Commit | Line | Data |
---|---|---|
a4bd217b JG |
1 | /* |
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> | |
6 | * | |
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. | |
10 | * | |
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. | |
15 | * | |
16 | * Implementation of a Physical Block-device target for Open-channel SSDs. | |
17 | * | |
18 | */ | |
19 | ||
20 | #ifndef PBLK_H_ | |
21 | #define PBLK_H_ | |
22 | ||
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> | |
31 | ||
32 | #include <linux/lightnvm.h> | |
33 | ||
34 | /* Run only GC if less than 1/X blocks are free */ | |
35 | #define GC_LIMIT_INVERSE 5 | |
36 | #define GC_TIME_MSECS 1000 | |
37 | ||
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) | |
42 | ||
ef576494 JG |
43 | #define PBLK_NR_CLOSE_JOBS (4) |
44 | ||
a4bd217b JG |
45 | #define PBLK_CACHE_NAME_LEN (DISK_NAME_LEN + 16) |
46 | ||
47 | #define PBLK_COMMAND_TIMEOUT_MS 30000 | |
48 | ||
49 | /* Max 512 LUNs per device */ | |
50 | #define PBLK_MAX_LUNS_BITMAP (4) | |
51 | ||
52 | #define NR_PHY_IN_LOG (PBLK_EXPOSED_PAGE_SIZE / PBLK_SECTOR) | |
53 | ||
0d880398 | 54 | /* Static pool sizes */ |
b84ae4a8 JG |
55 | #define PBLK_GEN_WS_POOL_SIZE (2) |
56 | ||
e2cddf20 JG |
57 | enum { |
58 | PBLK_READ = READ, | |
59 | PBLK_WRITE = WRITE,/* Write from write buffer */ | |
60 | PBLK_WRITE_INT, /* Internal write - no write buffer */ | |
61 | PBLK_ERASE, | |
62 | }; | |
63 | ||
a4bd217b JG |
64 | enum { |
65 | /* IO Types */ | |
66 | PBLK_IOTYPE_USER = 1 << 0, | |
67 | PBLK_IOTYPE_GC = 1 << 1, | |
68 | ||
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, | |
74 | }; | |
75 | ||
76 | enum { | |
77 | PBLK_BLK_ST_OPEN = 0x1, | |
78 | PBLK_BLK_ST_CLOSED = 0x2, | |
79 | }; | |
80 | ||
b20ba1bc JG |
81 | struct pblk_sec_meta { |
82 | u64 reserved; | |
83 | __le64 lba; | |
84 | }; | |
85 | ||
a4bd217b JG |
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. | |
88 | */ | |
b20ba1bc | 89 | #define PBLK_GC_NR_LISTS 3 |
a4bd217b JG |
90 | |
91 | enum { | |
92 | PBLK_RL_HIGH = 1, | |
93 | PBLK_RL_MID = 2, | |
94 | PBLK_RL_LOW = 3, | |
95 | }; | |
96 | ||
a4bd217b | 97 | #define pblk_dma_meta_size (sizeof(struct pblk_sec_meta) * PBLK_MAX_REQ_ADDRS) |
a4809fee | 98 | #define pblk_dma_ppa_size (sizeof(u64) * PBLK_MAX_REQ_ADDRS) |
a4bd217b | 99 | |
084ec9ba | 100 | /* write buffer completion context */ |
a4bd217b JG |
101 | struct pblk_c_ctx { |
102 | struct list_head list; /* Head for out-of-order completion */ | |
103 | ||
104 | unsigned long *lun_bitmap; /* Luns used on current request */ | |
105 | unsigned int sentry; | |
106 | unsigned int nr_valid; | |
107 | unsigned int nr_padded; | |
108 | }; | |
109 | ||
a4809fee | 110 | /* read context */ |
084ec9ba JG |
111 | struct pblk_g_ctx { |
112 | void *private; | |
a4809fee | 113 | u64 lba; |
a4bd217b JG |
114 | }; |
115 | ||
ee8d5c1a JG |
116 | /* Pad context */ |
117 | struct pblk_pad_rq { | |
118 | struct pblk *pblk; | |
119 | struct completion wait; | |
120 | struct kref ref; | |
121 | }; | |
122 | ||
a4bd217b JG |
123 | /* Recovery context */ |
124 | struct pblk_rec_ctx { | |
125 | struct pblk *pblk; | |
126 | struct nvm_rq *rqd; | |
127 | struct list_head failed; | |
128 | struct work_struct ws_rec; | |
129 | }; | |
130 | ||
131 | /* Write context */ | |
132 | struct pblk_w_ctx { | |
133 | struct bio_list bios; /* Original bios - used for completion | |
134 | * in REQ_FUA, REQ_FLUSH case | |
135 | */ | |
ef697902 | 136 | u64 lba; /* Logic addr. associated with entry */ |
a4bd217b JG |
137 | struct ppa_addr ppa; /* Physic addr. associated with entry */ |
138 | int flags; /* Write context flags */ | |
139 | }; | |
140 | ||
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 */ | |
146 | }; | |
147 | ||
148 | #define EMPTY_ENTRY (~0U) | |
149 | ||
150 | struct pblk_rb_pages { | |
151 | struct page *pages; | |
152 | int order; | |
153 | struct list_head list; | |
154 | }; | |
155 | ||
156 | struct pblk_rb { | |
157 | struct pblk_rb_entry *entries; /* Ring buffer entries */ | |
158 | unsigned int mem; /* Write offset - points to next | |
159 | * writable entry in memory | |
160 | */ | |
161 | unsigned int subm; /* Read offset - points to last entry | |
162 | * that has been submitted to the media | |
163 | * to be persisted | |
164 | */ | |
165 | unsigned int sync; /* Synced - backpointer that signals | |
166 | * the last submitted entry that has | |
167 | * been successfully persisted to media | |
168 | */ | |
169 | unsigned int sync_point; /* Sync point - last entry that must be | |
170 | * flushed to the media. Used with | |
171 | * REQ_FLUSH and REQ_FUA | |
172 | */ | |
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 | |
176 | * of a cacheline | |
177 | */ | |
178 | unsigned int nr_entries; /* Number of entries in write buffer - | |
179 | * must be a power of two | |
180 | */ | |
181 | unsigned int seg_size; /* Size of the data segments being | |
182 | * stored on each entry. Typically this | |
183 | * will be 4KB | |
184 | */ | |
185 | ||
186 | struct list_head pages; /* List of data pages */ | |
187 | ||
188 | spinlock_t w_lock; /* Write lock */ | |
189 | spinlock_t s_lock; /* Sync lock */ | |
190 | ||
191 | #ifdef CONFIG_NVM_DEBUG | |
192 | atomic_t inflight_sync_point; /* Not served REQ_FLUSH | REQ_FUA */ | |
193 | #endif | |
194 | }; | |
195 | ||
196 | #define PBLK_RECOVERY_SECTORS 16 | |
197 | ||
198 | struct pblk_lun { | |
199 | struct ppa_addr bppa; | |
200 | ||
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. | |
204 | */ | |
205 | ||
206 | struct semaphore wr_sem; | |
207 | }; | |
208 | ||
209 | struct pblk_gc_rq { | |
210 | struct pblk_line *line; | |
211 | void *data; | |
d340121e | 212 | u64 paddr_list[PBLK_MAX_REQ_ADDRS]; |
b20ba1bc | 213 | u64 lba_list[PBLK_MAX_REQ_ADDRS]; |
a4bd217b JG |
214 | int nr_secs; |
215 | int secs_to_gc; | |
216 | struct list_head list; | |
217 | }; | |
218 | ||
219 | struct pblk_gc { | |
b20ba1bc JG |
220 | /* These states are not protected by a lock since (i) they are in the |
221 | * fast path, and (ii) they are not critical. | |
222 | */ | |
a4bd217b JG |
223 | int gc_active; |
224 | int gc_enabled; | |
225 | int gc_forced; | |
a4bd217b JG |
226 | |
227 | struct task_struct *gc_ts; | |
228 | struct task_struct *gc_writer_ts; | |
b20ba1bc JG |
229 | struct task_struct *gc_reader_ts; |
230 | ||
231 | struct workqueue_struct *gc_line_reader_wq; | |
a4bd217b | 232 | struct workqueue_struct *gc_reader_wq; |
b20ba1bc | 233 | |
a4bd217b JG |
234 | struct timer_list gc_timer; |
235 | ||
b20ba1bc | 236 | struct semaphore gc_sem; |
d6b992f7 HH |
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 | |
240 | */ | |
a4bd217b | 241 | int w_entries; |
b20ba1bc | 242 | |
a4bd217b | 243 | struct list_head w_list; |
b20ba1bc | 244 | struct list_head r_list; |
a4bd217b JG |
245 | |
246 | spinlock_t lock; | |
247 | spinlock_t w_lock; | |
b20ba1bc | 248 | spinlock_t r_lock; |
a4bd217b JG |
249 | }; |
250 | ||
251 | struct pblk_rl { | |
252 | unsigned int high; /* Upper threshold for rate limiter (free run - | |
253 | * user I/O rate limiter | |
254 | */ | |
255 | unsigned int low; /* Lower threshold for rate limiter (user I/O | |
256 | * rate limiter - stall) | |
257 | */ | |
258 | unsigned int high_pw; /* High rounded up as a power of 2 */ | |
259 | ||
b20ba1bc JG |
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 */ | |
a4bd217b JG |
262 | |
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 | |
269 | * 64, i.e., 256kb. | |
270 | */ | |
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 */ | |
a4bd217b JG |
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 */ | |
da67e68f | 276 | int rb_max_io; /* Maximum size for an I/O giving the config */ |
588726d3 JG |
277 | |
278 | atomic_t rb_user_cnt; /* User I/O buffer counter */ | |
a4bd217b | 279 | atomic_t rb_gc_cnt; /* GC I/O buffer counter */ |
588726d3 | 280 | atomic_t rb_space; /* Space limit in case of reaching capacity */ |
a4bd217b | 281 | |
b20ba1bc JG |
282 | int rsv_blocks; /* Reserved blocks for GC */ |
283 | ||
a4bd217b | 284 | int rb_user_active; |
b20ba1bc JG |
285 | int rb_gc_active; |
286 | ||
a4bd217b JG |
287 | struct timer_list u_timer; |
288 | ||
289 | unsigned long long nr_secs; | |
290 | unsigned long total_blocks; | |
291 | atomic_t free_blocks; | |
292 | }; | |
293 | ||
a4bd217b JG |
294 | #define PBLK_LINE_EMPTY (~0U) |
295 | ||
296 | enum { | |
297 | /* Line Types */ | |
298 | PBLK_LINETYPE_FREE = 0, | |
299 | PBLK_LINETYPE_LOG = 1, | |
300 | PBLK_LINETYPE_DATA = 2, | |
301 | ||
302 | /* Line state */ | |
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, | |
309 | ||
310 | /* GC group */ | |
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, | |
317 | }; | |
318 | ||
319 | #define PBLK_MAGIC 0x70626c6b /*pblk*/ | |
c79819bc | 320 | #define SMETA_VERSION cpu_to_le16(1) |
a4bd217b JG |
321 | |
322 | struct line_header { | |
323 | __le32 crc; | |
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 */ | |
329 | }; | |
330 | ||
331 | struct line_smeta { | |
332 | struct line_header header; | |
333 | ||
334 | __le32 crc; /* Full structure including struct crc */ | |
335 | /* Previous line metadata */ | |
336 | __le32 prev_id; /* Line id for previous line */ | |
337 | ||
338 | /* Current line metadata */ | |
339 | __le64 seq_nr; /* Sequence number for current line */ | |
340 | ||
341 | /* Active writers */ | |
342 | __le32 window_wr_lun; /* Number of parallel LUNs to write */ | |
343 | ||
344 | __le32 rsvd[2]; | |
dd2a4343 JG |
345 | |
346 | __le64 lun_bitmap[]; | |
a4bd217b JG |
347 | }; |
348 | ||
349 | /* | |
dd2a4343 JG |
350 | * Metadata layout in media: |
351 | * First sector: | |
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) | |
a4bd217b JG |
358 | */ |
359 | struct line_emeta { | |
360 | struct line_header header; | |
361 | ||
362 | __le32 crc; /* Full structure including struct crc */ | |
363 | ||
364 | /* Previous line metadata */ | |
365 | __le32 prev_id; /* Line id for prev line */ | |
366 | ||
367 | /* Current line metadata */ | |
368 | __le64 seq_nr; /* Sequence number for current line */ | |
369 | ||
370 | /* Active writers */ | |
371 | __le32 window_wr_lun; /* Number of parallel LUNs to write */ | |
372 | ||
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 */ | |
dd2a4343 JG |
377 | __le64 bb_bitmap[]; /* Updated bad block bitmap for line */ |
378 | }; | |
379 | ||
380 | struct pblk_emeta { | |
381 | struct line_emeta *buf; /* emeta buffer in media format */ | |
382 | int mem; /* Write offset - points to next | |
383 | * writable entry in memory | |
384 | */ | |
385 | atomic_t sync; /* Synced - backpointer that signals the | |
386 | * last entry that has been successfully | |
387 | * persisted to media | |
388 | */ | |
389 | unsigned int nr_entries; /* Number of emeta entries */ | |
390 | }; | |
391 | ||
392 | struct pblk_smeta { | |
393 | struct line_smeta *buf; /* smeta buffer in persistent format */ | |
a4bd217b JG |
394 | }; |
395 | ||
396 | struct pblk_line { | |
397 | struct pblk *pblk; | |
398 | unsigned int id; /* Line number corresponds to the | |
399 | * block line | |
400 | */ | |
401 | unsigned int seq_nr; /* Unique line sequence number */ | |
402 | ||
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 */ | |
407 | ||
408 | unsigned long *lun_bitmap; /* Bitmap for LUNs mapped in line */ | |
409 | ||
dd2a4343 JG |
410 | struct pblk_smeta *smeta; /* Start metadata */ |
411 | struct pblk_emeta *emeta; /* End medatada */ | |
412 | ||
a4bd217b | 413 | int meta_line; /* Metadata line id */ |
dd2a4343 JG |
414 | int meta_distance; /* Distance between data and metadata */ |
415 | ||
a4bd217b JG |
416 | u64 smeta_ssec; /* Sector where smeta starts */ |
417 | u64 emeta_ssec; /* Sector where emeta starts */ | |
418 | ||
419 | unsigned int sec_in_line; /* Number of usable secs in line */ | |
420 | ||
a44f53fa | 421 | atomic_t blk_in_line; /* Number of good blocks in line */ |
a4bd217b JG |
422 | unsigned long *blk_bitmap; /* Bitmap for valid/invalid blocks */ |
423 | unsigned long *erase_bitmap; /* Bitmap for erased blocks */ | |
424 | ||
425 | unsigned long *map_bitmap; /* Bitmap for mapped sectors in line */ | |
426 | unsigned long *invalid_bitmap; /* Bitmap for invalid sectors in line */ | |
427 | ||
a44f53fa | 428 | atomic_t left_eblks; /* Blocks left for erasing */ |
a4bd217b JG |
429 | atomic_t left_seblks; /* Blocks left for sync erasing */ |
430 | ||
431 | int left_msecs; /* Sectors left for mapping */ | |
a4bd217b | 432 | unsigned int cur_sec; /* Sector map pointer */ |
dd2a4343 JG |
433 | unsigned int nr_valid_lbas; /* Number of valid lbas in line */ |
434 | ||
435 | __le32 *vsc; /* Valid sector count in line */ | |
a4bd217b JG |
436 | |
437 | struct kref ref; /* Write buffer L2P references */ | |
438 | ||
439 | spinlock_t lock; /* Necessary for invalid_bitmap only */ | |
440 | }; | |
441 | ||
a44f53fa | 442 | #define PBLK_DATA_LINES 4 |
a4bd217b | 443 | |
dd2a4343 | 444 | enum { |
a4bd217b JG |
445 | PBLK_KMALLOC_META = 1, |
446 | PBLK_VMALLOC_META = 2, | |
447 | }; | |
448 | ||
dd2a4343 JG |
449 | enum { |
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 */ | |
a4bd217b JG |
453 | }; |
454 | ||
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 */ | |
458 | ||
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 */ | |
463 | ||
464 | /* GC lists - use gc_lock */ | |
b20ba1bc | 465 | struct list_head *gc_lists[PBLK_GC_NR_LISTS]; |
a4bd217b JG |
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 */ | |
469 | ||
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 */ | |
472 | ||
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 */ | |
477 | ||
dd2a4343 JG |
478 | struct list_head emeta_list; /* Lines queued to schedule emeta */ |
479 | ||
480 | __le32 *vsc_list; /* Valid sector counts for all lines */ | |
481 | ||
a4bd217b | 482 | /* Metadata allocation type: VMALLOC | KMALLOC */ |
a4bd217b JG |
483 | int emeta_alloc_type; |
484 | ||
485 | /* Pre-allocated metadata for data lines */ | |
dd2a4343 JG |
486 | struct pblk_smeta *sline_meta[PBLK_DATA_LINES]; |
487 | struct pblk_emeta *eline_meta[PBLK_DATA_LINES]; | |
a4bd217b JG |
488 | unsigned long meta_bitmap; |
489 | ||
490 | /* Helpers for fast bitmap calculations */ | |
491 | unsigned long *bb_template; | |
492 | unsigned long *bb_aux; | |
493 | ||
494 | unsigned long d_seq_nr; /* Data line unique sequence number */ | |
495 | unsigned long l_seq_nr; /* Log line unique sequence number */ | |
496 | ||
497 | spinlock_t free_lock; | |
dd2a4343 | 498 | spinlock_t close_lock; |
a4bd217b JG |
499 | spinlock_t gc_lock; |
500 | }; | |
501 | ||
502 | struct pblk_line_meta { | |
503 | unsigned int smeta_len; /* Total length for smeta */ | |
dd2a4343 JG |
504 | unsigned int smeta_sec; /* Sectors needed for smeta */ |
505 | ||
506 | unsigned int emeta_len[4]; /* Lengths for emeta: | |
507 | * [0]: Total length | |
508 | * [1]: struct line_emeta length | |
509 | * [2]: L2P portion length | |
510 | * [3]: vsc list length | |
511 | */ | |
512 | unsigned int emeta_sec[4]; /* Sectors needed for emeta. Same layout | |
513 | * as emeta_len | |
514 | */ | |
515 | ||
a4bd217b | 516 | unsigned int emeta_bb; /* Boundary for bb that affects emeta */ |
dd2a4343 JG |
517 | |
518 | unsigned int vsc_list_len; /* Length for vsc list */ | |
a4bd217b JG |
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 */ | |
522 | ||
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 */ | |
dd2a4343 | 525 | unsigned int dsec_per_line; /* Number of data sectors in a line */ |
a4bd217b JG |
526 | unsigned int min_blk_line; /* Min. number of good blocks in line */ |
527 | ||
528 | unsigned int mid_thrs; /* Threshold for GC mid list */ | |
529 | unsigned int high_thrs; /* Threshold for GC high list */ | |
dd2a4343 JG |
530 | |
531 | unsigned int meta_distance; /* Distance between data and metadata */ | |
a4bd217b JG |
532 | }; |
533 | ||
534 | struct pblk_addr_format { | |
535 | u64 ch_mask; | |
536 | u64 lun_mask; | |
537 | u64 pln_mask; | |
538 | u64 blk_mask; | |
539 | u64 pg_mask; | |
540 | u64 sec_mask; | |
541 | u8 ch_offset; | |
542 | u8 lun_offset; | |
543 | u8 pln_offset; | |
544 | u8 blk_offset; | |
545 | u8 pg_offset; | |
546 | u8 sec_offset; | |
547 | }; | |
548 | ||
588726d3 JG |
549 | enum { |
550 | PBLK_STATE_RUNNING = 0, | |
551 | PBLK_STATE_STOPPING = 1, | |
552 | PBLK_STATE_RECOVERING = 2, | |
553 | PBLK_STATE_STOPPED = 3, | |
554 | }; | |
555 | ||
a4bd217b JG |
556 | struct pblk { |
557 | struct nvm_tgt_dev *dev; | |
558 | struct gendisk *disk; | |
559 | ||
560 | struct kobject kobj; | |
561 | ||
562 | struct pblk_lun *luns; | |
563 | ||
564 | struct pblk_line *lines; /* Line array */ | |
565 | struct pblk_line_mgmt l_mg; /* Line management */ | |
566 | struct pblk_line_meta lm; /* Line metadata */ | |
567 | ||
568 | int ppaf_bitsize; | |
569 | struct pblk_addr_format ppaf; | |
570 | ||
571 | struct pblk_rb rwb; | |
572 | ||
588726d3 JG |
573 | int state; /* pblk line state */ |
574 | ||
a4bd217b JG |
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. | |
579 | */ | |
580 | ||
581 | sector_t capacity; /* Device capacity when bad blocks are subtracted */ | |
582 | int over_pct; /* Percentage of device used for over-provisioning */ | |
583 | ||
584 | /* pblk provisioning values. Used by rate limiter */ | |
585 | struct pblk_rl rl; | |
586 | ||
c2e9f5d4 | 587 | int sec_per_write; |
a4bd217b JG |
588 | |
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 */ | |
a4bd217b | 599 | atomic_long_t inflight_reads; /* Inflight sector read requests */ |
db7ada33 | 600 | atomic_long_t cache_reads; /* Read requests that hit the cache */ |
a4bd217b JG |
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 */ | |
605 | #endif | |
606 | ||
607 | spinlock_t lock; | |
608 | ||
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; | |
615 | ||
588726d3 JG |
616 | atomic_t inflight_io; /* General inflight I/O counter */ |
617 | ||
a4bd217b JG |
618 | struct task_struct *writer_ts; |
619 | ||
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. | |
623 | */ | |
624 | unsigned char *trans_map; | |
625 | spinlock_t trans_lock; | |
626 | ||
627 | struct list_head compl_list; | |
628 | ||
bd432417 | 629 | mempool_t *page_bio_pool; |
b84ae4a8 | 630 | mempool_t *gen_ws_pool; |
a4bd217b | 631 | mempool_t *rec_pool; |
0d880398 | 632 | mempool_t *r_rq_pool; |
a4bd217b | 633 | mempool_t *w_rq_pool; |
0d880398 | 634 | mempool_t *e_rq_pool; |
a4bd217b | 635 | |
ef576494 JG |
636 | struct workqueue_struct *close_wq; |
637 | struct workqueue_struct *bb_wq; | |
7bd4d370 | 638 | struct workqueue_struct *r_end_wq; |
ef576494 | 639 | |
a4bd217b JG |
640 | struct timer_list wtimer; |
641 | ||
642 | struct pblk_gc gc; | |
643 | }; | |
644 | ||
645 | struct pblk_line_ws { | |
646 | struct pblk *pblk; | |
647 | struct pblk_line *line; | |
648 | void *priv; | |
649 | struct work_struct ws; | |
650 | }; | |
651 | ||
084ec9ba | 652 | #define pblk_g_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_g_ctx)) |
a4bd217b JG |
653 | #define pblk_w_rq_size (sizeof(struct nvm_rq) + sizeof(struct pblk_c_ctx)) |
654 | ||
655 | /* | |
656 | * pblk ring buffer operations | |
657 | */ | |
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, | |
665 | unsigned int *pos); | |
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, | |
d340121e JG |
669 | struct pblk_w_ctx w_ctx, struct pblk_line *line, |
670 | u64 paddr, unsigned int pos); | |
a4bd217b | 671 | struct pblk_w_ctx *pblk_rb_w_ctx(struct pblk_rb *rb, unsigned int pos); |
588726d3 | 672 | void pblk_rb_flush(struct pblk_rb *rb); |
a4bd217b JG |
673 | |
674 | void pblk_rb_sync_l2p(struct pblk_rb *rb); | |
d624f371 | 675 | unsigned int pblk_rb_read_to_bio(struct pblk_rb *rb, struct nvm_rq *rqd, |
875d94f3 JG |
676 | unsigned int pos, unsigned int nr_entries, |
677 | unsigned int count); | |
a4bd217b JG |
678 | unsigned int pblk_rb_read_to_bio_list(struct pblk_rb *rb, struct bio *bio, |
679 | struct list_head *list, | |
680 | unsigned int max); | |
681 | int pblk_rb_copy_to_bio(struct pblk_rb *rb, struct bio *bio, sector_t lba, | |
75cb8e93 | 682 | struct ppa_addr ppa, int bio_iter, bool advanced_bio); |
a4bd217b JG |
683 | unsigned int pblk_rb_read_commit(struct pblk_rb *rb, unsigned int entries); |
684 | ||
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_sync_point_count(struct pblk_rb *rb); | |
691 | ||
692 | unsigned int pblk_rb_read_count(struct pblk_rb *rb); | |
ee8d5c1a | 693 | unsigned int pblk_rb_sync_count(struct pblk_rb *rb); |
a4bd217b JG |
694 | unsigned int pblk_rb_wrap_pos(struct pblk_rb *rb, unsigned int pos); |
695 | ||
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); | |
700 | ||
701 | /* | |
702 | * pblk core | |
703 | */ | |
67bf26a3 JG |
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); | |
c2e9f5d4 | 706 | void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write); |
a4bd217b JG |
707 | int pblk_setup_w_rec_rq(struct pblk *pblk, struct nvm_rq *rqd, |
708 | struct pblk_c_ctx *c_ctx); | |
a4bd217b JG |
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); | |
1a94b2d4 | 713 | int pblk_submit_io_sync(struct pblk *pblk, struct nvm_rq *rqd); |
dd2a4343 | 714 | int pblk_submit_meta_io(struct pblk *pblk, struct pblk_line *meta_line); |
a4bd217b JG |
715 | struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data, |
716 | unsigned int nr_secs, unsigned int len, | |
de54e703 | 717 | int alloc_type, gfp_t gfp_mask); |
a4bd217b JG |
718 | struct pblk_line *pblk_line_get(struct pblk *pblk); |
719 | struct pblk_line *pblk_line_get_first_data(struct pblk *pblk); | |
21d22871 | 720 | struct pblk_line *pblk_line_replace_data(struct pblk *pblk); |
a4bd217b JG |
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); | |
d624f371 | 724 | struct pblk_line *pblk_line_get_erase(struct pblk *pblk); |
a4bd217b JG |
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); | |
dd2a4343 | 728 | void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line); |
a4bd217b | 729 | void pblk_line_close(struct pblk *pblk, struct pblk_line *line); |
dd2a4343 | 730 | void pblk_line_close_ws(struct work_struct *work); |
588726d3 | 731 | void pblk_pipeline_stop(struct pblk *pblk); |
b84ae4a8 JG |
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); | |
a4bd217b JG |
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); | |
dd2a4343 JG |
737 | int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line, |
738 | void *emeta_buf); | |
a4bd217b JG |
739 | int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr erase_ppa); |
740 | void pblk_line_put(struct kref *ref); | |
7bd4d370 | 741 | void pblk_line_put_wq(struct kref *ref); |
a4bd217b | 742 | struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line); |
dd2a4343 JG |
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); | |
a4bd217b | 745 | u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs); |
dd2a4343 | 746 | u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs); |
a4bd217b JG |
747 | int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail, |
748 | unsigned long secs_to_flush); | |
3eaa11e2 | 749 | void pblk_up_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas); |
a4bd217b JG |
750 | void pblk_down_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas, |
751 | unsigned long *lun_bitmap); | |
3eaa11e2 | 752 | void pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas); |
a4bd217b JG |
753 | void pblk_up_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas, |
754 | unsigned long *lun_bitmap); | |
a4bd217b JG |
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, | |
757 | int nr_pages); | |
a4bd217b JG |
758 | void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off, |
759 | int nr_pages); | |
760 | void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa); | |
0880a9aa JG |
761 | void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line, |
762 | u64 paddr); | |
a4bd217b JG |
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, | |
d340121e | 769 | struct pblk_line *gc_line, u64 paddr); |
a4bd217b JG |
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); | |
774 | ||
775 | /* | |
776 | * pblk user I/O write path | |
777 | */ | |
778 | int pblk_write_to_cache(struct pblk *pblk, struct bio *bio, | |
779 | unsigned long flags); | |
d340121e | 780 | int pblk_write_gc_to_cache(struct pblk *pblk, struct pblk_gc_rq *gc_rq); |
a4bd217b JG |
781 | |
782 | /* | |
783 | * pblk map | |
784 | */ | |
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, | |
790 | unsigned int off); | |
791 | ||
792 | /* | |
793 | * pblk write thread | |
794 | */ | |
795 | int pblk_write_ts(void *data); | |
87c1d2d3 | 796 | void pblk_write_timer_fn(struct timer_list *t); |
a4bd217b JG |
797 | void pblk_write_should_kick(struct pblk *pblk); |
798 | ||
799 | /* | |
800 | * pblk read path | |
801 | */ | |
b25d5237 | 802 | extern struct bio_set *pblk_bio_set; |
a4bd217b | 803 | int pblk_submit_read(struct pblk *pblk, struct bio *bio); |
d340121e | 804 | int pblk_submit_read_gc(struct pblk *pblk, struct pblk_gc_rq *gc_rq); |
a4bd217b JG |
805 | /* |
806 | * pblk recovery | |
807 | */ | |
808 | void pblk_submit_rec(struct work_struct *work); | |
809 | struct pblk_line *pblk_recov_l2p(struct pblk *pblk); | |
588726d3 | 810 | int pblk_recov_pad(struct pblk *pblk); |
a4bd217b JG |
811 | __le64 *pblk_recov_get_lba_list(struct pblk *pblk, struct line_emeta *emeta); |
812 | int pblk_recov_setup_rq(struct pblk *pblk, struct pblk_c_ctx *c_ctx, | |
813 | struct pblk_rec_ctx *recovery, u64 *comp_bits, | |
814 | unsigned int comp); | |
815 | ||
816 | /* | |
817 | * pblk gc | |
818 | */ | |
b20ba1bc | 819 | #define PBLK_GC_MAX_READERS 8 /* Max number of outstanding GC reader jobs */ |
3627896a | 820 | #define PBLK_GC_RQ_QD 128 /* Queue depth for inflight GC requests */ |
b20ba1bc JG |
821 | #define PBLK_GC_L_QD 4 /* Queue depth for inflight GC lines */ |
822 | #define PBLK_GC_RSV_LINE 1 /* Reserved lines for GC */ | |
a4bd217b JG |
823 | |
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); | |
03661b5f | 828 | void pblk_gc_should_kick(struct pblk *pblk); |
37ce33d5 | 829 | void pblk_gc_free_full_lines(struct pblk *pblk); |
a4bd217b JG |
830 | void pblk_gc_sysfs_state_show(struct pblk *pblk, int *gc_enabled, |
831 | int *gc_active); | |
b20ba1bc | 832 | int pblk_gc_sysfs_force(struct pblk *pblk, int force); |
a4bd217b JG |
833 | |
834 | /* | |
835 | * pblk rate limiter | |
836 | */ | |
837 | void pblk_rl_init(struct pblk_rl *rl, int budget); | |
838 | void pblk_rl_free(struct pblk_rl *rl); | |
03661b5f | 839 | void pblk_rl_update_rates(struct pblk_rl *rl); |
b20ba1bc | 840 | int pblk_rl_high_thrs(struct pblk_rl *rl); |
a4bd217b JG |
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); | |
588726d3 | 843 | void pblk_rl_inserted(struct pblk_rl *rl, int nr_entries); |
a4bd217b JG |
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); | |
da67e68f | 848 | int pblk_rl_max_io(struct pblk_rl *rl); |
a4bd217b JG |
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); | |
588726d3 | 851 | int pblk_rl_is_limit(struct pblk_rl *rl); |
a4bd217b JG |
852 | |
853 | /* | |
854 | * pblk sysfs | |
855 | */ | |
856 | int pblk_sysfs_init(struct gendisk *tdisk); | |
857 | void pblk_sysfs_exit(struct gendisk *tdisk); | |
858 | ||
859 | static inline void *pblk_malloc(size_t size, int type, gfp_t flags) | |
860 | { | |
861 | if (type == PBLK_KMALLOC_META) | |
862 | return kmalloc(size, flags); | |
863 | return vmalloc(size); | |
864 | } | |
865 | ||
866 | static inline void pblk_mfree(void *ptr, int type) | |
867 | { | |
868 | if (type == PBLK_KMALLOC_META) | |
869 | kfree(ptr); | |
870 | else | |
871 | vfree(ptr); | |
872 | } | |
873 | ||
874 | static inline struct nvm_rq *nvm_rq_from_c_ctx(void *c_ctx) | |
875 | { | |
876 | return c_ctx - sizeof(struct nvm_rq); | |
877 | } | |
878 | ||
dd2a4343 JG |
879 | static inline void *emeta_to_bb(struct line_emeta *emeta) |
880 | { | |
881 | return emeta->bb_bitmap; | |
882 | } | |
883 | ||
884 | static inline void *emeta_to_lbas(struct pblk *pblk, struct line_emeta *emeta) | |
885 | { | |
886 | return ((void *)emeta + pblk->lm.emeta_len[1]); | |
887 | } | |
888 | ||
889 | static inline void *emeta_to_vsc(struct pblk *pblk, struct line_emeta *emeta) | |
a4bd217b | 890 | { |
dd2a4343 | 891 | return (emeta_to_lbas(pblk, emeta) + pblk->lm.emeta_len[2]); |
a4bd217b JG |
892 | } |
893 | ||
b20ba1bc JG |
894 | static inline int pblk_line_vsc(struct pblk_line *line) |
895 | { | |
d340121e | 896 | return le32_to_cpu(*line->vsc); |
b20ba1bc JG |
897 | } |
898 | ||
a4bd217b JG |
899 | #define NVM_MEM_PAGE_WRITE (8) |
900 | ||
901 | static inline int pblk_pad_distance(struct pblk *pblk) | |
902 | { | |
903 | struct nvm_tgt_dev *dev = pblk->dev; | |
904 | struct nvm_geo *geo = &dev->geo; | |
905 | ||
fae7fae4 | 906 | return NVM_MEM_PAGE_WRITE * geo->all_luns * geo->sec_per_pl; |
a4bd217b JG |
907 | } |
908 | ||
b1bcfda1 | 909 | static inline int pblk_ppa_to_line(struct ppa_addr p) |
a4bd217b JG |
910 | { |
911 | return p.g.blk; | |
912 | } | |
913 | ||
b1bcfda1 | 914 | static inline int pblk_ppa_to_pos(struct nvm_geo *geo, struct ppa_addr p) |
a4bd217b | 915 | { |
b1bcfda1 | 916 | return p.g.lun * geo->nr_chnls + p.g.ch; |
a4bd217b JG |
917 | } |
918 | ||
b1bcfda1 JG |
919 | static inline struct ppa_addr addr_to_gen_ppa(struct pblk *pblk, u64 paddr, |
920 | u64 line_id) | |
a4bd217b | 921 | { |
b1bcfda1 JG |
922 | struct ppa_addr ppa; |
923 | ||
924 | ppa.ppa = 0; | |
925 | ppa.g.blk = line_id; | |
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; | |
931 | ||
932 | return ppa; | |
a4bd217b JG |
933 | } |
934 | ||
b1bcfda1 JG |
935 | static inline u64 pblk_dev_ppa_to_line_addr(struct pblk *pblk, |
936 | struct ppa_addr p) | |
a4bd217b | 937 | { |
b1bcfda1 JG |
938 | u64 paddr; |
939 | ||
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; | |
945 | ||
946 | return paddr; | |
a4bd217b JG |
947 | } |
948 | ||
949 | static inline struct ppa_addr pblk_ppa32_to_ppa64(struct pblk *pblk, u32 ppa32) | |
950 | { | |
951 | struct ppa_addr ppa64; | |
952 | ||
953 | ppa64.ppa = 0; | |
954 | ||
955 | if (ppa32 == -1) { | |
956 | ppa64.ppa = ADDR_EMPTY; | |
957 | } else if (ppa32 & (1U << 31)) { | |
958 | ppa64.c.line = ppa32 & ((~0U) >> 1); | |
959 | ppa64.c.is_cached = 1; | |
960 | } else { | |
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; | |
973 | } | |
974 | ||
975 | return ppa64; | |
976 | } | |
977 | ||
a4bd217b JG |
978 | static inline u32 pblk_ppa64_to_ppa32(struct pblk *pblk, struct ppa_addr ppa64) |
979 | { | |
980 | u32 ppa32 = 0; | |
981 | ||
982 | if (ppa64.ppa == ADDR_EMPTY) { | |
983 | ppa32 = ~0U; | |
984 | } else if (ppa64.c.is_cached) { | |
985 | ppa32 |= ppa64.c.line; | |
986 | ppa32 |= 1U << 31; | |
987 | } else { | |
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; | |
994 | } | |
995 | ||
996 | return ppa32; | |
997 | } | |
998 | ||
b1bcfda1 JG |
999 | static inline struct ppa_addr pblk_trans_map_get(struct pblk *pblk, |
1000 | sector_t lba) | |
a4bd217b | 1001 | { |
b1bcfda1 JG |
1002 | struct ppa_addr ppa; |
1003 | ||
a4bd217b JG |
1004 | if (pblk->ppaf_bitsize < 32) { |
1005 | u32 *map = (u32 *)pblk->trans_map; | |
1006 | ||
b1bcfda1 | 1007 | ppa = pblk_ppa32_to_ppa64(pblk, map[lba]); |
a4bd217b | 1008 | } else { |
b1bcfda1 | 1009 | struct ppa_addr *map = (struct ppa_addr *)pblk->trans_map; |
a4bd217b | 1010 | |
b1bcfda1 | 1011 | ppa = map[lba]; |
a4bd217b | 1012 | } |
b1bcfda1 JG |
1013 | |
1014 | return ppa; | |
a4bd217b JG |
1015 | } |
1016 | ||
b1bcfda1 JG |
1017 | static inline void pblk_trans_map_set(struct pblk *pblk, sector_t lba, |
1018 | struct ppa_addr ppa) | |
a4bd217b | 1019 | { |
b1bcfda1 JG |
1020 | if (pblk->ppaf_bitsize < 32) { |
1021 | u32 *map = (u32 *)pblk->trans_map; | |
a4bd217b | 1022 | |
b1bcfda1 JG |
1023 | map[lba] = pblk_ppa64_to_ppa32(pblk, ppa); |
1024 | } else { | |
1025 | u64 *map = (u64 *)pblk->trans_map; | |
a4bd217b | 1026 | |
b1bcfda1 JG |
1027 | map[lba] = ppa.ppa; |
1028 | } | |
a4bd217b JG |
1029 | } |
1030 | ||
1031 | static inline int pblk_ppa_empty(struct ppa_addr ppa_addr) | |
1032 | { | |
1033 | return (ppa_addr.ppa == ADDR_EMPTY); | |
1034 | } | |
1035 | ||
1036 | static inline void pblk_ppa_set_empty(struct ppa_addr *ppa_addr) | |
1037 | { | |
1038 | ppa_addr->ppa = ADDR_EMPTY; | |
1039 | } | |
1040 | ||
07698466 JG |
1041 | static inline bool pblk_ppa_comp(struct ppa_addr lppa, struct ppa_addr rppa) |
1042 | { | |
1043 | if (lppa.ppa == rppa.ppa) | |
1044 | return true; | |
1045 | ||
1046 | return false; | |
1047 | } | |
1048 | ||
a4bd217b JG |
1049 | static inline int pblk_addr_in_cache(struct ppa_addr ppa) |
1050 | { | |
1051 | return (ppa.ppa != ADDR_EMPTY && ppa.c.is_cached); | |
1052 | } | |
1053 | ||
1054 | static inline int pblk_addr_to_cacheline(struct ppa_addr ppa) | |
1055 | { | |
1056 | return ppa.c.line; | |
1057 | } | |
1058 | ||
1059 | static inline struct ppa_addr pblk_cacheline_to_addr(int addr) | |
1060 | { | |
1061 | struct ppa_addr p; | |
1062 | ||
1063 | p.c.line = addr; | |
1064 | p.c.is_cached = 1; | |
1065 | ||
1066 | return p; | |
1067 | } | |
1068 | ||
a4bd217b | 1069 | static inline u32 pblk_calc_meta_header_crc(struct pblk *pblk, |
dd2a4343 | 1070 | struct line_header *header) |
a4bd217b JG |
1071 | { |
1072 | u32 crc = ~(u32)0; | |
1073 | ||
dd2a4343 | 1074 | crc = crc32_le(crc, (unsigned char *)header + sizeof(crc), |
a4bd217b JG |
1075 | sizeof(struct line_header) - sizeof(crc)); |
1076 | ||
1077 | return crc; | |
1078 | } | |
1079 | ||
1080 | static inline u32 pblk_calc_smeta_crc(struct pblk *pblk, | |
1081 | struct line_smeta *smeta) | |
1082 | { | |
1083 | struct pblk_line_meta *lm = &pblk->lm; | |
1084 | u32 crc = ~(u32)0; | |
1085 | ||
1086 | crc = crc32_le(crc, (unsigned char *)smeta + | |
1087 | sizeof(struct line_header) + sizeof(crc), | |
1088 | lm->smeta_len - | |
1089 | sizeof(struct line_header) - sizeof(crc)); | |
1090 | ||
1091 | return crc; | |
1092 | } | |
1093 | ||
1094 | static inline u32 pblk_calc_emeta_crc(struct pblk *pblk, | |
1095 | struct line_emeta *emeta) | |
1096 | { | |
1097 | struct pblk_line_meta *lm = &pblk->lm; | |
1098 | u32 crc = ~(u32)0; | |
1099 | ||
1100 | crc = crc32_le(crc, (unsigned char *)emeta + | |
1101 | sizeof(struct line_header) + sizeof(crc), | |
dd2a4343 | 1102 | lm->emeta_len[0] - |
a4bd217b JG |
1103 | sizeof(struct line_header) - sizeof(crc)); |
1104 | ||
1105 | return crc; | |
1106 | } | |
1107 | ||
1108 | static inline int pblk_set_progr_mode(struct pblk *pblk, int type) | |
1109 | { | |
1110 | struct nvm_tgt_dev *dev = pblk->dev; | |
1111 | struct nvm_geo *geo = &dev->geo; | |
1112 | int flags; | |
1113 | ||
1114 | flags = geo->plane_mode >> 1; | |
1115 | ||
e2cddf20 | 1116 | if (type == PBLK_WRITE) |
a4bd217b JG |
1117 | flags |= NVM_IO_SCRAMBLE_ENABLE; |
1118 | ||
1119 | return flags; | |
1120 | } | |
1121 | ||
f9c10152 JG |
1122 | enum { |
1123 | PBLK_READ_RANDOM = 0, | |
1124 | PBLK_READ_SEQUENTIAL = 1, | |
1125 | }; | |
1126 | ||
1127 | static inline int pblk_set_read_mode(struct pblk *pblk, int type) | |
1128 | { | |
1129 | struct nvm_tgt_dev *dev = pblk->dev; | |
1130 | struct nvm_geo *geo = &dev->geo; | |
1131 | int flags; | |
1132 | ||
1133 | flags = NVM_IO_SUSPEND | NVM_IO_SCRAMBLE_ENABLE; | |
1134 | if (type == PBLK_READ_SEQUENTIAL) | |
1135 | flags |= geo->plane_mode >> 1; | |
1136 | ||
1137 | return flags; | |
1138 | } | |
1139 | ||
1140 | static inline int pblk_io_aligned(struct pblk *pblk, int nr_secs) | |
a4bd217b | 1141 | { |
f9c10152 | 1142 | return !(nr_secs % pblk->min_write_pgs); |
a4bd217b JG |
1143 | } |
1144 | ||
1145 | #ifdef CONFIG_NVM_DEBUG | |
1146 | static inline void print_ppa(struct ppa_addr *p, char *msg, int error) | |
1147 | { | |
1148 | if (p->c.is_cached) { | |
1149 | pr_err("ppa: (%s: %x) cache line: %llu\n", | |
1150 | msg, error, (u64)p->c.line); | |
1151 | } else { | |
1152 | pr_err("ppa: (%s: %x):ch:%d,lun:%d,blk:%d,pg:%d,pl:%d,sec:%d\n", | |
1153 | msg, error, | |
1154 | p->g.ch, p->g.lun, p->g.blk, | |
1155 | p->g.pg, p->g.pl, p->g.sec); | |
1156 | } | |
1157 | } | |
1158 | ||
1159 | static inline void pblk_print_failed_rqd(struct pblk *pblk, struct nvm_rq *rqd, | |
1160 | int error) | |
1161 | { | |
1162 | int bit = -1; | |
1163 | ||
1164 | if (rqd->nr_ppas == 1) { | |
1165 | print_ppa(&rqd->ppa_addr, "rqd", error); | |
1166 | return; | |
1167 | } | |
1168 | ||
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); | |
1172 | } | |
1173 | ||
1174 | pr_err("error:%d, ppa_status:%llx\n", error, rqd->ppa_status); | |
1175 | } | |
a4bd217b JG |
1176 | |
1177 | static inline int pblk_boundary_ppa_checks(struct nvm_tgt_dev *tgt_dev, | |
1178 | struct ppa_addr *ppas, int nr_ppas) | |
1179 | { | |
1180 | struct nvm_geo *geo = &tgt_dev->geo; | |
1181 | struct ppa_addr *ppa; | |
1182 | int i; | |
1183 | ||
1184 | for (i = 0; i < nr_ppas; i++) { | |
1185 | ppa = &ppas[i]; | |
1186 | ||
1187 | if (!ppa->c.is_cached && | |
1188 | ppa->g.ch < geo->nr_chnls && | |
fae7fae4 | 1189 | ppa->g.lun < geo->nr_luns && |
a4bd217b | 1190 | ppa->g.pl < geo->nr_planes && |
fae7fae4 MB |
1191 | ppa->g.blk < geo->nr_chks && |
1192 | ppa->g.pg < geo->ws_per_chk && | |
a4bd217b JG |
1193 | ppa->g.sec < geo->sec_per_pg) |
1194 | continue; | |
1195 | ||
a4bd217b | 1196 | print_ppa(ppa, "boundary", i); |
1a94b2d4 | 1197 | |
a4bd217b JG |
1198 | return 1; |
1199 | } | |
1200 | return 0; | |
1201 | } | |
1202 | ||
1a94b2d4 JG |
1203 | static inline int pblk_check_io(struct pblk *pblk, struct nvm_rq *rqd) |
1204 | { | |
1205 | struct nvm_tgt_dev *dev = pblk->dev; | |
1206 | struct ppa_addr *ppa_list; | |
1207 | ||
1208 | ppa_list = (rqd->nr_ppas > 1) ? rqd->ppa_list : &rqd->ppa_addr; | |
1209 | ||
1210 | if (pblk_boundary_ppa_checks(dev, ppa_list, rqd->nr_ppas)) { | |
1211 | WARN_ON(1); | |
1212 | return -EINVAL; | |
1213 | } | |
1214 | ||
1215 | if (rqd->opcode == NVM_OP_PWRITE) { | |
1216 | struct pblk_line *line; | |
1217 | struct ppa_addr ppa; | |
1218 | int i; | |
1219 | ||
1220 | for (i = 0; i < rqd->nr_ppas; i++) { | |
1221 | ppa = ppa_list[i]; | |
b1bcfda1 | 1222 | line = &pblk->lines[pblk_ppa_to_line(ppa)]; |
1a94b2d4 JG |
1223 | |
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); | |
1228 | WARN_ON(1); | |
1229 | spin_unlock(&line->lock); | |
1230 | return -EINVAL; | |
1231 | } | |
1232 | spin_unlock(&line->lock); | |
1233 | } | |
1234 | } | |
1235 | ||
1236 | return 0; | |
1237 | } | |
1238 | #endif | |
1239 | ||
a4bd217b JG |
1240 | static inline int pblk_boundary_paddr_checks(struct pblk *pblk, u64 paddr) |
1241 | { | |
1242 | struct pblk_line_meta *lm = &pblk->lm; | |
1243 | ||
1244 | if (paddr > lm->sec_per_line) | |
1245 | return 1; | |
1246 | ||
1247 | return 0; | |
1248 | } | |
1249 | ||
1250 | static inline unsigned int pblk_get_bi_idx(struct bio *bio) | |
1251 | { | |
1252 | return bio->bi_iter.bi_idx; | |
1253 | } | |
1254 | ||
1255 | static inline sector_t pblk_get_lba(struct bio *bio) | |
1256 | { | |
1257 | return bio->bi_iter.bi_sector / NR_PHY_IN_LOG; | |
1258 | } | |
1259 | ||
1260 | static inline unsigned int pblk_get_secs(struct bio *bio) | |
1261 | { | |
1262 | return bio->bi_iter.bi_size / PBLK_EXPOSED_PAGE_SIZE; | |
1263 | } | |
1264 | ||
a4bd217b JG |
1265 | static inline void pblk_setup_uuid(struct pblk *pblk) |
1266 | { | |
1267 | uuid_le uuid; | |
1268 | ||
1269 | uuid_le_gen(&uuid); | |
1270 | memcpy(pblk->instance_uuid, uuid.b, 16); | |
1271 | } | |
1272 | #endif /* PBLK_H_ */ |