mm: update get_user_pages_longterm to migrate pages allocated from CMA region
[linux-2.6-block.git] / drivers / lightnvm / pblk.h
CommitLineData
02a1520d 1/* SPDX-License-Identifier: GPL-2.0 */
a4bd217b
JG
2/*
3 * Copyright (C) 2015 IT University of Copenhagen (rrpc.h)
4 * Copyright (C) 2016 CNEX Labs
5 * Initial release: Matias Bjorling <matias@cnexlabs.com>
6 * Write buffering: Javier Gonzalez <javier@cnexlabs.com>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License version
10 * 2 as published by the Free Software Foundation.
11 *
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
16 *
17 * Implementation of a Physical Block-device target for Open-channel SSDs.
18 *
19 */
20
21#ifndef PBLK_H_
22#define PBLK_H_
23
24#include <linux/blkdev.h>
25#include <linux/blk-mq.h>
26#include <linux/bio.h>
27#include <linux/module.h>
28#include <linux/kthread.h>
29#include <linux/vmalloc.h>
30#include <linux/crc32.h>
31#include <linux/uuid.h>
32
33#include <linux/lightnvm.h>
34
35/* Run only GC if less than 1/X blocks are free */
36#define GC_LIMIT_INVERSE 5
37#define GC_TIME_MSECS 1000
38
39#define PBLK_SECTOR (512)
40#define PBLK_EXPOSED_PAGE_SIZE (4096)
a4bd217b 41
ef576494
JG
42#define PBLK_NR_CLOSE_JOBS (4)
43
a4bd217b
JG
44#define PBLK_CACHE_NAME_LEN (DISK_NAME_LEN + 16)
45
46#define PBLK_COMMAND_TIMEOUT_MS 30000
47
48/* Max 512 LUNs per device */
49#define PBLK_MAX_LUNS_BITMAP (4)
50
51#define NR_PHY_IN_LOG (PBLK_EXPOSED_PAGE_SIZE / PBLK_SECTOR)
52
0d880398 53/* Static pool sizes */
b84ae4a8
JG
54#define PBLK_GEN_WS_POOL_SIZE (2)
55
e5392739
JG
56#define PBLK_DEFAULT_OP (11)
57
e2cddf20
JG
58enum {
59 PBLK_READ = READ,
60 PBLK_WRITE = WRITE,/* Write from write buffer */
61 PBLK_WRITE_INT, /* Internal write - no write buffer */
8f554597 62 PBLK_READ_RECOV, /* Recovery read - errors allowed */
e2cddf20
JG
63 PBLK_ERASE,
64};
65
a4bd217b
JG
66enum {
67 /* IO Types */
68 PBLK_IOTYPE_USER = 1 << 0,
69 PBLK_IOTYPE_GC = 1 << 1,
70
71 /* Write buffer flags */
72 PBLK_FLUSH_ENTRY = 1 << 2,
73 PBLK_WRITTEN_DATA = 1 << 3,
74 PBLK_SUBMITTED_ENTRY = 1 << 4,
75 PBLK_WRITABLE_ENTRY = 1 << 5,
76};
77
78enum {
79 PBLK_BLK_ST_OPEN = 0x1,
80 PBLK_BLK_ST_CLOSED = 0x2,
81};
82
4209c31c
HH
83enum {
84 PBLK_CHUNK_RESET_START,
85 PBLK_CHUNK_RESET_DONE,
86 PBLK_CHUNK_RESET_FAILED,
87};
88
b20ba1bc
JG
89struct pblk_sec_meta {
90 u64 reserved;
91 __le64 lba;
92};
93
a4bd217b
JG
94/* The number of GC lists and the rate-limiter states go together. This way the
95 * rate-limiter can dictate how much GC is needed based on resource utilization.
96 */
48b8d208 97#define PBLK_GC_NR_LISTS 4
a4bd217b
JG
98
99enum {
48b8d208
HH
100 PBLK_RL_OFF = 0,
101 PBLK_RL_WERR = 1,
102 PBLK_RL_HIGH = 2,
103 PBLK_RL_MID = 3,
104 PBLK_RL_LOW = 4
a4bd217b
JG
105};
106
afdc23c9 107#define pblk_dma_ppa_size (sizeof(u64) * NVM_MAX_VLBA)
a4bd217b 108
084ec9ba 109/* write buffer completion context */
a4bd217b
JG
110struct pblk_c_ctx {
111 struct list_head list; /* Head for out-of-order completion */
112
113 unsigned long *lun_bitmap; /* Luns used on current request */
114 unsigned int sentry;
115 unsigned int nr_valid;
116 unsigned int nr_padded;
117};
118
a4809fee 119/* read context */
084ec9ba
JG
120struct pblk_g_ctx {
121 void *private;
998ba629 122 unsigned long start_time;
a4809fee 123 u64 lba;
a4bd217b
JG
124};
125
11f6ad69
HL
126/* partial read context */
127struct pblk_pr_ctx {
128 struct bio *orig_bio;
129 DECLARE_BITMAP(bitmap, NVM_MAX_VLBA);
130 unsigned int orig_nr_secs;
131 unsigned int bio_init_idx;
132 void *ppa_ptr;
133 dma_addr_t dma_ppa_list;
dd439496
IK
134 __le64 lba_list_mem[NVM_MAX_VLBA];
135 __le64 lba_list_media[NVM_MAX_VLBA];
11f6ad69
HL
136};
137
ee8d5c1a
JG
138/* Pad context */
139struct pblk_pad_rq {
140 struct pblk *pblk;
141 struct completion wait;
142 struct kref ref;
143};
144
a4bd217b
JG
145/* Recovery context */
146struct pblk_rec_ctx {
147 struct pblk *pblk;
148 struct nvm_rq *rqd;
a4bd217b
JG
149 struct work_struct ws_rec;
150};
151
152/* Write context */
153struct pblk_w_ctx {
154 struct bio_list bios; /* Original bios - used for completion
155 * in REQ_FUA, REQ_FLUSH case
156 */
ef697902 157 u64 lba; /* Logic addr. associated with entry */
a4bd217b
JG
158 struct ppa_addr ppa; /* Physic addr. associated with entry */
159 int flags; /* Write context flags */
160};
161
162struct pblk_rb_entry {
163 struct ppa_addr cacheline; /* Cacheline for this entry */
164 void *data; /* Pointer to data on this entry */
165 struct pblk_w_ctx w_ctx; /* Context for this entry */
166 struct list_head index; /* List head to enable indexes */
167};
168
169#define EMPTY_ENTRY (~0U)
170
171struct pblk_rb_pages {
172 struct page *pages;
173 int order;
174 struct list_head list;
175};
176
177struct pblk_rb {
178 struct pblk_rb_entry *entries; /* Ring buffer entries */
179 unsigned int mem; /* Write offset - points to next
180 * writable entry in memory
181 */
182 unsigned int subm; /* Read offset - points to last entry
183 * that has been submitted to the media
184 * to be persisted
185 */
186 unsigned int sync; /* Synced - backpointer that signals
187 * the last submitted entry that has
188 * been successfully persisted to media
189 */
8154d296 190 unsigned int flush_point; /* Sync point - last entry that must be
a4bd217b
JG
191 * flushed to the media. Used with
192 * REQ_FLUSH and REQ_FUA
193 */
194 unsigned int l2p_update; /* l2p update point - next entry for
195 * which l2p mapping will be updated to
196 * contain a device ppa address (instead
197 * of a cacheline
198 */
199 unsigned int nr_entries; /* Number of entries in write buffer -
200 * must be a power of two
201 */
202 unsigned int seg_size; /* Size of the data segments being
203 * stored on each entry. Typically this
204 * will be 4KB
205 */
206
766c8ceb
JG
207 unsigned int back_thres; /* Threshold that shall be maintained by
208 * the backpointer in order to respect
209 * geo->mw_cunits on a per chunk basis
210 */
211
a4bd217b
JG
212 struct list_head pages; /* List of data pages */
213
214 spinlock_t w_lock; /* Write lock */
215 spinlock_t s_lock; /* Sync lock */
216
880eda54 217#ifdef CONFIG_NVM_PBLK_DEBUG
8154d296 218 atomic_t inflight_flush_point; /* Not served REQ_FLUSH | REQ_FUA */
a4bd217b
JG
219#endif
220};
221
222#define PBLK_RECOVERY_SECTORS 16
223
224struct pblk_lun {
225 struct ppa_addr bppa;
a4bd217b
JG
226 struct semaphore wr_sem;
227};
228
229struct pblk_gc_rq {
230 struct pblk_line *line;
231 void *data;
afdc23c9
MB
232 u64 paddr_list[NVM_MAX_VLBA];
233 u64 lba_list[NVM_MAX_VLBA];
a4bd217b
JG
234 int nr_secs;
235 int secs_to_gc;
236 struct list_head list;
237};
238
239struct pblk_gc {
b20ba1bc
JG
240 /* These states are not protected by a lock since (i) they are in the
241 * fast path, and (ii) they are not critical.
242 */
a4bd217b
JG
243 int gc_active;
244 int gc_enabled;
245 int gc_forced;
a4bd217b
JG
246
247 struct task_struct *gc_ts;
248 struct task_struct *gc_writer_ts;
b20ba1bc
JG
249 struct task_struct *gc_reader_ts;
250
251 struct workqueue_struct *gc_line_reader_wq;
a4bd217b 252 struct workqueue_struct *gc_reader_wq;
b20ba1bc 253
a4bd217b
JG
254 struct timer_list gc_timer;
255
b20ba1bc 256 struct semaphore gc_sem;
d6b992f7
HH
257 atomic_t read_inflight_gc; /* Number of lines with inflight GC reads */
258 atomic_t pipeline_gc; /* Number of lines in the GC pipeline -
259 * started reads to finished writes
260 */
a4bd217b 261 int w_entries;
b20ba1bc 262
a4bd217b 263 struct list_head w_list;
b20ba1bc 264 struct list_head r_list;
a4bd217b
JG
265
266 spinlock_t lock;
267 spinlock_t w_lock;
b20ba1bc 268 spinlock_t r_lock;
a4bd217b
JG
269};
270
271struct pblk_rl {
272 unsigned int high; /* Upper threshold for rate limiter (free run -
273 * user I/O rate limiter
274 */
a4bd217b
JG
275 unsigned int high_pw; /* High rounded up as a power of 2 */
276
b20ba1bc
JG
277#define PBLK_USER_HIGH_THRS 8 /* Begin write limit at 12% available blks */
278#define PBLK_USER_LOW_THRS 10 /* Aggressive GC at 10% available blocks */
a4bd217b
JG
279
280 int rb_windows_pw; /* Number of rate windows in the write buffer
281 * given as a power-of-2. This guarantees that
282 * when user I/O is being rate limited, there
283 * will be reserved enough space for the GC to
284 * place its payload. A window is of
285 * pblk->max_write_pgs size, which in NVMe is
286 * 64, i.e., 256kb.
287 */
288 int rb_budget; /* Total number of entries available for I/O */
289 int rb_user_max; /* Max buffer entries available for user I/O */
a4bd217b
JG
290 int rb_gc_max; /* Max buffer entries available for GC I/O */
291 int rb_gc_rsv; /* Reserved buffer entries for GC I/O */
292 int rb_state; /* Rate-limiter current state */
da67e68f 293 int rb_max_io; /* Maximum size for an I/O giving the config */
588726d3
JG
294
295 atomic_t rb_user_cnt; /* User I/O buffer counter */
a4bd217b 296 atomic_t rb_gc_cnt; /* GC I/O buffer counter */
588726d3 297 atomic_t rb_space; /* Space limit in case of reaching capacity */
a4bd217b 298
b20ba1bc
JG
299 int rsv_blocks; /* Reserved blocks for GC */
300
a4bd217b 301 int rb_user_active;
b20ba1bc
JG
302 int rb_gc_active;
303
48b8d208
HH
304 atomic_t werr_lines; /* Number of write error lines that needs gc */
305
a4bd217b
JG
306 struct timer_list u_timer;
307
308 unsigned long long nr_secs;
309 unsigned long total_blocks;
a7689938
JG
310
311 atomic_t free_blocks; /* Total number of free blocks (+ OP) */
312 atomic_t free_user_blocks; /* Number of user free blocks (no OP) */
a4bd217b
JG
313};
314
a4bd217b
JG
315#define PBLK_LINE_EMPTY (~0U)
316
317enum {
318 /* Line Types */
319 PBLK_LINETYPE_FREE = 0,
320 PBLK_LINETYPE_LOG = 1,
321 PBLK_LINETYPE_DATA = 2,
322
323 /* Line state */
32ef9412 324 PBLK_LINESTATE_NEW = 9,
a4bd217b
JG
325 PBLK_LINESTATE_FREE = 10,
326 PBLK_LINESTATE_OPEN = 11,
327 PBLK_LINESTATE_CLOSED = 12,
328 PBLK_LINESTATE_GC = 13,
329 PBLK_LINESTATE_BAD = 14,
330 PBLK_LINESTATE_CORRUPT = 15,
331
332 /* GC group */
333 PBLK_LINEGC_NONE = 20,
334 PBLK_LINEGC_EMPTY = 21,
335 PBLK_LINEGC_LOW = 22,
336 PBLK_LINEGC_MID = 23,
337 PBLK_LINEGC_HIGH = 24,
338 PBLK_LINEGC_FULL = 25,
48b8d208 339 PBLK_LINEGC_WERR = 26
a4bd217b
JG
340};
341
342#define PBLK_MAGIC 0x70626c6b /*pblk*/
d0ab0b1a
HH
343
344/* emeta/smeta persistent storage format versions:
345 * Changes in major version requires offline migration.
346 * Changes in minor version are handled automatically during
347 * recovery.
348 */
349
350#define SMETA_VERSION_MAJOR (0)
351#define SMETA_VERSION_MINOR (1)
352
353#define EMETA_VERSION_MAJOR (0)
76758390 354#define EMETA_VERSION_MINOR (2)
a4bd217b
JG
355
356struct line_header {
357 __le32 crc;
358 __le32 identifier; /* pblk identifier */
359 __u8 uuid[16]; /* instance uuid */
360 __le16 type; /* line type */
d0ab0b1a
HH
361 __u8 version_major; /* version major */
362 __u8 version_minor; /* version minor */
a4bd217b
JG
363 __le32 id; /* line id for current line */
364};
365
366struct line_smeta {
367 struct line_header header;
368
369 __le32 crc; /* Full structure including struct crc */
370 /* Previous line metadata */
371 __le32 prev_id; /* Line id for previous line */
372
373 /* Current line metadata */
374 __le64 seq_nr; /* Sequence number for current line */
375
376 /* Active writers */
377 __le32 window_wr_lun; /* Number of parallel LUNs to write */
378
379 __le32 rsvd[2];
dd2a4343
JG
380
381 __le64 lun_bitmap[];
a4bd217b
JG
382};
383
76758390 384
a4bd217b 385/*
dd2a4343
JG
386 * Metadata layout in media:
387 * First sector:
388 * 1. struct line_emeta
389 * 2. bad block bitmap (u64 * window_wr_lun)
76758390 390 * 3. write amplification counters
dd2a4343
JG
391 * Mid sectors (start at lbas_sector):
392 * 3. nr_lbas (u64) forming lba list
393 * Last sectors (start at vsc_sector):
394 * 4. u32 valid sector count (vsc) for all lines (~0U: free line)
a4bd217b
JG
395 */
396struct line_emeta {
397 struct line_header header;
398
399 __le32 crc; /* Full structure including struct crc */
400
401 /* Previous line metadata */
402 __le32 prev_id; /* Line id for prev line */
403
404 /* Current line metadata */
405 __le64 seq_nr; /* Sequence number for current line */
406
407 /* Active writers */
408 __le32 window_wr_lun; /* Number of parallel LUNs to write */
409
410 /* Bookkeeping for recovery */
411 __le32 next_id; /* Line id for next line */
412 __le64 nr_lbas; /* Number of lbas mapped in line */
413 __le64 nr_valid_lbas; /* Number of valid lbas mapped in line */
76758390
HH
414 __le64 bb_bitmap[]; /* Updated bad block bitmap for line */
415};
416
417
418/* Write amplification counters stored on media */
419struct wa_counters {
420 __le64 user; /* Number of user written sectors */
421 __le64 gc; /* Number of sectors written by GC*/
422 __le64 pad; /* Number of padded sectors */
dd2a4343
JG
423};
424
425struct pblk_emeta {
426 struct line_emeta *buf; /* emeta buffer in media format */
427 int mem; /* Write offset - points to next
428 * writable entry in memory
429 */
430 atomic_t sync; /* Synced - backpointer that signals the
431 * last entry that has been successfully
432 * persisted to media
433 */
434 unsigned int nr_entries; /* Number of emeta entries */
435};
436
437struct pblk_smeta {
438 struct line_smeta *buf; /* smeta buffer in persistent format */
a4bd217b
JG
439};
440
48b8d208
HH
441struct pblk_w_err_gc {
442 int has_write_err;
443 __le64 *lba_list;
444};
445
a4bd217b
JG
446struct pblk_line {
447 struct pblk *pblk;
448 unsigned int id; /* Line number corresponds to the
449 * block line
450 */
451 unsigned int seq_nr; /* Unique line sequence number */
452
453 int state; /* PBLK_LINESTATE_X */
454 int type; /* PBLK_LINETYPE_X */
455 int gc_group; /* PBLK_LINEGC_X */
456 struct list_head list; /* Free, GC lists */
457
458 unsigned long *lun_bitmap; /* Bitmap for LUNs mapped in line */
459
32ef9412
JG
460 struct nvm_chk_meta *chks; /* Chunks forming line */
461
dd2a4343
JG
462 struct pblk_smeta *smeta; /* Start metadata */
463 struct pblk_emeta *emeta; /* End medatada */
464
a4bd217b 465 int meta_line; /* Metadata line id */
dd2a4343
JG
466 int meta_distance; /* Distance between data and metadata */
467
a4bd217b
JG
468 u64 smeta_ssec; /* Sector where smeta starts */
469 u64 emeta_ssec; /* Sector where emeta starts */
470
471 unsigned int sec_in_line; /* Number of usable secs in line */
472
a44f53fa 473 atomic_t blk_in_line; /* Number of good blocks in line */
a4bd217b
JG
474 unsigned long *blk_bitmap; /* Bitmap for valid/invalid blocks */
475 unsigned long *erase_bitmap; /* Bitmap for erased blocks */
476
477 unsigned long *map_bitmap; /* Bitmap for mapped sectors in line */
478 unsigned long *invalid_bitmap; /* Bitmap for invalid sectors in line */
479
a44f53fa 480 atomic_t left_eblks; /* Blocks left for erasing */
a4bd217b
JG
481 atomic_t left_seblks; /* Blocks left for sync erasing */
482
483 int left_msecs; /* Sectors left for mapping */
a4bd217b 484 unsigned int cur_sec; /* Sector map pointer */
dd2a4343
JG
485 unsigned int nr_valid_lbas; /* Number of valid lbas in line */
486
487 __le32 *vsc; /* Valid sector count in line */
a4bd217b
JG
488
489 struct kref ref; /* Write buffer L2P references */
490
48b8d208
HH
491 struct pblk_w_err_gc *w_err_gc; /* Write error gc recovery metadata */
492
a4bd217b
JG
493 spinlock_t lock; /* Necessary for invalid_bitmap only */
494};
495
a44f53fa 496#define PBLK_DATA_LINES 4
a4bd217b 497
dd2a4343 498enum {
a4bd217b
JG
499 PBLK_KMALLOC_META = 1,
500 PBLK_VMALLOC_META = 2,
501};
502
dd2a4343
JG
503enum {
504 PBLK_EMETA_TYPE_HEADER = 1, /* struct line_emeta first sector */
505 PBLK_EMETA_TYPE_LLBA = 2, /* lba list - type: __le64 */
506 PBLK_EMETA_TYPE_VSC = 3, /* vsc list - type: __le32 */
a4bd217b
JG
507};
508
509struct pblk_line_mgmt {
510 int nr_lines; /* Total number of full lines */
511 int nr_free_lines; /* Number of full lines in free list */
512
513 /* Free lists - use free_lock */
514 struct list_head free_list; /* Full lines ready to use */
515 struct list_head corrupt_list; /* Full lines corrupted */
516 struct list_head bad_list; /* Full lines bad */
517
518 /* GC lists - use gc_lock */
b20ba1bc 519 struct list_head *gc_lists[PBLK_GC_NR_LISTS];
a4bd217b
JG
520 struct list_head gc_high_list; /* Full lines ready to GC, high isc */
521 struct list_head gc_mid_list; /* Full lines ready to GC, mid isc */
522 struct list_head gc_low_list; /* Full lines ready to GC, low isc */
523
48b8d208
HH
524 struct list_head gc_werr_list; /* Write err recovery list */
525
a4bd217b
JG
526 struct list_head gc_full_list; /* Full lines ready to GC, no valid */
527 struct list_head gc_empty_list; /* Full lines close, all valid */
528
529 struct pblk_line *log_line; /* Current FTL log line */
530 struct pblk_line *data_line; /* Current data line */
531 struct pblk_line *log_next; /* Next FTL log line */
532 struct pblk_line *data_next; /* Next data line */
533
dd2a4343
JG
534 struct list_head emeta_list; /* Lines queued to schedule emeta */
535
536 __le32 *vsc_list; /* Valid sector counts for all lines */
537
a4bd217b 538 /* Metadata allocation type: VMALLOC | KMALLOC */
a4bd217b
JG
539 int emeta_alloc_type;
540
541 /* Pre-allocated metadata for data lines */
dd2a4343
JG
542 struct pblk_smeta *sline_meta[PBLK_DATA_LINES];
543 struct pblk_emeta *eline_meta[PBLK_DATA_LINES];
a4bd217b
JG
544 unsigned long meta_bitmap;
545
53d82db6
HH
546 /* Cache and mempool for map/invalid bitmaps */
547 struct kmem_cache *bitmap_cache;
548 mempool_t *bitmap_pool;
549
a4bd217b
JG
550 /* Helpers for fast bitmap calculations */
551 unsigned long *bb_template;
552 unsigned long *bb_aux;
553
554 unsigned long d_seq_nr; /* Data line unique sequence number */
555 unsigned long l_seq_nr; /* Log line unique sequence number */
556
557 spinlock_t free_lock;
dd2a4343 558 spinlock_t close_lock;
a4bd217b
JG
559 spinlock_t gc_lock;
560};
561
562struct pblk_line_meta {
563 unsigned int smeta_len; /* Total length for smeta */
dd2a4343
JG
564 unsigned int smeta_sec; /* Sectors needed for smeta */
565
566 unsigned int emeta_len[4]; /* Lengths for emeta:
76758390
HH
567 * [0]: Total
568 * [1]: struct line_emeta +
569 * bb_bitmap + struct wa_counters
570 * [2]: L2P portion
571 * [3]: vsc
dd2a4343
JG
572 */
573 unsigned int emeta_sec[4]; /* Sectors needed for emeta. Same layout
574 * as emeta_len
575 */
576
a4bd217b 577 unsigned int emeta_bb; /* Boundary for bb that affects emeta */
dd2a4343
JG
578
579 unsigned int vsc_list_len; /* Length for vsc list */
a4bd217b
JG
580 unsigned int sec_bitmap_len; /* Length for sector bitmap in line */
581 unsigned int blk_bitmap_len; /* Length for block bitmap in line */
582 unsigned int lun_bitmap_len; /* Length for lun bitmap in line */
583
584 unsigned int blk_per_line; /* Number of blocks in a full line */
585 unsigned int sec_per_line; /* Number of sectors in a line */
dd2a4343 586 unsigned int dsec_per_line; /* Number of data sectors in a line */
a4bd217b
JG
587 unsigned int min_blk_line; /* Min. number of good blocks in line */
588
589 unsigned int mid_thrs; /* Threshold for GC mid list */
590 unsigned int high_thrs; /* Threshold for GC high list */
dd2a4343
JG
591
592 unsigned int meta_distance; /* Distance between data and metadata */
a4bd217b
JG
593};
594
588726d3
JG
595enum {
596 PBLK_STATE_RUNNING = 0,
597 PBLK_STATE_STOPPING = 1,
598 PBLK_STATE_RECOVERING = 2,
599 PBLK_STATE_STOPPED = 3,
600};
601
3b2a3ad1
JG
602/* Internal format to support not power-of-2 device formats */
603struct pblk_addrf {
604 /* gen to dev */
605 int sec_stripe;
606 int ch_stripe;
607 int lun_stripe;
608
609 /* dev to gen */
610 int sec_lun_stripe;
611 int sec_ws_stripe;
612};
613
a4bd217b
JG
614struct pblk {
615 struct nvm_tgt_dev *dev;
616 struct gendisk *disk;
617
618 struct kobject kobj;
619
620 struct pblk_lun *luns;
621
622 struct pblk_line *lines; /* Line array */
623 struct pblk_line_mgmt l_mg; /* Line management */
624 struct pblk_line_meta lm; /* Line metadata */
625
3b2a3ad1
JG
626 struct nvm_addrf addrf; /* Aligned address format */
627 struct pblk_addrf uaddrf; /* Unaligned address format */
bb845ae4 628 int addrf_len;
a4bd217b
JG
629
630 struct pblk_rb rwb;
631
588726d3
JG
632 int state; /* pblk line state */
633
a4bd217b 634 int min_write_pgs; /* Minimum amount of pages required by controller */
55d8ec35 635 int min_write_pgs_data; /* Minimum amount of payload pages */
a4bd217b 636 int max_write_pgs; /* Maximum amount of pages supported by controller */
faa79f27 637 int oob_meta_size; /* Size of OOB sector metadata */
a4bd217b
JG
638
639 sector_t capacity; /* Device capacity when bad blocks are subtracted */
a7689938
JG
640
641 int op; /* Percentage of device used for over-provisioning */
642 int op_blks; /* Number of blocks used for over-provisioning */
a4bd217b
JG
643
644 /* pblk provisioning values. Used by rate limiter */
645 struct pblk_rl rl;
646
c2e9f5d4 647 int sec_per_write;
a4bd217b
JG
648
649 unsigned char instance_uuid[16];
76758390
HH
650
651 /* Persistent write amplification counters, 4kb sector I/Os */
652 atomic64_t user_wa; /* Sectors written by user */
653 atomic64_t gc_wa; /* Sectors written by GC */
654 atomic64_t pad_wa; /* Padded sectors written */
655
656 /* Reset values for delta write amplification measurements */
657 u64 user_rst_wa;
658 u64 gc_rst_wa;
659 u64 pad_rst_wa;
660
5d149bfa
HH
661 /* Counters used for calculating padding distribution */
662 atomic64_t *pad_dist; /* Padding distribution buckets */
663 u64 nr_flush_rst; /* Flushes reset value for pad dist.*/
664 atomic64_t nr_flush; /* Number of flush/fua I/O */
665
880eda54 666#ifdef CONFIG_NVM_PBLK_DEBUG
76758390 667 /* Non-persistent debug counters, 4kb sector I/Os */
a4bd217b
JG
668 atomic_long_t inflight_writes; /* Inflight writes (user and gc) */
669 atomic_long_t padded_writes; /* Sectors padded due to flush/fua */
670 atomic_long_t padded_wb; /* Sectors padded in write buffer */
a4bd217b
JG
671 atomic_long_t req_writes; /* Sectors stored on write buffer */
672 atomic_long_t sub_writes; /* Sectors submitted from buffer */
673 atomic_long_t sync_writes; /* Sectors synced to media */
a4bd217b 674 atomic_long_t inflight_reads; /* Inflight sector read requests */
db7ada33 675 atomic_long_t cache_reads; /* Read requests that hit the cache */
a4bd217b
JG
676 atomic_long_t sync_reads; /* Completed sector read requests */
677 atomic_long_t recov_writes; /* Sectors submitted from recovery */
678 atomic_long_t recov_gc_writes; /* Sectors submitted from write GC */
679 atomic_long_t recov_gc_reads; /* Sectors submitted from read GC */
680#endif
681
682 spinlock_t lock;
683
684 atomic_long_t read_failed;
685 atomic_long_t read_empty;
686 atomic_long_t read_high_ecc;
687 atomic_long_t read_failed_gc;
688 atomic_long_t write_failed;
689 atomic_long_t erase_failed;
690
588726d3
JG
691 atomic_t inflight_io; /* General inflight I/O counter */
692
a4bd217b
JG
693 struct task_struct *writer_ts;
694
695 /* Simple translation map of logical addresses to physical addresses.
696 * The logical addresses is known by the host system, while the physical
697 * addresses are used when writing to the disk block device.
698 */
699 unsigned char *trans_map;
700 spinlock_t trans_lock;
701
702 struct list_head compl_list;
703
6a3abf5b
HH
704 spinlock_t resubmit_lock; /* Resubmit list lock */
705 struct list_head resubmit_list; /* Resubmit list for failed writes*/
706
b906bbb6
KO
707 mempool_t page_bio_pool;
708 mempool_t gen_ws_pool;
709 mempool_t rec_pool;
710 mempool_t r_rq_pool;
711 mempool_t w_rq_pool;
712 mempool_t e_rq_pool;
a4bd217b 713
ef576494
JG
714 struct workqueue_struct *close_wq;
715 struct workqueue_struct *bb_wq;
7bd4d370 716 struct workqueue_struct *r_end_wq;
ef576494 717
a4bd217b
JG
718 struct timer_list wtimer;
719
720 struct pblk_gc gc;
721};
722
723struct pblk_line_ws {
724 struct pblk *pblk;
725 struct pblk_line *line;
726 void *priv;
727 struct work_struct ws;
728};
729
084ec9ba 730#define pblk_g_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_g_ctx))
a4bd217b
JG
731#define pblk_w_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_c_ctx))
732
4e495a46
MB
733#define pblk_err(pblk, fmt, ...) \
734 pr_err("pblk %s: " fmt, pblk->disk->disk_name, ##__VA_ARGS__)
735#define pblk_info(pblk, fmt, ...) \
736 pr_info("pblk %s: " fmt, pblk->disk->disk_name, ##__VA_ARGS__)
737#define pblk_warn(pblk, fmt, ...) \
738 pr_warn("pblk %s: " fmt, pblk->disk->disk_name, ##__VA_ARGS__)
739#define pblk_debug(pblk, fmt, ...) \
740 pr_debug("pblk %s: " fmt, pblk->disk->disk_name, ##__VA_ARGS__)
741
a4bd217b
JG
742/*
743 * pblk ring buffer operations
744 */
766c8ceb
JG
745int pblk_rb_init(struct pblk_rb *rb, unsigned int size, unsigned int threshold,
746 unsigned int seg_sz);
a4bd217b
JG
747int pblk_rb_may_write_user(struct pblk_rb *rb, struct bio *bio,
748 unsigned int nr_entries, unsigned int *pos);
749int pblk_rb_may_write_gc(struct pblk_rb *rb, unsigned int nr_entries,
750 unsigned int *pos);
751void pblk_rb_write_entry_user(struct pblk_rb *rb, void *data,
752 struct pblk_w_ctx w_ctx, unsigned int pos);
753void pblk_rb_write_entry_gc(struct pblk_rb *rb, void *data,
d340121e
JG
754 struct pblk_w_ctx w_ctx, struct pblk_line *line,
755 u64 paddr, unsigned int pos);
a4bd217b 756struct pblk_w_ctx *pblk_rb_w_ctx(struct pblk_rb *rb, unsigned int pos);
588726d3 757void pblk_rb_flush(struct pblk_rb *rb);
a4bd217b
JG
758
759void pblk_rb_sync_l2p(struct pblk_rb *rb);
d624f371 760unsigned int pblk_rb_read_to_bio(struct pblk_rb *rb, struct nvm_rq *rqd,
875d94f3
JG
761 unsigned int pos, unsigned int nr_entries,
762 unsigned int count);
a4bd217b 763int pblk_rb_copy_to_bio(struct pblk_rb *rb, struct bio *bio, sector_t lba,
75cb8e93 764 struct ppa_addr ppa, int bio_iter, bool advanced_bio);
a4bd217b
JG
765unsigned int pblk_rb_read_commit(struct pblk_rb *rb, unsigned int entries);
766
767unsigned int pblk_rb_sync_init(struct pblk_rb *rb, unsigned long *flags);
768unsigned int pblk_rb_sync_advance(struct pblk_rb *rb, unsigned int nr_entries);
40b8657d
JG
769unsigned int pblk_rb_ptr_wrap(struct pblk_rb *rb, unsigned int p,
770 unsigned int nr_entries);
a4bd217b 771void pblk_rb_sync_end(struct pblk_rb *rb, unsigned long *flags);
8154d296 772unsigned int pblk_rb_flush_point_count(struct pblk_rb *rb);
a4bd217b
JG
773
774unsigned int pblk_rb_read_count(struct pblk_rb *rb);
ee8d5c1a 775unsigned int pblk_rb_sync_count(struct pblk_rb *rb);
a4bd217b
JG
776unsigned int pblk_rb_wrap_pos(struct pblk_rb *rb, unsigned int pos);
777
778int pblk_rb_tear_down_check(struct pblk_rb *rb);
779int pblk_rb_pos_oob(struct pblk_rb *rb, u64 pos);
9bd1f875 780void pblk_rb_free(struct pblk_rb *rb);
a4bd217b
JG
781ssize_t pblk_rb_sysfs(struct pblk_rb *rb, char *buf);
782
783/*
784 * pblk core
785 */
67bf26a3
JG
786struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int type);
787void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int type);
45dcf29b
JG
788int pblk_alloc_rqd_meta(struct pblk *pblk, struct nvm_rq *rqd);
789void pblk_free_rqd_meta(struct pblk *pblk, struct nvm_rq *rqd);
c2e9f5d4 790void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write);
a4bd217b
JG
791int pblk_setup_w_rec_rq(struct pblk *pblk, struct nvm_rq *rqd,
792 struct pblk_c_ctx *c_ctx);
a4bd217b 793void pblk_discard(struct pblk *pblk, struct bio *bio);
aff3fb18 794struct nvm_chk_meta *pblk_get_chunk_meta(struct pblk *pblk);
32ef9412
JG
795struct nvm_chk_meta *pblk_chunk_get_off(struct pblk *pblk,
796 struct nvm_chk_meta *lp,
797 struct ppa_addr ppa);
a4bd217b
JG
798void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd);
799void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd);
800int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd);
1a94b2d4 801int pblk_submit_io_sync(struct pblk *pblk, struct nvm_rq *rqd);
253babc3 802int pblk_submit_io_sync_sem(struct pblk *pblk, struct nvm_rq *rqd);
dd2a4343 803int pblk_submit_meta_io(struct pblk *pblk, struct pblk_line *meta_line);
4c44abf4 804void pblk_check_chunk_state_update(struct pblk *pblk, struct nvm_rq *rqd);
a4bd217b
JG
805struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
806 unsigned int nr_secs, unsigned int len,
de54e703 807 int alloc_type, gfp_t gfp_mask);
a4bd217b
JG
808struct pblk_line *pblk_line_get(struct pblk *pblk);
809struct pblk_line *pblk_line_get_first_data(struct pblk *pblk);
21d22871 810struct pblk_line *pblk_line_replace_data(struct pblk *pblk);
ae14cc04
MB
811void pblk_ppa_to_line_put(struct pblk *pblk, struct ppa_addr ppa);
812void pblk_rq_to_line_put(struct pblk *pblk, struct nvm_rq *rqd);
a4bd217b
JG
813int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line);
814void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line);
815struct pblk_line *pblk_line_get_data(struct pblk *pblk);
d624f371 816struct pblk_line *pblk_line_get_erase(struct pblk *pblk);
a4bd217b
JG
817int pblk_line_erase(struct pblk *pblk, struct pblk_line *line);
818int pblk_line_is_full(struct pblk_line *line);
8e55c07b 819void pblk_line_free(struct pblk_line *line);
dd2a4343 820void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line);
a4bd217b 821void pblk_line_close(struct pblk *pblk, struct pblk_line *line);
dd2a4343 822void pblk_line_close_ws(struct work_struct *work);
588726d3 823void pblk_pipeline_stop(struct pblk *pblk);
a7c9e910
JG
824void __pblk_pipeline_stop(struct pblk *pblk);
825void __pblk_pipeline_flush(struct pblk *pblk);
b84ae4a8
JG
826void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
827 void (*work)(struct work_struct *), gfp_t gfp_mask,
828 struct workqueue_struct *wq);
a4bd217b 829u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line);
af3fac16
JG
830int pblk_line_smeta_read(struct pblk *pblk, struct pblk_line *line);
831int pblk_line_emeta_read(struct pblk *pblk, struct pblk_line *line,
dd2a4343 832 void *emeta_buf);
a4bd217b
JG
833int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr erase_ppa);
834void pblk_line_put(struct kref *ref);
7bd4d370 835void pblk_line_put_wq(struct kref *ref);
a4bd217b 836struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line);
dd2a4343
JG
837u64 pblk_lookup_page(struct pblk *pblk, struct pblk_line *line);
838void pblk_dealloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
a4bd217b 839u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
dd2a4343 840u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs);
a4bd217b 841int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
55d8ec35 842 unsigned long secs_to_flush, bool skip_meta);
43241cfe 843void pblk_down_rq(struct pblk *pblk, struct ppa_addr ppa,
a4bd217b 844 unsigned long *lun_bitmap);
43241cfe
MB
845void pblk_down_chunk(struct pblk *pblk, struct ppa_addr ppa);
846void pblk_up_chunk(struct pblk *pblk, struct ppa_addr ppa);
e99e802f 847void pblk_up_rq(struct pblk *pblk, unsigned long *lun_bitmap);
a4bd217b
JG
848int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
849 int nr_pages);
a4bd217b
JG
850void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
851 int nr_pages);
852void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa);
0880a9aa
JG
853void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
854 u64 paddr);
a4bd217b
JG
855void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa);
856void pblk_update_map_cache(struct pblk *pblk, sector_t lba,
857 struct ppa_addr ppa);
858void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
859 struct ppa_addr ppa, struct ppa_addr entry_line);
860int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa,
d340121e 861 struct pblk_line *gc_line, u64 paddr);
a4bd217b
JG
862void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
863 u64 *lba_list, int nr_secs);
864void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
865 sector_t blba, int nr_secs);
55d8ec35
IK
866void *pblk_get_meta_for_writes(struct pblk *pblk, struct nvm_rq *rqd);
867void pblk_get_packed_meta(struct pblk *pblk, struct nvm_rq *rqd);
a4bd217b
JG
868
869/*
870 * pblk user I/O write path
871 */
872int pblk_write_to_cache(struct pblk *pblk, struct bio *bio,
873 unsigned long flags);
d340121e 874int pblk_write_gc_to_cache(struct pblk *pblk, struct pblk_gc_rq *gc_rq);
a4bd217b
JG
875
876/*
877 * pblk map
878 */
525f7bb2 879int pblk_map_erase_rq(struct pblk *pblk, struct nvm_rq *rqd,
a4bd217b
JG
880 unsigned int sentry, unsigned long *lun_bitmap,
881 unsigned int valid_secs, struct ppa_addr *erase_ppa);
525f7bb2 882int pblk_map_rq(struct pblk *pblk, struct nvm_rq *rqd, unsigned int sentry,
a4bd217b
JG
883 unsigned long *lun_bitmap, unsigned int valid_secs,
884 unsigned int off);
885
886/*
887 * pblk write thread
888 */
889int pblk_write_ts(void *data);
87c1d2d3 890void pblk_write_timer_fn(struct timer_list *t);
a4bd217b 891void pblk_write_should_kick(struct pblk *pblk);
cc9c9a00 892void pblk_write_kick(struct pblk *pblk);
a4bd217b
JG
893
894/*
895 * pblk read path
896 */
b906bbb6 897extern struct bio_set pblk_bio_set;
a4bd217b 898int pblk_submit_read(struct pblk *pblk, struct bio *bio);
d340121e 899int pblk_submit_read_gc(struct pblk *pblk, struct pblk_gc_rq *gc_rq);
a4bd217b
JG
900/*
901 * pblk recovery
902 */
a4bd217b 903struct pblk_line *pblk_recov_l2p(struct pblk *pblk);
588726d3 904int pblk_recov_pad(struct pblk *pblk);
06bc072b 905int pblk_recov_check_emeta(struct pblk *pblk, struct line_emeta *emeta);
a4bd217b
JG
906
907/*
908 * pblk gc
909 */
b20ba1bc 910#define PBLK_GC_MAX_READERS 8 /* Max number of outstanding GC reader jobs */
3627896a 911#define PBLK_GC_RQ_QD 128 /* Queue depth for inflight GC requests */
b20ba1bc 912#define PBLK_GC_L_QD 4 /* Queue depth for inflight GC lines */
a4bd217b
JG
913
914int pblk_gc_init(struct pblk *pblk);
a7c9e910 915void pblk_gc_exit(struct pblk *pblk, bool graceful);
a4bd217b
JG
916void pblk_gc_should_start(struct pblk *pblk);
917void pblk_gc_should_stop(struct pblk *pblk);
03661b5f 918void pblk_gc_should_kick(struct pblk *pblk);
37ce33d5 919void pblk_gc_free_full_lines(struct pblk *pblk);
a4bd217b
JG
920void pblk_gc_sysfs_state_show(struct pblk *pblk, int *gc_enabled,
921 int *gc_active);
b20ba1bc 922int pblk_gc_sysfs_force(struct pblk *pblk, int force);
a4bd217b
JG
923
924/*
925 * pblk rate limiter
926 */
927void pblk_rl_init(struct pblk_rl *rl, int budget);
928void pblk_rl_free(struct pblk_rl *rl);
03661b5f 929void pblk_rl_update_rates(struct pblk_rl *rl);
b20ba1bc 930int pblk_rl_high_thrs(struct pblk_rl *rl);
a4bd217b 931unsigned long pblk_rl_nr_free_blks(struct pblk_rl *rl);
a7689938 932unsigned long pblk_rl_nr_user_free_blks(struct pblk_rl *rl);
a4bd217b 933int pblk_rl_user_may_insert(struct pblk_rl *rl, int nr_entries);
588726d3 934void pblk_rl_inserted(struct pblk_rl *rl, int nr_entries);
a4bd217b
JG
935void pblk_rl_user_in(struct pblk_rl *rl, int nr_entries);
936int pblk_rl_gc_may_insert(struct pblk_rl *rl, int nr_entries);
937void pblk_rl_gc_in(struct pblk_rl *rl, int nr_entries);
938void pblk_rl_out(struct pblk_rl *rl, int nr_user, int nr_gc);
da67e68f 939int pblk_rl_max_io(struct pblk_rl *rl);
a4bd217b 940void pblk_rl_free_lines_inc(struct pblk_rl *rl, struct pblk_line *line);
a7689938
JG
941void pblk_rl_free_lines_dec(struct pblk_rl *rl, struct pblk_line *line,
942 bool used);
588726d3 943int pblk_rl_is_limit(struct pblk_rl *rl);
a4bd217b 944
48b8d208
HH
945void pblk_rl_werr_line_in(struct pblk_rl *rl);
946void pblk_rl_werr_line_out(struct pblk_rl *rl);
947
a4bd217b
JG
948/*
949 * pblk sysfs
950 */
951int pblk_sysfs_init(struct gendisk *tdisk);
952void pblk_sysfs_exit(struct gendisk *tdisk);
953
954static inline void *pblk_malloc(size_t size, int type, gfp_t flags)
955{
956 if (type == PBLK_KMALLOC_META)
957 return kmalloc(size, flags);
958 return vmalloc(size);
959}
960
961static inline void pblk_mfree(void *ptr, int type)
962{
963 if (type == PBLK_KMALLOC_META)
964 kfree(ptr);
965 else
966 vfree(ptr);
967}
968
969static inline struct nvm_rq *nvm_rq_from_c_ctx(void *c_ctx)
970{
971 return c_ctx - sizeof(struct nvm_rq);
972}
973
dd2a4343
JG
974static inline void *emeta_to_bb(struct line_emeta *emeta)
975{
976 return emeta->bb_bitmap;
977}
978
76758390
HH
979static inline void *emeta_to_wa(struct pblk_line_meta *lm,
980 struct line_emeta *emeta)
981{
982 return emeta->bb_bitmap + lm->blk_bitmap_len;
983}
984
dd2a4343
JG
985static inline void *emeta_to_lbas(struct pblk *pblk, struct line_emeta *emeta)
986{
987 return ((void *)emeta + pblk->lm.emeta_len[1]);
988}
989
990static inline void *emeta_to_vsc(struct pblk *pblk, struct line_emeta *emeta)
a4bd217b 991{
dd2a4343 992 return (emeta_to_lbas(pblk, emeta) + pblk->lm.emeta_len[2]);
a4bd217b
JG
993}
994
b20ba1bc
JG
995static inline int pblk_line_vsc(struct pblk_line *line)
996{
d340121e 997 return le32_to_cpu(*line->vsc);
b20ba1bc
JG
998}
999
cb21665c 1000static inline int pblk_ppa_to_line_id(struct ppa_addr p)
a4bd217b 1001{
69471513 1002 return p.a.blk;
a4bd217b
JG
1003}
1004
cb21665c
JG
1005static inline struct pblk_line *pblk_ppa_to_line(struct pblk *pblk,
1006 struct ppa_addr p)
1007{
1008 return &pblk->lines[pblk_ppa_to_line_id(p)];
1009}
1010
b1bcfda1 1011static inline int pblk_ppa_to_pos(struct nvm_geo *geo, struct ppa_addr p)
a4bd217b 1012{
69471513 1013 return p.a.lun * geo->num_ch + p.a.ch;
a4bd217b
JG
1014}
1015
b1bcfda1
JG
1016static inline struct ppa_addr addr_to_gen_ppa(struct pblk *pblk, u64 paddr,
1017 u64 line_id)
a4bd217b 1018{
3b2a3ad1
JG
1019 struct nvm_tgt_dev *dev = pblk->dev;
1020 struct nvm_geo *geo = &dev->geo;
b1bcfda1
JG
1021 struct ppa_addr ppa;
1022
3b2a3ad1
JG
1023 if (geo->version == NVM_OCSSD_SPEC_12) {
1024 struct nvm_addrf_12 *ppaf = (struct nvm_addrf_12 *)&pblk->addrf;
1025
1026 ppa.ppa = 0;
1027 ppa.g.blk = line_id;
1028 ppa.g.pg = (paddr & ppaf->pg_mask) >> ppaf->pg_offset;
1029 ppa.g.lun = (paddr & ppaf->lun_mask) >> ppaf->lun_offset;
1030 ppa.g.ch = (paddr & ppaf->ch_mask) >> ppaf->ch_offset;
1031 ppa.g.pl = (paddr & ppaf->pln_mask) >> ppaf->pln_offset;
1032 ppa.g.sec = (paddr & ppaf->sec_mask) >> ppaf->sec_offset;
1033 } else {
1034 struct pblk_addrf *uaddrf = &pblk->uaddrf;
1035 int secs, chnls, luns;
1036
1037 ppa.ppa = 0;
1038
1039 ppa.m.chk = line_id;
1040
1041 paddr = div_u64_rem(paddr, uaddrf->sec_stripe, &secs);
1042 ppa.m.sec = secs;
1043
1044 paddr = div_u64_rem(paddr, uaddrf->ch_stripe, &chnls);
1045 ppa.m.grp = chnls;
1046
1047 paddr = div_u64_rem(paddr, uaddrf->lun_stripe, &luns);
1048 ppa.m.pu = luns;
1049
1050 ppa.m.sec += uaddrf->sec_stripe * paddr;
1051 }
b1bcfda1
JG
1052
1053 return ppa;
a4bd217b
JG
1054}
1055
2cf99bbd
JG
1056static inline struct nvm_chk_meta *pblk_dev_ppa_to_chunk(struct pblk *pblk,
1057 struct ppa_addr p)
1058{
1059 struct nvm_tgt_dev *dev = pblk->dev;
1060 struct nvm_geo *geo = &dev->geo;
cb21665c 1061 struct pblk_line *line = pblk_ppa_to_line(pblk, p);
2cf99bbd
JG
1062 int pos = pblk_ppa_to_pos(geo, p);
1063
1064 return &line->chks[pos];
1065}
1066
1067static inline u64 pblk_dev_ppa_to_chunk_addr(struct pblk *pblk,
1068 struct ppa_addr p)
1069{
1070 struct nvm_tgt_dev *dev = pblk->dev;
1071
1072 return dev_to_chunk_addr(dev->parent, &pblk->addrf, p);
1073}
1074
b1bcfda1
JG
1075static inline u64 pblk_dev_ppa_to_line_addr(struct pblk *pblk,
1076 struct ppa_addr p)
a4bd217b 1077{
3b2a3ad1
JG
1078 struct nvm_tgt_dev *dev = pblk->dev;
1079 struct nvm_geo *geo = &dev->geo;
b1bcfda1
JG
1080 u64 paddr;
1081
3b2a3ad1
JG
1082 if (geo->version == NVM_OCSSD_SPEC_12) {
1083 struct nvm_addrf_12 *ppaf = (struct nvm_addrf_12 *)&pblk->addrf;
1084
1085 paddr = (u64)p.g.ch << ppaf->ch_offset;
1086 paddr |= (u64)p.g.lun << ppaf->lun_offset;
1087 paddr |= (u64)p.g.pg << ppaf->pg_offset;
1088 paddr |= (u64)p.g.pl << ppaf->pln_offset;
1089 paddr |= (u64)p.g.sec << ppaf->sec_offset;
1090 } else {
1091 struct pblk_addrf *uaddrf = &pblk->uaddrf;
1092 u64 secs = p.m.sec;
1093 int sec_stripe;
1094
1095 paddr = (u64)p.m.grp * uaddrf->sec_stripe;
1096 paddr += (u64)p.m.pu * uaddrf->sec_lun_stripe;
1097
1098 secs = div_u64_rem(secs, uaddrf->sec_stripe, &sec_stripe);
1099 paddr += secs * uaddrf->sec_ws_stripe;
1100 paddr += sec_stripe;
1101 }
b1bcfda1
JG
1102
1103 return paddr;
a4bd217b
JG
1104}
1105
1106static inline struct ppa_addr pblk_ppa32_to_ppa64(struct pblk *pblk, u32 ppa32)
1107{
7f985f9a 1108 struct nvm_tgt_dev *dev = pblk->dev;
a4bd217b 1109
7f985f9a 1110 return nvm_ppa32_to_ppa64(dev->parent, &pblk->addrf, ppa32);
a4bd217b
JG
1111}
1112
a4bd217b
JG
1113static inline u32 pblk_ppa64_to_ppa32(struct pblk *pblk, struct ppa_addr ppa64)
1114{
7f985f9a 1115 struct nvm_tgt_dev *dev = pblk->dev;
a4bd217b 1116
7f985f9a 1117 return nvm_ppa64_to_ppa32(dev->parent, &pblk->addrf, ppa64);
a4bd217b
JG
1118}
1119
b1bcfda1
JG
1120static inline struct ppa_addr pblk_trans_map_get(struct pblk *pblk,
1121 sector_t lba)
a4bd217b 1122{
b1bcfda1
JG
1123 struct ppa_addr ppa;
1124
bb845ae4 1125 if (pblk->addrf_len < 32) {
a4bd217b
JG
1126 u32 *map = (u32 *)pblk->trans_map;
1127
b1bcfda1 1128 ppa = pblk_ppa32_to_ppa64(pblk, map[lba]);
a4bd217b 1129 } else {
b1bcfda1 1130 struct ppa_addr *map = (struct ppa_addr *)pblk->trans_map;
a4bd217b 1131
b1bcfda1 1132 ppa = map[lba];
a4bd217b 1133 }
b1bcfda1
JG
1134
1135 return ppa;
a4bd217b
JG
1136}
1137
b1bcfda1
JG
1138static inline void pblk_trans_map_set(struct pblk *pblk, sector_t lba,
1139 struct ppa_addr ppa)
a4bd217b 1140{
bb845ae4 1141 if (pblk->addrf_len < 32) {
b1bcfda1 1142 u32 *map = (u32 *)pblk->trans_map;
a4bd217b 1143
b1bcfda1
JG
1144 map[lba] = pblk_ppa64_to_ppa32(pblk, ppa);
1145 } else {
1146 u64 *map = (u64 *)pblk->trans_map;
a4bd217b 1147
b1bcfda1
JG
1148 map[lba] = ppa.ppa;
1149 }
a4bd217b
JG
1150}
1151
1152static inline int pblk_ppa_empty(struct ppa_addr ppa_addr)
1153{
1154 return (ppa_addr.ppa == ADDR_EMPTY);
1155}
1156
1157static inline void pblk_ppa_set_empty(struct ppa_addr *ppa_addr)
1158{
1159 ppa_addr->ppa = ADDR_EMPTY;
1160}
1161
07698466
JG
1162static inline bool pblk_ppa_comp(struct ppa_addr lppa, struct ppa_addr rppa)
1163{
8b7bc849 1164 return (lppa.ppa == rppa.ppa);
07698466
JG
1165}
1166
a4bd217b
JG
1167static inline int pblk_addr_in_cache(struct ppa_addr ppa)
1168{
1169 return (ppa.ppa != ADDR_EMPTY && ppa.c.is_cached);
1170}
1171
1172static inline int pblk_addr_to_cacheline(struct ppa_addr ppa)
1173{
1174 return ppa.c.line;
1175}
1176
1177static inline struct ppa_addr pblk_cacheline_to_addr(int addr)
1178{
1179 struct ppa_addr p;
1180
1181 p.c.line = addr;
1182 p.c.is_cached = 1;
1183
1184 return p;
1185}
1186
a4bd217b 1187static inline u32 pblk_calc_meta_header_crc(struct pblk *pblk,
dd2a4343 1188 struct line_header *header)
a4bd217b
JG
1189{
1190 u32 crc = ~(u32)0;
1191
dd2a4343 1192 crc = crc32_le(crc, (unsigned char *)header + sizeof(crc),
a4bd217b
JG
1193 sizeof(struct line_header) - sizeof(crc));
1194
1195 return crc;
1196}
1197
1198static inline u32 pblk_calc_smeta_crc(struct pblk *pblk,
1199 struct line_smeta *smeta)
1200{
1201 struct pblk_line_meta *lm = &pblk->lm;
1202 u32 crc = ~(u32)0;
1203
1204 crc = crc32_le(crc, (unsigned char *)smeta +
1205 sizeof(struct line_header) + sizeof(crc),
1206 lm->smeta_len -
1207 sizeof(struct line_header) - sizeof(crc));
1208
1209 return crc;
1210}
1211
1212static inline u32 pblk_calc_emeta_crc(struct pblk *pblk,
1213 struct line_emeta *emeta)
1214{
1215 struct pblk_line_meta *lm = &pblk->lm;
1216 u32 crc = ~(u32)0;
1217
1218 crc = crc32_le(crc, (unsigned char *)emeta +
1219 sizeof(struct line_header) + sizeof(crc),
dd2a4343 1220 lm->emeta_len[0] -
a4bd217b
JG
1221 sizeof(struct line_header) - sizeof(crc));
1222
1223 return crc;
1224}
1225
f9c10152 1226static inline int pblk_io_aligned(struct pblk *pblk, int nr_secs)
a4bd217b 1227{
f9c10152 1228 return !(nr_secs % pblk->min_write_pgs);
a4bd217b
JG
1229}
1230
880eda54 1231#ifdef CONFIG_NVM_PBLK_DEBUG
4e495a46 1232static inline void print_ppa(struct pblk *pblk, struct ppa_addr *p,
3b2a3ad1 1233 char *msg, int error)
a4bd217b 1234{
4e495a46
MB
1235 struct nvm_geo *geo = &pblk->dev->geo;
1236
a4bd217b 1237 if (p->c.is_cached) {
4e495a46 1238 pblk_err(pblk, "ppa: (%s: %x) cache line: %llu\n",
a4bd217b 1239 msg, error, (u64)p->c.line);
3b2a3ad1 1240 } else if (geo->version == NVM_OCSSD_SPEC_12) {
4e495a46 1241 pblk_err(pblk, "ppa: (%s: %x):ch:%d,lun:%d,blk:%d,pg:%d,pl:%d,sec:%d\n",
a4bd217b
JG
1242 msg, error,
1243 p->g.ch, p->g.lun, p->g.blk,
1244 p->g.pg, p->g.pl, p->g.sec);
3b2a3ad1 1245 } else {
4e495a46 1246 pblk_err(pblk, "ppa: (%s: %x):ch:%d,lun:%d,chk:%d,sec:%d\n",
3b2a3ad1
JG
1247 msg, error,
1248 p->m.grp, p->m.pu, p->m.chk, p->m.sec);
a4bd217b
JG
1249 }
1250}
1251
1252static inline void pblk_print_failed_rqd(struct pblk *pblk, struct nvm_rq *rqd,
1253 int error)
1254{
1255 int bit = -1;
1256
1257 if (rqd->nr_ppas == 1) {
4e495a46 1258 print_ppa(pblk, &rqd->ppa_addr, "rqd", error);
a4bd217b
JG
1259 return;
1260 }
1261
1262 while ((bit = find_next_bit((void *)&rqd->ppa_status, rqd->nr_ppas,
1263 bit + 1)) < rqd->nr_ppas) {
4e495a46 1264 print_ppa(pblk, &rqd->ppa_list[bit], "rqd", error);
a4bd217b
JG
1265 }
1266
4e495a46 1267 pblk_err(pblk, "error:%d, ppa_status:%llx\n", error, rqd->ppa_status);
a4bd217b 1268}
a4bd217b
JG
1269
1270static inline int pblk_boundary_ppa_checks(struct nvm_tgt_dev *tgt_dev,
1271 struct ppa_addr *ppas, int nr_ppas)
1272{
1273 struct nvm_geo *geo = &tgt_dev->geo;
1274 struct ppa_addr *ppa;
1275 int i;
1276
1277 for (i = 0; i < nr_ppas; i++) {
1278 ppa = &ppas[i];
1279
3b2a3ad1
JG
1280 if (geo->version == NVM_OCSSD_SPEC_12) {
1281 if (!ppa->c.is_cached &&
1282 ppa->g.ch < geo->num_ch &&
1283 ppa->g.lun < geo->num_lun &&
1284 ppa->g.pl < geo->num_pln &&
1285 ppa->g.blk < geo->num_chk &&
1286 ppa->g.pg < geo->num_pg &&
1287 ppa->g.sec < geo->ws_min)
1288 continue;
1289 } else {
1290 if (!ppa->c.is_cached &&
1291 ppa->m.grp < geo->num_ch &&
1292 ppa->m.pu < geo->num_lun &&
1293 ppa->m.chk < geo->num_chk &&
1294 ppa->m.sec < geo->clba)
1295 continue;
1296 }
1297
4e495a46 1298 print_ppa(tgt_dev->q->queuedata, ppa, "boundary", i);
1a94b2d4 1299
a4bd217b
JG
1300 return 1;
1301 }
1302 return 0;
1303}
1304
1a94b2d4
JG
1305static inline int pblk_check_io(struct pblk *pblk, struct nvm_rq *rqd)
1306{
1307 struct nvm_tgt_dev *dev = pblk->dev;
d68a9344 1308 struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
1a94b2d4
JG
1309
1310 if (pblk_boundary_ppa_checks(dev, ppa_list, rqd->nr_ppas)) {
1311 WARN_ON(1);
1312 return -EINVAL;
1313 }
1314
1315 if (rqd->opcode == NVM_OP_PWRITE) {
1316 struct pblk_line *line;
1a94b2d4
JG
1317 int i;
1318
1319 for (i = 0; i < rqd->nr_ppas; i++) {
cb21665c 1320 line = pblk_ppa_to_line(pblk, ppa_list[i]);
1a94b2d4
JG
1321
1322 spin_lock(&line->lock);
1323 if (line->state != PBLK_LINESTATE_OPEN) {
4e495a46 1324 pblk_err(pblk, "bad ppa: line:%d,state:%d\n",
1a94b2d4
JG
1325 line->id, line->state);
1326 WARN_ON(1);
1327 spin_unlock(&line->lock);
1328 return -EINVAL;
1329 }
1330 spin_unlock(&line->lock);
1331 }
1332 }
1333
1334 return 0;
1335}
1336#endif
1337
a4bd217b
JG
1338static inline int pblk_boundary_paddr_checks(struct pblk *pblk, u64 paddr)
1339{
1340 struct pblk_line_meta *lm = &pblk->lm;
1341
1342 if (paddr > lm->sec_per_line)
1343 return 1;
1344
1345 return 0;
1346}
1347
1348static inline unsigned int pblk_get_bi_idx(struct bio *bio)
1349{
1350 return bio->bi_iter.bi_idx;
1351}
1352
1353static inline sector_t pblk_get_lba(struct bio *bio)
1354{
1355 return bio->bi_iter.bi_sector / NR_PHY_IN_LOG;
1356}
1357
1358static inline unsigned int pblk_get_secs(struct bio *bio)
1359{
1360 return bio->bi_iter.bi_size / PBLK_EXPOSED_PAGE_SIZE;
1361}
1362
a4bd217b
JG
1363static inline void pblk_setup_uuid(struct pblk *pblk)
1364{
1365 uuid_le uuid;
1366
1367 uuid_le_gen(&uuid);
1368 memcpy(pblk->instance_uuid, uuid.b, 16);
1369}
4c44abf4
HH
1370
1371static inline char *pblk_disk_name(struct pblk *pblk)
1372{
1373 struct gendisk *disk = pblk->disk;
1374
1375 return disk->disk_name;
1376}
3bcebc5b
HH
1377
1378static inline unsigned int pblk_get_min_chks(struct pblk *pblk)
1379{
1380 struct pblk_line_meta *lm = &pblk->lm;
1381 /* In a worst-case scenario every line will have OP invalid sectors.
1382 * We will then need a minimum of 1/OP lines to free up a single line
1383 */
1384
1385 return DIV_ROUND_UP(100, pblk->op) * lm->blk_per_line;
faa79f27 1386}
3bcebc5b 1387
faa79f27
IK
1388static inline struct pblk_sec_meta *pblk_get_meta(struct pblk *pblk,
1389 void *meta, int index)
1390{
2c4d5356
IK
1391 return meta +
1392 max_t(int, sizeof(struct pblk_sec_meta), pblk->oob_meta_size)
1393 * index;
3bcebc5b 1394}
24828d05
IK
1395
1396static inline int pblk_dma_meta_size(struct pblk *pblk)
1397{
2c4d5356
IK
1398 return max_t(int, sizeof(struct pblk_sec_meta), pblk->oob_meta_size)
1399 * NVM_MAX_VLBA;
24828d05 1400}
55d8ec35
IK
1401
1402static inline int pblk_is_oob_meta_supported(struct pblk *pblk)
1403{
1404 return pblk->oob_meta_size >= sizeof(struct pblk_sec_meta);
1405}
a4bd217b 1406#endif /* PBLK_H_ */