]>
Commit | Line | Data |
---|---|---|
02a1520d | 1 | // SPDX-License-Identifier: GPL-2.0 |
a4bd217b JG |
2 | /* |
3 | * Copyright (C) 2015 IT University of Copenhagen (rrpc.c) | |
4 | * Copyright (C) 2016 CNEX Labs | |
5 | * Initial release: Javier Gonzalez <javier@cnexlabs.com> | |
6 | * Matias Bjorling <matias@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 | * pblk-init.c - pblk's initialization. | |
20 | */ | |
21 | ||
22 | #include "pblk.h" | |
4c44abf4 | 23 | #include "pblk-trace.h" |
a4bd217b | 24 | |
21ff1399 | 25 | static unsigned int write_buffer_size; |
4a828884 MD |
26 | |
27 | module_param(write_buffer_size, uint, 0644); | |
28 | MODULE_PARM_DESC(write_buffer_size, "number of entries in a write buffer"); | |
29 | ||
1864de94 HH |
30 | struct pblk_global_caches { |
31 | struct kmem_cache *ws; | |
32 | struct kmem_cache *rec; | |
33 | struct kmem_cache *g_rq; | |
34 | struct kmem_cache *w_rq; | |
35 | ||
36 | struct kref kref; | |
37 | ||
38 | struct mutex mutex; /* Ensures consistency between | |
39 | * caches and kref | |
40 | */ | |
41 | }; | |
42 | ||
43 | static struct pblk_global_caches pblk_caches = { | |
44 | .mutex = __MUTEX_INITIALIZER(pblk_caches.mutex), | |
45 | .kref = KREF_INIT(0), | |
46 | }; | |
47 | ||
b906bbb6 | 48 | struct bio_set pblk_bio_set; |
a4bd217b | 49 | |
a4bd217b JG |
50 | static blk_qc_t pblk_make_rq(struct request_queue *q, struct bio *bio) |
51 | { | |
52 | struct pblk *pblk = q->queuedata; | |
53 | ||
54 | if (bio_op(bio) == REQ_OP_DISCARD) { | |
55 | pblk_discard(pblk, bio); | |
56 | if (!(bio->bi_opf & REQ_PREFLUSH)) { | |
57 | bio_endio(bio); | |
58 | return BLK_QC_T_NONE; | |
59 | } | |
60 | } | |
61 | ||
05038712 CK |
62 | /* Read requests must be <= 256kb due to NVMe's 64 bit completion bitmap |
63 | * constraint. Writes can be of arbitrary size. | |
64 | */ | |
65 | if (bio_data_dir(bio) == READ) { | |
66 | blk_queue_split(q, &bio); | |
3e03f632 | 67 | pblk_submit_read(pblk, bio); |
05038712 CK |
68 | } else { |
69 | /* Prevent deadlock in the case of a modest LUN configuration | |
70 | * and large user I/Os. Unless stalled, the rate limiter | |
71 | * leaves at least 256KB available for user I/O. | |
72 | */ | |
73 | if (pblk_get_secs(bio) > pblk_rl_max_io(&pblk->rl)) | |
74 | blk_queue_split(q, &bio); | |
75 | ||
3e03f632 | 76 | pblk_write_to_cache(pblk, bio, PBLK_IOTYPE_USER); |
a4bd217b JG |
77 | } |
78 | ||
79 | return BLK_QC_T_NONE; | |
80 | } | |
81 | ||
c5586192 HH |
82 | static size_t pblk_trans_map_size(struct pblk *pblk) |
83 | { | |
84 | int entry_size = 8; | |
85 | ||
bb845ae4 | 86 | if (pblk->addrf_len < 32) |
c5586192 HH |
87 | entry_size = 4; |
88 | ||
847a3a27 | 89 | return entry_size * pblk->capacity; |
c5586192 HH |
90 | } |
91 | ||
880eda54 | 92 | #ifdef CONFIG_NVM_PBLK_DEBUG |
c5586192 HH |
93 | static u32 pblk_l2p_crc(struct pblk *pblk) |
94 | { | |
95 | size_t map_size; | |
96 | u32 crc = ~(u32)0; | |
97 | ||
98 | map_size = pblk_trans_map_size(pblk); | |
99 | crc = crc32_le(crc, pblk->trans_map, map_size); | |
100 | return crc; | |
101 | } | |
102 | #endif | |
103 | ||
a4bd217b JG |
104 | static void pblk_l2p_free(struct pblk *pblk) |
105 | { | |
106 | vfree(pblk->trans_map); | |
107 | } | |
108 | ||
43d47127 JG |
109 | static int pblk_l2p_recover(struct pblk *pblk, bool factory_init) |
110 | { | |
111 | struct pblk_line *line = NULL; | |
112 | ||
113 | if (factory_init) { | |
7e0a0847 | 114 | guid_gen(&pblk->instance_uuid); |
43d47127 JG |
115 | } else { |
116 | line = pblk_recov_l2p(pblk); | |
117 | if (IS_ERR(line)) { | |
4e495a46 | 118 | pblk_err(pblk, "could not recover l2p table\n"); |
43d47127 JG |
119 | return -EFAULT; |
120 | } | |
121 | } | |
122 | ||
880eda54 | 123 | #ifdef CONFIG_NVM_PBLK_DEBUG |
4e495a46 | 124 | pblk_info(pblk, "init: L2P CRC: %x\n", pblk_l2p_crc(pblk)); |
43d47127 JG |
125 | #endif |
126 | ||
127 | /* Free full lines directly as GC has not been started yet */ | |
128 | pblk_gc_free_full_lines(pblk); | |
129 | ||
130 | if (!line) { | |
131 | /* Configure next line for user data */ | |
132 | line = pblk_line_get_first_data(pblk); | |
1d8b33e0 | 133 | if (!line) |
43d47127 | 134 | return -EFAULT; |
43d47127 JG |
135 | } |
136 | ||
137 | return 0; | |
138 | } | |
139 | ||
140 | static int pblk_l2p_init(struct pblk *pblk, bool factory_init) | |
a4bd217b JG |
141 | { |
142 | sector_t i; | |
143 | struct ppa_addr ppa; | |
c5586192 | 144 | size_t map_size; |
1d8b33e0 | 145 | int ret = 0; |
a4bd217b | 146 | |
c5586192 | 147 | map_size = pblk_trans_map_size(pblk); |
6e46b8b2 IK |
148 | pblk->trans_map = __vmalloc(map_size, GFP_KERNEL | __GFP_NOWARN |
149 | | __GFP_RETRY_MAYFAIL | __GFP_HIGHMEM, | |
150 | PAGE_KERNEL); | |
151 | if (!pblk->trans_map) { | |
152 | pblk_err(pblk, "failed to allocate L2P (need %zu of memory)\n", | |
153 | map_size); | |
a4bd217b | 154 | return -ENOMEM; |
6e46b8b2 | 155 | } |
a4bd217b JG |
156 | |
157 | pblk_ppa_set_empty(&ppa); | |
158 | ||
847a3a27 | 159 | for (i = 0; i < pblk->capacity; i++) |
a4bd217b JG |
160 | pblk_trans_map_set(pblk, i, ppa); |
161 | ||
1d8b33e0 JG |
162 | ret = pblk_l2p_recover(pblk, factory_init); |
163 | if (ret) | |
164 | vfree(pblk->trans_map); | |
165 | ||
166 | return ret; | |
a4bd217b JG |
167 | } |
168 | ||
169 | static void pblk_rwb_free(struct pblk *pblk) | |
170 | { | |
171 | if (pblk_rb_tear_down_check(&pblk->rwb)) | |
4e495a46 | 172 | pblk_err(pblk, "write buffer error on tear down\n"); |
a4bd217b | 173 | |
9bd1f875 | 174 | pblk_rb_free(&pblk->rwb); |
a4bd217b JG |
175 | } |
176 | ||
177 | static int pblk_rwb_init(struct pblk *pblk) | |
178 | { | |
179 | struct nvm_tgt_dev *dev = pblk->dev; | |
180 | struct nvm_geo *geo = &dev->geo; | |
9bd1f875 | 181 | unsigned long buffer_size; |
766c8ceb | 182 | int pgs_in_buffer, threshold; |
a4bd217b | 183 | |
766c8ceb | 184 | threshold = geo->mw_cunits * geo->all_luns; |
d672d92d JG |
185 | pgs_in_buffer = (max(geo->mw_cunits, geo->ws_opt) + geo->ws_opt) |
186 | * geo->all_luns; | |
ffc03fb7 MD |
187 | |
188 | if (write_buffer_size && (write_buffer_size > pgs_in_buffer)) | |
4a828884 MD |
189 | buffer_size = write_buffer_size; |
190 | else | |
ffc03fb7 | 191 | buffer_size = pgs_in_buffer; |
4a828884 | 192 | |
766c8ceb | 193 | return pblk_rb_init(&pblk->rwb, buffer_size, threshold, geo->csecs); |
a4bd217b JG |
194 | } |
195 | ||
4e495a46 MB |
196 | static int pblk_set_addrf_12(struct pblk *pblk, struct nvm_geo *geo, |
197 | struct nvm_addrf_12 *dst) | |
a4bd217b | 198 | { |
e46f4e48 JG |
199 | struct nvm_addrf_12 *src = (struct nvm_addrf_12 *)&geo->addrf; |
200 | int power_len; | |
a4bd217b JG |
201 | |
202 | /* Re-calculate channel and lun format to adapt to configuration */ | |
a40afad9 JG |
203 | power_len = get_count_order(geo->num_ch); |
204 | if (1 << power_len != geo->num_ch) { | |
4e495a46 | 205 | pblk_err(pblk, "supports only power-of-two channel config.\n"); |
a4bd217b JG |
206 | return -EINVAL; |
207 | } | |
e46f4e48 | 208 | dst->ch_len = power_len; |
a4bd217b | 209 | |
a40afad9 JG |
210 | power_len = get_count_order(geo->num_lun); |
211 | if (1 << power_len != geo->num_lun) { | |
4e495a46 | 212 | pblk_err(pblk, "supports only power-of-two LUN config.\n"); |
a4bd217b JG |
213 | return -EINVAL; |
214 | } | |
e46f4e48 JG |
215 | dst->lun_len = power_len; |
216 | ||
217 | dst->blk_len = src->blk_len; | |
218 | dst->pg_len = src->pg_len; | |
219 | dst->pln_len = src->pln_len; | |
a40afad9 | 220 | dst->sec_len = src->sec_len; |
e46f4e48 | 221 | |
a40afad9 JG |
222 | dst->sec_offset = 0; |
223 | dst->pln_offset = dst->sec_len; | |
e46f4e48 JG |
224 | dst->ch_offset = dst->pln_offset + dst->pln_len; |
225 | dst->lun_offset = dst->ch_offset + dst->ch_len; | |
226 | dst->pg_offset = dst->lun_offset + dst->lun_len; | |
227 | dst->blk_offset = dst->pg_offset + dst->pg_len; | |
228 | ||
a40afad9 | 229 | dst->sec_mask = ((1ULL << dst->sec_len) - 1) << dst->sec_offset; |
e46f4e48 JG |
230 | dst->pln_mask = ((1ULL << dst->pln_len) - 1) << dst->pln_offset; |
231 | dst->ch_mask = ((1ULL << dst->ch_len) - 1) << dst->ch_offset; | |
232 | dst->lun_mask = ((1ULL << dst->lun_len) - 1) << dst->lun_offset; | |
233 | dst->pg_mask = ((1ULL << dst->pg_len) - 1) << dst->pg_offset; | |
234 | dst->blk_mask = ((1ULL << dst->blk_len) - 1) << dst->blk_offset; | |
235 | ||
236 | return dst->blk_offset + src->blk_len; | |
237 | } | |
238 | ||
3b2a3ad1 JG |
239 | static int pblk_set_addrf_20(struct nvm_geo *geo, struct nvm_addrf *adst, |
240 | struct pblk_addrf *udst) | |
241 | { | |
242 | struct nvm_addrf *src = &geo->addrf; | |
243 | ||
244 | adst->ch_len = get_count_order(geo->num_ch); | |
245 | adst->lun_len = get_count_order(geo->num_lun); | |
246 | adst->chk_len = src->chk_len; | |
247 | adst->sec_len = src->sec_len; | |
248 | ||
249 | adst->sec_offset = 0; | |
250 | adst->ch_offset = adst->sec_len; | |
251 | adst->lun_offset = adst->ch_offset + adst->ch_len; | |
252 | adst->chk_offset = adst->lun_offset + adst->lun_len; | |
253 | ||
254 | adst->sec_mask = ((1ULL << adst->sec_len) - 1) << adst->sec_offset; | |
255 | adst->chk_mask = ((1ULL << adst->chk_len) - 1) << adst->chk_offset; | |
256 | adst->lun_mask = ((1ULL << adst->lun_len) - 1) << adst->lun_offset; | |
257 | adst->ch_mask = ((1ULL << adst->ch_len) - 1) << adst->ch_offset; | |
258 | ||
259 | udst->sec_stripe = geo->ws_opt; | |
260 | udst->ch_stripe = geo->num_ch; | |
261 | udst->lun_stripe = geo->num_lun; | |
262 | ||
263 | udst->sec_lun_stripe = udst->sec_stripe * udst->ch_stripe; | |
264 | udst->sec_ws_stripe = udst->sec_lun_stripe * udst->lun_stripe; | |
265 | ||
266 | return adst->chk_offset + adst->chk_len; | |
267 | } | |
268 | ||
bb845ae4 | 269 | static int pblk_set_addrf(struct pblk *pblk) |
e46f4e48 JG |
270 | { |
271 | struct nvm_tgt_dev *dev = pblk->dev; | |
272 | struct nvm_geo *geo = &dev->geo; | |
273 | int mod; | |
274 | ||
3b2a3ad1 JG |
275 | switch (geo->version) { |
276 | case NVM_OCSSD_SPEC_12: | |
277 | div_u64_rem(geo->clba, pblk->min_write_pgs, &mod); | |
278 | if (mod) { | |
4e495a46 | 279 | pblk_err(pblk, "bad configuration of sectors/pages\n"); |
3b2a3ad1 JG |
280 | return -EINVAL; |
281 | } | |
282 | ||
4e495a46 MB |
283 | pblk->addrf_len = pblk_set_addrf_12(pblk, geo, |
284 | (void *)&pblk->addrf); | |
3b2a3ad1 JG |
285 | break; |
286 | case NVM_OCSSD_SPEC_20: | |
287 | pblk->addrf_len = pblk_set_addrf_20(geo, (void *)&pblk->addrf, | |
4e495a46 | 288 | &pblk->uaddrf); |
3b2a3ad1 JG |
289 | break; |
290 | default: | |
4e495a46 | 291 | pblk_err(pblk, "OCSSD revision not supported (%d)\n", |
3b2a3ad1 | 292 | geo->version); |
e46f4e48 JG |
293 | return -EINVAL; |
294 | } | |
295 | ||
a4bd217b JG |
296 | return 0; |
297 | } | |
298 | ||
1864de94 | 299 | static int pblk_create_global_caches(void) |
a4bd217b | 300 | { |
1864de94 HH |
301 | |
302 | pblk_caches.ws = kmem_cache_create("pblk_blk_ws", | |
a4bd217b | 303 | sizeof(struct pblk_line_ws), 0, 0, NULL); |
1864de94 | 304 | if (!pblk_caches.ws) |
a4bd217b | 305 | return -ENOMEM; |
a4bd217b | 306 | |
1864de94 | 307 | pblk_caches.rec = kmem_cache_create("pblk_rec", |
a4bd217b | 308 | sizeof(struct pblk_rec_ctx), 0, 0, NULL); |
1864de94 HH |
309 | if (!pblk_caches.rec) |
310 | goto fail_destroy_ws; | |
a4bd217b | 311 | |
1864de94 | 312 | pblk_caches.g_rq = kmem_cache_create("pblk_g_rq", pblk_g_rq_size, |
a4bd217b | 313 | 0, 0, NULL); |
1864de94 HH |
314 | if (!pblk_caches.g_rq) |
315 | goto fail_destroy_rec; | |
a4bd217b | 316 | |
1864de94 | 317 | pblk_caches.w_rq = kmem_cache_create("pblk_w_rq", pblk_w_rq_size, |
a4bd217b | 318 | 0, 0, NULL); |
1864de94 HH |
319 | if (!pblk_caches.w_rq) |
320 | goto fail_destroy_g_rq; | |
a4bd217b JG |
321 | |
322 | return 0; | |
1864de94 HH |
323 | |
324 | fail_destroy_g_rq: | |
325 | kmem_cache_destroy(pblk_caches.g_rq); | |
326 | fail_destroy_rec: | |
327 | kmem_cache_destroy(pblk_caches.rec); | |
328 | fail_destroy_ws: | |
329 | kmem_cache_destroy(pblk_caches.ws); | |
330 | ||
331 | return -ENOMEM; | |
332 | } | |
333 | ||
334 | static int pblk_get_global_caches(void) | |
335 | { | |
42bd0384 | 336 | int ret = 0; |
1864de94 HH |
337 | |
338 | mutex_lock(&pblk_caches.mutex); | |
339 | ||
42bd0384 JG |
340 | if (kref_get_unless_zero(&pblk_caches.kref)) |
341 | goto out; | |
1864de94 HH |
342 | |
343 | ret = pblk_create_global_caches(); | |
1864de94 | 344 | if (!ret) |
42bd0384 | 345 | kref_init(&pblk_caches.kref); |
1864de94 | 346 | |
42bd0384 | 347 | out: |
1864de94 | 348 | mutex_unlock(&pblk_caches.mutex); |
1864de94 HH |
349 | return ret; |
350 | } | |
351 | ||
352 | static void pblk_destroy_global_caches(struct kref *ref) | |
353 | { | |
354 | struct pblk_global_caches *c; | |
355 | ||
356 | c = container_of(ref, struct pblk_global_caches, kref); | |
357 | ||
358 | kmem_cache_destroy(c->ws); | |
359 | kmem_cache_destroy(c->rec); | |
360 | kmem_cache_destroy(c->g_rq); | |
361 | kmem_cache_destroy(c->w_rq); | |
a4bd217b JG |
362 | } |
363 | ||
1864de94 | 364 | static void pblk_put_global_caches(void) |
22a4e061 | 365 | { |
1864de94 HH |
366 | mutex_lock(&pblk_caches.mutex); |
367 | kref_put(&pblk_caches.kref, pblk_destroy_global_caches); | |
368 | mutex_unlock(&pblk_caches.mutex); | |
22a4e061 RP |
369 | } |
370 | ||
a4bd217b JG |
371 | static int pblk_core_init(struct pblk *pblk) |
372 | { | |
373 | struct nvm_tgt_dev *dev = pblk->dev; | |
374 | struct nvm_geo *geo = &dev->geo; | |
b906bbb6 | 375 | int ret, max_write_ppas; |
43d47127 JG |
376 | |
377 | atomic64_set(&pblk->user_wa, 0); | |
378 | atomic64_set(&pblk->pad_wa, 0); | |
379 | atomic64_set(&pblk->gc_wa, 0); | |
380 | pblk->user_rst_wa = 0; | |
381 | pblk->pad_rst_wa = 0; | |
382 | pblk->gc_rst_wa = 0; | |
383 | ||
384 | atomic64_set(&pblk->nr_flush, 0); | |
385 | pblk->nr_flush_rst = 0; | |
a4bd217b | 386 | |
8bbd45d0 | 387 | pblk->min_write_pgs = geo->ws_opt; |
55d8ec35 | 388 | pblk->min_write_pgs_data = pblk->min_write_pgs; |
43d47127 JG |
389 | max_write_ppas = pblk->min_write_pgs * geo->all_luns; |
390 | pblk->max_write_pgs = min_t(int, max_write_ppas, NVM_MAX_VLBA); | |
8a57fc38 ZW |
391 | pblk->max_write_pgs = min_t(int, pblk->max_write_pgs, |
392 | queue_max_hw_sectors(dev->q) / (geo->csecs >> SECTOR_SHIFT)); | |
43d47127 JG |
393 | pblk_set_sec_per_write(pblk, pblk->min_write_pgs); |
394 | ||
faa79f27 | 395 | pblk->oob_meta_size = geo->sos; |
55d8ec35 IK |
396 | if (!pblk_is_oob_meta_supported(pblk)) { |
397 | /* For drives which does not have OOB metadata feature | |
398 | * in order to support recovery feature we need to use | |
399 | * so called packed metadata. Packed metada will store | |
400 | * the same information as OOB metadata (l2p table mapping, | |
401 | * but in the form of the single page at the end of | |
402 | * every write request. | |
403 | */ | |
404 | if (pblk->min_write_pgs | |
405 | * sizeof(struct pblk_sec_meta) > PAGE_SIZE) { | |
406 | /* We want to keep all the packed metadata on single | |
407 | * page per write requests. So we need to ensure that | |
408 | * it will fit. | |
409 | * | |
410 | * This is more like sanity check, since there is | |
411 | * no device with such a big minimal write size | |
412 | * (above 1 metabytes). | |
413 | */ | |
414 | pblk_err(pblk, "Not supported min write size\n"); | |
415 | return -EINVAL; | |
416 | } | |
417 | /* For packed meta approach we do some simplification. | |
418 | * On read path we always issue requests which size | |
419 | * equal to max_write_pgs, with all pages filled with | |
420 | * user payload except of last one page which will be | |
421 | * filled with packed metadata. | |
422 | */ | |
423 | pblk->max_write_pgs = pblk->min_write_pgs; | |
424 | pblk->min_write_pgs_data = pblk->min_write_pgs - 1; | |
faa79f27 IK |
425 | } |
426 | ||
6396bb22 | 427 | pblk->pad_dist = kcalloc(pblk->min_write_pgs - 1, sizeof(atomic64_t), |
43d47127 JG |
428 | GFP_KERNEL); |
429 | if (!pblk->pad_dist) | |
a4bd217b JG |
430 | return -ENOMEM; |
431 | ||
1864de94 | 432 | if (pblk_get_global_caches()) |
43d47127 JG |
433 | goto fail_free_pad_dist; |
434 | ||
b84ae4a8 | 435 | /* Internal bios can be at most the sectors signaled by the device. */ |
b906bbb6 KO |
436 | ret = mempool_init_page_pool(&pblk->page_bio_pool, NVM_MAX_VLBA, 0); |
437 | if (ret) | |
22a4e061 | 438 | goto free_global_caches; |
a4bd217b | 439 | |
b906bbb6 | 440 | ret = mempool_init_slab_pool(&pblk->gen_ws_pool, PBLK_GEN_WS_POOL_SIZE, |
1864de94 | 441 | pblk_caches.ws); |
b906bbb6 | 442 | if (ret) |
bd432417 | 443 | goto free_page_bio_pool; |
a4bd217b | 444 | |
b906bbb6 | 445 | ret = mempool_init_slab_pool(&pblk->rec_pool, geo->all_luns, |
1864de94 | 446 | pblk_caches.rec); |
b906bbb6 | 447 | if (ret) |
b84ae4a8 | 448 | goto free_gen_ws_pool; |
a4bd217b | 449 | |
b906bbb6 | 450 | ret = mempool_init_slab_pool(&pblk->r_rq_pool, geo->all_luns, |
1864de94 | 451 | pblk_caches.g_rq); |
b906bbb6 | 452 | if (ret) |
a4bd217b JG |
453 | goto free_rec_pool; |
454 | ||
b906bbb6 | 455 | ret = mempool_init_slab_pool(&pblk->e_rq_pool, geo->all_luns, |
1864de94 | 456 | pblk_caches.g_rq); |
b906bbb6 | 457 | if (ret) |
0d880398 JG |
458 | goto free_r_rq_pool; |
459 | ||
b906bbb6 | 460 | ret = mempool_init_slab_pool(&pblk->w_rq_pool, geo->all_luns, |
1864de94 | 461 | pblk_caches.w_rq); |
b906bbb6 | 462 | if (ret) |
0d880398 | 463 | goto free_e_rq_pool; |
a4bd217b | 464 | |
ef576494 JG |
465 | pblk->close_wq = alloc_workqueue("pblk-close-wq", |
466 | WQ_MEM_RECLAIM | WQ_UNBOUND, PBLK_NR_CLOSE_JOBS); | |
467 | if (!pblk->close_wq) | |
e72ec1d3 | 468 | goto free_w_rq_pool; |
a4bd217b | 469 | |
ef576494 JG |
470 | pblk->bb_wq = alloc_workqueue("pblk-bb-wq", |
471 | WQ_MEM_RECLAIM | WQ_UNBOUND, 0); | |
472 | if (!pblk->bb_wq) | |
473 | goto free_close_wq; | |
474 | ||
7bd4d370 JG |
475 | pblk->r_end_wq = alloc_workqueue("pblk-read-end-wq", |
476 | WQ_MEM_RECLAIM | WQ_UNBOUND, 0); | |
477 | if (!pblk->r_end_wq) | |
ef576494 | 478 | goto free_bb_wq; |
a4bd217b | 479 | |
bb845ae4 | 480 | if (pblk_set_addrf(pblk)) |
7bd4d370 JG |
481 | goto free_r_end_wq; |
482 | ||
a4bd217b | 483 | INIT_LIST_HEAD(&pblk->compl_list); |
6a3abf5b | 484 | INIT_LIST_HEAD(&pblk->resubmit_list); |
43d47127 | 485 | |
a4bd217b JG |
486 | return 0; |
487 | ||
7bd4d370 JG |
488 | free_r_end_wq: |
489 | destroy_workqueue(pblk->r_end_wq); | |
ef576494 JG |
490 | free_bb_wq: |
491 | destroy_workqueue(pblk->bb_wq); | |
492 | free_close_wq: | |
493 | destroy_workqueue(pblk->close_wq); | |
a4bd217b | 494 | free_w_rq_pool: |
b906bbb6 | 495 | mempool_exit(&pblk->w_rq_pool); |
0d880398 | 496 | free_e_rq_pool: |
b906bbb6 | 497 | mempool_exit(&pblk->e_rq_pool); |
0d880398 | 498 | free_r_rq_pool: |
b906bbb6 | 499 | mempool_exit(&pblk->r_rq_pool); |
a4bd217b | 500 | free_rec_pool: |
b906bbb6 | 501 | mempool_exit(&pblk->rec_pool); |
b84ae4a8 | 502 | free_gen_ws_pool: |
b906bbb6 | 503 | mempool_exit(&pblk->gen_ws_pool); |
bd432417 | 504 | free_page_bio_pool: |
b906bbb6 | 505 | mempool_exit(&pblk->page_bio_pool); |
22a4e061 | 506 | free_global_caches: |
1864de94 | 507 | pblk_put_global_caches(); |
43d47127 JG |
508 | fail_free_pad_dist: |
509 | kfree(pblk->pad_dist); | |
a4bd217b JG |
510 | return -ENOMEM; |
511 | } | |
512 | ||
513 | static void pblk_core_free(struct pblk *pblk) | |
514 | { | |
ef576494 JG |
515 | if (pblk->close_wq) |
516 | destroy_workqueue(pblk->close_wq); | |
517 | ||
7bd4d370 JG |
518 | if (pblk->r_end_wq) |
519 | destroy_workqueue(pblk->r_end_wq); | |
520 | ||
ef576494 JG |
521 | if (pblk->bb_wq) |
522 | destroy_workqueue(pblk->bb_wq); | |
a4bd217b | 523 | |
b906bbb6 KO |
524 | mempool_exit(&pblk->page_bio_pool); |
525 | mempool_exit(&pblk->gen_ws_pool); | |
526 | mempool_exit(&pblk->rec_pool); | |
527 | mempool_exit(&pblk->r_rq_pool); | |
528 | mempool_exit(&pblk->e_rq_pool); | |
529 | mempool_exit(&pblk->w_rq_pool); | |
a4bd217b | 530 | |
1864de94 | 531 | pblk_put_global_caches(); |
43d47127 | 532 | kfree(pblk->pad_dist); |
a4bd217b JG |
533 | } |
534 | ||
e411b331 JG |
535 | static void pblk_line_mg_free(struct pblk *pblk) |
536 | { | |
537 | struct pblk_line_mgmt *l_mg = &pblk->l_mg; | |
538 | int i; | |
539 | ||
540 | kfree(l_mg->bb_template); | |
541 | kfree(l_mg->bb_aux); | |
542 | kfree(l_mg->vsc_list); | |
543 | ||
544 | for (i = 0; i < PBLK_DATA_LINES; i++) { | |
545 | kfree(l_mg->sline_meta[i]); | |
546 | pblk_mfree(l_mg->eline_meta[i]->buf, l_mg->emeta_alloc_type); | |
547 | kfree(l_mg->eline_meta[i]); | |
548 | } | |
53d82db6 HH |
549 | |
550 | mempool_destroy(l_mg->bitmap_pool); | |
551 | kmem_cache_destroy(l_mg->bitmap_cache); | |
e411b331 JG |
552 | } |
553 | ||
48b8d208 HH |
554 | static void pblk_line_meta_free(struct pblk_line_mgmt *l_mg, |
555 | struct pblk_line *line) | |
dffdd960 | 556 | { |
48b8d208 HH |
557 | struct pblk_w_err_gc *w_err_gc = line->w_err_gc; |
558 | ||
dffdd960 JG |
559 | kfree(line->blk_bitmap); |
560 | kfree(line->erase_bitmap); | |
32ef9412 | 561 | kfree(line->chks); |
48b8d208 HH |
562 | |
563 | pblk_mfree(w_err_gc->lba_list, l_mg->emeta_alloc_type); | |
564 | kfree(w_err_gc); | |
dffdd960 JG |
565 | } |
566 | ||
a4bd217b JG |
567 | static void pblk_lines_free(struct pblk *pblk) |
568 | { | |
569 | struct pblk_line_mgmt *l_mg = &pblk->l_mg; | |
570 | struct pblk_line *line; | |
571 | int i; | |
572 | ||
a4bd217b JG |
573 | for (i = 0; i < l_mg->nr_lines; i++) { |
574 | line = &pblk->lines[i]; | |
575 | ||
8e55c07b | 576 | pblk_line_free(line); |
48b8d208 | 577 | pblk_line_meta_free(l_mg, line); |
a4bd217b | 578 | } |
43d47127 JG |
579 | |
580 | pblk_line_mg_free(pblk); | |
581 | ||
582 | kfree(pblk->luns); | |
583 | kfree(pblk->lines); | |
a4bd217b JG |
584 | } |
585 | ||
43d47127 | 586 | static int pblk_luns_init(struct pblk *pblk) |
a4bd217b JG |
587 | { |
588 | struct nvm_tgt_dev *dev = pblk->dev; | |
589 | struct nvm_geo *geo = &dev->geo; | |
590 | struct pblk_lun *rlun; | |
e411b331 | 591 | int i; |
a4bd217b JG |
592 | |
593 | /* TODO: Implement unbalanced LUN support */ | |
a40afad9 | 594 | if (geo->num_lun < 0) { |
4e495a46 | 595 | pblk_err(pblk, "unbalanced LUN config.\n"); |
a4bd217b JG |
596 | return -EINVAL; |
597 | } | |
598 | ||
fae7fae4 MB |
599 | pblk->luns = kcalloc(geo->all_luns, sizeof(struct pblk_lun), |
600 | GFP_KERNEL); | |
a4bd217b JG |
601 | if (!pblk->luns) |
602 | return -ENOMEM; | |
603 | ||
fae7fae4 | 604 | for (i = 0; i < geo->all_luns; i++) { |
a4bd217b | 605 | /* Stripe across channels */ |
a40afad9 JG |
606 | int ch = i % geo->num_ch; |
607 | int lun_raw = i / geo->num_ch; | |
608 | int lunid = lun_raw + ch * geo->num_lun; | |
a4bd217b JG |
609 | |
610 | rlun = &pblk->luns[i]; | |
43d47127 | 611 | rlun->bppa = dev->luns[lunid]; |
a4bd217b JG |
612 | |
613 | sema_init(&rlun->wr_sem, 1); | |
a4bd217b JG |
614 | } |
615 | ||
616 | return 0; | |
617 | } | |
618 | ||
a4bd217b | 619 | /* See comment over struct line_emeta definition */ |
dd2a4343 | 620 | static unsigned int calc_emeta_len(struct pblk *pblk) |
a4bd217b | 621 | { |
dd2a4343 JG |
622 | struct pblk_line_meta *lm = &pblk->lm; |
623 | struct pblk_line_mgmt *l_mg = &pblk->l_mg; | |
624 | struct nvm_tgt_dev *dev = pblk->dev; | |
625 | struct nvm_geo *geo = &dev->geo; | |
626 | ||
627 | /* Round to sector size so that lba_list starts on its own sector */ | |
628 | lm->emeta_sec[1] = DIV_ROUND_UP( | |
76758390 | 629 | sizeof(struct line_emeta) + lm->blk_bitmap_len + |
e46f4e48 JG |
630 | sizeof(struct wa_counters), geo->csecs); |
631 | lm->emeta_len[1] = lm->emeta_sec[1] * geo->csecs; | |
dd2a4343 JG |
632 | |
633 | /* Round to sector size so that vsc_list starts on its own sector */ | |
634 | lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0]; | |
635 | lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64), | |
e46f4e48 JG |
636 | geo->csecs); |
637 | lm->emeta_len[2] = lm->emeta_sec[2] * geo->csecs; | |
dd2a4343 JG |
638 | |
639 | lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32), | |
e46f4e48 JG |
640 | geo->csecs); |
641 | lm->emeta_len[3] = lm->emeta_sec[3] * geo->csecs; | |
dd2a4343 JG |
642 | |
643 | lm->vsc_list_len = l_mg->nr_lines * sizeof(u32); | |
644 | ||
645 | return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]); | |
a4bd217b JG |
646 | } |
647 | ||
3bcebc5b | 648 | static int pblk_set_provision(struct pblk *pblk, int nr_free_chks) |
a4bd217b JG |
649 | { |
650 | struct nvm_tgt_dev *dev = pblk->dev; | |
a7689938 JG |
651 | struct pblk_line_mgmt *l_mg = &pblk->l_mg; |
652 | struct pblk_line_meta *lm = &pblk->lm; | |
a4bd217b JG |
653 | struct nvm_geo *geo = &dev->geo; |
654 | sector_t provisioned; | |
55d8ec35 | 655 | int sec_meta, blk_meta, clba; |
3bcebc5b | 656 | int minimum; |
a4bd217b | 657 | |
e5392739 JG |
658 | if (geo->op == NVM_TARGET_DEFAULT_OP) |
659 | pblk->op = PBLK_DEFAULT_OP; | |
660 | else | |
661 | pblk->op = geo->op; | |
a4bd217b | 662 | |
3bcebc5b HH |
663 | minimum = pblk_get_min_chks(pblk); |
664 | provisioned = nr_free_chks; | |
a7689938 | 665 | provisioned *= (100 - pblk->op); |
a4bd217b JG |
666 | sector_div(provisioned, 100); |
667 | ||
3bcebc5b HH |
668 | if ((nr_free_chks - provisioned) < minimum) { |
669 | if (geo->op != NVM_TARGET_DEFAULT_OP) { | |
670 | pblk_err(pblk, "OP too small to create a sane instance\n"); | |
671 | return -EINTR; | |
672 | } | |
673 | ||
674 | /* If the user did not specify an OP value, and PBLK_DEFAULT_OP | |
675 | * is not enough, calculate and set sane value | |
676 | */ | |
677 | ||
678 | provisioned = nr_free_chks - minimum; | |
679 | pblk->op = (100 * minimum) / nr_free_chks; | |
680 | pblk_info(pblk, "Default OP insufficient, adjusting OP to %d\n", | |
681 | pblk->op); | |
682 | } | |
683 | ||
684 | pblk->op_blks = nr_free_chks - provisioned; | |
a7689938 | 685 | |
a4bd217b JG |
686 | /* Internally pblk manages all free blocks, but all calculations based |
687 | * on user capacity consider only provisioned blocks | |
688 | */ | |
3bcebc5b | 689 | pblk->rl.total_blocks = nr_free_chks; |
a7689938 JG |
690 | |
691 | /* Consider sectors used for metadata */ | |
692 | sec_meta = (lm->smeta_sec + lm->emeta_sec[0]) * l_mg->nr_free_lines; | |
e46f4e48 | 693 | blk_meta = DIV_ROUND_UP(sec_meta, geo->clba); |
a7689938 | 694 | |
55d8ec35 IK |
695 | clba = (geo->clba / pblk->min_write_pgs) * pblk->min_write_pgs_data; |
696 | pblk->capacity = (provisioned - blk_meta) * clba; | |
a7689938 | 697 | |
3bcebc5b HH |
698 | atomic_set(&pblk->rl.free_blocks, nr_free_chks); |
699 | atomic_set(&pblk->rl.free_user_blocks, nr_free_chks); | |
700 | ||
701 | return 0; | |
a4bd217b JG |
702 | } |
703 | ||
aff3fb18 | 704 | static int pblk_setup_line_meta_chk(struct pblk *pblk, struct pblk_line *line, |
32ef9412 JG |
705 | struct nvm_chk_meta *meta) |
706 | { | |
707 | struct nvm_tgt_dev *dev = pblk->dev; | |
708 | struct nvm_geo *geo = &dev->geo; | |
709 | struct pblk_line_meta *lm = &pblk->lm; | |
710 | int i, nr_bad_chks = 0; | |
711 | ||
712 | for (i = 0; i < lm->blk_per_line; i++) { | |
713 | struct pblk_lun *rlun = &pblk->luns[i]; | |
714 | struct nvm_chk_meta *chunk; | |
715 | struct nvm_chk_meta *chunk_meta; | |
716 | struct ppa_addr ppa; | |
717 | int pos; | |
718 | ||
719 | ppa = rlun->bppa; | |
720 | pos = pblk_ppa_to_pos(geo, ppa); | |
721 | chunk = &line->chks[pos]; | |
722 | ||
723 | ppa.m.chk = line->id; | |
724 | chunk_meta = pblk_chunk_get_off(pblk, meta, ppa); | |
725 | ||
726 | chunk->state = chunk_meta->state; | |
727 | chunk->type = chunk_meta->type; | |
728 | chunk->wi = chunk_meta->wi; | |
729 | chunk->slba = chunk_meta->slba; | |
730 | chunk->cnlb = chunk_meta->cnlb; | |
731 | chunk->wp = chunk_meta->wp; | |
732 | ||
4c44abf4 HH |
733 | trace_pblk_chunk_state(pblk_disk_name(pblk), &ppa, |
734 | chunk->state); | |
735 | ||
32ef9412 JG |
736 | if (chunk->type & NVM_CHK_TP_SZ_SPEC) { |
737 | WARN_ONCE(1, "pblk: custom-sized chunks unsupported\n"); | |
738 | continue; | |
739 | } | |
740 | ||
6f9c9607 JG |
741 | if (!(chunk->state & NVM_CHK_ST_OFFLINE)) |
742 | continue; | |
743 | ||
32ef9412 JG |
744 | set_bit(pos, line->blk_bitmap); |
745 | nr_bad_chks++; | |
746 | } | |
747 | ||
748 | return nr_bad_chks; | |
749 | } | |
750 | ||
751 | static long pblk_setup_line_meta(struct pblk *pblk, struct pblk_line *line, | |
752 | void *chunk_meta, int line_id) | |
753 | { | |
32ef9412 JG |
754 | struct pblk_line_mgmt *l_mg = &pblk->l_mg; |
755 | struct pblk_line_meta *lm = &pblk->lm; | |
756 | long nr_bad_chks, chk_in_line; | |
757 | ||
758 | line->pblk = pblk; | |
759 | line->id = line_id; | |
760 | line->type = PBLK_LINETYPE_FREE; | |
761 | line->state = PBLK_LINESTATE_NEW; | |
762 | line->gc_group = PBLK_LINEGC_NONE; | |
763 | line->vsc = &l_mg->vsc_list[line_id]; | |
764 | spin_lock_init(&line->lock); | |
765 | ||
aff3fb18 | 766 | nr_bad_chks = pblk_setup_line_meta_chk(pblk, line, chunk_meta); |
32ef9412 JG |
767 | |
768 | chk_in_line = lm->blk_per_line - nr_bad_chks; | |
769 | if (nr_bad_chks < 0 || nr_bad_chks > lm->blk_per_line || | |
770 | chk_in_line < lm->min_blk_line) { | |
771 | line->state = PBLK_LINESTATE_BAD; | |
772 | list_add_tail(&line->list, &l_mg->bad_list); | |
773 | return 0; | |
774 | } | |
775 | ||
776 | atomic_set(&line->blk_in_line, chk_in_line); | |
777 | list_add_tail(&line->list, &l_mg->free_list); | |
778 | l_mg->nr_free_lines++; | |
779 | ||
780 | return chk_in_line; | |
781 | } | |
782 | ||
783 | static int pblk_alloc_line_meta(struct pblk *pblk, struct pblk_line *line) | |
43d47127 JG |
784 | { |
785 | struct pblk_line_meta *lm = &pblk->lm; | |
786 | ||
787 | line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL); | |
788 | if (!line->blk_bitmap) | |
789 | return -ENOMEM; | |
790 | ||
791 | line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL); | |
48b8d208 HH |
792 | if (!line->erase_bitmap) |
793 | goto free_blk_bitmap; | |
794 | ||
43d47127 | 795 | |
6da2ec56 KC |
796 | line->chks = kmalloc_array(lm->blk_per_line, |
797 | sizeof(struct nvm_chk_meta), GFP_KERNEL); | |
48b8d208 HH |
798 | if (!line->chks) |
799 | goto free_erase_bitmap; | |
800 | ||
801 | line->w_err_gc = kzalloc(sizeof(struct pblk_w_err_gc), GFP_KERNEL); | |
802 | if (!line->w_err_gc) | |
803 | goto free_chks; | |
43d47127 JG |
804 | |
805 | return 0; | |
48b8d208 HH |
806 | |
807 | free_chks: | |
808 | kfree(line->chks); | |
809 | free_erase_bitmap: | |
810 | kfree(line->erase_bitmap); | |
811 | free_blk_bitmap: | |
812 | kfree(line->blk_bitmap); | |
813 | return -ENOMEM; | |
43d47127 JG |
814 | } |
815 | ||
816 | static int pblk_line_mg_init(struct pblk *pblk) | |
dd2a4343 | 817 | { |
43d47127 JG |
818 | struct nvm_tgt_dev *dev = pblk->dev; |
819 | struct nvm_geo *geo = &dev->geo; | |
dd2a4343 JG |
820 | struct pblk_line_mgmt *l_mg = &pblk->l_mg; |
821 | struct pblk_line_meta *lm = &pblk->lm; | |
43d47127 JG |
822 | int i, bb_distance; |
823 | ||
a40afad9 | 824 | l_mg->nr_lines = geo->num_chk; |
43d47127 JG |
825 | l_mg->log_line = l_mg->data_line = NULL; |
826 | l_mg->l_seq_nr = l_mg->d_seq_nr = 0; | |
827 | l_mg->nr_free_lines = 0; | |
828 | bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES); | |
829 | ||
830 | INIT_LIST_HEAD(&l_mg->free_list); | |
831 | INIT_LIST_HEAD(&l_mg->corrupt_list); | |
832 | INIT_LIST_HEAD(&l_mg->bad_list); | |
833 | INIT_LIST_HEAD(&l_mg->gc_full_list); | |
834 | INIT_LIST_HEAD(&l_mg->gc_high_list); | |
835 | INIT_LIST_HEAD(&l_mg->gc_mid_list); | |
836 | INIT_LIST_HEAD(&l_mg->gc_low_list); | |
837 | INIT_LIST_HEAD(&l_mg->gc_empty_list); | |
48b8d208 | 838 | INIT_LIST_HEAD(&l_mg->gc_werr_list); |
43d47127 JG |
839 | |
840 | INIT_LIST_HEAD(&l_mg->emeta_list); | |
841 | ||
48b8d208 HH |
842 | l_mg->gc_lists[0] = &l_mg->gc_werr_list; |
843 | l_mg->gc_lists[1] = &l_mg->gc_high_list; | |
844 | l_mg->gc_lists[2] = &l_mg->gc_mid_list; | |
845 | l_mg->gc_lists[3] = &l_mg->gc_low_list; | |
43d47127 JG |
846 | |
847 | spin_lock_init(&l_mg->free_lock); | |
848 | spin_lock_init(&l_mg->close_lock); | |
849 | spin_lock_init(&l_mg->gc_lock); | |
850 | ||
851 | l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL); | |
852 | if (!l_mg->vsc_list) | |
853 | goto fail; | |
854 | ||
855 | l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL); | |
856 | if (!l_mg->bb_template) | |
857 | goto fail_free_vsc_list; | |
858 | ||
859 | l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL); | |
860 | if (!l_mg->bb_aux) | |
861 | goto fail_free_bb_template; | |
dd2a4343 JG |
862 | |
863 | /* smeta is always small enough to fit on a kmalloc memory allocation, | |
864 | * emeta depends on the number of LUNs allocated to the pblk instance | |
865 | */ | |
dd2a4343 JG |
866 | for (i = 0; i < PBLK_DATA_LINES; i++) { |
867 | l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL); | |
868 | if (!l_mg->sline_meta[i]) | |
869 | goto fail_free_smeta; | |
870 | } | |
871 | ||
53d82db6 HH |
872 | l_mg->bitmap_cache = kmem_cache_create("pblk_lm_bitmap", |
873 | lm->sec_bitmap_len, 0, 0, NULL); | |
874 | if (!l_mg->bitmap_cache) | |
875 | goto fail_free_smeta; | |
876 | ||
877 | /* the bitmap pool is used for both valid and map bitmaps */ | |
878 | l_mg->bitmap_pool = mempool_create_slab_pool(PBLK_DATA_LINES * 2, | |
879 | l_mg->bitmap_cache); | |
880 | if (!l_mg->bitmap_pool) | |
881 | goto fail_destroy_bitmap_cache; | |
882 | ||
dd2a4343 JG |
883 | /* emeta allocates three different buffers for managing metadata with |
884 | * in-memory and in-media layouts | |
885 | */ | |
886 | for (i = 0; i < PBLK_DATA_LINES; i++) { | |
887 | struct pblk_emeta *emeta; | |
888 | ||
889 | emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL); | |
890 | if (!emeta) | |
891 | goto fail_free_emeta; | |
892 | ||
893 | if (lm->emeta_len[0] > KMALLOC_MAX_CACHE_SIZE) { | |
894 | l_mg->emeta_alloc_type = PBLK_VMALLOC_META; | |
895 | ||
896 | emeta->buf = vmalloc(lm->emeta_len[0]); | |
897 | if (!emeta->buf) { | |
898 | kfree(emeta); | |
899 | goto fail_free_emeta; | |
900 | } | |
901 | ||
902 | emeta->nr_entries = lm->emeta_sec[0]; | |
903 | l_mg->eline_meta[i] = emeta; | |
904 | } else { | |
905 | l_mg->emeta_alloc_type = PBLK_KMALLOC_META; | |
906 | ||
907 | emeta->buf = kmalloc(lm->emeta_len[0], GFP_KERNEL); | |
908 | if (!emeta->buf) { | |
909 | kfree(emeta); | |
910 | goto fail_free_emeta; | |
911 | } | |
912 | ||
913 | emeta->nr_entries = lm->emeta_sec[0]; | |
914 | l_mg->eline_meta[i] = emeta; | |
915 | } | |
916 | } | |
917 | ||
dd2a4343 JG |
918 | for (i = 0; i < l_mg->nr_lines; i++) |
919 | l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY); | |
920 | ||
43d47127 JG |
921 | bb_distance = (geo->all_luns) * geo->ws_opt; |
922 | for (i = 0; i < lm->sec_per_line; i += bb_distance) | |
923 | bitmap_set(l_mg->bb_template, i, geo->ws_opt); | |
924 | ||
dd2a4343 JG |
925 | return 0; |
926 | ||
927 | fail_free_emeta: | |
928 | while (--i >= 0) { | |
c9d84b35 RP |
929 | if (l_mg->emeta_alloc_type == PBLK_VMALLOC_META) |
930 | vfree(l_mg->eline_meta[i]->buf); | |
931 | else | |
932 | kfree(l_mg->eline_meta[i]->buf); | |
f680f19a | 933 | kfree(l_mg->eline_meta[i]); |
dd2a4343 | 934 | } |
53d82db6 HH |
935 | |
936 | mempool_destroy(l_mg->bitmap_pool); | |
937 | fail_destroy_bitmap_cache: | |
938 | kmem_cache_destroy(l_mg->bitmap_cache); | |
dd2a4343 JG |
939 | fail_free_smeta: |
940 | for (i = 0; i < PBLK_DATA_LINES; i++) | |
f680f19a | 941 | kfree(l_mg->sline_meta[i]); |
43d47127 JG |
942 | kfree(l_mg->bb_aux); |
943 | fail_free_bb_template: | |
944 | kfree(l_mg->bb_template); | |
945 | fail_free_vsc_list: | |
946 | kfree(l_mg->vsc_list); | |
947 | fail: | |
dd2a4343 JG |
948 | return -ENOMEM; |
949 | } | |
950 | ||
43d47127 | 951 | static int pblk_line_meta_init(struct pblk *pblk) |
a4bd217b JG |
952 | { |
953 | struct nvm_tgt_dev *dev = pblk->dev; | |
954 | struct nvm_geo *geo = &dev->geo; | |
a4bd217b | 955 | struct pblk_line_meta *lm = &pblk->lm; |
a4bd217b | 956 | unsigned int smeta_len, emeta_len; |
43d47127 | 957 | int i; |
a4bd217b | 958 | |
e46f4e48 | 959 | lm->sec_per_line = geo->clba * geo->all_luns; |
fae7fae4 MB |
960 | lm->blk_per_line = geo->all_luns; |
961 | lm->blk_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long); | |
a4bd217b | 962 | lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long); |
fae7fae4 | 963 | lm->lun_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long); |
27b97872 RP |
964 | lm->mid_thrs = lm->sec_per_line / 2; |
965 | lm->high_thrs = lm->sec_per_line / 4; | |
fae7fae4 | 966 | lm->meta_distance = (geo->all_luns / 2) * pblk->min_write_pgs; |
a4bd217b JG |
967 | |
968 | /* Calculate necessary pages for smeta. See comment over struct | |
969 | * line_smeta definition | |
970 | */ | |
a4bd217b JG |
971 | i = 1; |
972 | add_smeta_page: | |
e46f4e48 JG |
973 | lm->smeta_sec = i * geo->ws_opt; |
974 | lm->smeta_len = lm->smeta_sec * geo->csecs; | |
a4bd217b | 975 | |
dd2a4343 | 976 | smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len; |
a4bd217b JG |
977 | if (smeta_len > lm->smeta_len) { |
978 | i++; | |
979 | goto add_smeta_page; | |
980 | } | |
981 | ||
982 | /* Calculate necessary pages for emeta. See comment over struct | |
983 | * line_emeta definition | |
984 | */ | |
985 | i = 1; | |
986 | add_emeta_page: | |
e46f4e48 JG |
987 | lm->emeta_sec[0] = i * geo->ws_opt; |
988 | lm->emeta_len[0] = lm->emeta_sec[0] * geo->csecs; | |
a4bd217b | 989 | |
dd2a4343 JG |
990 | emeta_len = calc_emeta_len(pblk); |
991 | if (emeta_len > lm->emeta_len[0]) { | |
a4bd217b JG |
992 | i++; |
993 | goto add_emeta_page; | |
994 | } | |
a4bd217b | 995 | |
fae7fae4 | 996 | lm->emeta_bb = geo->all_luns > i ? geo->all_luns - i : 0; |
21d22871 JG |
997 | |
998 | lm->min_blk_line = 1; | |
fae7fae4 | 999 | if (geo->all_luns > 1) |
21d22871 | 1000 | lm->min_blk_line += DIV_ROUND_UP(lm->smeta_sec + |
e46f4e48 | 1001 | lm->emeta_sec[0], geo->clba); |
21d22871 | 1002 | |
b5e063a2 | 1003 | if (lm->min_blk_line > lm->blk_per_line) { |
4e495a46 | 1004 | pblk_err(pblk, "config. not supported. Min. LUN in line:%d\n", |
b5e063a2 | 1005 | lm->blk_per_line); |
e411b331 | 1006 | return -EINVAL; |
b5e063a2 | 1007 | } |
a4bd217b | 1008 | |
43d47127 JG |
1009 | return 0; |
1010 | } | |
1011 | ||
1012 | static int pblk_lines_init(struct pblk *pblk) | |
1013 | { | |
1014 | struct pblk_line_mgmt *l_mg = &pblk->l_mg; | |
43d47127 | 1015 | struct pblk_line *line; |
32ef9412 | 1016 | void *chunk_meta; |
3bcebc5b | 1017 | int nr_free_chks = 0; |
43d47127 JG |
1018 | int i, ret; |
1019 | ||
1020 | ret = pblk_line_meta_init(pblk); | |
dd2a4343 | 1021 | if (ret) |
e411b331 | 1022 | return ret; |
a4bd217b | 1023 | |
43d47127 JG |
1024 | ret = pblk_line_mg_init(pblk); |
1025 | if (ret) | |
1026 | return ret; | |
1027 | ||
1028 | ret = pblk_luns_init(pblk); | |
1029 | if (ret) | |
a4bd217b JG |
1030 | goto fail_free_meta; |
1031 | ||
aff3fb18 | 1032 | chunk_meta = pblk_get_chunk_meta(pblk); |
32ef9412 JG |
1033 | if (IS_ERR(chunk_meta)) { |
1034 | ret = PTR_ERR(chunk_meta); | |
43d47127 | 1035 | goto fail_free_luns; |
1c6286f2 | 1036 | } |
a4bd217b | 1037 | |
a4bd217b JG |
1038 | pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line), |
1039 | GFP_KERNEL); | |
1c6286f2 DC |
1040 | if (!pblk->lines) { |
1041 | ret = -ENOMEM; | |
32ef9412 | 1042 | goto fail_free_chunk_meta; |
e411b331 JG |
1043 | } |
1044 | ||
a4bd217b JG |
1045 | for (i = 0; i < l_mg->nr_lines; i++) { |
1046 | line = &pblk->lines[i]; | |
1047 | ||
32ef9412 | 1048 | ret = pblk_alloc_line_meta(pblk, line); |
dffdd960 | 1049 | if (ret) |
43d47127 | 1050 | goto fail_free_lines; |
dffdd960 | 1051 | |
32ef9412 | 1052 | nr_free_chks += pblk_setup_line_meta(pblk, line, chunk_meta, i); |
f2937232 HH |
1053 | |
1054 | trace_pblk_line_state(pblk_disk_name(pblk), line->id, | |
1055 | line->state); | |
a4bd217b JG |
1056 | } |
1057 | ||
2deeefc0 | 1058 | if (!nr_free_chks) { |
4e495a46 | 1059 | pblk_err(pblk, "too many bad blocks prevent for sane instance\n"); |
a70985f8 WY |
1060 | ret = -EINTR; |
1061 | goto fail_free_lines; | |
2deeefc0 JG |
1062 | } |
1063 | ||
3bcebc5b HH |
1064 | ret = pblk_set_provision(pblk, nr_free_chks); |
1065 | if (ret) | |
1066 | goto fail_free_lines; | |
a4bd217b | 1067 | |
090ee26f | 1068 | vfree(chunk_meta); |
a4bd217b | 1069 | return 0; |
e411b331 | 1070 | |
43d47127 | 1071 | fail_free_lines: |
dffdd960 | 1072 | while (--i >= 0) |
48b8d208 | 1073 | pblk_line_meta_free(l_mg, &pblk->lines[i]); |
43d47127 | 1074 | kfree(pblk->lines); |
32ef9412 | 1075 | fail_free_chunk_meta: |
0934ce87 | 1076 | vfree(chunk_meta); |
43d47127 JG |
1077 | fail_free_luns: |
1078 | kfree(pblk->luns); | |
a4bd217b | 1079 | fail_free_meta: |
e411b331 | 1080 | pblk_line_mg_free(pblk); |
a4bd217b JG |
1081 | |
1082 | return ret; | |
1083 | } | |
1084 | ||
1085 | static int pblk_writer_init(struct pblk *pblk) | |
1086 | { | |
a4bd217b JG |
1087 | pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t"); |
1088 | if (IS_ERR(pblk->writer_ts)) { | |
cc4f5ba1 JG |
1089 | int err = PTR_ERR(pblk->writer_ts); |
1090 | ||
1091 | if (err != -EINTR) | |
4e495a46 | 1092 | pblk_err(pblk, "could not allocate writer kthread (%d)\n", |
cc4f5ba1 JG |
1093 | err); |
1094 | return err; | |
a4bd217b JG |
1095 | } |
1096 | ||
cc4f5ba1 JG |
1097 | timer_setup(&pblk->wtimer, pblk_write_timer_fn, 0); |
1098 | mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100)); | |
1099 | ||
a4bd217b JG |
1100 | return 0; |
1101 | } | |
1102 | ||
1103 | static void pblk_writer_stop(struct pblk *pblk) | |
1104 | { | |
ee8d5c1a JG |
1105 | /* The pipeline must be stopped and the write buffer emptied before the |
1106 | * write thread is stopped | |
1107 | */ | |
1108 | WARN(pblk_rb_read_count(&pblk->rwb), | |
1109 | "Stopping not fully persisted write buffer\n"); | |
1110 | ||
1111 | WARN(pblk_rb_sync_count(&pblk->rwb), | |
1112 | "Stopping not fully synced write buffer\n"); | |
1113 | ||
7be970b2 | 1114 | del_timer_sync(&pblk->wtimer); |
a4bd217b JG |
1115 | if (pblk->writer_ts) |
1116 | kthread_stop(pblk->writer_ts); | |
a4bd217b JG |
1117 | } |
1118 | ||
1119 | static void pblk_free(struct pblk *pblk) | |
1120 | { | |
a4bd217b | 1121 | pblk_lines_free(pblk); |
a4bd217b | 1122 | pblk_l2p_free(pblk); |
43d47127 JG |
1123 | pblk_rwb_free(pblk); |
1124 | pblk_core_free(pblk); | |
a4bd217b JG |
1125 | |
1126 | kfree(pblk); | |
1127 | } | |
1128 | ||
a7c9e910 | 1129 | static void pblk_tear_down(struct pblk *pblk, bool graceful) |
a4bd217b | 1130 | { |
a7c9e910 JG |
1131 | if (graceful) |
1132 | __pblk_pipeline_flush(pblk); | |
1133 | __pblk_pipeline_stop(pblk); | |
a4bd217b JG |
1134 | pblk_writer_stop(pblk); |
1135 | pblk_rb_sync_l2p(&pblk->rwb); | |
a4bd217b JG |
1136 | pblk_rl_free(&pblk->rl); |
1137 | ||
4e495a46 | 1138 | pblk_debug(pblk, "consistent tear down (graceful:%d)\n", graceful); |
a4bd217b JG |
1139 | } |
1140 | ||
a7c9e910 | 1141 | static void pblk_exit(void *private, bool graceful) |
a4bd217b JG |
1142 | { |
1143 | struct pblk *pblk = private; | |
1144 | ||
a7c9e910 JG |
1145 | pblk_gc_exit(pblk, graceful); |
1146 | pblk_tear_down(pblk, graceful); | |
c5586192 | 1147 | |
880eda54 | 1148 | #ifdef CONFIG_NVM_PBLK_DEBUG |
4e495a46 | 1149 | pblk_info(pblk, "exit: L2P CRC: %x\n", pblk_l2p_crc(pblk)); |
c5586192 HH |
1150 | #endif |
1151 | ||
a4bd217b | 1152 | pblk_free(pblk); |
a4bd217b JG |
1153 | } |
1154 | ||
1155 | static sector_t pblk_capacity(void *private) | |
1156 | { | |
1157 | struct pblk *pblk = private; | |
1158 | ||
1159 | return pblk->capacity * NR_PHY_IN_LOG; | |
1160 | } | |
1161 | ||
1162 | static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk, | |
1163 | int flags) | |
1164 | { | |
1165 | struct nvm_geo *geo = &dev->geo; | |
1166 | struct request_queue *bqueue = dev->q; | |
1167 | struct request_queue *tqueue = tdisk->queue; | |
1168 | struct pblk *pblk; | |
1169 | int ret; | |
1170 | ||
4e495a46 MB |
1171 | pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL); |
1172 | if (!pblk) | |
1173 | return ERR_PTR(-ENOMEM); | |
1174 | ||
1175 | pblk->dev = dev; | |
1176 | pblk->disk = tdisk; | |
1177 | pblk->state = PBLK_STATE_RUNNING; | |
1b0dd0bf | 1178 | trace_pblk_state(pblk_disk_name(pblk), pblk->state); |
4e495a46 MB |
1179 | pblk->gc.gc_enabled = 0; |
1180 | ||
3b2a3ad1 JG |
1181 | if (!(geo->version == NVM_OCSSD_SPEC_12 || |
1182 | geo->version == NVM_OCSSD_SPEC_20)) { | |
4e495a46 | 1183 | pblk_err(pblk, "OCSSD version not supported (%u)\n", |
7ad5039e | 1184 | geo->version); |
4e495a46 | 1185 | kfree(pblk); |
7ad5039e JG |
1186 | return ERR_PTR(-EINVAL); |
1187 | } | |
1188 | ||
a16816b9 IK |
1189 | if (geo->ext) { |
1190 | pblk_err(pblk, "extended metadata not supported\n"); | |
1191 | kfree(pblk); | |
1192 | return ERR_PTR(-EINVAL); | |
1193 | } | |
1194 | ||
6a3abf5b | 1195 | spin_lock_init(&pblk->resubmit_lock); |
a4bd217b JG |
1196 | spin_lock_init(&pblk->trans_lock); |
1197 | spin_lock_init(&pblk->lock); | |
1198 | ||
880eda54 | 1199 | #ifdef CONFIG_NVM_PBLK_DEBUG |
a4bd217b JG |
1200 | atomic_long_set(&pblk->inflight_writes, 0); |
1201 | atomic_long_set(&pblk->padded_writes, 0); | |
1202 | atomic_long_set(&pblk->padded_wb, 0); | |
a4bd217b JG |
1203 | atomic_long_set(&pblk->req_writes, 0); |
1204 | atomic_long_set(&pblk->sub_writes, 0); | |
1205 | atomic_long_set(&pblk->sync_writes, 0); | |
a4bd217b | 1206 | atomic_long_set(&pblk->inflight_reads, 0); |
db7ada33 | 1207 | atomic_long_set(&pblk->cache_reads, 0); |
a4bd217b JG |
1208 | atomic_long_set(&pblk->sync_reads, 0); |
1209 | atomic_long_set(&pblk->recov_writes, 0); | |
1210 | atomic_long_set(&pblk->recov_writes, 0); | |
1211 | atomic_long_set(&pblk->recov_gc_writes, 0); | |
a1121176 | 1212 | atomic_long_set(&pblk->recov_gc_reads, 0); |
a4bd217b JG |
1213 | #endif |
1214 | ||
1215 | atomic_long_set(&pblk->read_failed, 0); | |
1216 | atomic_long_set(&pblk->read_empty, 0); | |
1217 | atomic_long_set(&pblk->read_high_ecc, 0); | |
1218 | atomic_long_set(&pblk->read_failed_gc, 0); | |
1219 | atomic_long_set(&pblk->write_failed, 0); | |
1220 | atomic_long_set(&pblk->erase_failed, 0); | |
1221 | ||
43d47127 | 1222 | ret = pblk_core_init(pblk); |
a4bd217b | 1223 | if (ret) { |
4e495a46 | 1224 | pblk_err(pblk, "could not initialize core\n"); |
a4bd217b JG |
1225 | goto fail; |
1226 | } | |
1227 | ||
1228 | ret = pblk_lines_init(pblk); | |
1229 | if (ret) { | |
4e495a46 | 1230 | pblk_err(pblk, "could not initialize lines\n"); |
43d47127 | 1231 | goto fail_free_core; |
5d149bfa HH |
1232 | } |
1233 | ||
43d47127 | 1234 | ret = pblk_rwb_init(pblk); |
a4bd217b | 1235 | if (ret) { |
4e495a46 | 1236 | pblk_err(pblk, "could not initialize write buffer\n"); |
43d47127 | 1237 | goto fail_free_lines; |
a4bd217b JG |
1238 | } |
1239 | ||
43d47127 | 1240 | ret = pblk_l2p_init(pblk, flags & NVM_TARGET_FACTORY); |
a4bd217b | 1241 | if (ret) { |
4e495a46 | 1242 | pblk_err(pblk, "could not initialize maps\n"); |
43d47127 | 1243 | goto fail_free_rwb; |
a4bd217b JG |
1244 | } |
1245 | ||
1246 | ret = pblk_writer_init(pblk); | |
1247 | if (ret) { | |
cc4f5ba1 | 1248 | if (ret != -EINTR) |
4e495a46 | 1249 | pblk_err(pblk, "could not initialize write thread\n"); |
43d47127 | 1250 | goto fail_free_l2p; |
a4bd217b JG |
1251 | } |
1252 | ||
1253 | ret = pblk_gc_init(pblk); | |
1254 | if (ret) { | |
4e495a46 | 1255 | pblk_err(pblk, "could not initialize gc\n"); |
a4bd217b JG |
1256 | goto fail_stop_writer; |
1257 | } | |
1258 | ||
1259 | /* inherit the size from the underlying device */ | |
1260 | blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue)); | |
1261 | blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue)); | |
1262 | ||
1263 | blk_queue_write_cache(tqueue, true, false); | |
1264 | ||
e46f4e48 | 1265 | tqueue->limits.discard_granularity = geo->clba * geo->csecs; |
a4bd217b JG |
1266 | tqueue->limits.discard_alignment = 0; |
1267 | blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9); | |
8b904b5b | 1268 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, tqueue); |
a4bd217b | 1269 | |
4e495a46 | 1270 | pblk_info(pblk, "luns:%u, lines:%d, secs:%llu, buf entries:%u\n", |
fae7fae4 | 1271 | geo->all_luns, pblk->l_mg.nr_lines, |
847a3a27 | 1272 | (unsigned long long)pblk->capacity, |
a4bd217b JG |
1273 | pblk->rwb.nr_entries); |
1274 | ||
1275 | wake_up_process(pblk->writer_ts); | |
03661b5f HH |
1276 | |
1277 | /* Check if we need to start GC */ | |
1278 | pblk_gc_should_kick(pblk); | |
1279 | ||
a4bd217b JG |
1280 | return pblk; |
1281 | ||
1282 | fail_stop_writer: | |
1283 | pblk_writer_stop(pblk); | |
a4bd217b JG |
1284 | fail_free_l2p: |
1285 | pblk_l2p_free(pblk); | |
43d47127 JG |
1286 | fail_free_rwb: |
1287 | pblk_rwb_free(pblk); | |
1288 | fail_free_lines: | |
1289 | pblk_lines_free(pblk); | |
a4bd217b JG |
1290 | fail_free_core: |
1291 | pblk_core_free(pblk); | |
a4bd217b JG |
1292 | fail: |
1293 | kfree(pblk); | |
1294 | return ERR_PTR(ret); | |
1295 | } | |
1296 | ||
1297 | /* physical block device target */ | |
1298 | static struct nvm_tgt_type tt_pblk = { | |
1299 | .name = "pblk", | |
1300 | .version = {1, 0, 0}, | |
1301 | ||
1302 | .make_rq = pblk_make_rq, | |
1303 | .capacity = pblk_capacity, | |
1304 | ||
1305 | .init = pblk_init, | |
1306 | .exit = pblk_exit, | |
1307 | ||
1308 | .sysfs_init = pblk_sysfs_init, | |
1309 | .sysfs_exit = pblk_sysfs_exit, | |
90014829 | 1310 | .owner = THIS_MODULE, |
a4bd217b JG |
1311 | }; |
1312 | ||
1313 | static int __init pblk_module_init(void) | |
1314 | { | |
b25d5237 N |
1315 | int ret; |
1316 | ||
b906bbb6 KO |
1317 | ret = bioset_init(&pblk_bio_set, BIO_POOL_SIZE, 0, 0); |
1318 | if (ret) | |
1319 | return ret; | |
b25d5237 N |
1320 | ret = nvm_register_tgt_type(&tt_pblk); |
1321 | if (ret) | |
b906bbb6 | 1322 | bioset_exit(&pblk_bio_set); |
b25d5237 | 1323 | return ret; |
a4bd217b JG |
1324 | } |
1325 | ||
1326 | static void pblk_module_exit(void) | |
1327 | { | |
b906bbb6 | 1328 | bioset_exit(&pblk_bio_set); |
a4bd217b JG |
1329 | nvm_unregister_tgt_type(&tt_pblk); |
1330 | } | |
1331 | ||
1332 | module_init(pblk_module_init); | |
1333 | module_exit(pblk_module_exit); | |
1334 | MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>"); | |
1335 | MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>"); | |
1336 | MODULE_LICENSE("GPL v2"); | |
1337 | MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs"); |