xref: /spdk/lib/ftl/ftl_l2p_cache.c (revision 307b8c112ffd90a26d53dd15fad67bd9038ef526)
1 /*   SPDX-License-Identifier: BSD-3-Clause
2  *   Copyright (c) Intel Corporation.
3  *   All rights reserved.
4  */
5 
6 #include "spdk/stdinc.h"
7 #include "spdk/cpuset.h"
8 #include "spdk/queue.h"
9 #include "spdk/thread.h"
10 #include "spdk/event.h"
11 #include "spdk/ftl.h"
12 #include "spdk/conf.h"
13 #include "spdk/env.h"
14 #include "spdk/util.h"
15 
16 #include "ftl_core.h"
17 #include "ftl_l2p_cache.h"
18 #include "ftl_layout.h"
19 #include "ftl_nv_cache_io.h"
20 #include "mngt/ftl_mngt_steps.h"
21 #include "utils/ftl_defs.h"
22 #include "utils/ftl_addr_utils.h"
23 
24 struct ftl_l2p_cache_page_io_ctx {
25 	struct ftl_l2p_cache *cache;
26 	uint64_t updates;
27 	spdk_bdev_io_completion_cb cb;
28 	struct spdk_bdev_io_wait_entry bdev_io_wait;
29 };
30 
31 enum ftl_l2p_page_state {
32 	L2P_CACHE_PAGE_INIT,		/* Page in memory not initialized from disk page */
33 	L2P_CACHE_PAGE_READY,		/* Page initialized from disk */
34 	L2P_CACHE_PAGE_FLUSHING,	/* Page is being flushed to disk and removed from memory */
35 	L2P_CACHE_PAGE_PERSISTING,	/* Page is being flushed to disk and not removed from memory */
36 	L2P_CACHE_PAGE_CLEARING,	/* Page is being initialized with INVALID addresses */
37 	L2P_CACHE_PAGE_CORRUPTED	/* Page corrupted */
38 };
39 
40 struct ftl_l2p_page {
41 	uint64_t updates; /* Number of times an L2P entry was updated in the page since it was last persisted */
42 	TAILQ_HEAD(, ftl_l2p_page_wait_ctx) ppe_list; /* for deferred pins */
43 	TAILQ_ENTRY(ftl_l2p_page) list_entry;
44 	uint64_t page_no;
45 	enum ftl_l2p_page_state state;
46 	uint64_t pin_ref_cnt;
47 	struct ftl_l2p_cache_page_io_ctx ctx;
48 	bool on_lru_list;
49 	void *page_buffer;
50 	uint64_t ckpt_seq_id;
51 	ftl_df_obj_id obj_id;
52 };
53 
54 struct ftl_l2p_page_set;
55 
56 struct ftl_l2p_page_wait_ctx {
57 	uint16_t	pg_pin_issued;
58 	uint16_t	pg_pin_completed;
59 	struct ftl_l2p_page_set *parent;
60 	uint64_t	pg_no;
61 	TAILQ_ENTRY(ftl_l2p_page_wait_ctx) list_entry;
62 };
63 
64 /* A L2P page contains 1024 4B entries (or 512 8B ones for big drives).
65  * Currently internal IO will only pin 1 LBA at a time, so only one entry should be needed.
66  * User IO is split on internal xfer_size boundaries, which is currently set to 1MiB (256 blocks),
67  * so one entry should also be enough.
68  * TODO: We should probably revisit this though, when/if the xfer_size is based on io requirements of the
69  * bottom device (e.g. RAID5F), since then big IOs (especially unaligned ones) could potentially break this.
70  */
71 #define L2P_MAX_PAGES_TO_PIN 4
72 struct ftl_l2p_page_set {
73 	uint16_t to_pin_cnt;
74 	uint16_t pinned_cnt;
75 	uint16_t pin_fault_cnt;
76 	uint8_t locked;
77 	uint8_t deferred;
78 	struct ftl_l2p_pin_ctx *pin_ctx;
79 	TAILQ_ENTRY(ftl_l2p_page_set) list_entry;
80 	struct ftl_l2p_page_wait_ctx entry[L2P_MAX_PAGES_TO_PIN];
81 };
82 
83 struct ftl_l2p_l1_map_entry {
84 	ftl_df_obj_id page_obj_id;
85 };
86 
87 enum ftl_l2p_cache_state {
88 	L2P_CACHE_INIT,
89 	L2P_CACHE_RUNNING,
90 	L2P_CACHE_IN_SHUTDOWN,
91 	L2P_CACHE_SHUTDOWN_DONE,
92 };
93 
94 struct ftl_l2p_cache_process_ctx {
95 	int status;
96 	ftl_l2p_cb cb;
97 	void *cb_ctx;
98 	uint64_t idx;
99 	uint64_t qd;
100 };
101 
102 struct ftl_l2p_cache {
103 	struct spdk_ftl_dev *dev;
104 	struct ftl_l2p_l1_map_entry *l2_mapping;
105 	struct ftl_md *l2_md;
106 	struct ftl_md *l2_ctx_md;
107 	struct ftl_mempool *l2_ctx_pool;
108 	struct ftl_md *l1_md;
109 
110 	TAILQ_HEAD(l2p_lru_list, ftl_l2p_page) lru_list;
111 	/* TODO: A lot of / and % operations are done on this value, consider adding a shift based field and calculactions instead */
112 	uint64_t lbas_in_page;
113 	uint64_t num_pages;		/* num pages to hold the entire L2P */
114 
115 	uint64_t ios_in_flight;		/* Currently in flight IOs, to determine l2p shutdown readiness */
116 	enum ftl_l2p_cache_state state;
117 	uint32_t l2_pgs_avail;
118 	uint32_t l2_pgs_evicting;
119 	uint32_t l2_pgs_resident_max;
120 	uint32_t evict_keep;
121 	struct ftl_mempool *page_sets_pool;
122 	TAILQ_HEAD(, ftl_l2p_page_set) deferred_page_set_list; /* for deferred page sets */
123 
124 	/* Process unmap in backgorund */
125 	struct {
126 #define FTL_L2P_MAX_LAZY_UNMAP_QD 1
127 		/* Unmap queue depth */
128 		uint32_t qd;
129 		/* Currently processed page */
130 		uint64_t page_no;
131 		/* Context for page pinning */
132 		struct ftl_l2p_pin_ctx pin_ctx;
133 	} lazy_unmap;
134 
135 	/* This is a context for a management process */
136 	struct ftl_l2p_cache_process_ctx mctx;
137 
138 	/* MD layout cache: Offset on a device in FTL_BLOCK_SIZE unit */
139 	uint64_t cache_layout_offset;
140 
141 	/* MD layout cache: Device of region */
142 	struct spdk_bdev_desc *cache_layout_bdev_desc;
143 
144 	/* MD layout cache: IO channel of region */
145 	struct spdk_io_channel *cache_layout_ioch;
146 };
147 
148 typedef void (*ftl_l2p_cache_clear_cb)(struct ftl_l2p_cache *cache, int status, void *ctx_page);
149 typedef void (*ftl_l2p_cache_persist_cb)(struct ftl_l2p_cache *cache, int status, void *ctx_page);
150 typedef void (*ftl_l2p_cache_sync_cb)(struct spdk_ftl_dev *dev, int status, void *page,
151 				      void *user_ctx);
152 
153 static bool page_set_is_done(struct ftl_l2p_page_set *page_set);
154 static void page_set_end(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache,
155 			 struct ftl_l2p_page_set *page_set);
156 static void page_out_io_retry(void *arg);
157 static void page_in_io_retry(void *arg);
158 
159 static inline void
160 ftl_l2p_page_queue_wait_ctx(struct ftl_l2p_page *page,
161 			    struct ftl_l2p_page_wait_ctx *ppe)
162 {
163 	TAILQ_INSERT_TAIL(&page->ppe_list, ppe, list_entry);
164 }
165 
166 static inline uint64_t
167 ftl_l2p_cache_get_l1_page_size(void)
168 {
169 	return 1UL << 12;
170 }
171 
172 static inline uint64_t
173 ftl_l2p_cache_get_lbas_in_page(struct ftl_l2p_cache *cache)
174 {
175 	return cache->lbas_in_page;
176 }
177 
178 static inline size_t
179 ftl_l2p_cache_get_page_all_size(void)
180 {
181 	return sizeof(struct ftl_l2p_page) + ftl_l2p_cache_get_l1_page_size();
182 }
183 
184 static void
185 ftl_l2p_cache_lru_remove_page(struct ftl_l2p_cache *cache, struct ftl_l2p_page *page)
186 {
187 	assert(page);
188 	assert(page->on_lru_list);
189 
190 	TAILQ_REMOVE(&cache->lru_list, page, list_entry);
191 	page->on_lru_list = false;
192 }
193 
194 static void
195 ftl_l2p_cache_lru_add_page(struct ftl_l2p_cache *cache, struct ftl_l2p_page *page)
196 {
197 	assert(page);
198 	assert(!page->on_lru_list);
199 
200 	TAILQ_INSERT_HEAD(&cache->lru_list, page, list_entry);
201 
202 	page->on_lru_list = true;
203 }
204 
205 static void
206 ftl_l2p_cache_lru_promote_page(struct ftl_l2p_cache *cache, struct ftl_l2p_page *page)
207 {
208 	if (!page->on_lru_list) {
209 		return;
210 	}
211 
212 	ftl_l2p_cache_lru_remove_page(cache, page);
213 	ftl_l2p_cache_lru_add_page(cache, page);
214 }
215 
216 static inline void
217 ftl_l2p_cache_page_insert(struct ftl_l2p_cache *cache, struct ftl_l2p_page *page)
218 {
219 	struct ftl_l2p_l1_map_entry *me = cache->l2_mapping;
220 	assert(me);
221 
222 	assert(me[page->page_no].page_obj_id == FTL_DF_OBJ_ID_INVALID);
223 	me[page->page_no].page_obj_id = page->obj_id;
224 }
225 
226 static void
227 ftl_l2p_cache_page_remove(struct ftl_l2p_cache *cache, struct ftl_l2p_page *page)
228 {
229 	struct ftl_l2p_l1_map_entry *me = cache->l2_mapping;
230 	assert(me);
231 	assert(me[page->page_no].page_obj_id != FTL_DF_OBJ_ID_INVALID);
232 	assert(TAILQ_EMPTY(&page->ppe_list));
233 
234 	me[page->page_no].page_obj_id = FTL_DF_OBJ_ID_INVALID;
235 	cache->l2_pgs_avail++;
236 	ftl_mempool_put(cache->l2_ctx_pool, page);
237 }
238 
239 static inline struct ftl_l2p_page *
240 ftl_l2p_cache_get_coldest_page(struct ftl_l2p_cache *cache)
241 {
242 	return TAILQ_LAST(&cache->lru_list, l2p_lru_list);
243 }
244 
245 static inline struct ftl_l2p_page *
246 ftl_l2p_cache_get_hotter_page(struct ftl_l2p_page *page)
247 {
248 	return TAILQ_PREV(page, l2p_lru_list, list_entry);
249 }
250 
251 static inline uint64_t
252 ftl_l2p_cache_page_get_bdev_offset(struct ftl_l2p_cache *cache,
253 				   struct ftl_l2p_page *page)
254 {
255 	return cache->cache_layout_offset + page->page_no;
256 }
257 
258 static inline struct spdk_bdev_desc *
259 ftl_l2p_cache_get_bdev_desc(struct ftl_l2p_cache *cache)
260 {
261 	return cache->cache_layout_bdev_desc;
262 }
263 
264 static inline struct spdk_io_channel *
265 ftl_l2p_cache_get_bdev_iochannel(struct ftl_l2p_cache *cache)
266 {
267 	return cache->cache_layout_ioch;
268 }
269 
270 static struct ftl_l2p_page *
271 ftl_l2p_cache_page_alloc(struct ftl_l2p_cache *cache, size_t page_no)
272 {
273 	struct ftl_l2p_page *page = ftl_mempool_get(cache->l2_ctx_pool);
274 	ftl_bug(!page);
275 
276 	cache->l2_pgs_avail--;
277 
278 	memset(page, 0, sizeof(*page));
279 
280 	page->obj_id = ftl_mempool_get_df_obj_id(cache->l2_ctx_pool, page);
281 
282 	page->page_buffer = (char *)ftl_md_get_buffer(cache->l1_md) + ftl_mempool_get_df_obj_index(
283 				    cache->l2_ctx_pool, page) * FTL_BLOCK_SIZE;
284 
285 	TAILQ_INIT(&page->ppe_list);
286 
287 	page->page_no = page_no;
288 	page->state = L2P_CACHE_PAGE_INIT;
289 
290 	return page;
291 }
292 
293 static inline bool
294 ftl_l2p_cache_page_can_remove(struct ftl_l2p_page *page)
295 {
296 	return (!page->updates &&
297 		page->state != L2P_CACHE_PAGE_INIT &&
298 		!page->pin_ref_cnt);
299 }
300 
301 static inline ftl_addr
302 ftl_l2p_cache_get_addr(struct spdk_ftl_dev *dev,
303 		       struct ftl_l2p_cache *cache, struct ftl_l2p_page *page, uint64_t lba)
304 {
305 	return ftl_addr_load(dev, page->page_buffer, lba % cache->lbas_in_page);
306 }
307 
308 static inline void
309 ftl_l2p_cache_set_addr(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache,
310 		       struct ftl_l2p_page *page, uint64_t lba, ftl_addr addr)
311 {
312 	ftl_addr_store(dev, page->page_buffer, lba % cache->lbas_in_page, addr);
313 }
314 
315 static void
316 ftl_l2p_page_set_invalid(struct spdk_ftl_dev *dev, struct ftl_l2p_page *page)
317 {
318 	ftl_addr addr;
319 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
320 	uint64_t naddr;
321 
322 	page->updates++;
323 
324 	naddr = ftl_l2p_cache_get_lbas_in_page(cache);
325 	for (uint64_t i = 0; i < naddr; i++) {
326 		addr = ftl_addr_load(dev, page->page_buffer, i);
327 		if (addr == FTL_ADDR_INVALID) {
328 			continue;
329 		}
330 
331 		ftl_invalidate_addr(dev, addr);
332 		ftl_l2p_cache_set_addr(dev, cache, page, i, FTL_ADDR_INVALID);
333 	}
334 }
335 
336 static inline void
337 ftl_l2p_cache_page_pin(struct ftl_l2p_cache *cache, struct ftl_l2p_page *page)
338 {
339 	page->pin_ref_cnt++;
340 	/* Pinned pages can't be evicted (since L2P sets/gets will be executed on it), so remove them from LRU */
341 	if (page->on_lru_list) {
342 		ftl_l2p_cache_lru_remove_page(cache, page);
343 	}
344 }
345 
346 static inline void
347 ftl_l2p_cache_page_unpin(struct ftl_l2p_cache *cache, struct ftl_l2p_page *page)
348 {
349 	page->pin_ref_cnt--;
350 	if (!page->pin_ref_cnt && !page->on_lru_list && page->state != L2P_CACHE_PAGE_FLUSHING) {
351 		/* L2P_CACHE_PAGE_FLUSHING: the page is currently being evicted.
352 		 * In such a case, the page can't be returned to the rank list, because
353 		 * the ongoing eviction will remove it if no pg updates had happened.
354 		 * Moreover, the page could make it to the top of the rank list and be
355 		 * selected for another eviction, while the ongoing one did not finish yet.
356 		 *
357 		 * Depending on the page updates tracker, the page will be evicted
358 		 * or returned to the rank list in context of the eviction completion
359 		 * cb - see page_out_io_complete().
360 		 */
361 		ftl_l2p_cache_lru_add_page(cache, page);
362 	}
363 }
364 
365 static inline bool
366 ftl_l2p_cache_page_can_evict(struct ftl_l2p_page *page)
367 {
368 	return (page->state == L2P_CACHE_PAGE_FLUSHING ||
369 		page->state == L2P_CACHE_PAGE_PERSISTING ||
370 		page->state == L2P_CACHE_PAGE_INIT ||
371 		page->pin_ref_cnt) ? false : true;
372 }
373 
374 static bool
375 ftl_l2p_cache_evict_continue(struct ftl_l2p_cache *cache)
376 {
377 	return cache->l2_pgs_avail + cache->l2_pgs_evicting < cache->evict_keep;
378 }
379 
380 static void *
381 _ftl_l2p_cache_init(struct spdk_ftl_dev *dev, size_t addr_size, uint64_t l2p_size)
382 {
383 	struct ftl_l2p_cache *cache;
384 	uint64_t l2_pages = spdk_divide_round_up(l2p_size, ftl_l2p_cache_get_l1_page_size());
385 	size_t l2_size = l2_pages * sizeof(struct ftl_l2p_l1_map_entry);
386 
387 	cache = calloc(1, sizeof(struct ftl_l2p_cache));
388 	if (cache == NULL) {
389 		return NULL;
390 	}
391 	cache->dev = dev;
392 
393 	cache->l2_md = ftl_md_create(dev,
394 				     spdk_divide_round_up(l2_size, FTL_BLOCK_SIZE), 0,
395 				     FTL_L2P_CACHE_MD_NAME_L2,
396 				     ftl_md_create_shm_flags(dev), NULL);
397 
398 	if (cache->l2_md == NULL) {
399 		goto fail_l2_md;
400 	}
401 	cache->l2_mapping = ftl_md_get_buffer(cache->l2_md);
402 
403 	cache->lbas_in_page = dev->layout.l2p.lbas_in_page;
404 	cache->num_pages = l2_pages;
405 
406 	return cache;
407 fail_l2_md:
408 	free(cache);
409 	return NULL;
410 }
411 
412 static struct ftl_l2p_page *
413 get_l2p_page_by_df_id(struct ftl_l2p_cache *cache, size_t page_no)
414 {
415 	struct ftl_l2p_l1_map_entry *me = cache->l2_mapping;
416 	ftl_df_obj_id obj_id = me[page_no].page_obj_id;
417 
418 	if (obj_id != FTL_DF_OBJ_ID_INVALID) {
419 		return ftl_mempool_get_df_ptr(cache->l2_ctx_pool, obj_id);
420 	}
421 
422 	return NULL;
423 }
424 
425 int
426 ftl_l2p_cache_init(struct spdk_ftl_dev *dev)
427 {
428 	uint64_t l2p_size = dev->num_lbas * dev->layout.l2p.addr_size;
429 	struct ftl_l2p_cache *cache;
430 	const struct ftl_layout_region *reg;
431 	void *l2p = _ftl_l2p_cache_init(dev, dev->layout.l2p.addr_size, l2p_size);
432 	size_t page_sets_pool_size = 1 << 15;
433 	size_t max_resident_size, max_resident_pgs;
434 
435 	if (!l2p) {
436 		return -1;
437 	}
438 	dev->l2p = l2p;
439 
440 	cache = (struct ftl_l2p_cache *)dev->l2p;
441 	cache->page_sets_pool = ftl_mempool_create(page_sets_pool_size,
442 				sizeof(struct ftl_l2p_page_set),
443 				64, SPDK_ENV_SOCKET_ID_ANY);
444 	if (!cache->page_sets_pool) {
445 		return -1;
446 	}
447 
448 	max_resident_size = dev->conf.l2p_dram_limit << 20;
449 	max_resident_pgs = max_resident_size / ftl_l2p_cache_get_page_all_size();
450 
451 	if (max_resident_pgs > cache->num_pages) {
452 		SPDK_NOTICELOG("l2p memory limit higher than entire L2P size\n");
453 		max_resident_pgs = cache->num_pages;
454 	}
455 
456 	/* Round down max res pgs to the nearest # of l2/l1 pgs */
457 	max_resident_size = max_resident_pgs * ftl_l2p_cache_get_page_all_size();
458 	SPDK_NOTICELOG("l2p maximum resident size is: %"PRIu64" (of %"PRIu64") MiB\n",
459 		       max_resident_size >> 20, dev->conf.l2p_dram_limit);
460 
461 	TAILQ_INIT(&cache->deferred_page_set_list);
462 	TAILQ_INIT(&cache->lru_list);
463 
464 	cache->l2_ctx_md = ftl_md_create(dev,
465 					 spdk_divide_round_up(max_resident_pgs * SPDK_ALIGN_CEIL(sizeof(struct ftl_l2p_page), 64),
466 							 FTL_BLOCK_SIZE), 0, FTL_L2P_CACHE_MD_NAME_L2_CTX, ftl_md_create_shm_flags(dev), NULL);
467 
468 	if (cache->l2_ctx_md == NULL) {
469 		return -1;
470 	}
471 
472 	cache->l2_pgs_resident_max = max_resident_pgs;
473 	cache->l2_pgs_avail = max_resident_pgs;
474 	cache->l2_pgs_evicting = 0;
475 	cache->l2_ctx_pool = ftl_mempool_create_ext(ftl_md_get_buffer(cache->l2_ctx_md),
476 			     max_resident_pgs, sizeof(struct ftl_l2p_page), 64);
477 
478 	if (cache->l2_ctx_pool == NULL) {
479 		return -1;
480 	}
481 
482 #define FTL_L2P_CACHE_PAGE_AVAIL_MAX            16UL << 10
483 #define FTL_L2P_CACHE_PAGE_AVAIL_RATIO          5UL
484 	cache->evict_keep = spdk_divide_round_up(cache->num_pages * FTL_L2P_CACHE_PAGE_AVAIL_RATIO, 100);
485 	cache->evict_keep = spdk_min(FTL_L2P_CACHE_PAGE_AVAIL_MAX, cache->evict_keep);
486 
487 	if (!ftl_fast_startup(dev) && !ftl_fast_recovery(dev)) {
488 		memset(cache->l2_mapping, (int)FTL_DF_OBJ_ID_INVALID, ftl_md_get_buffer_size(cache->l2_md));
489 		ftl_mempool_initialize_ext(cache->l2_ctx_pool);
490 	}
491 
492 	cache->l1_md = ftl_md_create(dev,
493 				     max_resident_pgs, 0,
494 				     FTL_L2P_CACHE_MD_NAME_L1,
495 				     ftl_md_create_shm_flags(dev), NULL);
496 
497 	if (cache->l1_md == NULL) {
498 		return -1;
499 	}
500 
501 	/* Cache MD layout */
502 	reg = &dev->layout.region[FTL_LAYOUT_REGION_TYPE_L2P];
503 	cache->cache_layout_offset = reg->current.offset;
504 	cache->cache_layout_bdev_desc = reg->bdev_desc;
505 	cache->cache_layout_ioch = reg->ioch;
506 
507 	cache->state = L2P_CACHE_RUNNING;
508 	return 0;
509 }
510 
511 static void
512 ftl_l2p_cache_deinit_l2(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache)
513 {
514 	ftl_md_destroy(cache->l2_ctx_md, ftl_md_destroy_shm_flags(dev));
515 	cache->l2_ctx_md = NULL;
516 
517 	ftl_mempool_destroy_ext(cache->l2_ctx_pool);
518 	cache->l2_ctx_pool = NULL;
519 
520 	ftl_md_destroy(cache->l1_md, ftl_md_destroy_shm_flags(dev));
521 	cache->l1_md = NULL;
522 
523 	ftl_mempool_destroy(cache->page_sets_pool);
524 	cache->page_sets_pool = NULL;
525 }
526 
527 static void
528 _ftl_l2p_cache_deinit(struct spdk_ftl_dev *dev)
529 {
530 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
531 
532 	ftl_l2p_cache_deinit_l2(dev, cache);
533 	ftl_md_destroy(cache->l2_md, ftl_md_destroy_shm_flags(dev));
534 	free(cache);
535 }
536 
537 void
538 ftl_l2p_cache_deinit(struct spdk_ftl_dev *dev)
539 {
540 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
541 
542 	if (!cache) {
543 		return;
544 	}
545 	assert(cache->state == L2P_CACHE_SHUTDOWN_DONE || cache->state == L2P_CACHE_INIT);
546 
547 	_ftl_l2p_cache_deinit(dev);
548 	dev->l2p = 0;
549 }
550 
551 static void
552 process_init_ctx(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache,
553 		 ftl_l2p_cb cb, void *cb_ctx)
554 {
555 	struct ftl_l2p_cache_process_ctx *ctx = &cache->mctx;
556 
557 	assert(NULL == ctx->cb_ctx);
558 	assert(0 == cache->l2_pgs_evicting);
559 
560 	memset(ctx, 0, sizeof(*ctx));
561 
562 	ctx->cb = cb;
563 	ctx->cb_ctx = cb_ctx;
564 }
565 
566 static void
567 process_finish(struct ftl_l2p_cache *cache)
568 {
569 	struct ftl_l2p_cache_process_ctx ctx = cache->mctx;
570 
571 	assert(cache->l2_pgs_avail == cache->l2_pgs_resident_max);
572 	assert(0 == ctx.qd);
573 
574 	memset(&cache->mctx, 0, sizeof(cache->mctx));
575 	ctx.cb(cache->dev, ctx.status, ctx.cb_ctx);
576 }
577 
578 static void process_page_out_retry(void *_page);
579 static void process_persist(struct ftl_l2p_cache *cache);
580 
581 static void
582 process_page_in(struct ftl_l2p_page *page, spdk_bdev_io_completion_cb cb)
583 {
584 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)page->ctx.cache;
585 	int rc;
586 
587 	assert(page->page_buffer);
588 
589 	rc = ftl_nv_cache_bdev_read_blocks_with_md(cache->dev, ftl_l2p_cache_get_bdev_desc(cache),
590 			ftl_l2p_cache_get_bdev_iochannel(cache),
591 			page->page_buffer, NULL, ftl_l2p_cache_page_get_bdev_offset(cache, page),
592 			1, cb, page);
593 
594 	if (rc) {
595 		cb(NULL, false, page);
596 	}
597 }
598 
599 static void
600 process_persist_page_out_cb(struct spdk_bdev_io *bdev_io, bool success, void *arg)
601 {
602 	struct ftl_l2p_page *page = arg;
603 	struct ftl_l2p_cache *cache = page->ctx.cache;
604 	struct spdk_ftl_dev *dev = cache->dev;
605 	struct ftl_l2p_cache_process_ctx *ctx = &cache->mctx;
606 
607 	assert(bdev_io);
608 	ftl_stats_bdev_io_completed(dev, FTL_STATS_TYPE_L2P, bdev_io);
609 	spdk_bdev_free_io(bdev_io);
610 
611 	if (!success) {
612 		ctx->status = -EIO;
613 	}
614 
615 	if (ftl_bitmap_get(dev->unmap_map, ctx->idx)) {
616 		/*
617 		 * Page had been unmapped, in persist path before IO, it was invalidated entirely
618 		 * now clear unmap flag
619 		 */
620 		ftl_bitmap_clear(dev->unmap_map, page->page_no);
621 	}
622 	ftl_l2p_cache_page_remove(cache, page);
623 
624 	ctx->qd--;
625 	process_persist(cache);
626 }
627 
628 static void
629 process_page_out(struct ftl_l2p_page *page, spdk_bdev_io_completion_cb cb)
630 {
631 	struct spdk_bdev *bdev;
632 	struct spdk_bdev_io_wait_entry *bdev_io_wait;
633 	struct ftl_l2p_cache *cache = page->ctx.cache;
634 	struct spdk_ftl_dev *dev = cache->dev;
635 	int rc;
636 
637 	assert(page->page_buffer);
638 
639 	rc = ftl_nv_cache_bdev_write_blocks_with_md(dev, ftl_l2p_cache_get_bdev_desc(cache),
640 			ftl_l2p_cache_get_bdev_iochannel(cache),
641 			page->page_buffer, NULL, ftl_l2p_cache_page_get_bdev_offset(cache, page),
642 			1, cb, page);
643 
644 	if (spdk_likely(0 == rc)) {
645 		return;
646 	}
647 
648 	if (rc == -ENOMEM) {
649 		bdev = spdk_bdev_desc_get_bdev(ftl_l2p_cache_get_bdev_desc(cache));
650 		bdev_io_wait = &page->ctx.bdev_io_wait;
651 		bdev_io_wait->bdev = bdev;
652 		bdev_io_wait->cb_fn = process_page_out_retry;
653 		bdev_io_wait->cb_arg = page;
654 		page->ctx.cb = cb;
655 
656 		rc = spdk_bdev_queue_io_wait(bdev, ftl_l2p_cache_get_bdev_iochannel(cache), bdev_io_wait);
657 		ftl_bug(rc);
658 	} else {
659 		ftl_abort();
660 	}
661 }
662 
663 static void
664 process_page_out_retry(void *_page)
665 {
666 	struct ftl_l2p_page *page = _page;
667 
668 	process_page_out(page, page->ctx.cb);
669 }
670 
671 static void process_unmap(struct ftl_l2p_cache *cache);
672 
673 static void
674 process_unmap_page_out_cb(struct spdk_bdev_io *bdev_io, bool success, void *ctx_page)
675 {
676 	struct ftl_l2p_page *page = (struct ftl_l2p_page *)ctx_page;
677 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)page->ctx.cache;
678 	struct spdk_ftl_dev *dev = cache->dev;
679 	struct ftl_l2p_cache_process_ctx *ctx = &cache->mctx;
680 
681 	assert(bdev_io);
682 	ftl_stats_bdev_io_completed(dev, FTL_STATS_TYPE_L2P, bdev_io);
683 	spdk_bdev_free_io(bdev_io);
684 
685 	if (!success) {
686 		ctx->status = -EIO;
687 	}
688 
689 	assert(!page->on_lru_list);
690 	assert(ftl_bitmap_get(dev->unmap_map, page->page_no));
691 	ftl_bitmap_clear(dev->unmap_map, page->page_no);
692 	ftl_l2p_cache_page_remove(cache, page);
693 
694 	ctx->qd--;
695 	process_unmap(cache);
696 }
697 
698 static void
699 process_unmap_page_in_cb(struct spdk_bdev_io *bdev_io, bool success, void *ctx_page)
700 {
701 	struct ftl_l2p_page *page = (struct ftl_l2p_page *)ctx_page;
702 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)page->ctx.cache;
703 	struct spdk_ftl_dev *dev = cache->dev;
704 	struct ftl_l2p_cache_process_ctx *ctx = &cache->mctx;
705 
706 	if (bdev_io) {
707 		ftl_stats_bdev_io_completed(dev, FTL_STATS_TYPE_L2P, bdev_io);
708 		spdk_bdev_free_io(bdev_io);
709 	}
710 	if (success) {
711 		assert(ftl_bitmap_get(dev->unmap_map, page->page_no));
712 		ftl_l2p_page_set_invalid(dev, page);
713 		process_page_out(page, process_unmap_page_out_cb);
714 	} else {
715 		ctx->status = -EIO;
716 		ctx->qd--;
717 		process_unmap(cache);
718 	}
719 }
720 
721 static void
722 process_unmap(struct ftl_l2p_cache *cache)
723 {
724 	struct ftl_l2p_cache_process_ctx *ctx = &cache->mctx;
725 
726 	while (ctx->idx < cache->num_pages && ctx->qd < 64) {
727 		struct ftl_l2p_page *page;
728 
729 		if (!ftl_bitmap_get(cache->dev->unmap_map, ctx->idx)) {
730 			/* Page had not been unmapped, continue */
731 			ctx->idx++;
732 			continue;
733 		}
734 
735 		/* All pages were removed in persist phase */
736 		assert(get_l2p_page_by_df_id(cache, ctx->idx) == NULL);
737 
738 		/* Allocate page to invalidate it */
739 		page = ftl_l2p_cache_page_alloc(cache, ctx->idx);
740 		if (!page) {
741 			/* All pages utilized so far, continue when they will be back available */
742 			assert(ctx->qd);
743 			break;
744 		}
745 
746 		page->state = L2P_CACHE_PAGE_CLEARING;
747 		page->ctx.cache = cache;
748 
749 		ftl_l2p_cache_page_insert(cache, page);
750 		process_page_in(page, process_unmap_page_in_cb);
751 
752 		ctx->qd++;
753 		ctx->idx++;
754 	}
755 
756 	if (0 == ctx->qd) {
757 		process_finish(cache);
758 	}
759 }
760 
761 void
762 ftl_l2p_cache_unmap(struct spdk_ftl_dev *dev, ftl_l2p_cb cb, void *cb_ctx)
763 {
764 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
765 
766 	process_init_ctx(dev, cache, cb, cb_ctx);
767 	process_unmap(cache);
768 }
769 
770 static void
771 clear_cb(struct spdk_ftl_dev *dev, struct ftl_md *md, int status)
772 {
773 	ftl_l2p_cb cb = md->owner.private;
774 	void *cb_cntx = md->owner.cb_ctx;
775 
776 	cb(dev, status, cb_cntx);
777 }
778 
779 void
780 ftl_l2p_cache_clear(struct spdk_ftl_dev *dev, ftl_l2p_cb cb, void *cb_ctx)
781 {
782 	struct ftl_md *md = dev->layout.md[FTL_LAYOUT_REGION_TYPE_L2P];
783 	ftl_addr invalid_addr = FTL_ADDR_INVALID;
784 
785 	md->cb =  clear_cb;
786 	md->owner.cb_ctx = cb_ctx;
787 	md->owner.private = cb;
788 
789 	ftl_md_clear(md, invalid_addr, NULL);
790 }
791 
792 static void
793 l2p_shm_restore_clean(struct spdk_ftl_dev *dev)
794 {
795 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
796 	struct ftl_l2p_l1_map_entry *me = cache->l2_mapping;
797 	struct ftl_l2p_page *page;
798 	ftl_df_obj_id obj_id;
799 	uint64_t page_no;
800 
801 	for (page_no = 0; page_no < cache->num_pages; ++page_no) {
802 		obj_id = me[page_no].page_obj_id;
803 		if (obj_id == FTL_DF_OBJ_ID_INVALID) {
804 			continue;
805 		}
806 
807 		page = ftl_mempool_claim_df(cache->l2_ctx_pool, obj_id);
808 		assert(page);
809 		assert(page->obj_id == ftl_mempool_get_df_obj_id(cache->l2_ctx_pool, page));
810 		assert(page->page_no == page_no);
811 		assert(page->state != L2P_CACHE_PAGE_INIT);
812 		assert(page->state != L2P_CACHE_PAGE_CLEARING);
813 		assert(cache->l2_pgs_avail > 0);
814 		cache->l2_pgs_avail--;
815 
816 		page->page_buffer = (char *)ftl_md_get_buffer(cache->l1_md) + ftl_mempool_get_df_obj_index(
817 					    cache->l2_ctx_pool, page) * FTL_BLOCK_SIZE;
818 
819 		TAILQ_INIT(&page->ppe_list);
820 
821 		page->pin_ref_cnt = 0;
822 		page->on_lru_list = 0;
823 		memset(&page->ctx, 0, sizeof(page->ctx));
824 
825 		ftl_l2p_cache_lru_add_page(cache, page);
826 	}
827 
828 	ftl_mempool_initialize_ext(cache->l2_ctx_pool);
829 }
830 
831 static void
832 l2p_shm_restore_dirty(struct spdk_ftl_dev *dev)
833 {
834 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
835 	struct ftl_l2p_l1_map_entry *me = cache->l2_mapping;
836 	struct ftl_l2p_page *page;
837 	ftl_df_obj_id obj_id;
838 	uint64_t page_no;
839 
840 	for (page_no = 0; page_no < cache->num_pages; ++page_no) {
841 		obj_id = me[page_no].page_obj_id;
842 		if (obj_id == FTL_DF_OBJ_ID_INVALID) {
843 			continue;
844 		}
845 
846 		page = ftl_mempool_claim_df(cache->l2_ctx_pool, obj_id);
847 		assert(page);
848 		assert(page->obj_id == ftl_mempool_get_df_obj_id(cache->l2_ctx_pool, page));
849 		assert(page->page_no == page_no);
850 		assert(page->state != L2P_CACHE_PAGE_CLEARING);
851 		assert(cache->l2_pgs_avail > 0);
852 		cache->l2_pgs_avail--;
853 
854 		if (page->state == L2P_CACHE_PAGE_INIT) {
855 			me[page_no].page_obj_id = FTL_DF_OBJ_ID_INVALID;
856 			cache->l2_pgs_avail++;
857 			ftl_mempool_release_df(cache->l2_ctx_pool, obj_id);
858 			continue;
859 		}
860 
861 		page->state = L2P_CACHE_PAGE_READY;
862 		/* Assume page is dirty after crash */
863 		page->updates = 1;
864 		page->page_buffer = (char *)ftl_md_get_buffer(cache->l1_md) + ftl_mempool_get_df_obj_index(
865 					    cache->l2_ctx_pool, page) * FTL_BLOCK_SIZE;
866 
867 		TAILQ_INIT(&page->ppe_list);
868 
869 		page->pin_ref_cnt = 0;
870 		page->on_lru_list = 0;
871 		memset(&page->ctx, 0, sizeof(page->ctx));
872 
873 		ftl_l2p_cache_lru_add_page(cache, page);
874 	}
875 
876 	ftl_mempool_initialize_ext(cache->l2_ctx_pool);
877 }
878 
879 void
880 ftl_l2p_cache_restore(struct spdk_ftl_dev *dev, ftl_l2p_cb cb, void *cb_ctx)
881 {
882 	if (ftl_fast_startup(dev)) {
883 		l2p_shm_restore_clean(dev);
884 	}
885 
886 	if (ftl_fast_recovery(dev)) {
887 		l2p_shm_restore_dirty(dev);
888 	}
889 
890 	cb(dev, 0, cb_ctx);
891 }
892 
893 static void
894 process_persist(struct ftl_l2p_cache *cache)
895 {
896 	struct ftl_l2p_cache_process_ctx *ctx = &cache->mctx;
897 	struct spdk_ftl_dev *dev = cache->dev;
898 
899 	while (ctx->idx < cache->num_pages && ctx->qd < 64) {
900 		struct ftl_l2p_page *page = get_l2p_page_by_df_id(cache, ctx->idx);
901 		ctx->idx++;
902 
903 		if (!page) {
904 			continue;
905 		}
906 
907 		/* Finished unmap if the page was marked */
908 		if (ftl_bitmap_get(dev->unmap_map, ctx->idx)) {
909 			ftl_l2p_page_set_invalid(dev, page);
910 		}
911 
912 		if (page->on_lru_list) {
913 			ftl_l2p_cache_lru_remove_page(cache, page);
914 		}
915 
916 		if (page->updates) {
917 			/* Need to persist the page */
918 			page->state = L2P_CACHE_PAGE_PERSISTING;
919 			page->ctx.cache = cache;
920 			ctx->qd++;
921 			process_page_out(page, process_persist_page_out_cb);
922 		} else {
923 			ftl_l2p_cache_page_remove(cache, page);
924 		}
925 	}
926 
927 	if (0 == ctx->qd) {
928 		process_finish(cache);
929 	}
930 }
931 
932 void
933 ftl_l2p_cache_persist(struct spdk_ftl_dev *dev, ftl_l2p_cb cb, void *cb_ctx)
934 {
935 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
936 
937 	process_init_ctx(dev, cache, cb, cb_ctx);
938 	process_persist(cache);
939 }
940 
941 bool
942 ftl_l2p_cache_is_halted(struct spdk_ftl_dev *dev)
943 {
944 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
945 
946 	return cache->state == L2P_CACHE_SHUTDOWN_DONE;
947 }
948 
949 void
950 ftl_l2p_cache_halt(struct spdk_ftl_dev *dev)
951 {
952 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
953 
954 	if (cache->state != L2P_CACHE_SHUTDOWN_DONE) {
955 		cache->state = L2P_CACHE_IN_SHUTDOWN;
956 		if (!cache->ios_in_flight && !cache->l2_pgs_evicting) {
957 			cache->state = L2P_CACHE_SHUTDOWN_DONE;
958 		}
959 	}
960 }
961 
962 static inline struct ftl_l2p_page *
963 get_page(struct ftl_l2p_cache *cache, uint64_t lba)
964 {
965 	return get_l2p_page_by_df_id(cache, lba / cache->lbas_in_page);
966 }
967 
968 static inline void
969 ftl_l2p_cache_init_page_set(struct ftl_l2p_page_set *page_set, struct ftl_l2p_pin_ctx *pin_ctx)
970 {
971 	page_set->to_pin_cnt = 0;
972 	page_set->pinned_cnt = 0;
973 	page_set->pin_fault_cnt = 0;
974 	page_set->locked = 0;
975 	page_set->deferred = 0;
976 	page_set->pin_ctx = pin_ctx;
977 }
978 
979 static inline bool
980 ftl_l2p_cache_running(struct ftl_l2p_cache *cache)
981 {
982 	return cache->state == L2P_CACHE_RUNNING;
983 }
984 
985 static inline bool
986 ftl_l2p_cache_page_is_pinnable(struct ftl_l2p_page *page)
987 {
988 	return page->state != L2P_CACHE_PAGE_INIT;
989 }
990 
991 void
992 ftl_l2p_cache_pin(struct spdk_ftl_dev *dev, struct ftl_l2p_pin_ctx *pin_ctx)
993 {
994 	assert(dev->num_lbas >= pin_ctx->lba + pin_ctx->count);
995 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
996 	struct ftl_l2p_page_set *page_set;
997 	bool defer_pin = false;
998 
999 	/* Calculate first and last page to pin, count of them */
1000 	uint64_t start = pin_ctx->lba / cache->lbas_in_page;
1001 	uint64_t end = (pin_ctx->lba + pin_ctx->count - 1) / cache->lbas_in_page;
1002 	uint64_t count = end - start + 1;
1003 	uint64_t i;
1004 
1005 	if (spdk_unlikely(count > L2P_MAX_PAGES_TO_PIN)) {
1006 		ftl_l2p_pin_complete(dev, -E2BIG, pin_ctx);
1007 		return;
1008 	}
1009 
1010 	/* Get and initialize page sets */
1011 	assert(ftl_l2p_cache_running(cache));
1012 	page_set = ftl_mempool_get(cache->page_sets_pool);
1013 	if (!page_set) {
1014 		ftl_l2p_pin_complete(dev, -EAGAIN, pin_ctx);
1015 		return;
1016 	}
1017 	ftl_l2p_cache_init_page_set(page_set, pin_ctx);
1018 
1019 	struct ftl_l2p_page_wait_ctx *entry = page_set->entry;
1020 	for (i = start; i <= end; i++, entry++) {
1021 		struct ftl_l2p_page *page;
1022 		entry->parent = page_set;
1023 		entry->pg_no = i;
1024 		entry->pg_pin_completed = false;
1025 		entry->pg_pin_issued = false;
1026 
1027 		page_set->to_pin_cnt++;
1028 
1029 		/* Try get page and pin */
1030 		page = get_l2p_page_by_df_id(cache, i);
1031 		if (page) {
1032 			if (ftl_l2p_cache_page_is_pinnable(page)) {
1033 				/* Page available and we can pin it */
1034 				page_set->pinned_cnt++;
1035 				entry->pg_pin_issued = true;
1036 				entry->pg_pin_completed = true;
1037 				ftl_l2p_cache_page_pin(cache, page);
1038 			} else {
1039 				/* The page is being loaded */
1040 				/* Queue the page pin entry to be executed on page in */
1041 				ftl_l2p_page_queue_wait_ctx(page, entry);
1042 				entry->pg_pin_issued = true;
1043 			}
1044 		} else {
1045 			/* The page is not in the cache, queue the page_set to page in */
1046 			defer_pin = true;
1047 		}
1048 	}
1049 
1050 	/* Check if page set is done */
1051 	if (page_set_is_done(page_set)) {
1052 		page_set_end(dev, cache, page_set);
1053 	} else if (defer_pin) {
1054 		TAILQ_INSERT_TAIL(&cache->deferred_page_set_list, page_set, list_entry);
1055 		page_set->deferred = 1;
1056 	}
1057 }
1058 
1059 void
1060 ftl_l2p_cache_unpin(struct spdk_ftl_dev *dev, uint64_t lba, uint64_t count)
1061 {
1062 	assert(dev->num_lbas >= lba + count);
1063 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
1064 	struct ftl_l2p_page *page;
1065 	uint64_t start = lba / cache->lbas_in_page;
1066 	uint64_t end = (lba + count - 1) / cache->lbas_in_page;
1067 	uint64_t i;
1068 
1069 	assert(count);
1070 	assert(start < cache->num_pages);
1071 	assert(end < cache->num_pages);
1072 
1073 	for (i = start; i <= end; i++) {
1074 		page = get_l2p_page_by_df_id(cache, i);
1075 		ftl_bug(!page);
1076 		ftl_l2p_cache_page_unpin(cache, page);
1077 	}
1078 }
1079 
1080 ftl_addr
1081 ftl_l2p_cache_get(struct spdk_ftl_dev *dev, uint64_t lba)
1082 {
1083 	assert(dev->num_lbas > lba);
1084 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
1085 	struct ftl_l2p_page *page = get_page(cache, lba);
1086 	ftl_addr addr;
1087 
1088 	ftl_bug(!page);
1089 	assert(ftl_l2p_cache_running(cache));
1090 	assert(page->pin_ref_cnt);
1091 
1092 	if (ftl_bitmap_get(dev->unmap_map, page->page_no)) {
1093 		ftl_l2p_page_set_invalid(dev, page);
1094 		ftl_bitmap_clear(dev->unmap_map, page->page_no);
1095 	}
1096 
1097 	ftl_l2p_cache_lru_promote_page(cache, page);
1098 	addr = ftl_l2p_cache_get_addr(dev, cache, page, lba);
1099 
1100 	return addr;
1101 }
1102 
1103 void
1104 ftl_l2p_cache_set(struct spdk_ftl_dev *dev, uint64_t lba, ftl_addr addr)
1105 {
1106 	assert(dev->num_lbas > lba);
1107 	struct ftl_l2p_cache *cache = (struct ftl_l2p_cache *)dev->l2p;
1108 	struct ftl_l2p_page *page = get_page(cache, lba);
1109 
1110 	ftl_bug(!page);
1111 	assert(ftl_l2p_cache_running(cache));
1112 	assert(page->pin_ref_cnt);
1113 
1114 	if (ftl_bitmap_get(dev->unmap_map, page->page_no)) {
1115 		ftl_l2p_page_set_invalid(dev, page);
1116 		ftl_bitmap_clear(dev->unmap_map, page->page_no);
1117 	}
1118 
1119 	page->updates++;
1120 	ftl_l2p_cache_lru_promote_page(cache, page);
1121 	ftl_l2p_cache_set_addr(dev, cache, page, lba, addr);
1122 }
1123 
1124 static struct ftl_l2p_page *
1125 page_allocate(struct ftl_l2p_cache *cache, uint64_t page_no)
1126 {
1127 	struct ftl_l2p_page *page = ftl_l2p_cache_page_alloc(cache, page_no);
1128 	ftl_l2p_cache_page_insert(cache, page);
1129 
1130 	return page;
1131 }
1132 
1133 static bool
1134 page_set_is_done(struct ftl_l2p_page_set *page_set)
1135 {
1136 	if (page_set->locked) {
1137 		return false;
1138 	}
1139 
1140 	assert(page_set->pinned_cnt + page_set->pin_fault_cnt <= page_set->to_pin_cnt);
1141 	return (page_set->pinned_cnt + page_set->pin_fault_cnt == page_set->to_pin_cnt);
1142 }
1143 
1144 static void
1145 page_set_unpin(struct ftl_l2p_cache *cache, struct ftl_l2p_page_set *page_set)
1146 {
1147 	uint64_t i;
1148 	struct ftl_l2p_page_wait_ctx *pentry = page_set->entry;
1149 
1150 	for (i = 0; i < page_set->to_pin_cnt; i++, pentry++) {
1151 		struct ftl_l2p_page *pinned_page;
1152 
1153 		if (false == pentry->pg_pin_completed) {
1154 			continue;
1155 		}
1156 
1157 		pinned_page = get_l2p_page_by_df_id(cache, pentry->pg_no);
1158 		ftl_bug(!pinned_page);
1159 
1160 		ftl_l2p_cache_page_unpin(cache, pinned_page);
1161 	}
1162 }
1163 
1164 static void
1165 page_set_end(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache,
1166 	     struct ftl_l2p_page_set *page_set)
1167 {
1168 	if (spdk_likely(0 == page_set->pin_fault_cnt)) {
1169 		ftl_l2p_pin_complete(dev, 0, page_set->pin_ctx);
1170 	} else {
1171 		page_set_unpin(cache, page_set);
1172 		ftl_l2p_pin_complete(dev, -EIO, page_set->pin_ctx);
1173 	}
1174 
1175 	if (page_set->deferred) {
1176 		TAILQ_REMOVE(&cache->deferred_page_set_list, page_set, list_entry);
1177 	}
1178 
1179 	assert(0 == page_set->locked);
1180 	ftl_mempool_put(cache->page_sets_pool, page_set);
1181 }
1182 
1183 static void
1184 page_in_io_complete(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache,
1185 		    struct ftl_l2p_page *page, bool success)
1186 {
1187 	struct ftl_l2p_page_set *page_set;
1188 	struct ftl_l2p_page_wait_ctx *pentry;
1189 
1190 	cache->ios_in_flight--;
1191 
1192 	assert(0 == page->pin_ref_cnt);
1193 	assert(L2P_CACHE_PAGE_INIT == page->state);
1194 	assert(false == page->on_lru_list);
1195 
1196 	if (spdk_likely(success)) {
1197 		page->state = L2P_CACHE_PAGE_READY;
1198 	}
1199 
1200 	while ((pentry = TAILQ_FIRST(&page->ppe_list))) {
1201 		TAILQ_REMOVE(&page->ppe_list, pentry, list_entry);
1202 
1203 		page_set = pentry->parent;
1204 
1205 		assert(false == pentry->pg_pin_completed);
1206 
1207 		if (success) {
1208 			ftl_l2p_cache_page_pin(cache, page);
1209 			page_set->pinned_cnt++;
1210 			pentry->pg_pin_completed = true;
1211 		} else {
1212 			page_set->pin_fault_cnt++;
1213 		}
1214 
1215 		/* Check if page_set is done */
1216 		if (page_set_is_done(page_set)) {
1217 			page_set_end(dev, cache, page_set);
1218 		}
1219 	}
1220 
1221 	if (spdk_unlikely(!success)) {
1222 		ftl_bug(page->on_lru_list);
1223 		ftl_l2p_cache_page_remove(cache, page);
1224 	}
1225 }
1226 
1227 static void
1228 page_in_io_cb(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
1229 {
1230 	struct ftl_l2p_page *page = cb_arg;
1231 	struct ftl_l2p_cache *cache = page->ctx.cache;
1232 	struct spdk_ftl_dev *dev = cache->dev;
1233 
1234 	ftl_stats_bdev_io_completed(dev, FTL_STATS_TYPE_L2P, bdev_io);
1235 	spdk_bdev_free_io(bdev_io);
1236 	page_in_io_complete(dev, cache, page, success);
1237 }
1238 
1239 static void
1240 page_in_io(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache, struct ftl_l2p_page *page)
1241 {
1242 	struct spdk_io_channel *ioch;
1243 	struct spdk_bdev *bdev;
1244 	struct spdk_bdev_io_wait_entry *bdev_io_wait;
1245 	int rc;
1246 	page->ctx.cache = cache;
1247 
1248 	rc = ftl_nv_cache_bdev_read_blocks_with_md(cache->dev, ftl_l2p_cache_get_bdev_desc(cache),
1249 			ftl_l2p_cache_get_bdev_iochannel(cache),
1250 			page->page_buffer, NULL, ftl_l2p_cache_page_get_bdev_offset(cache, page),
1251 			1, page_in_io_cb, page);
1252 	cache->ios_in_flight++;
1253 	if (spdk_likely(0 == rc)) {
1254 		return;
1255 	}
1256 
1257 	if (rc == -ENOMEM) {
1258 		ioch = ftl_l2p_cache_get_bdev_iochannel(cache);
1259 		bdev = spdk_bdev_desc_get_bdev(ftl_l2p_cache_get_bdev_desc(cache));
1260 		bdev_io_wait = &page->ctx.bdev_io_wait;
1261 		bdev_io_wait->bdev = bdev;
1262 		bdev_io_wait->cb_fn = page_in_io_retry;
1263 		bdev_io_wait->cb_arg = page;
1264 
1265 		rc = spdk_bdev_queue_io_wait(bdev, ioch, bdev_io_wait);
1266 		ftl_bug(rc);
1267 	} else {
1268 		ftl_abort();
1269 	}
1270 }
1271 
1272 static void
1273 page_in_io_retry(void *arg)
1274 {
1275 	struct ftl_l2p_page *page = arg;
1276 	struct ftl_l2p_cache *cache = page->ctx.cache;
1277 	struct spdk_ftl_dev *dev = cache->dev;
1278 
1279 	cache->ios_in_flight--;
1280 	page_in_io(dev, cache, page);
1281 }
1282 
1283 static void
1284 page_in(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache,
1285 	struct ftl_l2p_page_set *page_set, struct ftl_l2p_page_wait_ctx *pentry)
1286 {
1287 	struct ftl_l2p_page *page;
1288 	bool page_in = false;
1289 
1290 	/* Get page */
1291 	page = get_l2p_page_by_df_id(cache, pentry->pg_no);
1292 	if (!page) {
1293 		/* Page not allocated yet, do it */
1294 		page = page_allocate(cache, pentry->pg_no);
1295 		page_in = true;
1296 	}
1297 
1298 	if (ftl_l2p_cache_page_is_pinnable(page)) {
1299 		ftl_l2p_cache_page_pin(cache, page);
1300 		page_set->pinned_cnt++;
1301 		pentry->pg_pin_issued = true;
1302 		pentry->pg_pin_completed = true;
1303 	} else {
1304 		pentry->pg_pin_issued = true;
1305 		ftl_l2p_page_queue_wait_ctx(page, pentry);
1306 	}
1307 
1308 	if (page_in) {
1309 		page_in_io(dev, cache, page);
1310 	}
1311 }
1312 
1313 static int
1314 ftl_l2p_cache_process_page_sets(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache)
1315 {
1316 	struct ftl_l2p_page_set *page_set;
1317 	struct ftl_l2p_page_wait_ctx *pentry;
1318 	uint64_t i;
1319 
1320 	page_set = TAILQ_FIRST(&cache->deferred_page_set_list);
1321 	if (!page_set) {
1322 		/* No page_set */
1323 		return -ECHILD;
1324 	}
1325 
1326 	if (page_set->to_pin_cnt > cache->l2_pgs_avail) {
1327 		/* No enough page to pin, wait */
1328 		return -EBUSY;
1329 	}
1330 	if (cache->ios_in_flight > 512) {
1331 		/* Too big QD */
1332 		return -EBUSY;
1333 	}
1334 
1335 	TAILQ_REMOVE(&cache->deferred_page_set_list, page_set, list_entry);
1336 	page_set->deferred = 0;
1337 	page_set->locked = 1;
1338 
1339 	/* Now we can start pinning */
1340 	pentry = page_set->entry;
1341 	for (i = 0; i < page_set->to_pin_cnt; i++, pentry++) {
1342 		if (!pentry->pg_pin_issued) {
1343 			page_in(dev, cache, page_set, pentry);
1344 		}
1345 	}
1346 
1347 	page_set->locked = 0;
1348 
1349 	/* Check if page_set is done */
1350 	if (page_set_is_done(page_set)) {
1351 		page_set_end(dev, cache, page_set);
1352 	}
1353 
1354 	return 0;
1355 }
1356 
1357 static struct ftl_l2p_page *
1358 eviction_get_page(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache)
1359 {
1360 	uint64_t i = 0;
1361 	struct ftl_l2p_page *page = ftl_l2p_cache_get_coldest_page(cache);
1362 
1363 	while (page) {
1364 		ftl_bug(L2P_CACHE_PAGE_READY != page->state);
1365 		ftl_bug(page->pin_ref_cnt);
1366 
1367 		if (ftl_l2p_cache_page_can_evict(page)) {
1368 			ftl_l2p_cache_lru_remove_page(cache, page);
1369 			return page;
1370 		}
1371 
1372 		/*
1373 		 * Practically only one iteration is needed to find a page. It is because
1374 		 * the rank of pages contains only ready and unpinned pages
1375 		 */
1376 		ftl_bug(++i > 1024);
1377 
1378 		page = ftl_l2p_cache_get_hotter_page(page);
1379 	}
1380 
1381 	return NULL;
1382 }
1383 
1384 static void
1385 page_out_io_complete(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache,
1386 		     struct ftl_l2p_page *page, bool success)
1387 {
1388 	cache->l2_pgs_evicting--;
1389 
1390 	ftl_bug(page->ctx.updates > page->updates);
1391 	ftl_bug(!TAILQ_EMPTY(&page->ppe_list));
1392 	ftl_bug(page->on_lru_list);
1393 
1394 	if (spdk_likely(success)) {
1395 		page->updates -= page->ctx.updates;
1396 	}
1397 
1398 	if (success && ftl_l2p_cache_page_can_remove(page)) {
1399 		ftl_l2p_cache_page_remove(cache, page);
1400 	} else {
1401 		if (!page->pin_ref_cnt) {
1402 			ftl_l2p_cache_lru_add_page(cache, page);
1403 		}
1404 		page->state = L2P_CACHE_PAGE_READY;
1405 	}
1406 }
1407 
1408 static void
1409 page_out_io_cb(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
1410 {
1411 	struct ftl_l2p_page *page = cb_arg;
1412 	struct ftl_l2p_cache *cache = page->ctx.cache;
1413 	struct spdk_ftl_dev *dev = cache->dev;
1414 
1415 	ftl_stats_bdev_io_completed(dev, FTL_STATS_TYPE_L2P, bdev_io);
1416 	spdk_bdev_free_io(bdev_io);
1417 	page_out_io_complete(dev, cache, page, success);
1418 }
1419 
1420 static void
1421 page_out_io(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache,
1422 	    struct ftl_l2p_page *page)
1423 {
1424 	struct spdk_io_channel *ioch;
1425 	struct spdk_bdev *bdev;
1426 	struct spdk_bdev_io_wait_entry *bdev_io_wait;
1427 	int rc;
1428 
1429 	page->ctx.cache = cache;
1430 
1431 	rc = ftl_nv_cache_bdev_write_blocks_with_md(dev, ftl_l2p_cache_get_bdev_desc(cache),
1432 			ftl_l2p_cache_get_bdev_iochannel(cache),
1433 			page->page_buffer, NULL, ftl_l2p_cache_page_get_bdev_offset(cache, page),
1434 			1, page_out_io_cb, page);
1435 
1436 	cache->l2_pgs_evicting++;
1437 	if (spdk_likely(0 == rc)) {
1438 		return;
1439 	}
1440 
1441 	if (rc == -ENOMEM) {
1442 		ioch = ftl_l2p_cache_get_bdev_iochannel(cache);
1443 		bdev = spdk_bdev_desc_get_bdev(ftl_l2p_cache_get_bdev_desc(cache));
1444 		bdev_io_wait = &page->ctx.bdev_io_wait;
1445 		bdev_io_wait->bdev = bdev;
1446 		bdev_io_wait->cb_fn = page_out_io_retry;
1447 		bdev_io_wait->cb_arg = page;
1448 
1449 		rc = spdk_bdev_queue_io_wait(bdev, ioch, bdev_io_wait);
1450 		ftl_bug(rc);
1451 	} else {
1452 		ftl_abort();
1453 	}
1454 }
1455 
1456 static void
1457 page_out_io_retry(void *arg)
1458 {
1459 	struct ftl_l2p_page *page = arg;
1460 	struct ftl_l2p_cache *cache = page->ctx.cache;
1461 	struct spdk_ftl_dev *dev = cache->dev;
1462 
1463 	cache->l2_pgs_evicting--;
1464 	page_out_io(dev, cache, page);
1465 }
1466 
1467 static void
1468 ftl_l2p_cache_process_eviction(struct spdk_ftl_dev *dev, struct ftl_l2p_cache *cache)
1469 {
1470 	struct ftl_l2p_page *page;
1471 
1472 	if (!ftl_l2p_cache_evict_continue(cache)) {
1473 		return;
1474 	}
1475 
1476 	if (cache->l2_pgs_evicting > 512) {
1477 		return;
1478 	}
1479 
1480 	page = eviction_get_page(dev, cache);
1481 	if (spdk_unlikely(!page)) {
1482 		return;
1483 	}
1484 
1485 	if (page->updates) {
1486 		page->state = L2P_CACHE_PAGE_FLUSHING;
1487 		page->ctx.updates = page->updates;
1488 		page_out_io(dev, cache, page);
1489 	} else {
1490 		/* Page clean and we can remove it */
1491 		ftl_l2p_cache_page_remove(cache, page);
1492 	}
1493 }
1494 
1495 static void
1496 ftl_l2p_lazy_unmap_process_cb(struct spdk_ftl_dev *dev, int status, struct ftl_l2p_pin_ctx *pin_ctx)
1497 {
1498 	struct ftl_l2p_cache *cache = dev->l2p;
1499 
1500 	cache->lazy_unmap.qd--;
1501 
1502 	/* We will retry on next ftl_l2p_lazy_unmap_process */
1503 	if (spdk_unlikely(status != 0)) {
1504 		return;
1505 	}
1506 
1507 	if (ftl_l2p_cache_running(cache)) {
1508 		ftl_l2p_cache_get(dev, pin_ctx->lba);
1509 	}
1510 
1511 	ftl_l2p_cache_unpin(dev, pin_ctx->lba, pin_ctx->count);
1512 }
1513 
1514 static void
1515 ftl_l2p_lazy_unmap_process(struct spdk_ftl_dev *dev)
1516 {
1517 	struct ftl_l2p_cache *cache = dev->l2p;
1518 	struct ftl_l2p_pin_ctx *pin_ctx;
1519 	uint64_t page_no;
1520 
1521 	if (spdk_likely(!dev->unmap_in_progress)) {
1522 		return;
1523 	}
1524 
1525 	if (cache->lazy_unmap.qd == FTL_L2P_MAX_LAZY_UNMAP_QD) {
1526 		return;
1527 	}
1528 
1529 	page_no = ftl_bitmap_find_first_set(dev->unmap_map, cache->lazy_unmap.page_no, UINT64_MAX);
1530 	if (page_no == UINT64_MAX) {
1531 		cache->lazy_unmap.page_no = 0;
1532 
1533 		/* Check unmap map from beginning to detect unprocessed unmaps */
1534 		page_no = ftl_bitmap_find_first_set(dev->unmap_map, cache->lazy_unmap.page_no, UINT64_MAX);
1535 		if (page_no == UINT64_MAX) {
1536 			dev->unmap_in_progress = false;
1537 			return;
1538 		}
1539 	}
1540 
1541 	cache->lazy_unmap.page_no = page_no;
1542 
1543 	pin_ctx = &cache->lazy_unmap.pin_ctx;
1544 
1545 	cache->lazy_unmap.qd++;
1546 	assert(cache->lazy_unmap.qd <= FTL_L2P_MAX_LAZY_UNMAP_QD);
1547 	assert(page_no < cache->num_pages);
1548 
1549 	pin_ctx->lba = page_no * cache->lbas_in_page;
1550 	pin_ctx->count = 1;
1551 	pin_ctx->cb = ftl_l2p_lazy_unmap_process_cb;
1552 	pin_ctx->cb_ctx = pin_ctx;
1553 
1554 	ftl_l2p_cache_pin(dev, pin_ctx);
1555 }
1556 
1557 void
1558 ftl_l2p_cache_process(struct spdk_ftl_dev *dev)
1559 {
1560 	struct ftl_l2p_cache *cache = dev->l2p;
1561 	int i;
1562 
1563 	if (spdk_unlikely(cache->state != L2P_CACHE_RUNNING)) {
1564 		return;
1565 	}
1566 
1567 	for (i = 0; i < 256; i++) {
1568 		if (ftl_l2p_cache_process_page_sets(dev, cache)) {
1569 			break;
1570 		}
1571 	}
1572 
1573 	ftl_l2p_cache_process_eviction(dev, cache);
1574 	ftl_l2p_lazy_unmap_process(dev);
1575 }
1576