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 /* Static pool sizes */
55 #define PBLK_GEN_WS_POOL_SIZE (2)
59 PBLK_WRITE = WRITE,/* Write from write buffer */
60 PBLK_WRITE_INT, /* Internal write - no write buffer */
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)
98 #define pblk_dma_ppa_size (sizeof(u64) * PBLK_MAX_REQ_ADDRS)
100 /* write buffer completion context */
102 struct list_head list; /* Head for out-of-order completion */
104 unsigned long *lun_bitmap; /* Luns used on current request */
106 unsigned int nr_valid;
107 unsigned int nr_padded;
119 struct completion wait;
123 /* Recovery context */
124 struct pblk_rec_ctx {
127 struct list_head failed;
128 struct work_struct ws_rec;
133 struct bio_list bios; /* Original bios - used for completion
134 * in REQ_FUA, REQ_FLUSH case
136 u64 lba; /* Logic addr. associated with entry */
137 struct ppa_addr ppa; /* Physic addr. associated with entry */
138 int flags; /* Write context flags */
141 struct pblk_rb_entry {
142 struct ppa_addr cacheline; /* Cacheline for this entry */
143 void *data; /* Pointer to data on this entry */
144 struct pblk_w_ctx w_ctx; /* Context for this entry */
145 struct list_head index; /* List head to enable indexes */
148 #define EMPTY_ENTRY (~0U)
150 struct pblk_rb_pages {
153 struct list_head list;
157 struct pblk_rb_entry *entries; /* Ring buffer entries */
158 unsigned int mem; /* Write offset - points to next
159 * writable entry in memory
161 unsigned int subm; /* Read offset - points to last entry
162 * that has been submitted to the media
165 unsigned int sync; /* Synced - backpointer that signals
166 * the last submitted entry that has
167 * been successfully persisted to media
169 unsigned int flush_point; /* Sync point - last entry that must be
170 * flushed to the media. Used with
171 * REQ_FLUSH and REQ_FUA
173 unsigned int l2p_update; /* l2p update point - next entry for
174 * which l2p mapping will be updated to
175 * contain a device ppa address (instead
178 unsigned int nr_entries; /* Number of entries in write buffer -
179 * must be a power of two
181 unsigned int seg_size; /* Size of the data segments being
182 * stored on each entry. Typically this
186 struct list_head pages; /* List of data pages */
188 spinlock_t w_lock; /* Write lock */
189 spinlock_t s_lock; /* Sync lock */
191 #ifdef CONFIG_NVM_DEBUG
192 atomic_t inflight_flush_point; /* Not served REQ_FLUSH | REQ_FUA */
196 #define PBLK_RECOVERY_SECTORS 16
199 struct ppa_addr bppa;
201 u8 *bb_list; /* Bad block list for LUN. Only used on
202 * bring up. Bad blocks are managed
203 * within lines on run-time.
206 struct semaphore wr_sem;
210 struct pblk_line *line;
212 u64 paddr_list[PBLK_MAX_REQ_ADDRS];
213 u64 lba_list[PBLK_MAX_REQ_ADDRS];
216 struct list_head list;
220 /* These states are not protected by a lock since (i) they are in the
221 * fast path, and (ii) they are not critical.
227 struct task_struct *gc_ts;
228 struct task_struct *gc_writer_ts;
229 struct task_struct *gc_reader_ts;
231 struct workqueue_struct *gc_line_reader_wq;
232 struct workqueue_struct *gc_reader_wq;
234 struct timer_list gc_timer;
236 struct semaphore gc_sem;
237 atomic_t read_inflight_gc; /* Number of lines with inflight GC reads */
238 atomic_t pipeline_gc; /* Number of lines in the GC pipeline -
239 * started reads to finished writes
243 struct list_head w_list;
244 struct list_head r_list;
252 unsigned int high; /* Upper threshold for rate limiter (free run -
253 * user I/O rate limiter
255 unsigned int low; /* Lower threshold for rate limiter (user I/O
256 * rate limiter - stall)
258 unsigned int high_pw; /* High rounded up as a power of 2 */
260 #define PBLK_USER_HIGH_THRS 8 /* Begin write limit at 12% available blks */
261 #define PBLK_USER_LOW_THRS 10 /* Aggressive GC at 10% available blocks */
263 int rb_windows_pw; /* Number of rate windows in the write buffer
264 * given as a power-of-2. This guarantees that
265 * when user I/O is being rate limited, there
266 * will be reserved enough space for the GC to
267 * place its payload. A window is of
268 * pblk->max_write_pgs size, which in NVMe is
271 int rb_budget; /* Total number of entries available for I/O */
272 int rb_user_max; /* Max buffer entries available for user I/O */
273 int rb_gc_max; /* Max buffer entries available for GC I/O */
274 int rb_gc_rsv; /* Reserved buffer entries for GC I/O */
275 int rb_state; /* Rate-limiter current state */
276 int rb_max_io; /* Maximum size for an I/O giving the config */
278 atomic_t rb_user_cnt; /* User I/O buffer counter */
279 atomic_t rb_gc_cnt; /* GC I/O buffer counter */
280 atomic_t rb_space; /* Space limit in case of reaching capacity */
282 int rsv_blocks; /* Reserved blocks for GC */
287 struct timer_list u_timer;
289 unsigned long long nr_secs;
290 unsigned long total_blocks;
291 atomic_t free_blocks;
294 #define PBLK_LINE_EMPTY (~0U)
298 PBLK_LINETYPE_FREE = 0,
299 PBLK_LINETYPE_LOG = 1,
300 PBLK_LINETYPE_DATA = 2,
303 PBLK_LINESTATE_FREE = 10,
304 PBLK_LINESTATE_OPEN = 11,
305 PBLK_LINESTATE_CLOSED = 12,
306 PBLK_LINESTATE_GC = 13,
307 PBLK_LINESTATE_BAD = 14,
308 PBLK_LINESTATE_CORRUPT = 15,
311 PBLK_LINEGC_NONE = 20,
312 PBLK_LINEGC_EMPTY = 21,
313 PBLK_LINEGC_LOW = 22,
314 PBLK_LINEGC_MID = 23,
315 PBLK_LINEGC_HIGH = 24,
316 PBLK_LINEGC_FULL = 25,
319 #define PBLK_MAGIC 0x70626c6b /*pblk*/
320 #define SMETA_VERSION cpu_to_le16(1)
324 __le32 identifier; /* pblk identifier */
325 __u8 uuid[16]; /* instance uuid */
326 __le16 type; /* line type */
327 __le16 version; /* type version */
328 __le32 id; /* line id for current line */
332 struct line_header header;
334 __le32 crc; /* Full structure including struct crc */
335 /* Previous line metadata */
336 __le32 prev_id; /* Line id for previous line */
338 /* Current line metadata */
339 __le64 seq_nr; /* Sequence number for current line */
342 __le32 window_wr_lun; /* Number of parallel LUNs to write */
350 * Metadata layout in media:
352 * 1. struct line_emeta
353 * 2. bad block bitmap (u64 * window_wr_lun)
354 * Mid sectors (start at lbas_sector):
355 * 3. nr_lbas (u64) forming lba list
356 * Last sectors (start at vsc_sector):
357 * 4. u32 valid sector count (vsc) for all lines (~0U: free line)
360 struct line_header header;
362 __le32 crc; /* Full structure including struct crc */
364 /* Previous line metadata */
365 __le32 prev_id; /* Line id for prev line */
367 /* Current line metadata */
368 __le64 seq_nr; /* Sequence number for current line */
371 __le32 window_wr_lun; /* Number of parallel LUNs to write */
373 /* Bookkeeping for recovery */
374 __le32 next_id; /* Line id for next line */
375 __le64 nr_lbas; /* Number of lbas mapped in line */
376 __le64 nr_valid_lbas; /* Number of valid lbas mapped in line */
377 __le64 bb_bitmap[]; /* Updated bad block bitmap for line */
381 struct line_emeta *buf; /* emeta buffer in media format */
382 int mem; /* Write offset - points to next
383 * writable entry in memory
385 atomic_t sync; /* Synced - backpointer that signals the
386 * last entry that has been successfully
389 unsigned int nr_entries; /* Number of emeta entries */
393 struct line_smeta *buf; /* smeta buffer in persistent format */
398 unsigned int id; /* Line number corresponds to the
401 unsigned int seq_nr; /* Unique line sequence number */
403 int state; /* PBLK_LINESTATE_X */
404 int type; /* PBLK_LINETYPE_X */
405 int gc_group; /* PBLK_LINEGC_X */
406 struct list_head list; /* Free, GC lists */
408 unsigned long *lun_bitmap; /* Bitmap for LUNs mapped in line */
410 struct pblk_smeta *smeta; /* Start metadata */
411 struct pblk_emeta *emeta; /* End medatada */
413 int meta_line; /* Metadata line id */
414 int meta_distance; /* Distance between data and metadata */
416 u64 smeta_ssec; /* Sector where smeta starts */
417 u64 emeta_ssec; /* Sector where emeta starts */
419 unsigned int sec_in_line; /* Number of usable secs in line */
421 atomic_t blk_in_line; /* Number of good blocks in line */
422 unsigned long *blk_bitmap; /* Bitmap for valid/invalid blocks */
423 unsigned long *erase_bitmap; /* Bitmap for erased blocks */
425 unsigned long *map_bitmap; /* Bitmap for mapped sectors in line */
426 unsigned long *invalid_bitmap; /* Bitmap for invalid sectors in line */
428 atomic_t left_eblks; /* Blocks left for erasing */
429 atomic_t left_seblks; /* Blocks left for sync erasing */
431 int left_msecs; /* Sectors left for mapping */
432 unsigned int cur_sec; /* Sector map pointer */
433 unsigned int nr_valid_lbas; /* Number of valid lbas in line */
435 __le32 *vsc; /* Valid sector count in line */
437 struct kref ref; /* Write buffer L2P references */
439 spinlock_t lock; /* Necessary for invalid_bitmap only */
442 #define PBLK_DATA_LINES 4
445 PBLK_KMALLOC_META = 1,
446 PBLK_VMALLOC_META = 2,
450 PBLK_EMETA_TYPE_HEADER = 1, /* struct line_emeta first sector */
451 PBLK_EMETA_TYPE_LLBA = 2, /* lba list - type: __le64 */
452 PBLK_EMETA_TYPE_VSC = 3, /* vsc list - type: __le32 */
455 struct pblk_line_mgmt {
456 int nr_lines; /* Total number of full lines */
457 int nr_free_lines; /* Number of full lines in free list */
459 /* Free lists - use free_lock */
460 struct list_head free_list; /* Full lines ready to use */
461 struct list_head corrupt_list; /* Full lines corrupted */
462 struct list_head bad_list; /* Full lines bad */
464 /* GC lists - use gc_lock */
465 struct list_head *gc_lists[PBLK_GC_NR_LISTS];
466 struct list_head gc_high_list; /* Full lines ready to GC, high isc */
467 struct list_head gc_mid_list; /* Full lines ready to GC, mid isc */
468 struct list_head gc_low_list; /* Full lines ready to GC, low isc */
470 struct list_head gc_full_list; /* Full lines ready to GC, no valid */
471 struct list_head gc_empty_list; /* Full lines close, all valid */
473 struct pblk_line *log_line; /* Current FTL log line */
474 struct pblk_line *data_line; /* Current data line */
475 struct pblk_line *log_next; /* Next FTL log line */
476 struct pblk_line *data_next; /* Next data line */
478 struct list_head emeta_list; /* Lines queued to schedule emeta */
480 __le32 *vsc_list; /* Valid sector counts for all lines */
482 /* Metadata allocation type: VMALLOC | KMALLOC */
483 int emeta_alloc_type;
485 /* Pre-allocated metadata for data lines */
486 struct pblk_smeta *sline_meta[PBLK_DATA_LINES];
487 struct pblk_emeta *eline_meta[PBLK_DATA_LINES];
488 unsigned long meta_bitmap;
490 /* Helpers for fast bitmap calculations */
491 unsigned long *bb_template;
492 unsigned long *bb_aux;
494 unsigned long d_seq_nr; /* Data line unique sequence number */
495 unsigned long l_seq_nr; /* Log line unique sequence number */
497 spinlock_t free_lock;
498 spinlock_t close_lock;
502 struct pblk_line_meta {
503 unsigned int smeta_len; /* Total length for smeta */
504 unsigned int smeta_sec; /* Sectors needed for smeta */
506 unsigned int emeta_len[4]; /* Lengths for emeta:
508 * [1]: struct line_emeta length
509 * [2]: L2P portion length
510 * [3]: vsc list length
512 unsigned int emeta_sec[4]; /* Sectors needed for emeta. Same layout
516 unsigned int emeta_bb; /* Boundary for bb that affects emeta */
518 unsigned int vsc_list_len; /* Length for vsc list */
519 unsigned int sec_bitmap_len; /* Length for sector bitmap in line */
520 unsigned int blk_bitmap_len; /* Length for block bitmap in line */
521 unsigned int lun_bitmap_len; /* Length for lun bitmap in line */
523 unsigned int blk_per_line; /* Number of blocks in a full line */
524 unsigned int sec_per_line; /* Number of sectors in a line */
525 unsigned int dsec_per_line; /* Number of data sectors in a line */
526 unsigned int min_blk_line; /* Min. number of good blocks in line */
528 unsigned int mid_thrs; /* Threshold for GC mid list */
529 unsigned int high_thrs; /* Threshold for GC high list */
531 unsigned int meta_distance; /* Distance between data and metadata */
534 struct pblk_addr_format {
550 PBLK_STATE_RUNNING = 0,
551 PBLK_STATE_STOPPING = 1,
552 PBLK_STATE_RECOVERING = 2,
553 PBLK_STATE_STOPPED = 3,
557 struct nvm_tgt_dev *dev;
558 struct gendisk *disk;
562 struct pblk_lun *luns;
564 struct pblk_line *lines; /* Line array */
565 struct pblk_line_mgmt l_mg; /* Line management */
566 struct pblk_line_meta lm; /* Line metadata */
569 struct pblk_addr_format ppaf;
573 int state; /* pblk line state */
575 int min_write_pgs; /* Minimum amount of pages required by controller */
576 int max_write_pgs; /* Maximum amount of pages supported by controller */
577 int pgs_in_buffer; /* Number of pages that need to be held in buffer to
578 * guarantee successful reads.
581 sector_t capacity; /* Device capacity when bad blocks are subtracted */
582 int over_pct; /* Percentage of device used for over-provisioning */
584 /* pblk provisioning values. Used by rate limiter */
589 unsigned char instance_uuid[16];
590 #ifdef CONFIG_NVM_DEBUG
591 /* All debug counters apply to 4kb sector I/Os */
592 atomic_long_t inflight_writes; /* Inflight writes (user and gc) */
593 atomic_long_t padded_writes; /* Sectors padded due to flush/fua */
594 atomic_long_t padded_wb; /* Sectors padded in write buffer */
595 atomic_long_t nr_flush; /* Number of flush/fua I/O */
596 atomic_long_t req_writes; /* Sectors stored on write buffer */
597 atomic_long_t sub_writes; /* Sectors submitted from buffer */
598 atomic_long_t sync_writes; /* Sectors synced to media */
599 atomic_long_t inflight_reads; /* Inflight sector read requests */
600 atomic_long_t cache_reads; /* Read requests that hit the cache */
601 atomic_long_t sync_reads; /* Completed sector read requests */
602 atomic_long_t recov_writes; /* Sectors submitted from recovery */
603 atomic_long_t recov_gc_writes; /* Sectors submitted from write GC */
604 atomic_long_t recov_gc_reads; /* Sectors submitted from read GC */
609 atomic_long_t read_failed;
610 atomic_long_t read_empty;
611 atomic_long_t read_high_ecc;
612 atomic_long_t read_failed_gc;
613 atomic_long_t write_failed;
614 atomic_long_t erase_failed;
616 atomic_t inflight_io; /* General inflight I/O counter */
618 struct task_struct *writer_ts;
620 /* Simple translation map of logical addresses to physical addresses.
621 * The logical addresses is known by the host system, while the physical
622 * addresses are used when writing to the disk block device.
624 unsigned char *trans_map;
625 spinlock_t trans_lock;
627 struct list_head compl_list;
629 mempool_t *page_bio_pool;
630 mempool_t *gen_ws_pool;
632 mempool_t *r_rq_pool;
633 mempool_t *w_rq_pool;
634 mempool_t *e_rq_pool;
636 struct workqueue_struct *close_wq;
637 struct workqueue_struct *bb_wq;
638 struct workqueue_struct *r_end_wq;
640 struct timer_list wtimer;
645 struct pblk_line_ws {
647 struct pblk_line *line;
649 struct work_struct ws;
652 #define pblk_g_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_g_ctx))
653 #define pblk_w_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_c_ctx))
656 * pblk ring buffer operations
658 int pblk_rb_init(struct pblk_rb *rb, struct pblk_rb_entry *rb_entry_base,
659 unsigned int power_size, unsigned int power_seg_sz);
660 unsigned int pblk_rb_calculate_size(unsigned int nr_entries);
661 void *pblk_rb_entries_ref(struct pblk_rb *rb);
662 int pblk_rb_may_write_user(struct pblk_rb *rb, struct bio *bio,
663 unsigned int nr_entries, unsigned int *pos);
664 int pblk_rb_may_write_gc(struct pblk_rb *rb, unsigned int nr_entries,
666 void pblk_rb_write_entry_user(struct pblk_rb *rb, void *data,
667 struct pblk_w_ctx w_ctx, unsigned int pos);
668 void pblk_rb_write_entry_gc(struct pblk_rb *rb, void *data,
669 struct pblk_w_ctx w_ctx, struct pblk_line *line,
670 u64 paddr, unsigned int pos);
671 struct pblk_w_ctx *pblk_rb_w_ctx(struct pblk_rb *rb, unsigned int pos);
672 void pblk_rb_flush(struct pblk_rb *rb);
674 void pblk_rb_sync_l2p(struct pblk_rb *rb);
675 unsigned int pblk_rb_read_to_bio(struct pblk_rb *rb, struct nvm_rq *rqd,
676 unsigned int pos, unsigned int nr_entries,
678 unsigned int pblk_rb_read_to_bio_list(struct pblk_rb *rb, struct bio *bio,
679 struct list_head *list,
681 int pblk_rb_copy_to_bio(struct pblk_rb *rb, struct bio *bio, sector_t lba,
682 struct ppa_addr ppa, int bio_iter, bool advanced_bio);
683 unsigned int pblk_rb_read_commit(struct pblk_rb *rb, unsigned int entries);
685 unsigned int pblk_rb_sync_init(struct pblk_rb *rb, unsigned long *flags);
686 unsigned int pblk_rb_sync_advance(struct pblk_rb *rb, unsigned int nr_entries);
687 struct pblk_rb_entry *pblk_rb_sync_scan_entry(struct pblk_rb *rb,
688 struct ppa_addr *ppa);
689 void pblk_rb_sync_end(struct pblk_rb *rb, unsigned long *flags);
690 unsigned int pblk_rb_flush_point_count(struct pblk_rb *rb);
692 unsigned int pblk_rb_read_count(struct pblk_rb *rb);
693 unsigned int pblk_rb_sync_count(struct pblk_rb *rb);
694 unsigned int pblk_rb_wrap_pos(struct pblk_rb *rb, unsigned int pos);
696 int pblk_rb_tear_down_check(struct pblk_rb *rb);
697 int pblk_rb_pos_oob(struct pblk_rb *rb, u64 pos);
698 void pblk_rb_data_free(struct pblk_rb *rb);
699 ssize_t pblk_rb_sysfs(struct pblk_rb *rb, char *buf);
704 struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int type);
705 void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int type);
706 void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write);
707 int pblk_setup_w_rec_rq(struct pblk *pblk, struct nvm_rq *rqd,
708 struct pblk_c_ctx *c_ctx);
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_io_sync(struct pblk *pblk, struct nvm_rq *rqd);
714 int pblk_submit_meta_io(struct pblk *pblk, struct pblk_line *meta_line);
715 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
716 unsigned int nr_secs, unsigned int len,
717 int alloc_type, gfp_t gfp_mask);
718 struct pblk_line *pblk_line_get(struct pblk *pblk);
719 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk);
720 struct pblk_line *pblk_line_replace_data(struct pblk *pblk);
721 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line);
722 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line);
723 struct pblk_line *pblk_line_get_data(struct pblk *pblk);
724 struct pblk_line *pblk_line_get_erase(struct pblk *pblk);
725 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line);
726 int pblk_line_is_full(struct pblk_line *line);
727 void pblk_line_free(struct pblk *pblk, struct pblk_line *line);
728 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line);
729 void pblk_line_close(struct pblk *pblk, struct pblk_line *line);
730 void pblk_line_close_ws(struct work_struct *work);
731 void pblk_pipeline_stop(struct pblk *pblk);
732 void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
733 void (*work)(struct work_struct *), gfp_t gfp_mask,
734 struct workqueue_struct *wq);
735 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line);
736 int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line);
737 int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line,
739 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr erase_ppa);
740 void pblk_line_put(struct kref *ref);
741 void pblk_line_put_wq(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_io_sync(struct nvm_rq *rqd);
756 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
758 void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
760 void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa);
761 void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
763 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa);
764 void pblk_update_map_cache(struct pblk *pblk, sector_t lba,
765 struct ppa_addr ppa);
766 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
767 struct ppa_addr ppa, struct ppa_addr entry_line);
768 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa,
769 struct pblk_line *gc_line, u64 paddr);
770 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
771 u64 *lba_list, int nr_secs);
772 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
773 sector_t blba, int nr_secs);
776 * pblk user I/O write path
778 int pblk_write_to_cache(struct pblk *pblk, struct bio *bio,
779 unsigned long flags);
780 int pblk_write_gc_to_cache(struct pblk *pblk, struct pblk_gc_rq *gc_rq);
785 void pblk_map_erase_rq(struct pblk *pblk, struct nvm_rq *rqd,
786 unsigned int sentry, unsigned long *lun_bitmap,
787 unsigned int valid_secs, struct ppa_addr *erase_ppa);
788 void pblk_map_rq(struct pblk *pblk, struct nvm_rq *rqd, unsigned int sentry,
789 unsigned long *lun_bitmap, unsigned int valid_secs,
795 int pblk_write_ts(void *data);
796 void pblk_write_timer_fn(struct timer_list *t);
797 void pblk_write_should_kick(struct pblk *pblk);
802 extern struct bio_set *pblk_bio_set;
803 int pblk_submit_read(struct pblk *pblk, struct bio *bio);
804 int pblk_submit_read_gc(struct pblk *pblk, struct pblk_gc_rq *gc_rq);
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 int pblk_recov_check_emeta(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_RQ_QD 128 /* Queue depth for inflight GC requests */
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_should_kick(struct pblk *pblk);
829 void pblk_gc_free_full_lines(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 void pblk_rl_update_rates(struct pblk_rl *rl);
840 int pblk_rl_high_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_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 int pblk_rl_is_limit(struct pblk_rl *rl);
856 int pblk_sysfs_init(struct gendisk *tdisk);
857 void pblk_sysfs_exit(struct gendisk *tdisk);
859 static inline void *pblk_malloc(size_t size, int type, gfp_t flags)
861 if (type == PBLK_KMALLOC_META)
862 return kmalloc(size, flags);
863 return vmalloc(size);
866 static inline void pblk_mfree(void *ptr, int type)
868 if (type == PBLK_KMALLOC_META)
874 static inline struct nvm_rq *nvm_rq_from_c_ctx(void *c_ctx)
876 return c_ctx - sizeof(struct nvm_rq);
879 static inline void *emeta_to_bb(struct line_emeta *emeta)
881 return emeta->bb_bitmap;
884 static inline void *emeta_to_lbas(struct pblk *pblk, struct line_emeta *emeta)
886 return ((void *)emeta + pblk->lm.emeta_len[1]);
889 static inline void *emeta_to_vsc(struct pblk *pblk, struct line_emeta *emeta)
891 return (emeta_to_lbas(pblk, emeta) + pblk->lm.emeta_len[2]);
894 static inline int pblk_line_vsc(struct pblk_line *line)
896 return le32_to_cpu(*line->vsc);
899 #define NVM_MEM_PAGE_WRITE (8)
901 static inline int pblk_pad_distance(struct pblk *pblk)
903 struct nvm_tgt_dev *dev = pblk->dev;
904 struct nvm_geo *geo = &dev->geo;
906 return NVM_MEM_PAGE_WRITE * geo->all_luns * geo->sec_per_pl;
909 static inline int pblk_ppa_to_line(struct ppa_addr p)
914 static inline int pblk_ppa_to_pos(struct nvm_geo *geo, struct ppa_addr p)
916 return p.g.lun * geo->nr_chnls + p.g.ch;
919 static inline struct ppa_addr addr_to_gen_ppa(struct pblk *pblk, u64 paddr,
926 ppa.g.pg = (paddr & pblk->ppaf.pg_mask) >> pblk->ppaf.pg_offset;
927 ppa.g.lun = (paddr & pblk->ppaf.lun_mask) >> pblk->ppaf.lun_offset;
928 ppa.g.ch = (paddr & pblk->ppaf.ch_mask) >> pblk->ppaf.ch_offset;
929 ppa.g.pl = (paddr & pblk->ppaf.pln_mask) >> pblk->ppaf.pln_offset;
930 ppa.g.sec = (paddr & pblk->ppaf.sec_mask) >> pblk->ppaf.sec_offset;
935 static inline u64 pblk_dev_ppa_to_line_addr(struct pblk *pblk,
940 paddr = (u64)p.g.pg << pblk->ppaf.pg_offset;
941 paddr |= (u64)p.g.lun << pblk->ppaf.lun_offset;
942 paddr |= (u64)p.g.ch << pblk->ppaf.ch_offset;
943 paddr |= (u64)p.g.pl << pblk->ppaf.pln_offset;
944 paddr |= (u64)p.g.sec << pblk->ppaf.sec_offset;
949 static inline struct ppa_addr pblk_ppa32_to_ppa64(struct pblk *pblk, u32 ppa32)
951 struct ppa_addr ppa64;
956 ppa64.ppa = ADDR_EMPTY;
957 } else if (ppa32 & (1U << 31)) {
958 ppa64.c.line = ppa32 & ((~0U) >> 1);
959 ppa64.c.is_cached = 1;
961 ppa64.g.blk = (ppa32 & pblk->ppaf.blk_mask) >>
962 pblk->ppaf.blk_offset;
963 ppa64.g.pg = (ppa32 & pblk->ppaf.pg_mask) >>
964 pblk->ppaf.pg_offset;
965 ppa64.g.lun = (ppa32 & pblk->ppaf.lun_mask) >>
966 pblk->ppaf.lun_offset;
967 ppa64.g.ch = (ppa32 & pblk->ppaf.ch_mask) >>
968 pblk->ppaf.ch_offset;
969 ppa64.g.pl = (ppa32 & pblk->ppaf.pln_mask) >>
970 pblk->ppaf.pln_offset;
971 ppa64.g.sec = (ppa32 & pblk->ppaf.sec_mask) >>
972 pblk->ppaf.sec_offset;
978 static inline u32 pblk_ppa64_to_ppa32(struct pblk *pblk, struct ppa_addr ppa64)
982 if (ppa64.ppa == ADDR_EMPTY) {
984 } else if (ppa64.c.is_cached) {
985 ppa32 |= ppa64.c.line;
988 ppa32 |= ppa64.g.blk << pblk->ppaf.blk_offset;
989 ppa32 |= ppa64.g.pg << pblk->ppaf.pg_offset;
990 ppa32 |= ppa64.g.lun << pblk->ppaf.lun_offset;
991 ppa32 |= ppa64.g.ch << pblk->ppaf.ch_offset;
992 ppa32 |= ppa64.g.pl << pblk->ppaf.pln_offset;
993 ppa32 |= ppa64.g.sec << pblk->ppaf.sec_offset;
999 static inline struct ppa_addr pblk_trans_map_get(struct pblk *pblk,
1002 struct ppa_addr ppa;
1004 if (pblk->ppaf_bitsize < 32) {
1005 u32 *map = (u32 *)pblk->trans_map;
1007 ppa = pblk_ppa32_to_ppa64(pblk, map[lba]);
1009 struct ppa_addr *map = (struct ppa_addr *)pblk->trans_map;
1017 static inline void pblk_trans_map_set(struct pblk *pblk, sector_t lba,
1018 struct ppa_addr ppa)
1020 if (pblk->ppaf_bitsize < 32) {
1021 u32 *map = (u32 *)pblk->trans_map;
1023 map[lba] = pblk_ppa64_to_ppa32(pblk, ppa);
1025 u64 *map = (u64 *)pblk->trans_map;
1031 static inline int pblk_ppa_empty(struct ppa_addr ppa_addr)
1033 return (ppa_addr.ppa == ADDR_EMPTY);
1036 static inline void pblk_ppa_set_empty(struct ppa_addr *ppa_addr)
1038 ppa_addr->ppa = ADDR_EMPTY;
1041 static inline bool pblk_ppa_comp(struct ppa_addr lppa, struct ppa_addr rppa)
1043 if (lppa.ppa == rppa.ppa)
1049 static inline int pblk_addr_in_cache(struct ppa_addr ppa)
1051 return (ppa.ppa != ADDR_EMPTY && ppa.c.is_cached);
1054 static inline int pblk_addr_to_cacheline(struct ppa_addr ppa)
1059 static inline struct ppa_addr pblk_cacheline_to_addr(int addr)
1069 static inline u32 pblk_calc_meta_header_crc(struct pblk *pblk,
1070 struct line_header *header)
1074 crc = crc32_le(crc, (unsigned char *)header + sizeof(crc),
1075 sizeof(struct line_header) - sizeof(crc));
1080 static inline u32 pblk_calc_smeta_crc(struct pblk *pblk,
1081 struct line_smeta *smeta)
1083 struct pblk_line_meta *lm = &pblk->lm;
1086 crc = crc32_le(crc, (unsigned char *)smeta +
1087 sizeof(struct line_header) + sizeof(crc),
1089 sizeof(struct line_header) - sizeof(crc));
1094 static inline u32 pblk_calc_emeta_crc(struct pblk *pblk,
1095 struct line_emeta *emeta)
1097 struct pblk_line_meta *lm = &pblk->lm;
1100 crc = crc32_le(crc, (unsigned char *)emeta +
1101 sizeof(struct line_header) + sizeof(crc),
1103 sizeof(struct line_header) - sizeof(crc));
1108 static inline int pblk_set_progr_mode(struct pblk *pblk, int type)
1110 struct nvm_tgt_dev *dev = pblk->dev;
1111 struct nvm_geo *geo = &dev->geo;
1114 flags = geo->plane_mode >> 1;
1116 if (type == PBLK_WRITE)
1117 flags |= NVM_IO_SCRAMBLE_ENABLE;
1123 PBLK_READ_RANDOM = 0,
1124 PBLK_READ_SEQUENTIAL = 1,
1127 static inline int pblk_set_read_mode(struct pblk *pblk, int type)
1129 struct nvm_tgt_dev *dev = pblk->dev;
1130 struct nvm_geo *geo = &dev->geo;
1133 flags = NVM_IO_SUSPEND | NVM_IO_SCRAMBLE_ENABLE;
1134 if (type == PBLK_READ_SEQUENTIAL)
1135 flags |= geo->plane_mode >> 1;
1140 static inline int pblk_io_aligned(struct pblk *pblk, int nr_secs)
1142 return !(nr_secs % pblk->min_write_pgs);
1145 #ifdef CONFIG_NVM_DEBUG
1146 static inline void print_ppa(struct ppa_addr *p, char *msg, int error)
1148 if (p->c.is_cached) {
1149 pr_err("ppa: (%s: %x) cache line: %llu\n",
1150 msg, error, (u64)p->c.line);
1152 pr_err("ppa: (%s: %x):ch:%d,lun:%d,blk:%d,pg:%d,pl:%d,sec:%d\n",
1154 p->g.ch, p->g.lun, p->g.blk,
1155 p->g.pg, p->g.pl, p->g.sec);
1159 static inline void pblk_print_failed_rqd(struct pblk *pblk, struct nvm_rq *rqd,
1164 if (rqd->nr_ppas == 1) {
1165 print_ppa(&rqd->ppa_addr, "rqd", error);
1169 while ((bit = find_next_bit((void *)&rqd->ppa_status, rqd->nr_ppas,
1170 bit + 1)) < rqd->nr_ppas) {
1171 print_ppa(&rqd->ppa_list[bit], "rqd", error);
1174 pr_err("error:%d, ppa_status:%llx\n", error, rqd->ppa_status);
1177 static inline int pblk_boundary_ppa_checks(struct nvm_tgt_dev *tgt_dev,
1178 struct ppa_addr *ppas, int nr_ppas)
1180 struct nvm_geo *geo = &tgt_dev->geo;
1181 struct ppa_addr *ppa;
1184 for (i = 0; i < nr_ppas; i++) {
1187 if (!ppa->c.is_cached &&
1188 ppa->g.ch < geo->nr_chnls &&
1189 ppa->g.lun < geo->nr_luns &&
1190 ppa->g.pl < geo->nr_planes &&
1191 ppa->g.blk < geo->nr_chks &&
1192 ppa->g.pg < geo->ws_per_chk &&
1193 ppa->g.sec < geo->sec_per_pg)
1196 print_ppa(ppa, "boundary", i);
1203 static inline int pblk_check_io(struct pblk *pblk, struct nvm_rq *rqd)
1205 struct nvm_tgt_dev *dev = pblk->dev;
1206 struct ppa_addr *ppa_list;
1208 ppa_list = (rqd->nr_ppas > 1) ? rqd->ppa_list : &rqd->ppa_addr;
1210 if (pblk_boundary_ppa_checks(dev, ppa_list, rqd->nr_ppas)) {
1215 if (rqd->opcode == NVM_OP_PWRITE) {
1216 struct pblk_line *line;
1217 struct ppa_addr ppa;
1220 for (i = 0; i < rqd->nr_ppas; i++) {
1222 line = &pblk->lines[pblk_ppa_to_line(ppa)];
1224 spin_lock(&line->lock);
1225 if (line->state != PBLK_LINESTATE_OPEN) {
1226 pr_err("pblk: bad ppa: line:%d,state:%d\n",
1227 line->id, line->state);
1229 spin_unlock(&line->lock);
1232 spin_unlock(&line->lock);
1240 static inline int pblk_boundary_paddr_checks(struct pblk *pblk, u64 paddr)
1242 struct pblk_line_meta *lm = &pblk->lm;
1244 if (paddr > lm->sec_per_line)
1250 static inline unsigned int pblk_get_bi_idx(struct bio *bio)
1252 return bio->bi_iter.bi_idx;
1255 static inline sector_t pblk_get_lba(struct bio *bio)
1257 return bio->bi_iter.bi_sector / NR_PHY_IN_LOG;
1260 static inline unsigned int pblk_get_secs(struct bio *bio)
1262 return bio->bi_iter.bi_size / PBLK_EXPOSED_PAGE_SIZE;
1265 static inline void pblk_setup_uuid(struct pblk *pblk)
1270 memcpy(pblk->instance_uuid, uuid.b, 16);
1272 #endif /* PBLK_H_ */