struct pblk_rb *rb,
681 struct ppa_addr *ppa);
682 void pblk_rb_sync_end(struct pblk_rb *rb, unsigned long *flags);
683 unsigned int pblk_rb_sync_point_count(struct pblk_rb *rb);
685 unsigned int pblk_rb_read_count(struct pblk_rb *rb);
686 unsigned int pblk_rb_sync_count(struct pblk_rb *rb);
687 unsigned int pblk_rb_wrap_pos(struct pblk_rb *rb, unsigned int pos);
689 int pblk_rb_tear_down_check(struct pblk_rb *rb);
690 int pblk_rb_pos_oob(struct pblk_rb *rb, u64 pos);
691 void pblk_rb_data_free(struct pblk_rb *rb);
692 ssize_t pblk_rb_sysfs(struct pblk_rb *rb, char *buf);
697 struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int rw);
698 void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write);
699 int pblk_setup_w_rec_rq(struct pblk *pblk, struct nvm_rq *rqd,
700 struct pblk_c_ctx *c_ctx);
701 void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int rw);
702 void pblk_wait_for_meta(struct pblk *pblk);
703 struct ppa_addr pblk_get_lba_map(struct pblk *pblk, sector_t lba);
704 void pblk_discard(struct pblk *pblk, struct bio *bio);
705 void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd);
706 void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd);
707 int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd);
708 int pblk_submit_meta_io(struct pblk *pblk, struct pblk_line *meta_line);
709 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
710 unsigned int nr_secs, unsigned int len,
711 int alloc_type, gfp_t gfp_mask);
712 struct pblk_line *pblk_line_get(struct pblk *pblk);
713 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk);
714 void pblk_line_replace_data(struct pblk *pblk);
715 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line);
716 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line);
717 struct pblk_line *pblk_line_get_data(struct pblk *pblk);
718 struct pblk_line *pblk_line_get_erase(struct pblk *pblk);
719 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line);
720 int pblk_line_is_full(struct pblk_line *line);
721 void pblk_line_free(struct pblk *pblk, struct pblk_line *line);
722 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line);
723 void pblk_line_close(struct pblk *pblk, struct pblk_line *line);
724 void pblk_line_close_meta_sync(struct pblk *pblk);
725 void pblk_line_close_ws(struct work_struct *work);
726 void pblk_pipeline_stop(struct pblk *pblk);
727 void pblk_line_mark_bb(struct work_struct *work);
728 void pblk_line_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
729 void (*work)(struct work_struct *),
730 struct workqueue_struct *wq);
731 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line);
732 int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line);
733 int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line,
735 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr erase_ppa);
736 void pblk_line_put(struct kref *ref);
737 struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line);
738 u64 pblk_lookup_page(struct pblk *pblk, struct pblk_line *line);
739 void pblk_dealloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
740 u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
741 u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
742 int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
743 unsigned long secs_to_flush);
744 void pblk_up_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas);
745 void pblk_down_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
746 unsigned long *lun_bitmap);
747 void pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas);
748 void pblk_up_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
749 unsigned long *lun_bitmap);
750 void pblk_end_bio_sync(struct bio *bio);
751 void pblk_end_io_sync(struct nvm_rq *rqd);
752 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
754 void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
756 void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa);
757 void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
759 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa);
760 void pblk_update_map_cache(struct pblk *pblk, sector_t lba,
761 struct ppa_addr ppa);
762 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
763 struct ppa_addr ppa, struct ppa_addr entry_line);
764 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa,
765 struct pblk_line *gc_line);
766 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
767 u64 *lba_list, int nr_secs);
768 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
769 sector_t blba, int nr_secs);
772 * pblk user I/O write path
774 int pblk_write_to_cache(struct pblk *pblk, struct bio *bio,
775 unsigned long flags);
776 int pblk_write_gc_to_cache(struct pblk *pblk, void *data, u64 *lba_list,
777 unsigned int nr_entries, unsigned int nr_rec_entries,
778 struct pblk_line *gc_line, unsigned long flags);
783 void pblk_map_erase_rq(struct pblk *pblk, struct nvm_rq *rqd,
784 unsigned int sentry, unsigned long *lun_bitmap,
785 unsigned int valid_secs, struct ppa_addr *erase_ppa);
786 void pblk_map_rq(struct pblk *pblk, struct nvm_rq *rqd, unsigned int sentry,
787 unsigned long *lun_bitmap, unsigned int valid_secs,
793 int pblk_write_ts(void *data);
794 void pblk_write_timer_fn(unsigned long data);
795 void pblk_write_should_kick(struct pblk *pblk);
800 extern struct bio_set *pblk_bio_set;
801 int pblk_submit_read(struct pblk *pblk, struct bio *bio);
802 int pblk_submit_read_gc(struct pblk *pblk, u64 *lba_list, void *data,
803 unsigned int nr_secs, unsigned int *secs_to_gc,
804 struct pblk_line *line);
808 void pblk_submit_rec(struct work_struct *work);
809 struct pblk_line *pblk_recov_l2p(struct pblk *pblk);
810 int pblk_recov_pad(struct pblk *pblk);
811 __le64 *pblk_recov_get_lba_list(struct pblk *pblk, struct line_emeta *emeta);
812 int pblk_recov_setup_rq(struct pblk *pblk, struct pblk_c_ctx *c_ctx,
813 struct pblk_rec_ctx *recovery, u64 *comp_bits,
819 #define PBLK_GC_MAX_READERS 8 /* Max number of outstanding GC reader jobs */
820 #define PBLK_GC_W_QD 128 /* Queue depth for inflight GC write I/Os */
821 #define PBLK_GC_L_QD 4 /* Queue depth for inflight GC lines */
822 #define PBLK_GC_RSV_LINE 1 /* Reserved lines for GC */
824 int pblk_gc_init(struct pblk *pblk);
825 void pblk_gc_exit(struct pblk *pblk);
826 void pblk_gc_should_start(struct pblk *pblk);
827 void pblk_gc_should_stop(struct pblk *pblk);
828 void pblk_gc_kick(struct pblk *pblk);
829 void pblk_gc_sysfs_state_show(struct pblk *pblk, int *gc_enabled,
831 int pblk_gc_sysfs_force(struct pblk *pblk, int force);
836 void pblk_rl_init(struct pblk_rl *rl, int budget);
837 void pblk_rl_free(struct pblk_rl *rl);
838 int pblk_rl_high_thrs(struct pblk_rl *rl);
839 int pblk_rl_low_thrs(struct pblk_rl *rl);
840 unsigned long pblk_rl_nr_free_blks(struct pblk_rl *rl);
841 int pblk_rl_user_may_insert(struct pblk_rl *rl, int nr_entries);
842 void pblk_rl_inserted(struct pblk_rl *rl, int nr_entries);
843 void pblk_rl_user_in(struct pblk_rl *rl, int nr_entries);
844 int pblk_rl_gc_may_insert(struct pblk_rl *rl, int nr_entries);
845 void pblk_rl_gc_in(struct pblk_rl *rl, int nr_entries);
846 void pblk_rl_out(struct pblk_rl *rl, int nr_user, int nr_gc);
847 int pblk_rl_sysfs_rate_show(struct pblk_rl *rl);
848 int pblk_rl_max_io(struct pblk_rl *rl);
849 void pblk_rl_free_lines_inc(struct pblk_rl *rl, struct pblk_line *line);
850 void pblk_rl_free_lines_dec(struct pblk_rl *rl, struct pblk_line *line);
851 void pblk_rl_set_space_limit(struct pblk_rl *rl, int entries_left);
852 int pblk_rl_is_limit(struct pblk_rl *rl);
857 int pblk_sysfs_init(struct gendisk *tdisk);
858 void pblk_sysfs_exit(struct gendisk *tdisk);
860 static inline void *pblk_malloc(size_t size, int type, gfp_t flags)
862 if (type == PBLK_KMALLOC_META)
863 return kmalloc(size, flags);
864 return vmalloc(size);
867 static inline void pblk_mfree(void *ptr, int type)
869 if (type == PBLK_KMALLOC_META)
875 static inline struct nvm_rq *nvm_rq_from_c_ctx(void *c_ctx)
877 return c_ctx - sizeof(struct nvm_rq);
880 static inline void *emeta_to_bb(struct line_emeta *emeta)
882 return emeta->bb_bitmap;
885 static inline void *emeta_to_lbas(struct pblk *pblk, struct line_emeta *emeta)
887 return ((void *)emeta + pblk->lm.emeta_len[1]);
890 static inline void *emeta_to_vsc(struct pblk *pblk, struct line_emeta *emeta)
892 return (emeta_to_lbas(pblk, emeta) + pblk->lm.emeta_len[2]);
895 static inline int pblk_line_vsc(struct pblk_line *line)
899 spin_lock(&line->lock);
900 vsc = le32_to_cpu(*line->vsc);
901 spin_unlock(&line->lock);
906 #define NVM_MEM_PAGE_WRITE (8)
908 static inline int pblk_pad_distance(struct pblk *pblk)
910 struct nvm_tgt_dev *dev = pblk->dev;
911 struct nvm_geo *geo = &dev->geo;
913 return NVM_MEM_PAGE_WRITE * geo->nr_luns * geo->sec_per_pl;
916 static inline int pblk_dev_ppa_to_line(struct ppa_addr p)
921 static inline int pblk_tgt_ppa_to_line(struct ppa_addr p)
926 static inline int pblk_ppa_to_pos(struct nvm_geo *geo, struct ppa_addr p)
928 return p.g.lun * geo->nr_chnls + p.g.ch;
931 /* A block within a line corresponds to the lun */
932 static inline int pblk_dev_ppa_to_pos(struct nvm_geo *geo, struct ppa_addr p)
934 return p.g.lun * geo->nr_chnls + p.g.ch;
937 static inline struct ppa_addr pblk_ppa32_to_ppa64(struct pblk *pblk, u32 ppa32)
939 struct ppa_addr ppa64;
944 ppa64.ppa = ADDR_EMPTY;
945 } else if (ppa32 & (1U << 31)) {
946 ppa64.c.line = ppa32 & ((~0U) >> 1);
947 ppa64.c.is_cached = 1;
949 ppa64.g.blk = (ppa32 & pblk->ppaf.blk_mask) >>
950 pblk->ppaf.blk_offset;
951 ppa64.g.pg = (ppa32 & pblk->ppaf.pg_mask) >>
952 pblk->ppaf.pg_offset;
953 ppa64.g.lun = (ppa32 & pblk->ppaf.lun_mask) >>
954 pblk->ppaf.lun_offset;
955 ppa64.g.ch = (ppa32 & pblk->ppaf.ch_mask) >>
956 pblk->ppaf.ch_offset;
957 ppa64.g.pl = (ppa32 & pblk->ppaf.pln_mask) >>
958 pblk->ppaf.pln_offset;
959 ppa64.g.sec = (ppa32 & pblk->ppaf.sec_mask) >>
960 pblk->ppaf.sec_offset;
966 static inline struct ppa_addr pblk_trans_map_get(struct pblk *pblk,
971 if (pblk->ppaf_bitsize < 32) {
972 u32 *map = (u32 *)pblk->trans_map;
974 ppa = pblk_ppa32_to_ppa64(pblk, map[lba]);
976 struct ppa_addr *map = (struct ppa_addr *)pblk->trans_map;
984 static inline u32 pblk_ppa64_to_ppa32(struct pblk *pblk, struct ppa_addr ppa64)
988 if (ppa64.ppa == ADDR_EMPTY) {
990 } else if (ppa64.c.is_cached) {
991 ppa32 |= ppa64.c.line;
994 ppa32 |= ppa64.g.blk << pblk->ppaf.blk_offset;
995 ppa32 |= ppa64.g.pg << pblk->ppaf.pg_offset;
996 ppa32 |= ppa64.g.lun << pblk->ppaf.lun_offset;
997 ppa32 |= ppa64.g.ch << pblk->ppaf.ch_offset;
998 ppa32 |= ppa64.g.pl << pblk->ppaf.pln_offset;
999 ppa32 |= ppa64.g.sec << pblk->ppaf.sec_offset;
1005 static inline void pblk_trans_map_set(struct pblk *pblk, sector_t lba,
1006 struct ppa_addr ppa)
1008 if (pblk->ppaf_bitsize < 32) {
1009 u32 *map = (u32 *)pblk->trans_map;
1011 map[lba] = pblk_ppa64_to_ppa32(pblk, ppa);
1013 u64 *map = (u64 *)pblk->trans_map;
1019 static inline u64 pblk_dev_ppa_to_line_addr(struct pblk *pblk,
1025 paddr |= (u64)p.g.pg << pblk->ppaf.pg_offset;
1026 paddr |= (u64)p.g.lun << pblk->ppaf.lun_offset;
1027 paddr |= (u64)p.g.ch << pblk->ppaf.ch_offset;
1028 paddr |= (u64)p.g.pl << pblk->ppaf.pln_offset;
1029 paddr |= (u64)p.g.sec << pblk->ppaf.sec_offset;
1034 static inline int pblk_ppa_empty(struct ppa_addr ppa_addr)
1036 return (ppa_addr.ppa == ADDR_EMPTY);
1039 static inline void pblk_ppa_set_empty(struct ppa_addr *ppa_addr)
1041 ppa_addr->ppa = ADDR_EMPTY;
1044 static inline bool pblk_ppa_comp(struct ppa_addr lppa, struct ppa_addr rppa)
1046 if (lppa.ppa == rppa.ppa)
1052 static inline int pblk_addr_in_cache(struct ppa_addr ppa)
1054 return (ppa.ppa != ADDR_EMPTY && ppa.c.is_cached);
1057 static inline int pblk_addr_to_cacheline(struct ppa_addr ppa)
1062 static inline struct ppa_addr pblk_cacheline_to_addr(int addr)
1072 static inline struct ppa_addr addr_to_gen_ppa(struct pblk *pblk, u64 paddr,
1075 struct ppa_addr ppa;
1078 ppa.g.blk = line_id;
1079 ppa.g.pg = (paddr & pblk->ppaf.pg_mask) >> pblk->ppaf.pg_offset;
1080 ppa.g.lun = (paddr & pblk->ppaf.lun_mask) >> pblk->ppaf.lun_offset;
1081 ppa.g.ch = (paddr & pblk->ppaf.ch_mask) >> pblk->ppaf.ch_offset;
1082 ppa.g.pl = (paddr & pblk->ppaf.pln_mask) >> pblk->ppaf.pln_offset;
1083 ppa.g.sec = (paddr & pblk->ppaf.sec_mask) >> pblk->ppaf.sec_offset;
1088 static inline struct ppa_addr addr_to_pblk_ppa(struct pblk *pblk, u64 paddr,
1091 struct ppa_addr ppa;
1093 ppa = addr_to_gen_ppa(pblk, paddr, line_id);
1098 static inline u32 pblk_calc_meta_header_crc(struct pblk *pblk,
1099 struct line_header *header)
1103 crc = crc32_le(crc, (unsigned char *)header + sizeof(crc),
1104 sizeof(struct line_header) - sizeof(crc));
1109 static inline u32 pblk_calc_smeta_crc(struct pblk *pblk,
1110 struct line_smeta *smeta)
1112 struct pblk_line_meta *lm = &pblk->lm;
1115 crc = crc32_le(crc, (unsigned char *)smeta +
1116 sizeof(struct line_header) + sizeof(crc),
1118 sizeof(struct line_header) - sizeof(crc));
1123 static inline u32 pblk_calc_emeta_crc(struct pblk *pblk,
1124 struct line_emeta *emeta)
1126 struct pblk_line_meta *lm = &pblk->lm;
1129 crc = crc32_le(crc, (unsigned char *)emeta +
1130 sizeof(struct line_header) + sizeof(crc),
1132 sizeof(struct line_header) - sizeof(crc));
1137 static inline int pblk_set_progr_mode(struct pblk *pblk, int type)
1139 struct nvm_tgt_dev *dev = pblk->dev;
1140 struct nvm_geo *geo = &dev->geo;
1143 flags = geo->plane_mode >> 1;
1146 flags |= NVM_IO_SCRAMBLE_ENABLE;
1152 PBLK_READ_RANDOM = 0,
1153 PBLK_READ_SEQUENTIAL = 1,
1156 static inline int pblk_set_read_mode(struct pblk *pblk, int type)
1158 struct nvm_tgt_dev *dev = pblk->dev;
1159 struct nvm_geo *geo = &dev->geo;
1162 flags = NVM_IO_SUSPEND | NVM_IO_SCRAMBLE_ENABLE;
1163 if (type == PBLK_READ_SEQUENTIAL)
1164 flags |= geo->plane_mode >> 1;
1169 static inline int pblk_io_aligned(struct pblk *pblk, int nr_secs)
1171 return !(nr_secs % pblk->min_write_pgs);
1174 #ifdef CONFIG_NVM_DEBUG
1175 static inline void print_ppa(struct ppa_addr *p, char *msg, int error)
1177 if (p->c.is_cached) {
1178 pr_err("ppa: (%s: %x) cache line: %llu\n",
1179 msg, error, (u64)p->c.line);
1181 pr_err("ppa: (%s: %x):ch:%d,lun:%d,blk:%d,pg:%d,pl:%d,sec:%d\n",
1183 p->g.ch, p->g.lun, p->g.blk,
1184 p->g.pg, p->g.pl, p->g.sec);
1188 static inline void pblk_print_failed_rqd(struct pblk *pblk, struct nvm_rq *rqd,
1193 if (rqd->nr_ppas == 1) {
1194 print_ppa(&rqd->ppa_addr, "rqd", error);
1198 while ((bit = find_next_bit((void *)&rqd->ppa_status, rqd->nr_ppas,
1199 bit + 1)) < rqd->nr_ppas) {
1200 print_ppa(&rqd->ppa_list[bit], "rqd", error);
1203 pr_err("error:%d, ppa_status:%llx\n", error, rqd->ppa_status);
1207 static inline int pblk_boundary_ppa_checks(struct nvm_tgt_dev *tgt_dev,
1208 struct ppa_addr *ppas, int nr_ppas)
1210 struct nvm_geo *geo = &tgt_dev->geo;
1211 struct ppa_addr *ppa;
1214 for (i = 0; i < nr_ppas; i++) {
1217 if (!ppa->c.is_cached &&
1218 ppa->g.ch < geo->nr_chnls &&
1219 ppa->g.lun < geo->luns_per_chnl &&
1220 ppa->g.pl < geo->nr_planes &&
1221 ppa->g.blk < geo->blks_per_lun &&
1222 ppa->g.pg < geo->pgs_per_blk &&
1223 ppa->g.sec < geo->sec_per_pg)
1226 #ifdef CONFIG_NVM_DEBUG
1227 print_ppa(ppa, "boundary", i);
1234 static inline int pblk_boundary_paddr_checks(struct pblk *pblk, u64 paddr)
1236 struct pblk_line_meta *lm = &pblk->lm;
1238 if (paddr > lm->sec_per_line)
1244 static inline unsigned int pblk_get_bi_idx(struct bio *bio)
1246 return bio->bi_iter.bi_idx;
1249 static inline sector_t pblk_get_lba(struct bio *bio)
1251 return bio->bi_iter.bi_sector / NR_PHY_IN_LOG;
1254 static inline unsigned int pblk_get_secs(struct bio *bio)
1256 return bio->bi_iter.bi_size / PBLK_EXPOSED_PAGE_SIZE;
1259 static inline sector_t pblk_get_sector(sector_t lba)
1261 return lba * NR_PHY_IN_LOG;
1264 static inline void pblk_setup_uuid(struct pblk *pblk)
1269 memcpy(pblk->instance_uuid, uuid.b, 16);
1271 #endif /* PBLK_H_ */