• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
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"
23 #include "pblk-trace.h"
24 
25 static unsigned int write_buffer_size;
26 
27 module_param(write_buffer_size, uint, 0644);
28 MODULE_PARM_DESC(write_buffer_size, "number of entries in a write buffer");
29 
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 
48 struct bio_set pblk_bio_set;
49 
pblk_make_rq(struct request_queue * q,struct bio * bio)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 
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);
67 		pblk_submit_read(pblk, bio);
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 
76 		pblk_write_to_cache(pblk, bio, PBLK_IOTYPE_USER);
77 	}
78 
79 	return BLK_QC_T_NONE;
80 }
81 
pblk_trans_map_size(struct pblk * pblk)82 static size_t pblk_trans_map_size(struct pblk *pblk)
83 {
84 	int entry_size = 8;
85 
86 	if (pblk->addrf_len < 32)
87 		entry_size = 4;
88 
89 	return entry_size * pblk->capacity;
90 }
91 
92 #ifdef CONFIG_NVM_PBLK_DEBUG
pblk_l2p_crc(struct pblk * pblk)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 
pblk_l2p_free(struct pblk * pblk)104 static void pblk_l2p_free(struct pblk *pblk)
105 {
106 	vfree(pblk->trans_map);
107 }
108 
pblk_l2p_recover(struct pblk * pblk,bool factory_init)109 static int pblk_l2p_recover(struct pblk *pblk, bool factory_init)
110 {
111 	struct pblk_line *line = NULL;
112 
113 	if (factory_init) {
114 		guid_gen(&pblk->instance_uuid);
115 	} else {
116 		line = pblk_recov_l2p(pblk);
117 		if (IS_ERR(line)) {
118 			pblk_err(pblk, "could not recover l2p table\n");
119 			return -EFAULT;
120 		}
121 	}
122 
123 #ifdef CONFIG_NVM_PBLK_DEBUG
124 	pblk_info(pblk, "init: L2P CRC: %x\n", pblk_l2p_crc(pblk));
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);
133 		if (!line)
134 			return -EFAULT;
135 	}
136 
137 	return 0;
138 }
139 
pblk_l2p_init(struct pblk * pblk,bool factory_init)140 static int pblk_l2p_init(struct pblk *pblk, bool factory_init)
141 {
142 	sector_t i;
143 	struct ppa_addr ppa;
144 	size_t map_size;
145 	int ret = 0;
146 
147 	map_size = pblk_trans_map_size(pblk);
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);
154 		return -ENOMEM;
155 	}
156 
157 	pblk_ppa_set_empty(&ppa);
158 
159 	for (i = 0; i < pblk->capacity; i++)
160 		pblk_trans_map_set(pblk, i, ppa);
161 
162 	ret = pblk_l2p_recover(pblk, factory_init);
163 	if (ret)
164 		vfree(pblk->trans_map);
165 
166 	return ret;
167 }
168 
pblk_rwb_free(struct pblk * pblk)169 static void pblk_rwb_free(struct pblk *pblk)
170 {
171 	if (pblk_rb_tear_down_check(&pblk->rwb))
172 		pblk_err(pblk, "write buffer error on tear down\n");
173 
174 	pblk_rb_free(&pblk->rwb);
175 }
176 
pblk_rwb_init(struct pblk * pblk)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;
181 	unsigned long buffer_size;
182 	int pgs_in_buffer, threshold;
183 
184 	threshold = geo->mw_cunits * geo->all_luns;
185 	pgs_in_buffer = (max(geo->mw_cunits, geo->ws_opt) + geo->ws_opt)
186 								* geo->all_luns;
187 
188 	if (write_buffer_size && (write_buffer_size > pgs_in_buffer))
189 		buffer_size = write_buffer_size;
190 	else
191 		buffer_size = pgs_in_buffer;
192 
193 	return pblk_rb_init(&pblk->rwb, buffer_size, threshold, geo->csecs);
194 }
195 
pblk_set_addrf_12(struct pblk * pblk,struct nvm_geo * geo,struct nvm_addrf_12 * dst)196 static int pblk_set_addrf_12(struct pblk *pblk, struct nvm_geo *geo,
197 			     struct nvm_addrf_12 *dst)
198 {
199 	struct nvm_addrf_12 *src = (struct nvm_addrf_12 *)&geo->addrf;
200 	int power_len;
201 
202 	/* Re-calculate channel and lun format to adapt to configuration */
203 	power_len = get_count_order(geo->num_ch);
204 	if (1 << power_len != geo->num_ch) {
205 		pblk_err(pblk, "supports only power-of-two channel config.\n");
206 		return -EINVAL;
207 	}
208 	dst->ch_len = power_len;
209 
210 	power_len = get_count_order(geo->num_lun);
211 	if (1 << power_len != geo->num_lun) {
212 		pblk_err(pblk, "supports only power-of-two LUN config.\n");
213 		return -EINVAL;
214 	}
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;
220 	dst->sec_len = src->sec_len;
221 
222 	dst->sec_offset = 0;
223 	dst->pln_offset = dst->sec_len;
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 
229 	dst->sec_mask = ((1ULL << dst->sec_len) - 1) << dst->sec_offset;
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 
pblk_set_addrf_20(struct nvm_geo * geo,struct nvm_addrf * adst,struct pblk_addrf * udst)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 
pblk_set_addrf(struct pblk * pblk)269 static int pblk_set_addrf(struct pblk *pblk)
270 {
271 	struct nvm_tgt_dev *dev = pblk->dev;
272 	struct nvm_geo *geo = &dev->geo;
273 	int mod;
274 
275 	switch (geo->version) {
276 	case NVM_OCSSD_SPEC_12:
277 		div_u64_rem(geo->clba, pblk->min_write_pgs, &mod);
278 		if (mod) {
279 			pblk_err(pblk, "bad configuration of sectors/pages\n");
280 			return -EINVAL;
281 		}
282 
283 		pblk->addrf_len = pblk_set_addrf_12(pblk, geo,
284 							(void *)&pblk->addrf);
285 		break;
286 	case NVM_OCSSD_SPEC_20:
287 		pblk->addrf_len = pblk_set_addrf_20(geo, (void *)&pblk->addrf,
288 							&pblk->uaddrf);
289 		break;
290 	default:
291 		pblk_err(pblk, "OCSSD revision not supported (%d)\n",
292 								geo->version);
293 		return -EINVAL;
294 	}
295 
296 	return 0;
297 }
298 
pblk_create_global_caches(void)299 static int pblk_create_global_caches(void)
300 {
301 
302 	pblk_caches.ws = kmem_cache_create("pblk_blk_ws",
303 				sizeof(struct pblk_line_ws), 0, 0, NULL);
304 	if (!pblk_caches.ws)
305 		return -ENOMEM;
306 
307 	pblk_caches.rec = kmem_cache_create("pblk_rec",
308 				sizeof(struct pblk_rec_ctx), 0, 0, NULL);
309 	if (!pblk_caches.rec)
310 		goto fail_destroy_ws;
311 
312 	pblk_caches.g_rq = kmem_cache_create("pblk_g_rq", pblk_g_rq_size,
313 				0, 0, NULL);
314 	if (!pblk_caches.g_rq)
315 		goto fail_destroy_rec;
316 
317 	pblk_caches.w_rq = kmem_cache_create("pblk_w_rq", pblk_w_rq_size,
318 				0, 0, NULL);
319 	if (!pblk_caches.w_rq)
320 		goto fail_destroy_g_rq;
321 
322 	return 0;
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 
pblk_get_global_caches(void)334 static int pblk_get_global_caches(void)
335 {
336 	int ret = 0;
337 
338 	mutex_lock(&pblk_caches.mutex);
339 
340 	if (kref_get_unless_zero(&pblk_caches.kref))
341 		goto out;
342 
343 	ret = pblk_create_global_caches();
344 	if (!ret)
345 		kref_init(&pblk_caches.kref);
346 
347 out:
348 	mutex_unlock(&pblk_caches.mutex);
349 	return ret;
350 }
351 
pblk_destroy_global_caches(struct kref * ref)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);
362 }
363 
pblk_put_global_caches(void)364 static void pblk_put_global_caches(void)
365 {
366 	mutex_lock(&pblk_caches.mutex);
367 	kref_put(&pblk_caches.kref, pblk_destroy_global_caches);
368 	mutex_unlock(&pblk_caches.mutex);
369 }
370 
pblk_core_init(struct pblk * pblk)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;
375 	int ret, max_write_ppas;
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;
386 
387 	pblk->min_write_pgs = geo->ws_opt;
388 	pblk->min_write_pgs_data = pblk->min_write_pgs;
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);
391 	pblk->max_write_pgs = min_t(int, pblk->max_write_pgs,
392 		queue_max_hw_sectors(dev->q) / (geo->csecs >> SECTOR_SHIFT));
393 	pblk_set_sec_per_write(pblk, pblk->min_write_pgs);
394 
395 	pblk->oob_meta_size = geo->sos;
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;
425 	}
426 
427 	pblk->pad_dist = kcalloc(pblk->min_write_pgs - 1, sizeof(atomic64_t),
428 								GFP_KERNEL);
429 	if (!pblk->pad_dist)
430 		return -ENOMEM;
431 
432 	if (pblk_get_global_caches())
433 		goto fail_free_pad_dist;
434 
435 	/* Internal bios can be at most the sectors signaled by the device. */
436 	ret = mempool_init_page_pool(&pblk->page_bio_pool, NVM_MAX_VLBA, 0);
437 	if (ret)
438 		goto free_global_caches;
439 
440 	ret = mempool_init_slab_pool(&pblk->gen_ws_pool, PBLK_GEN_WS_POOL_SIZE,
441 				     pblk_caches.ws);
442 	if (ret)
443 		goto free_page_bio_pool;
444 
445 	ret = mempool_init_slab_pool(&pblk->rec_pool, geo->all_luns,
446 				     pblk_caches.rec);
447 	if (ret)
448 		goto free_gen_ws_pool;
449 
450 	ret = mempool_init_slab_pool(&pblk->r_rq_pool, geo->all_luns,
451 				     pblk_caches.g_rq);
452 	if (ret)
453 		goto free_rec_pool;
454 
455 	ret = mempool_init_slab_pool(&pblk->e_rq_pool, geo->all_luns,
456 				     pblk_caches.g_rq);
457 	if (ret)
458 		goto free_r_rq_pool;
459 
460 	ret = mempool_init_slab_pool(&pblk->w_rq_pool, geo->all_luns,
461 				     pblk_caches.w_rq);
462 	if (ret)
463 		goto free_e_rq_pool;
464 
465 	pblk->close_wq = alloc_workqueue("pblk-close-wq",
466 			WQ_MEM_RECLAIM | WQ_UNBOUND, PBLK_NR_CLOSE_JOBS);
467 	if (!pblk->close_wq)
468 		goto free_w_rq_pool;
469 
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 
475 	pblk->r_end_wq = alloc_workqueue("pblk-read-end-wq",
476 			WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
477 	if (!pblk->r_end_wq)
478 		goto free_bb_wq;
479 
480 	if (pblk_set_addrf(pblk))
481 		goto free_r_end_wq;
482 
483 	INIT_LIST_HEAD(&pblk->compl_list);
484 	INIT_LIST_HEAD(&pblk->resubmit_list);
485 
486 	return 0;
487 
488 free_r_end_wq:
489 	destroy_workqueue(pblk->r_end_wq);
490 free_bb_wq:
491 	destroy_workqueue(pblk->bb_wq);
492 free_close_wq:
493 	destroy_workqueue(pblk->close_wq);
494 free_w_rq_pool:
495 	mempool_exit(&pblk->w_rq_pool);
496 free_e_rq_pool:
497 	mempool_exit(&pblk->e_rq_pool);
498 free_r_rq_pool:
499 	mempool_exit(&pblk->r_rq_pool);
500 free_rec_pool:
501 	mempool_exit(&pblk->rec_pool);
502 free_gen_ws_pool:
503 	mempool_exit(&pblk->gen_ws_pool);
504 free_page_bio_pool:
505 	mempool_exit(&pblk->page_bio_pool);
506 free_global_caches:
507 	pblk_put_global_caches();
508 fail_free_pad_dist:
509 	kfree(pblk->pad_dist);
510 	return -ENOMEM;
511 }
512 
pblk_core_free(struct pblk * pblk)513 static void pblk_core_free(struct pblk *pblk)
514 {
515 	if (pblk->close_wq)
516 		destroy_workqueue(pblk->close_wq);
517 
518 	if (pblk->r_end_wq)
519 		destroy_workqueue(pblk->r_end_wq);
520 
521 	if (pblk->bb_wq)
522 		destroy_workqueue(pblk->bb_wq);
523 
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);
530 
531 	pblk_put_global_caches();
532 	kfree(pblk->pad_dist);
533 }
534 
pblk_line_mg_free(struct pblk * pblk)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 		kvfree(l_mg->eline_meta[i]->buf);
547 		kfree(l_mg->eline_meta[i]);
548 	}
549 
550 	mempool_destroy(l_mg->bitmap_pool);
551 	kmem_cache_destroy(l_mg->bitmap_cache);
552 }
553 
pblk_line_meta_free(struct pblk_line_mgmt * l_mg,struct pblk_line * line)554 static void pblk_line_meta_free(struct pblk_line_mgmt *l_mg,
555 				struct pblk_line *line)
556 {
557 	struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
558 
559 	kfree(line->blk_bitmap);
560 	kfree(line->erase_bitmap);
561 	kfree(line->chks);
562 
563 	kvfree(w_err_gc->lba_list);
564 	kfree(w_err_gc);
565 }
566 
pblk_lines_free(struct pblk * pblk)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 
573 	for (i = 0; i < l_mg->nr_lines; i++) {
574 		line = &pblk->lines[i];
575 
576 		pblk_line_free(line);
577 		pblk_line_meta_free(l_mg, line);
578 	}
579 
580 	pblk_line_mg_free(pblk);
581 
582 	kfree(pblk->luns);
583 	kfree(pblk->lines);
584 }
585 
pblk_luns_init(struct pblk * pblk)586 static int pblk_luns_init(struct pblk *pblk)
587 {
588 	struct nvm_tgt_dev *dev = pblk->dev;
589 	struct nvm_geo *geo = &dev->geo;
590 	struct pblk_lun *rlun;
591 	int i;
592 
593 	/* TODO: Implement unbalanced LUN support */
594 	if (geo->num_lun < 0) {
595 		pblk_err(pblk, "unbalanced LUN config.\n");
596 		return -EINVAL;
597 	}
598 
599 	pblk->luns = kcalloc(geo->all_luns, sizeof(struct pblk_lun),
600 								GFP_KERNEL);
601 	if (!pblk->luns)
602 		return -ENOMEM;
603 
604 	for (i = 0; i < geo->all_luns; i++) {
605 		/* Stripe across channels */
606 		int ch = i % geo->num_ch;
607 		int lun_raw = i / geo->num_ch;
608 		int lunid = lun_raw + ch * geo->num_lun;
609 
610 		rlun = &pblk->luns[i];
611 		rlun->bppa = dev->luns[lunid];
612 
613 		sema_init(&rlun->wr_sem, 1);
614 	}
615 
616 	return 0;
617 }
618 
619 /* See comment over struct line_emeta definition */
calc_emeta_len(struct pblk * pblk)620 static unsigned int calc_emeta_len(struct pblk *pblk)
621 {
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(
629 			sizeof(struct line_emeta) + lm->blk_bitmap_len +
630 			sizeof(struct wa_counters), geo->csecs);
631 	lm->emeta_len[1] = lm->emeta_sec[1] * geo->csecs;
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),
636 			geo->csecs);
637 	lm->emeta_len[2] = lm->emeta_sec[2] * geo->csecs;
638 
639 	lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
640 			geo->csecs);
641 	lm->emeta_len[3] = lm->emeta_sec[3] * geo->csecs;
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]);
646 }
647 
pblk_set_provision(struct pblk * pblk,int nr_free_chks)648 static int pblk_set_provision(struct pblk *pblk, int nr_free_chks)
649 {
650 	struct nvm_tgt_dev *dev = pblk->dev;
651 	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
652 	struct pblk_line_meta *lm = &pblk->lm;
653 	struct nvm_geo *geo = &dev->geo;
654 	sector_t provisioned;
655 	int sec_meta, blk_meta, clba;
656 	int minimum;
657 
658 	if (geo->op == NVM_TARGET_DEFAULT_OP)
659 		pblk->op = PBLK_DEFAULT_OP;
660 	else
661 		pblk->op = geo->op;
662 
663 	minimum = pblk_get_min_chks(pblk);
664 	provisioned = nr_free_chks;
665 	provisioned *= (100 - pblk->op);
666 	sector_div(provisioned, 100);
667 
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;
685 
686 	/* Internally pblk manages all free blocks, but all calculations based
687 	 * on user capacity consider only provisioned blocks
688 	 */
689 	pblk->rl.total_blocks = nr_free_chks;
690 
691 	/* Consider sectors used for metadata */
692 	sec_meta = (lm->smeta_sec + lm->emeta_sec[0]) * l_mg->nr_free_lines;
693 	blk_meta = DIV_ROUND_UP(sec_meta, geo->clba);
694 
695 	clba = (geo->clba / pblk->min_write_pgs) * pblk->min_write_pgs_data;
696 	pblk->capacity = (provisioned - blk_meta) * clba;
697 
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;
702 }
703 
pblk_setup_line_meta_chk(struct pblk * pblk,struct pblk_line * line,struct nvm_chk_meta * meta)704 static int pblk_setup_line_meta_chk(struct pblk *pblk, struct pblk_line *line,
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 
733 		trace_pblk_chunk_state(pblk_disk_name(pblk), &ppa,
734 					chunk->state);
735 
736 		if (chunk->type & NVM_CHK_TP_SZ_SPEC) {
737 			WARN_ONCE(1, "pblk: custom-sized chunks unsupported\n");
738 			continue;
739 		}
740 
741 		if (!(chunk->state & NVM_CHK_ST_OFFLINE))
742 			continue;
743 
744 		set_bit(pos, line->blk_bitmap);
745 		nr_bad_chks++;
746 	}
747 
748 	return nr_bad_chks;
749 }
750 
pblk_setup_line_meta(struct pblk * pblk,struct pblk_line * line,void * chunk_meta,int line_id)751 static long pblk_setup_line_meta(struct pblk *pblk, struct pblk_line *line,
752 				 void *chunk_meta, int line_id)
753 {
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 
766 	nr_bad_chks = pblk_setup_line_meta_chk(pblk, line, chunk_meta);
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 
pblk_alloc_line_meta(struct pblk * pblk,struct pblk_line * line)783 static int pblk_alloc_line_meta(struct pblk *pblk, struct pblk_line *line)
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);
792 	if (!line->erase_bitmap)
793 		goto free_blk_bitmap;
794 
795 
796 	line->chks = kmalloc_array(lm->blk_per_line,
797 				   sizeof(struct nvm_chk_meta), GFP_KERNEL);
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;
804 
805 	return 0;
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;
814 }
815 
pblk_line_mg_init(struct pblk * pblk)816 static int pblk_line_mg_init(struct pblk *pblk)
817 {
818 	struct nvm_tgt_dev *dev = pblk->dev;
819 	struct nvm_geo *geo = &dev->geo;
820 	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
821 	struct pblk_line_meta *lm = &pblk->lm;
822 	int i, bb_distance;
823 
824 	l_mg->nr_lines = geo->num_chk;
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);
838 	INIT_LIST_HEAD(&l_mg->gc_werr_list);
839 
840 	INIT_LIST_HEAD(&l_mg->emeta_list);
841 
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;
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;
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 	 */
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 
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 
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 		emeta->buf = kvmalloc(lm->emeta_len[0], GFP_KERNEL);
894 		if (!emeta->buf) {
895 			kfree(emeta);
896 			goto fail_free_emeta;
897 		}
898 
899 		emeta->nr_entries = lm->emeta_sec[0];
900 		l_mg->eline_meta[i] = emeta;
901 	}
902 
903 	for (i = 0; i < l_mg->nr_lines; i++)
904 		l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
905 
906 	bb_distance = (geo->all_luns) * geo->ws_opt;
907 	for (i = 0; i < lm->sec_per_line; i += bb_distance)
908 		bitmap_set(l_mg->bb_template, i, geo->ws_opt);
909 
910 	return 0;
911 
912 fail_free_emeta:
913 	while (--i >= 0) {
914 		kvfree(l_mg->eline_meta[i]->buf);
915 		kfree(l_mg->eline_meta[i]);
916 	}
917 
918 	mempool_destroy(l_mg->bitmap_pool);
919 fail_destroy_bitmap_cache:
920 	kmem_cache_destroy(l_mg->bitmap_cache);
921 fail_free_smeta:
922 	for (i = 0; i < PBLK_DATA_LINES; i++)
923 		kfree(l_mg->sline_meta[i]);
924 	kfree(l_mg->bb_aux);
925 fail_free_bb_template:
926 	kfree(l_mg->bb_template);
927 fail_free_vsc_list:
928 	kfree(l_mg->vsc_list);
929 fail:
930 	return -ENOMEM;
931 }
932 
pblk_line_meta_init(struct pblk * pblk)933 static int pblk_line_meta_init(struct pblk *pblk)
934 {
935 	struct nvm_tgt_dev *dev = pblk->dev;
936 	struct nvm_geo *geo = &dev->geo;
937 	struct pblk_line_meta *lm = &pblk->lm;
938 	unsigned int smeta_len, emeta_len;
939 	int i;
940 
941 	lm->sec_per_line = geo->clba * geo->all_luns;
942 	lm->blk_per_line = geo->all_luns;
943 	lm->blk_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
944 	lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
945 	lm->lun_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
946 	lm->mid_thrs = lm->sec_per_line / 2;
947 	lm->high_thrs = lm->sec_per_line / 4;
948 	lm->meta_distance = (geo->all_luns / 2) * pblk->min_write_pgs;
949 
950 	/* Calculate necessary pages for smeta. See comment over struct
951 	 * line_smeta definition
952 	 */
953 	i = 1;
954 add_smeta_page:
955 	lm->smeta_sec = i * geo->ws_opt;
956 	lm->smeta_len = lm->smeta_sec * geo->csecs;
957 
958 	smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
959 	if (smeta_len > lm->smeta_len) {
960 		i++;
961 		goto add_smeta_page;
962 	}
963 
964 	/* Calculate necessary pages for emeta. See comment over struct
965 	 * line_emeta definition
966 	 */
967 	i = 1;
968 add_emeta_page:
969 	lm->emeta_sec[0] = i * geo->ws_opt;
970 	lm->emeta_len[0] = lm->emeta_sec[0] * geo->csecs;
971 
972 	emeta_len = calc_emeta_len(pblk);
973 	if (emeta_len > lm->emeta_len[0]) {
974 		i++;
975 		goto add_emeta_page;
976 	}
977 
978 	lm->emeta_bb = geo->all_luns > i ? geo->all_luns - i : 0;
979 
980 	lm->min_blk_line = 1;
981 	if (geo->all_luns > 1)
982 		lm->min_blk_line += DIV_ROUND_UP(lm->smeta_sec +
983 					lm->emeta_sec[0], geo->clba);
984 
985 	if (lm->min_blk_line > lm->blk_per_line) {
986 		pblk_err(pblk, "config. not supported. Min. LUN in line:%d\n",
987 							lm->blk_per_line);
988 		return -EINVAL;
989 	}
990 
991 	return 0;
992 }
993 
pblk_lines_init(struct pblk * pblk)994 static int pblk_lines_init(struct pblk *pblk)
995 {
996 	struct pblk_line_mgmt *l_mg = &pblk->l_mg;
997 	struct pblk_line *line;
998 	void *chunk_meta;
999 	int nr_free_chks = 0;
1000 	int i, ret;
1001 
1002 	ret = pblk_line_meta_init(pblk);
1003 	if (ret)
1004 		return ret;
1005 
1006 	ret = pblk_line_mg_init(pblk);
1007 	if (ret)
1008 		return ret;
1009 
1010 	ret = pblk_luns_init(pblk);
1011 	if (ret)
1012 		goto fail_free_meta;
1013 
1014 	chunk_meta = pblk_get_chunk_meta(pblk);
1015 	if (IS_ERR(chunk_meta)) {
1016 		ret = PTR_ERR(chunk_meta);
1017 		goto fail_free_luns;
1018 	}
1019 
1020 	pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
1021 								GFP_KERNEL);
1022 	if (!pblk->lines) {
1023 		ret = -ENOMEM;
1024 		goto fail_free_chunk_meta;
1025 	}
1026 
1027 	for (i = 0; i < l_mg->nr_lines; i++) {
1028 		line = &pblk->lines[i];
1029 
1030 		ret = pblk_alloc_line_meta(pblk, line);
1031 		if (ret)
1032 			goto fail_free_lines;
1033 
1034 		nr_free_chks += pblk_setup_line_meta(pblk, line, chunk_meta, i);
1035 
1036 		trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1037 								line->state);
1038 	}
1039 
1040 	if (!nr_free_chks) {
1041 		pblk_err(pblk, "too many bad blocks prevent for sane instance\n");
1042 		ret = -EINTR;
1043 		goto fail_free_lines;
1044 	}
1045 
1046 	ret = pblk_set_provision(pblk, nr_free_chks);
1047 	if (ret)
1048 		goto fail_free_lines;
1049 
1050 	vfree(chunk_meta);
1051 	return 0;
1052 
1053 fail_free_lines:
1054 	while (--i >= 0)
1055 		pblk_line_meta_free(l_mg, &pblk->lines[i]);
1056 	kfree(pblk->lines);
1057 fail_free_chunk_meta:
1058 	vfree(chunk_meta);
1059 fail_free_luns:
1060 	kfree(pblk->luns);
1061 fail_free_meta:
1062 	pblk_line_mg_free(pblk);
1063 
1064 	return ret;
1065 }
1066 
pblk_writer_init(struct pblk * pblk)1067 static int pblk_writer_init(struct pblk *pblk)
1068 {
1069 	pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
1070 	if (IS_ERR(pblk->writer_ts)) {
1071 		int err = PTR_ERR(pblk->writer_ts);
1072 
1073 		if (err != -EINTR)
1074 			pblk_err(pblk, "could not allocate writer kthread (%d)\n",
1075 					err);
1076 		return err;
1077 	}
1078 
1079 	timer_setup(&pblk->wtimer, pblk_write_timer_fn, 0);
1080 	mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
1081 
1082 	return 0;
1083 }
1084 
pblk_writer_stop(struct pblk * pblk)1085 static void pblk_writer_stop(struct pblk *pblk)
1086 {
1087 	/* The pipeline must be stopped and the write buffer emptied before the
1088 	 * write thread is stopped
1089 	 */
1090 	WARN(pblk_rb_read_count(&pblk->rwb),
1091 			"Stopping not fully persisted write buffer\n");
1092 
1093 	WARN(pblk_rb_sync_count(&pblk->rwb),
1094 			"Stopping not fully synced write buffer\n");
1095 
1096 	del_timer_sync(&pblk->wtimer);
1097 	if (pblk->writer_ts)
1098 		kthread_stop(pblk->writer_ts);
1099 }
1100 
pblk_free(struct pblk * pblk)1101 static void pblk_free(struct pblk *pblk)
1102 {
1103 	pblk_lines_free(pblk);
1104 	pblk_l2p_free(pblk);
1105 	pblk_rwb_free(pblk);
1106 	pblk_core_free(pblk);
1107 
1108 	kfree(pblk);
1109 }
1110 
pblk_tear_down(struct pblk * pblk,bool graceful)1111 static void pblk_tear_down(struct pblk *pblk, bool graceful)
1112 {
1113 	if (graceful)
1114 		__pblk_pipeline_flush(pblk);
1115 	__pblk_pipeline_stop(pblk);
1116 	pblk_writer_stop(pblk);
1117 	pblk_rb_sync_l2p(&pblk->rwb);
1118 	pblk_rl_free(&pblk->rl);
1119 
1120 	pblk_debug(pblk, "consistent tear down (graceful:%d)\n", graceful);
1121 }
1122 
pblk_exit(void * private,bool graceful)1123 static void pblk_exit(void *private, bool graceful)
1124 {
1125 	struct pblk *pblk = private;
1126 
1127 	pblk_gc_exit(pblk, graceful);
1128 	pblk_tear_down(pblk, graceful);
1129 
1130 #ifdef CONFIG_NVM_PBLK_DEBUG
1131 	pblk_info(pblk, "exit: L2P CRC: %x\n", pblk_l2p_crc(pblk));
1132 #endif
1133 
1134 	pblk_free(pblk);
1135 }
1136 
pblk_capacity(void * private)1137 static sector_t pblk_capacity(void *private)
1138 {
1139 	struct pblk *pblk = private;
1140 
1141 	return pblk->capacity * NR_PHY_IN_LOG;
1142 }
1143 
pblk_init(struct nvm_tgt_dev * dev,struct gendisk * tdisk,int flags)1144 static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
1145 		       int flags)
1146 {
1147 	struct nvm_geo *geo = &dev->geo;
1148 	struct request_queue *bqueue = dev->q;
1149 	struct request_queue *tqueue = tdisk->queue;
1150 	struct pblk *pblk;
1151 	int ret;
1152 
1153 	pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
1154 	if (!pblk)
1155 		return ERR_PTR(-ENOMEM);
1156 
1157 	pblk->dev = dev;
1158 	pblk->disk = tdisk;
1159 	pblk->state = PBLK_STATE_RUNNING;
1160 	trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1161 	pblk->gc.gc_enabled = 0;
1162 
1163 	if (!(geo->version == NVM_OCSSD_SPEC_12 ||
1164 					geo->version == NVM_OCSSD_SPEC_20)) {
1165 		pblk_err(pblk, "OCSSD version not supported (%u)\n",
1166 							geo->version);
1167 		kfree(pblk);
1168 		return ERR_PTR(-EINVAL);
1169 	}
1170 
1171 	if (geo->ext) {
1172 		pblk_err(pblk, "extended metadata not supported\n");
1173 		kfree(pblk);
1174 		return ERR_PTR(-EINVAL);
1175 	}
1176 
1177 	spin_lock_init(&pblk->resubmit_lock);
1178 	spin_lock_init(&pblk->trans_lock);
1179 	spin_lock_init(&pblk->lock);
1180 
1181 #ifdef CONFIG_NVM_PBLK_DEBUG
1182 	atomic_long_set(&pblk->inflight_writes, 0);
1183 	atomic_long_set(&pblk->padded_writes, 0);
1184 	atomic_long_set(&pblk->padded_wb, 0);
1185 	atomic_long_set(&pblk->req_writes, 0);
1186 	atomic_long_set(&pblk->sub_writes, 0);
1187 	atomic_long_set(&pblk->sync_writes, 0);
1188 	atomic_long_set(&pblk->inflight_reads, 0);
1189 	atomic_long_set(&pblk->cache_reads, 0);
1190 	atomic_long_set(&pblk->sync_reads, 0);
1191 	atomic_long_set(&pblk->recov_writes, 0);
1192 	atomic_long_set(&pblk->recov_writes, 0);
1193 	atomic_long_set(&pblk->recov_gc_writes, 0);
1194 	atomic_long_set(&pblk->recov_gc_reads, 0);
1195 #endif
1196 
1197 	atomic_long_set(&pblk->read_failed, 0);
1198 	atomic_long_set(&pblk->read_empty, 0);
1199 	atomic_long_set(&pblk->read_high_ecc, 0);
1200 	atomic_long_set(&pblk->read_failed_gc, 0);
1201 	atomic_long_set(&pblk->write_failed, 0);
1202 	atomic_long_set(&pblk->erase_failed, 0);
1203 
1204 	ret = pblk_core_init(pblk);
1205 	if (ret) {
1206 		pblk_err(pblk, "could not initialize core\n");
1207 		goto fail;
1208 	}
1209 
1210 	ret = pblk_lines_init(pblk);
1211 	if (ret) {
1212 		pblk_err(pblk, "could not initialize lines\n");
1213 		goto fail_free_core;
1214 	}
1215 
1216 	ret = pblk_rwb_init(pblk);
1217 	if (ret) {
1218 		pblk_err(pblk, "could not initialize write buffer\n");
1219 		goto fail_free_lines;
1220 	}
1221 
1222 	ret = pblk_l2p_init(pblk, flags & NVM_TARGET_FACTORY);
1223 	if (ret) {
1224 		pblk_err(pblk, "could not initialize maps\n");
1225 		goto fail_free_rwb;
1226 	}
1227 
1228 	ret = pblk_writer_init(pblk);
1229 	if (ret) {
1230 		if (ret != -EINTR)
1231 			pblk_err(pblk, "could not initialize write thread\n");
1232 		goto fail_free_l2p;
1233 	}
1234 
1235 	ret = pblk_gc_init(pblk);
1236 	if (ret) {
1237 		pblk_err(pblk, "could not initialize gc\n");
1238 		goto fail_stop_writer;
1239 	}
1240 
1241 	/* inherit the size from the underlying device */
1242 	blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1243 	blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1244 
1245 	blk_queue_write_cache(tqueue, true, false);
1246 
1247 	tqueue->limits.discard_granularity = geo->clba * geo->csecs;
1248 	tqueue->limits.discard_alignment = 0;
1249 	blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
1250 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, tqueue);
1251 
1252 	pblk_info(pblk, "luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
1253 			geo->all_luns, pblk->l_mg.nr_lines,
1254 			(unsigned long long)pblk->capacity,
1255 			pblk->rwb.nr_entries);
1256 
1257 	wake_up_process(pblk->writer_ts);
1258 
1259 	/* Check if we need to start GC */
1260 	pblk_gc_should_kick(pblk);
1261 
1262 	return pblk;
1263 
1264 fail_stop_writer:
1265 	pblk_writer_stop(pblk);
1266 fail_free_l2p:
1267 	pblk_l2p_free(pblk);
1268 fail_free_rwb:
1269 	pblk_rwb_free(pblk);
1270 fail_free_lines:
1271 	pblk_lines_free(pblk);
1272 fail_free_core:
1273 	pblk_core_free(pblk);
1274 fail:
1275 	kfree(pblk);
1276 	return ERR_PTR(ret);
1277 }
1278 
1279 /* physical block device target */
1280 static struct nvm_tgt_type tt_pblk = {
1281 	.name		= "pblk",
1282 	.version	= {1, 0, 0},
1283 
1284 	.make_rq	= pblk_make_rq,
1285 	.capacity	= pblk_capacity,
1286 
1287 	.init		= pblk_init,
1288 	.exit		= pblk_exit,
1289 
1290 	.sysfs_init	= pblk_sysfs_init,
1291 	.sysfs_exit	= pblk_sysfs_exit,
1292 	.owner		= THIS_MODULE,
1293 };
1294 
pblk_module_init(void)1295 static int __init pblk_module_init(void)
1296 {
1297 	int ret;
1298 
1299 	ret = bioset_init(&pblk_bio_set, BIO_POOL_SIZE, 0, 0);
1300 	if (ret)
1301 		return ret;
1302 	ret = nvm_register_tgt_type(&tt_pblk);
1303 	if (ret)
1304 		bioset_exit(&pblk_bio_set);
1305 	return ret;
1306 }
1307 
pblk_module_exit(void)1308 static void pblk_module_exit(void)
1309 {
1310 	bioset_exit(&pblk_bio_set);
1311 	nvm_unregister_tgt_type(&tt_pblk);
1312 }
1313 
1314 module_init(pblk_module_init);
1315 module_exit(pblk_module_exit);
1316 MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
1317 MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
1318 MODULE_LICENSE("GPL v2");
1319 MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");
1320