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 */
271 atomic_t rb_user_cnt; /* User I/O buffer counter */
272 atomic_t rb_gc_cnt; /* GC I/O buffer counter */
273 atomic_t rb_space; /* Space limit in case of reaching capacity */
275 int rsv_blocks; /* Reserved blocks for GC */
280 struct timer_list u_timer;
282 unsigned long long nr_secs;
283 unsigned long total_blocks;
284 atomic_t free_blocks;
287 #define PBLK_LINE_EMPTY (~0U)
291 PBLK_LINETYPE_FREE = 0,
292 PBLK_LINETYPE_LOG = 1,
293 PBLK_LINETYPE_DATA = 2,
296 PBLK_LINESTATE_FREE = 10,
297 PBLK_LINESTATE_OPEN = 11,
298 PBLK_LINESTATE_CLOSED = 12,
299 PBLK_LINESTATE_GC = 13,
300 PBLK_LINESTATE_BAD = 14,
301 PBLK_LINESTATE_CORRUPT = 15,
304 PBLK_LINEGC_NONE = 20,
305 PBLK_LINEGC_EMPTY = 21,
306 PBLK_LINEGC_LOW = 22,
307 PBLK_LINEGC_MID = 23,
308 PBLK_LINEGC_HIGH = 24,
309 PBLK_LINEGC_FULL = 25,
312 #define PBLK_MAGIC 0x70626c6b /*pblk*/
316 __le32 identifier; /* pblk identifier */
317 __u8 uuid[16]; /* instance uuid */
318 __le16 type; /* line type */
319 __le16 version; /* type version */
320 __le32 id; /* line id for current line */
324 struct line_header header;
326 __le32 crc; /* Full structure including struct crc */
327 /* Previous line metadata */
328 __le32 prev_id; /* Line id for previous line */
330 /* Current line metadata */
331 __le64 seq_nr; /* Sequence number for current line */
334 __le32 window_wr_lun; /* Number of parallel LUNs to write */
342 * Metadata layout in media:
344 * 1. struct line_emeta
345 * 2. bad block bitmap (u64 * window_wr_lun)
346 * Mid sectors (start at lbas_sector):
347 * 3. nr_lbas (u64) forming lba list
348 * Last sectors (start at vsc_sector):
349 * 4. u32 valid sector count (vsc) for all lines (~0U: free line)
352 struct line_header header;
354 __le32 crc; /* Full structure including struct crc */
356 /* Previous line metadata */
357 __le32 prev_id; /* Line id for prev line */
359 /* Current line metadata */
360 __le64 seq_nr; /* Sequence number for current line */
363 __le32 window_wr_lun; /* Number of parallel LUNs to write */
365 /* Bookkeeping for recovery */
366 __le32 next_id; /* Line id for next line */
367 __le64 nr_lbas; /* Number of lbas mapped in line */
368 __le64 nr_valid_lbas; /* Number of valid lbas mapped in line */
369 __le64 bb_bitmap[]; /* Updated bad block bitmap for line */
373 struct line_emeta *buf; /* emeta buffer in media format */
374 int mem; /* Write offset - points to next
375 * writable entry in memory
377 atomic_t sync; /* Synced - backpointer that signals the
378 * last entry that has been successfully
381 unsigned int nr_entries; /* Number of emeta entries */
385 struct line_smeta *buf; /* smeta buffer in persistent format */
390 unsigned int id; /* Line number corresponds to the
393 unsigned int seq_nr; /* Unique line sequence number */
395 int state; /* PBLK_LINESTATE_X */
396 int type; /* PBLK_LINETYPE_X */
397 int gc_group; /* PBLK_LINEGC_X */
398 struct list_head list; /* Free, GC lists */
400 unsigned long *lun_bitmap; /* Bitmap for LUNs mapped in line */
402 struct pblk_smeta *smeta; /* Start metadata */
403 struct pblk_emeta *emeta; /* End medatada */
405 int meta_line; /* Metadata line id */
406 int meta_distance; /* Distance between data and metadata */
408 u64 smeta_ssec; /* Sector where smeta starts */
409 u64 emeta_ssec; /* Sector where emeta starts */
411 unsigned int sec_in_line; /* Number of usable secs in line */
413 atomic_t blk_in_line; /* Number of good blocks in line */
414 unsigned long *blk_bitmap; /* Bitmap for valid/invalid blocks */
415 unsigned long *erase_bitmap; /* Bitmap for erased blocks */
417 unsigned long *map_bitmap; /* Bitmap for mapped sectors in line */
418 unsigned long *invalid_bitmap; /* Bitmap for invalid sectors in line */
420 atomic_t left_eblks; /* Blocks left for erasing */
421 atomic_t left_seblks; /* Blocks left for sync erasing */
423 int left_msecs; /* Sectors left for mapping */
424 unsigned int cur_sec; /* Sector map pointer */
425 unsigned int nr_valid_lbas; /* Number of valid lbas in line */
427 __le32 *vsc; /* Valid sector count in line */
429 struct kref ref; /* Write buffer L2P references */
431 spinlock_t lock; /* Necessary for invalid_bitmap only */
434 #define PBLK_DATA_LINES 4
437 PBLK_KMALLOC_META = 1,
438 PBLK_VMALLOC_META = 2,
442 PBLK_EMETA_TYPE_HEADER = 1, /* struct line_emeta first sector */
443 PBLK_EMETA_TYPE_LLBA = 2, /* lba list - type: __le64 */
444 PBLK_EMETA_TYPE_VSC = 3, /* vsc list - type: __le32 */
447 struct pblk_line_mgmt {
448 int nr_lines; /* Total number of full lines */
449 int nr_free_lines; /* Number of full lines in free list */
451 /* Free lists - use free_lock */
452 struct list_head free_list; /* Full lines ready to use */
453 struct list_head corrupt_list; /* Full lines corrupted */
454 struct list_head bad_list; /* Full lines bad */
456 /* GC lists - use gc_lock */
457 struct list_head *gc_lists[PBLK_GC_NR_LISTS];
458 struct list_head gc_high_list; /* Full lines ready to GC, high isc */
459 struct list_head gc_mid_list; /* Full lines ready to GC, mid isc */
460 struct list_head gc_low_list; /* Full lines ready to GC, low isc */
462 struct list_head gc_full_list; /* Full lines ready to GC, no valid */
463 struct list_head gc_empty_list; /* Full lines close, all valid */
465 struct pblk_line *log_line; /* Current FTL log line */
466 struct pblk_line *data_line; /* Current data line */
467 struct pblk_line *log_next; /* Next FTL log line */
468 struct pblk_line *data_next; /* Next data line */
470 struct list_head emeta_list; /* Lines queued to schedule emeta */
472 __le32 *vsc_list; /* Valid sector counts for all lines */
474 /* Metadata allocation type: VMALLOC | KMALLOC */
475 int emeta_alloc_type;
477 /* Pre-allocated metadata for data lines */
478 struct pblk_smeta *sline_meta[PBLK_DATA_LINES];
479 struct pblk_emeta *eline_meta[PBLK_DATA_LINES];
480 unsigned long meta_bitmap;
482 /* Helpers for fast bitmap calculations */
483 unsigned long *bb_template;
484 unsigned long *bb_aux;
486 unsigned long d_seq_nr; /* Data line unique sequence number */
487 unsigned long l_seq_nr; /* Log line unique sequence number */
489 spinlock_t free_lock;
490 spinlock_t close_lock;
494 struct pblk_line_meta {
495 unsigned int smeta_len; /* Total length for smeta */
496 unsigned int smeta_sec; /* Sectors needed for smeta */
498 unsigned int emeta_len[4]; /* Lengths for emeta:
500 * [1]: struct line_emeta length
501 * [2]: L2P portion length
502 * [3]: vsc list length
504 unsigned int emeta_sec[4]; /* Sectors needed for emeta. Same layout
508 unsigned int emeta_bb; /* Boundary for bb that affects emeta */
510 unsigned int vsc_list_len; /* Length for vsc list */
511 unsigned int sec_bitmap_len; /* Length for sector bitmap in line */
512 unsigned int blk_bitmap_len; /* Length for block bitmap in line */
513 unsigned int lun_bitmap_len; /* Length for lun bitmap in line */
515 unsigned int blk_per_line; /* Number of blocks in a full line */
516 unsigned int sec_per_line; /* Number of sectors in a line */
517 unsigned int dsec_per_line; /* Number of data sectors in a line */
518 unsigned int min_blk_line; /* Min. number of good blocks in line */
520 unsigned int mid_thrs; /* Threshold for GC mid list */
521 unsigned int high_thrs; /* Threshold for GC high list */
523 unsigned int meta_distance; /* Distance between data and metadata */
526 struct pblk_addr_format {
542 PBLK_STATE_RUNNING = 0,
543 PBLK_STATE_STOPPING = 1,
544 PBLK_STATE_RECOVERING = 2,
545 PBLK_STATE_STOPPED = 3,
549 struct nvm_tgt_dev *dev;
550 struct gendisk *disk;
554 struct pblk_lun *luns;
556 struct pblk_line *lines; /* Line array */
557 struct pblk_line_mgmt l_mg; /* Line management */
558 struct pblk_line_meta lm; /* Line metadata */
561 struct pblk_addr_format ppaf;
565 int state; /* pblk line state */
567 int min_write_pgs; /* Minimum amount of pages required by controller */
568 int max_write_pgs; /* Maximum amount of pages supported by controller */
569 int pgs_in_buffer; /* Number of pages that need to be held in buffer to
570 * guarantee successful reads.
573 sector_t capacity; /* Device capacity when bad blocks are subtracted */
574 int over_pct; /* Percentage of device used for over-provisioning */
576 /* pblk provisioning values. Used by rate limiter */
581 unsigned char instance_uuid[16];
582 #ifdef CONFIG_NVM_DEBUG
583 /* All debug counters apply to 4kb sector I/Os */
584 atomic_long_t inflight_writes; /* Inflight writes (user and gc) */
585 atomic_long_t padded_writes; /* Sectors padded due to flush/fua */
586 atomic_long_t padded_wb; /* Sectors padded in write buffer */
587 atomic_long_t nr_flush; /* Number of flush/fua I/O */
588 atomic_long_t req_writes; /* Sectors stored on write buffer */
589 atomic_long_t sub_writes; /* Sectors submitted from buffer */
590 atomic_long_t sync_writes; /* Sectors synced to media */
591 atomic_long_t inflight_reads; /* Inflight sector read requests */
592 atomic_long_t cache_reads; /* Read requests that hit the cache */
593 atomic_long_t sync_reads; /* Completed sector read requests */
594 atomic_long_t recov_writes; /* Sectors submitted from recovery */
595 atomic_long_t recov_gc_writes; /* Sectors submitted from write GC */
596 atomic_long_t recov_gc_reads; /* Sectors submitted from read GC */
601 atomic_long_t read_failed;
602 atomic_long_t read_empty;
603 atomic_long_t read_high_ecc;
604 atomic_long_t read_failed_gc;
605 atomic_long_t write_failed;
606 atomic_long_t erase_failed;
608 atomic_t inflight_io; /* General inflight I/O counter */
610 struct task_struct *writer_ts;
612 /* Simple translation map of logical addresses to physical addresses.
613 * The logical addresses is known by the host system, while the physical
614 * addresses are used when writing to the disk block device.
616 unsigned char *trans_map;
617 spinlock_t trans_lock;
619 struct list_head compl_list;
621 mempool_t *page_pool;
622 mempool_t *line_ws_pool;
624 mempool_t *g_rq_pool;
625 mempool_t *w_rq_pool;
626 mempool_t *line_meta_pool;
628 struct workqueue_struct *close_wq;
629 struct workqueue_struct *bb_wq;
631 struct timer_list wtimer;
636 struct pblk_line_ws {
638 struct pblk_line *line;
640 struct work_struct ws;
643 #define pblk_g_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_g_ctx))
644 #define pblk_w_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_c_ctx))
647 * pblk ring buffer operations
649 int pblk_rb_init(struct pblk_rb *rb, struct pblk_rb_entry *rb_entry_base,
650 unsigned int power_size, unsigned int power_seg_sz);
651 unsigned int pblk_rb_calculate_size(unsigned int nr_entries);
652 void *pblk_rb_entries_ref(struct pblk_rb *rb);
653 int pblk_rb_may_write_user(struct pblk_rb *rb, struct bio *bio,
654 unsigned int nr_entries, unsigned int *pos);
655 int pblk_rb_may_write_gc(struct pblk_rb *rb, unsigned int nr_entries,
657 void pblk_rb_write_entry_user(struct pblk_rb *rb, void *data,
658 struct pblk_w_ctx w_ctx, unsigned int pos);
659 void pblk_rb_write_entry_gc(struct pblk_rb *rb, void *data,
660 struct pblk_w_ctx w_ctx, struct pblk_line *gc_line,
662 struct pblk_w_ctx *pblk_rb_w_ctx(struct pblk_rb *rb, unsigned int pos);
663 void pblk_rb_flush(struct pblk_rb *rb);
665 void pblk_rb_sync_l2p(struct pblk_rb *rb);
666 unsigned int pblk_rb_read_to_bio(struct pblk_rb *rb, struct nvm_rq *rqd,
667 struct bio *bio, unsigned int pos,
668 unsigned int nr_entries, unsigned int count);
669 unsigned int pblk_rb_read_to_bio_list(struct pblk_rb *rb, struct bio *bio,
670 struct list_head *list,
672 int pblk_rb_copy_to_bio(struct pblk_rb *rb, struct bio *bio, sector_t lba,
673 struct ppa_addr ppa, int bio_iter);
674 unsigned int pblk_rb_read_commit(struct pblk_rb *rb, unsigned int entries);
676 unsigned int pblk_rb_sync_init(struct pblk_rb *rb, unsigned long *flags);
677 unsigned int pblk_rb_sync_advance(struct pblk_rb *rb, unsigned int nr_entries);
678 struct pblk_rb_entry *pblk_rb_sync_scan_entry(struct pblk_rb *rb,
679 struct ppa_addr *ppa);
680 void pblk_rb_sync_end(struct pblk_rb *rb, unsigned long *flags);
681 unsigned int pblk_rb_sync_point_count(struct pblk_rb *rb);
683 unsigned int pblk_rb_read_count(struct pblk_rb *rb);
684 unsigned int pblk_rb_sync_count(struct pblk_rb *rb);
685 unsigned int pblk_rb_wrap_pos(struct pblk_rb *rb, unsigned int pos);
687 int pblk_rb_tear_down_check(struct pblk_rb *rb);
688 int pblk_rb_pos_oob(struct pblk_rb *rb, u64 pos);
689 void pblk_rb_data_free(struct pblk_rb *rb);
690 ssize_t pblk_rb_sysfs(struct pblk_rb *rb, char *buf);
695 struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int rw);
696 void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write);
697 int pblk_setup_w_rec_rq(struct pblk *pblk, struct nvm_rq *rqd,
698 struct pblk_c_ctx *c_ctx);
699 void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int rw);
700 void pblk_wait_for_meta(struct pblk *pblk);
701 struct ppa_addr pblk_get_lba_map(struct pblk *pblk, sector_t lba);
702 void pblk_discard(struct pblk *pblk, struct bio *bio);
703 void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd);
704 void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd);
705 int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd);
706 int pblk_submit_meta_io(struct pblk *pblk, struct pblk_line *meta_line);
707 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
708 unsigned int nr_secs, unsigned int len,
709 int alloc_type, gfp_t gfp_mask);
710 struct pblk_line *pblk_line_get(struct pblk *pblk);
711 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk);
712 void pblk_line_replace_data(struct pblk *pblk);
713 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line);
714 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line);
715 struct pblk_line *pblk_line_get_data(struct pblk *pblk);
716 struct pblk_line *pblk_line_get_erase(struct pblk *pblk);
717 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line);
718 int pblk_line_is_full(struct pblk_line *line);
719 void pblk_line_free(struct pblk *pblk, struct pblk_line *line);
720 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line);
721 void pblk_line_close(struct pblk *pblk, struct pblk_line *line);
722 void pblk_line_close_meta_sync(struct pblk *pblk);
723 void pblk_line_close_ws(struct work_struct *work);
724 void pblk_pipeline_stop(struct pblk *pblk);
725 void pblk_line_mark_bb(struct work_struct *work);
726 void pblk_line_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
727 void (*work)(struct work_struct *),
728 struct workqueue_struct *wq);
729 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line);
730 int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line);
731 int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line,
733 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr erase_ppa);
734 void pblk_line_put(struct kref *ref);
735 struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line);
736 u64 pblk_lookup_page(struct pblk *pblk, struct pblk_line *line);
737 void pblk_dealloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
738 u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
739 u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
740 int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
741 unsigned long secs_to_flush);
742 void pblk_down_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
743 unsigned long *lun_bitmap);
744 void pblk_up_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
745 unsigned long *lun_bitmap);
746 void pblk_end_bio_sync(struct bio *bio);
747 void pblk_end_io_sync(struct nvm_rq *rqd);
748 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
750 void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
752 void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa);
753 void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
755 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa);
756 void pblk_update_map_cache(struct pblk *pblk, sector_t lba,
757 struct ppa_addr ppa);
758 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
759 struct ppa_addr ppa, struct ppa_addr entry_line);
760 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa,
761 struct pblk_line *gc_line);
762 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
763 u64 *lba_list, int nr_secs);
764 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
765 sector_t blba, int nr_secs);
768 * pblk user I/O write path
770 int pblk_write_to_cache(struct pblk *pblk, struct bio *bio,
771 unsigned long flags);
772 int pblk_write_gc_to_cache(struct pblk *pblk, void *data, u64 *lba_list,
773 unsigned int nr_entries, unsigned int nr_rec_entries,
774 struct pblk_line *gc_line, unsigned long flags);
779 void pblk_map_erase_rq(struct pblk *pblk, struct nvm_rq *rqd,
780 unsigned int sentry, unsigned long *lun_bitmap,
781 unsigned int valid_secs, struct ppa_addr *erase_ppa);
782 void pblk_map_rq(struct pblk *pblk, struct nvm_rq *rqd, unsigned int sentry,
783 unsigned long *lun_bitmap, unsigned int valid_secs,
789 int pblk_write_ts(void *data);
790 void pblk_write_timer_fn(unsigned long data);
791 void pblk_write_should_kick(struct pblk *pblk);
796 extern struct bio_set *pblk_bio_set;
797 int pblk_submit_read(struct pblk *pblk, struct bio *bio);
798 int pblk_submit_read_gc(struct pblk *pblk, u64 *lba_list, void *data,
799 unsigned int nr_secs, unsigned int *secs_to_gc,
800 struct pblk_line *line);
804 void pblk_submit_rec(struct work_struct *work);
805 struct pblk_line *pblk_recov_l2p(struct pblk *pblk);
806 int pblk_recov_pad(struct pblk *pblk);
807 __le64 *pblk_recov_get_lba_list(struct pblk *pblk, struct line_emeta *emeta);
808 int pblk_recov_setup_rq(struct pblk *pblk, struct pblk_c_ctx *c_ctx,
809 struct pblk_rec_ctx *recovery, u64 *comp_bits,
815 #define PBLK_GC_MAX_READERS 8 /* Max number of outstanding GC reader jobs */
816 #define PBLK_GC_W_QD 128 /* Queue depth for inflight GC write I/Os */
817 #define PBLK_GC_L_QD 4 /* Queue depth for inflight GC lines */
818 #define PBLK_GC_RSV_LINE 1 /* Reserved lines for GC */
820 int pblk_gc_init(struct pblk *pblk);
821 void pblk_gc_exit(struct pblk *pblk);
822 void pblk_gc_should_start(struct pblk *pblk);
823 void pblk_gc_should_stop(struct pblk *pblk);
824 void pblk_gc_should_kick(struct pblk *pblk);
825 void pblk_gc_kick(struct pblk *pblk);
826 void pblk_gc_sysfs_state_show(struct pblk *pblk, int *gc_enabled,
828 int pblk_gc_sysfs_force(struct pblk *pblk, int force);
833 void pblk_rl_init(struct pblk_rl *rl, int budget);
834 void pblk_rl_free(struct pblk_rl *rl);
835 int pblk_rl_high_thrs(struct pblk_rl *rl);
836 int pblk_rl_low_thrs(struct pblk_rl *rl);
837 unsigned long pblk_rl_nr_free_blks(struct pblk_rl *rl);
838 int pblk_rl_user_may_insert(struct pblk_rl *rl, int nr_entries);
839 void pblk_rl_inserted(struct pblk_rl *rl, int nr_entries);
840 void pblk_rl_user_in(struct pblk_rl *rl, int nr_entries);
841 int pblk_rl_gc_may_insert(struct pblk_rl *rl, int nr_entries);
842 void pblk_rl_gc_in(struct pblk_rl *rl, int nr_entries);
843 void pblk_rl_out(struct pblk_rl *rl, int nr_user, int nr_gc);
844 int pblk_rl_sysfs_rate_show(struct pblk_rl *rl);
845 void pblk_rl_free_lines_inc(struct pblk_rl *rl, struct pblk_line *line);
846 void pblk_rl_free_lines_dec(struct pblk_rl *rl, struct pblk_line *line);
847 void pblk_rl_set_space_limit(struct pblk_rl *rl, int entries_left);
848 int pblk_rl_is_limit(struct pblk_rl *rl);
853 int pblk_sysfs_init(struct gendisk *tdisk);
854 void pblk_sysfs_exit(struct gendisk *tdisk);
856 static inline void *pblk_malloc(size_t size, int type, gfp_t flags)
858 if (type == PBLK_KMALLOC_META)
859 return kmalloc(size, flags);
860 return vmalloc(size);
863 static inline void pblk_mfree(void *ptr, int type)
865 if (type == PBLK_KMALLOC_META)
871 static inline struct nvm_rq *nvm_rq_from_c_ctx(void *c_ctx)
873 return c_ctx - sizeof(struct nvm_rq);
876 static inline void *emeta_to_bb(struct line_emeta *emeta)
878 return emeta->bb_bitmap;
881 static inline void *emeta_to_lbas(struct pblk *pblk, struct line_emeta *emeta)
883 return ((void *)emeta + pblk->lm.emeta_len[1]);
886 static inline void *emeta_to_vsc(struct pblk *pblk, struct line_emeta *emeta)
888 return (emeta_to_lbas(pblk, emeta) + pblk->lm.emeta_len[2]);
891 static inline int pblk_line_vsc(struct pblk_line *line)
895 spin_lock(&line->lock);
896 vsc = le32_to_cpu(*line->vsc);
897 spin_unlock(&line->lock);
902 #define NVM_MEM_PAGE_WRITE (8)
904 static inline int pblk_pad_distance(struct pblk *pblk)
906 struct nvm_tgt_dev *dev = pblk->dev;
907 struct nvm_geo *geo = &dev->geo;
909 return NVM_MEM_PAGE_WRITE * geo->nr_luns * geo->sec_per_pl;
912 static inline int pblk_dev_ppa_to_line(struct ppa_addr p)
917 static inline int pblk_tgt_ppa_to_line(struct ppa_addr p)
922 static inline int pblk_ppa_to_pos(struct nvm_geo *geo, struct ppa_addr p)
924 return p.g.lun * geo->nr_chnls + p.g.ch;
927 /* A block within a line corresponds to the lun */
928 static inline int pblk_dev_ppa_to_pos(struct nvm_geo *geo, struct ppa_addr p)
930 return p.g.lun * geo->nr_chnls + p.g.ch;
933 static inline struct ppa_addr pblk_ppa32_to_ppa64(struct pblk *pblk, u32 ppa32)
935 struct ppa_addr ppa64;
940 ppa64.ppa = ADDR_EMPTY;
941 } else if (ppa32 & (1U << 31)) {
942 ppa64.c.line = ppa32 & ((~0U) >> 1);
943 ppa64.c.is_cached = 1;
945 ppa64.g.blk = (ppa32 & pblk->ppaf.blk_mask) >>
946 pblk->ppaf.blk_offset;
947 ppa64.g.pg = (ppa32 & pblk->ppaf.pg_mask) >>
948 pblk->ppaf.pg_offset;
949 ppa64.g.lun = (ppa32 & pblk->ppaf.lun_mask) >>
950 pblk->ppaf.lun_offset;
951 ppa64.g.ch = (ppa32 & pblk->ppaf.ch_mask) >>
952 pblk->ppaf.ch_offset;
953 ppa64.g.pl = (ppa32 & pblk->ppaf.pln_mask) >>
954 pblk->ppaf.pln_offset;
955 ppa64.g.sec = (ppa32 & pblk->ppaf.sec_mask) >>
956 pblk->ppaf.sec_offset;
962 static inline struct ppa_addr pblk_trans_map_get(struct pblk *pblk,
967 if (pblk->ppaf_bitsize < 32) {
968 u32 *map = (u32 *)pblk->trans_map;
970 ppa = pblk_ppa32_to_ppa64(pblk, map[lba]);
972 struct ppa_addr *map = (struct ppa_addr *)pblk->trans_map;
980 static inline u32 pblk_ppa64_to_ppa32(struct pblk *pblk, struct ppa_addr ppa64)
984 if (ppa64.ppa == ADDR_EMPTY) {
986 } else if (ppa64.c.is_cached) {
987 ppa32 |= ppa64.c.line;
990 ppa32 |= ppa64.g.blk << pblk->ppaf.blk_offset;
991 ppa32 |= ppa64.g.pg << pblk->ppaf.pg_offset;
992 ppa32 |= ppa64.g.lun << pblk->ppaf.lun_offset;
993 ppa32 |= ppa64.g.ch << pblk->ppaf.ch_offset;
994 ppa32 |= ppa64.g.pl << pblk->ppaf.pln_offset;
995 ppa32 |= ppa64.g.sec << pblk->ppaf.sec_offset;
1001 static inline void pblk_trans_map_set(struct pblk *pblk, sector_t lba,
1002 struct ppa_addr ppa)
1004 if (pblk->ppaf_bitsize < 32) {
1005 u32 *map = (u32 *)pblk->trans_map;
1007 map[lba] = pblk_ppa64_to_ppa32(pblk, ppa);
1009 u64 *map = (u64 *)pblk->trans_map;
1015 static inline u64 pblk_dev_ppa_to_line_addr(struct pblk *pblk,
1021 paddr |= (u64)p.g.pg << pblk->ppaf.pg_offset;
1022 paddr |= (u64)p.g.lun << pblk->ppaf.lun_offset;
1023 paddr |= (u64)p.g.ch << pblk->ppaf.ch_offset;
1024 paddr |= (u64)p.g.pl << pblk->ppaf.pln_offset;
1025 paddr |= (u64)p.g.sec << pblk->ppaf.sec_offset;
1030 static inline int pblk_ppa_empty(struct ppa_addr ppa_addr)
1032 return (ppa_addr.ppa == ADDR_EMPTY);
1035 static inline void pblk_ppa_set_empty(struct ppa_addr *ppa_addr)
1037 ppa_addr->ppa = ADDR_EMPTY;
1040 static inline bool pblk_ppa_comp(struct ppa_addr lppa, struct ppa_addr rppa)
1042 if (lppa.ppa == rppa.ppa)
1048 static inline int pblk_addr_in_cache(struct ppa_addr ppa)
1050 return (ppa.ppa != ADDR_EMPTY && ppa.c.is_cached);
1053 static inline int pblk_addr_to_cacheline(struct ppa_addr ppa)
1058 static inline struct ppa_addr pblk_cacheline_to_addr(int addr)
1068 static inline struct ppa_addr addr_to_gen_ppa(struct pblk *pblk, u64 paddr,
1071 struct ppa_addr ppa;
1074 ppa.g.blk = line_id;
1075 ppa.g.pg = (paddr & pblk->ppaf.pg_mask) >> pblk->ppaf.pg_offset;
1076 ppa.g.lun = (paddr & pblk->ppaf.lun_mask) >> pblk->ppaf.lun_offset;
1077 ppa.g.ch = (paddr & pblk->ppaf.ch_mask) >> pblk->ppaf.ch_offset;
1078 ppa.g.pl = (paddr & pblk->ppaf.pln_mask) >> pblk->ppaf.pln_offset;
1079 ppa.g.sec = (paddr & pblk->ppaf.sec_mask) >> pblk->ppaf.sec_offset;
1084 static inline struct ppa_addr addr_to_pblk_ppa(struct pblk *pblk, u64 paddr,
1087 struct ppa_addr ppa;
1089 ppa = addr_to_gen_ppa(pblk, paddr, line_id);
1094 static inline u32 pblk_calc_meta_header_crc(struct pblk *pblk,
1095 struct line_header *header)
1099 crc = crc32_le(crc, (unsigned char *)header + sizeof(crc),
1100 sizeof(struct line_header) - sizeof(crc));
1105 static inline u32 pblk_calc_smeta_crc(struct pblk *pblk,
1106 struct line_smeta *smeta)
1108 struct pblk_line_meta *lm = &pblk->lm;
1111 crc = crc32_le(crc, (unsigned char *)smeta +
1112 sizeof(struct line_header) + sizeof(crc),
1114 sizeof(struct line_header) - sizeof(crc));
1119 static inline u32 pblk_calc_emeta_crc(struct pblk *pblk,
1120 struct line_emeta *emeta)
1122 struct pblk_line_meta *lm = &pblk->lm;
1125 crc = crc32_le(crc, (unsigned char *)emeta +
1126 sizeof(struct line_header) + sizeof(crc),
1128 sizeof(struct line_header) - sizeof(crc));
1133 static inline int pblk_set_progr_mode(struct pblk *pblk, int type)
1135 struct nvm_tgt_dev *dev = pblk->dev;
1136 struct nvm_geo *geo = &dev->geo;
1139 flags = geo->plane_mode >> 1;
1142 flags |= NVM_IO_SCRAMBLE_ENABLE;
1148 PBLK_READ_RANDOM = 0,
1149 PBLK_READ_SEQUENTIAL = 1,
1152 static inline int pblk_set_read_mode(struct pblk *pblk, int type)
1154 struct nvm_tgt_dev *dev = pblk->dev;
1155 struct nvm_geo *geo = &dev->geo;
1158 flags = NVM_IO_SUSPEND | NVM_IO_SCRAMBLE_ENABLE;
1159 if (type == PBLK_READ_SEQUENTIAL)
1160 flags |= geo->plane_mode >> 1;
1165 static inline int pblk_io_aligned(struct pblk *pblk, int nr_secs)
1167 return !(nr_secs % pblk->min_write_pgs);
1170 #ifdef CONFIG_NVM_DEBUG
1171 static inline void print_ppa(struct ppa_addr *p, char *msg, int error)
1173 if (p->c.is_cached) {
1174 pr_err("ppa: (%s: %x) cache line: %llu\n",
1175 msg, error, (u64)p->c.line);
1177 pr_err("ppa: (%s: %x):ch:%d,lun:%d,blk:%d,pg:%d,pl:%d,sec:%d\n",
1179 p->g.ch, p->g.lun, p->g.blk,
1180 p->g.pg, p->g.pl, p->g.sec);
1184 static inline void pblk_print_failed_rqd(struct pblk *pblk, struct nvm_rq *rqd,
1189 if (rqd->nr_ppas == 1) {
1190 print_ppa(&rqd->ppa_addr, "rqd", error);
1194 while ((bit = find_next_bit((void *)&rqd->ppa_status, rqd->nr_ppas,
1195 bit + 1)) < rqd->nr_ppas) {
1196 print_ppa(&rqd->ppa_list[bit], "rqd", error);
1199 pr_err("error:%d, ppa_status:%llx\n", error, rqd->ppa_status);
1203 static inline int pblk_boundary_ppa_checks(struct nvm_tgt_dev *tgt_dev,
1204 struct ppa_addr *ppas, int nr_ppas)
1206 struct nvm_geo *geo = &tgt_dev->geo;
1207 struct ppa_addr *ppa;
1210 for (i = 0; i < nr_ppas; i++) {
1213 if (!ppa->c.is_cached &&
1214 ppa->g.ch < geo->nr_chnls &&
1215 ppa->g.lun < geo->luns_per_chnl &&
1216 ppa->g.pl < geo->nr_planes &&
1217 ppa->g.blk < geo->blks_per_lun &&
1218 ppa->g.pg < geo->pgs_per_blk &&
1219 ppa->g.sec < geo->sec_per_pg)
1222 #ifdef CONFIG_NVM_DEBUG
1223 print_ppa(ppa, "boundary", i);
1230 static inline int pblk_boundary_paddr_checks(struct pblk *pblk, u64 paddr)
1232 struct pblk_line_meta *lm = &pblk->lm;
1234 if (paddr > lm->sec_per_line)
1240 static inline unsigned int pblk_get_bi_idx(struct bio *bio)
1242 return bio->bi_iter.bi_idx;
1245 static inline sector_t pblk_get_lba(struct bio *bio)
1247 return bio->bi_iter.bi_sector / NR_PHY_IN_LOG;
1250 static inline unsigned int pblk_get_secs(struct bio *bio)
1252 return bio->bi_iter.bi_size / PBLK_EXPOSED_PAGE_SIZE;
1255 static inline sector_t pblk_get_sector(sector_t lba)
1257 return lba * NR_PHY_IN_LOG;
1260 static inline void pblk_setup_uuid(struct pblk *pblk)
1265 memcpy(pblk->instance_uuid, uuid.b, 16);
1267 #endif /* PBLK_H_ */