1 /*
2  * Copyright (C) 2016 CNEX Labs
3  * Initial release: Javier Gonzalez <javier@cnexlabs.com>
4  *
5  * Based upon the circular ringbuffer.
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License version
9  * 2 as published by the Free Software Foundation.
10  *
11  * This program is distributed in the hope that it will be useful, but
12  * WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * General Public License for more details.
15  *
16  * pblk-rb.c - pblk's write buffer
17  */
18 
19 #include <linux/circ_buf.h>
20 
21 #include "pblk.h"
22 
23 static DECLARE_RWSEM(pblk_rb_lock);
24 
pblk_rb_data_free(struct pblk_rb * rb)25 void pblk_rb_data_free(struct pblk_rb *rb)
26 {
27 	struct pblk_rb_pages *p, *t;
28 
29 	down_write(&pblk_rb_lock);
30 	list_for_each_entry_safe(p, t, &rb->pages, list) {
31 		free_pages((unsigned long)page_address(p->pages), p->order);
32 		list_del(&p->list);
33 		kfree(p);
34 	}
35 	up_write(&pblk_rb_lock);
36 }
37 
38 /*
39  * Initialize ring buffer. The data and metadata buffers must be previously
40  * allocated and their size must be a power of two
41  * (Documentation/core-api/circular-buffers.rst)
42  */
pblk_rb_init(struct pblk_rb * rb,struct pblk_rb_entry * rb_entry_base,unsigned int power_size,unsigned int power_seg_sz)43 int pblk_rb_init(struct pblk_rb *rb, struct pblk_rb_entry *rb_entry_base,
44 		 unsigned int power_size, unsigned int power_seg_sz)
45 {
46 	struct pblk *pblk = container_of(rb, struct pblk, rwb);
47 	unsigned int init_entry = 0;
48 	unsigned int alloc_order = power_size;
49 	unsigned int max_order = MAX_ORDER - 1;
50 	unsigned int order, iter;
51 
52 	down_write(&pblk_rb_lock);
53 	rb->entries = rb_entry_base;
54 	rb->seg_size = (1 << power_seg_sz);
55 	rb->nr_entries = (1 << power_size);
56 	rb->mem = rb->subm = rb->sync = rb->l2p_update = 0;
57 	rb->flush_point = EMPTY_ENTRY;
58 
59 	spin_lock_init(&rb->w_lock);
60 	spin_lock_init(&rb->s_lock);
61 
62 	INIT_LIST_HEAD(&rb->pages);
63 
64 	if (alloc_order >= max_order) {
65 		order = max_order;
66 		iter = (1 << (alloc_order - max_order));
67 	} else {
68 		order = alloc_order;
69 		iter = 1;
70 	}
71 
72 	do {
73 		struct pblk_rb_entry *entry;
74 		struct pblk_rb_pages *page_set;
75 		void *kaddr;
76 		unsigned long set_size;
77 		int i;
78 
79 		page_set = kmalloc(sizeof(struct pblk_rb_pages), GFP_KERNEL);
80 		if (!page_set) {
81 			up_write(&pblk_rb_lock);
82 			return -ENOMEM;
83 		}
84 
85 		page_set->order = order;
86 		page_set->pages = alloc_pages(GFP_KERNEL, order);
87 		if (!page_set->pages) {
88 			kfree(page_set);
89 			pblk_rb_data_free(rb);
90 			up_write(&pblk_rb_lock);
91 			return -ENOMEM;
92 		}
93 		kaddr = page_address(page_set->pages);
94 
95 		entry = &rb->entries[init_entry];
96 		entry->data = kaddr;
97 		entry->cacheline = pblk_cacheline_to_addr(init_entry++);
98 		entry->w_ctx.flags = PBLK_WRITABLE_ENTRY;
99 
100 		set_size = (1 << order);
101 		for (i = 1; i < set_size; i++) {
102 			entry = &rb->entries[init_entry];
103 			entry->cacheline = pblk_cacheline_to_addr(init_entry++);
104 			entry->data = kaddr + (i * rb->seg_size);
105 			entry->w_ctx.flags = PBLK_WRITABLE_ENTRY;
106 			bio_list_init(&entry->w_ctx.bios);
107 		}
108 
109 		list_add_tail(&page_set->list, &rb->pages);
110 		iter--;
111 	} while (iter > 0);
112 	up_write(&pblk_rb_lock);
113 
114 #ifdef CONFIG_NVM_PBLK_DEBUG
115 	atomic_set(&rb->inflight_flush_point, 0);
116 #endif
117 
118 	/*
119 	 * Initialize rate-limiter, which controls access to the write buffer
120 	 * but user and GC I/O
121 	 */
122 	pblk_rl_init(&pblk->rl, rb->nr_entries);
123 
124 	return 0;
125 }
126 
127 /*
128  * pblk_rb_calculate_size -- calculate the size of the write buffer
129  */
pblk_rb_calculate_size(unsigned int nr_entries)130 unsigned int pblk_rb_calculate_size(unsigned int nr_entries)
131 {
132 	/* Alloc a write buffer that can at least fit 128 entries */
133 	return (1 << max(get_count_order(nr_entries), 7));
134 }
135 
pblk_rb_entries_ref(struct pblk_rb * rb)136 void *pblk_rb_entries_ref(struct pblk_rb *rb)
137 {
138 	return rb->entries;
139 }
140 
clean_wctx(struct pblk_w_ctx * w_ctx)141 static void clean_wctx(struct pblk_w_ctx *w_ctx)
142 {
143 	int flags;
144 
145 	flags = READ_ONCE(w_ctx->flags);
146 	WARN_ONCE(!(flags & PBLK_SUBMITTED_ENTRY),
147 			"pblk: overwriting unsubmitted data\n");
148 
149 	/* Release flags on context. Protect from writes and reads */
150 	smp_store_release(&w_ctx->flags, PBLK_WRITABLE_ENTRY);
151 	pblk_ppa_set_empty(&w_ctx->ppa);
152 	w_ctx->lba = ADDR_EMPTY;
153 }
154 
155 #define pblk_rb_ring_count(head, tail, size) CIRC_CNT(head, tail, size)
156 #define pblk_rb_ring_space(rb, head, tail, size) \
157 					(CIRC_SPACE(head, tail, size))
158 
159 /*
160  * Buffer space is calculated with respect to the back pointer signaling
161  * synchronized entries to the media.
162  */
pblk_rb_space(struct pblk_rb * rb)163 static unsigned int pblk_rb_space(struct pblk_rb *rb)
164 {
165 	unsigned int mem = READ_ONCE(rb->mem);
166 	unsigned int sync = READ_ONCE(rb->sync);
167 
168 	return pblk_rb_ring_space(rb, mem, sync, rb->nr_entries);
169 }
170 
171 /*
172  * Buffer count is calculated with respect to the submission entry signaling the
173  * entries that are available to send to the media
174  */
pblk_rb_read_count(struct pblk_rb * rb)175 unsigned int pblk_rb_read_count(struct pblk_rb *rb)
176 {
177 	unsigned int mem = READ_ONCE(rb->mem);
178 	unsigned int subm = READ_ONCE(rb->subm);
179 
180 	return pblk_rb_ring_count(mem, subm, rb->nr_entries);
181 }
182 
pblk_rb_sync_count(struct pblk_rb * rb)183 unsigned int pblk_rb_sync_count(struct pblk_rb *rb)
184 {
185 	unsigned int mem = READ_ONCE(rb->mem);
186 	unsigned int sync = READ_ONCE(rb->sync);
187 
188 	return pblk_rb_ring_count(mem, sync, rb->nr_entries);
189 }
190 
pblk_rb_read_commit(struct pblk_rb * rb,unsigned int nr_entries)191 unsigned int pblk_rb_read_commit(struct pblk_rb *rb, unsigned int nr_entries)
192 {
193 	unsigned int subm;
194 
195 	subm = READ_ONCE(rb->subm);
196 	/* Commit read means updating submission pointer */
197 	smp_store_release(&rb->subm,
198 				(subm + nr_entries) & (rb->nr_entries - 1));
199 
200 	return subm;
201 }
202 
__pblk_rb_update_l2p(struct pblk_rb * rb,unsigned int to_update)203 static int __pblk_rb_update_l2p(struct pblk_rb *rb, unsigned int to_update)
204 {
205 	struct pblk *pblk = container_of(rb, struct pblk, rwb);
206 	struct pblk_line *line;
207 	struct pblk_rb_entry *entry;
208 	struct pblk_w_ctx *w_ctx;
209 	unsigned int user_io = 0, gc_io = 0;
210 	unsigned int i;
211 	int flags;
212 
213 	for (i = 0; i < to_update; i++) {
214 		entry = &rb->entries[rb->l2p_update];
215 		w_ctx = &entry->w_ctx;
216 
217 		flags = READ_ONCE(entry->w_ctx.flags);
218 		if (flags & PBLK_IOTYPE_USER)
219 			user_io++;
220 		else if (flags & PBLK_IOTYPE_GC)
221 			gc_io++;
222 		else
223 			WARN(1, "pblk: unknown IO type\n");
224 
225 		pblk_update_map_dev(pblk, w_ctx->lba, w_ctx->ppa,
226 							entry->cacheline);
227 
228 		line = &pblk->lines[pblk_ppa_to_line(w_ctx->ppa)];
229 		kref_put(&line->ref, pblk_line_put);
230 		clean_wctx(w_ctx);
231 		rb->l2p_update = (rb->l2p_update + 1) & (rb->nr_entries - 1);
232 	}
233 
234 	pblk_rl_out(&pblk->rl, user_io, gc_io);
235 
236 	return 0;
237 }
238 
239 /*
240  * When we move the l2p_update pointer, we update the l2p table - lookups will
241  * point to the physical address instead of to the cacheline in the write buffer
242  * from this moment on.
243  */
pblk_rb_update_l2p(struct pblk_rb * rb,unsigned int nr_entries,unsigned int mem,unsigned int sync)244 static int pblk_rb_update_l2p(struct pblk_rb *rb, unsigned int nr_entries,
245 			      unsigned int mem, unsigned int sync)
246 {
247 	unsigned int space, count;
248 	int ret = 0;
249 
250 	lockdep_assert_held(&rb->w_lock);
251 
252 	/* Update l2p only as buffer entries are being overwritten */
253 	space = pblk_rb_ring_space(rb, mem, rb->l2p_update, rb->nr_entries);
254 	if (space > nr_entries)
255 		goto out;
256 
257 	count = nr_entries - space;
258 	/* l2p_update used exclusively under rb->w_lock */
259 	ret = __pblk_rb_update_l2p(rb, count);
260 
261 out:
262 	return ret;
263 }
264 
265 /*
266  * Update the l2p entry for all sectors stored on the write buffer. This means
267  * that all future lookups to the l2p table will point to a device address, not
268  * to the cacheline in the write buffer.
269  */
pblk_rb_sync_l2p(struct pblk_rb * rb)270 void pblk_rb_sync_l2p(struct pblk_rb *rb)
271 {
272 	unsigned int sync;
273 	unsigned int to_update;
274 
275 	spin_lock(&rb->w_lock);
276 
277 	/* Protect from reads and writes */
278 	sync = smp_load_acquire(&rb->sync);
279 
280 	to_update = pblk_rb_ring_count(sync, rb->l2p_update, rb->nr_entries);
281 	__pblk_rb_update_l2p(rb, to_update);
282 
283 	spin_unlock(&rb->w_lock);
284 }
285 
286 /*
287  * Write @nr_entries to ring buffer from @data buffer if there is enough space.
288  * Typically, 4KB data chunks coming from a bio will be copied to the ring
289  * buffer, thus the write will fail if not all incoming data can be copied.
290  *
291  */
__pblk_rb_write_entry(struct pblk_rb * rb,void * data,struct pblk_w_ctx w_ctx,struct pblk_rb_entry * entry)292 static void __pblk_rb_write_entry(struct pblk_rb *rb, void *data,
293 				  struct pblk_w_ctx w_ctx,
294 				  struct pblk_rb_entry *entry)
295 {
296 	memcpy(entry->data, data, rb->seg_size);
297 
298 	entry->w_ctx.lba = w_ctx.lba;
299 	entry->w_ctx.ppa = w_ctx.ppa;
300 }
301 
pblk_rb_write_entry_user(struct pblk_rb * rb,void * data,struct pblk_w_ctx w_ctx,unsigned int ring_pos)302 void pblk_rb_write_entry_user(struct pblk_rb *rb, void *data,
303 			      struct pblk_w_ctx w_ctx, unsigned int ring_pos)
304 {
305 	struct pblk *pblk = container_of(rb, struct pblk, rwb);
306 	struct pblk_rb_entry *entry;
307 	int flags;
308 
309 	entry = &rb->entries[ring_pos];
310 	flags = READ_ONCE(entry->w_ctx.flags);
311 #ifdef CONFIG_NVM_PBLK_DEBUG
312 	/* Caller must guarantee that the entry is free */
313 	BUG_ON(!(flags & PBLK_WRITABLE_ENTRY));
314 #endif
315 
316 	__pblk_rb_write_entry(rb, data, w_ctx, entry);
317 
318 	pblk_update_map_cache(pblk, w_ctx.lba, entry->cacheline);
319 	flags = w_ctx.flags | PBLK_WRITTEN_DATA;
320 
321 	/* Release flags on write context. Protect from writes */
322 	smp_store_release(&entry->w_ctx.flags, flags);
323 }
324 
pblk_rb_write_entry_gc(struct pblk_rb * rb,void * data,struct pblk_w_ctx w_ctx,struct pblk_line * line,u64 paddr,unsigned int ring_pos)325 void pblk_rb_write_entry_gc(struct pblk_rb *rb, void *data,
326 			    struct pblk_w_ctx w_ctx, struct pblk_line *line,
327 			    u64 paddr, unsigned int ring_pos)
328 {
329 	struct pblk *pblk = container_of(rb, struct pblk, rwb);
330 	struct pblk_rb_entry *entry;
331 	int flags;
332 
333 	entry = &rb->entries[ring_pos];
334 	flags = READ_ONCE(entry->w_ctx.flags);
335 #ifdef CONFIG_NVM_PBLK_DEBUG
336 	/* Caller must guarantee that the entry is free */
337 	BUG_ON(!(flags & PBLK_WRITABLE_ENTRY));
338 #endif
339 
340 	__pblk_rb_write_entry(rb, data, w_ctx, entry);
341 
342 	if (!pblk_update_map_gc(pblk, w_ctx.lba, entry->cacheline, line, paddr))
343 		entry->w_ctx.lba = ADDR_EMPTY;
344 
345 	flags = w_ctx.flags | PBLK_WRITTEN_DATA;
346 
347 	/* Release flags on write context. Protect from writes */
348 	smp_store_release(&entry->w_ctx.flags, flags);
349 }
350 
pblk_rb_flush_point_set(struct pblk_rb * rb,struct bio * bio,unsigned int pos)351 static int pblk_rb_flush_point_set(struct pblk_rb *rb, struct bio *bio,
352 				   unsigned int pos)
353 {
354 	struct pblk_rb_entry *entry;
355 	unsigned int sync, flush_point;
356 
357 	pblk_rb_sync_init(rb, NULL);
358 	sync = READ_ONCE(rb->sync);
359 
360 	if (pos == sync) {
361 		pblk_rb_sync_end(rb, NULL);
362 		return 0;
363 	}
364 
365 #ifdef CONFIG_NVM_PBLK_DEBUG
366 	atomic_inc(&rb->inflight_flush_point);
367 #endif
368 
369 	flush_point = (pos == 0) ? (rb->nr_entries - 1) : (pos - 1);
370 	entry = &rb->entries[flush_point];
371 
372 	/* Protect flush points */
373 	smp_store_release(&rb->flush_point, flush_point);
374 
375 	if (bio)
376 		bio_list_add(&entry->w_ctx.bios, bio);
377 
378 	pblk_rb_sync_end(rb, NULL);
379 
380 	return bio ? 1 : 0;
381 }
382 
__pblk_rb_may_write(struct pblk_rb * rb,unsigned int nr_entries,unsigned int * pos)383 static int __pblk_rb_may_write(struct pblk_rb *rb, unsigned int nr_entries,
384 			       unsigned int *pos)
385 {
386 	unsigned int mem;
387 	unsigned int sync;
388 
389 	sync = READ_ONCE(rb->sync);
390 	mem = READ_ONCE(rb->mem);
391 
392 	if (pblk_rb_ring_space(rb, mem, sync, rb->nr_entries) < nr_entries)
393 		return 0;
394 
395 	if (pblk_rb_update_l2p(rb, nr_entries, mem, sync))
396 		return 0;
397 
398 	*pos = mem;
399 
400 	return 1;
401 }
402 
pblk_rb_may_write(struct pblk_rb * rb,unsigned int nr_entries,unsigned int * pos)403 static int pblk_rb_may_write(struct pblk_rb *rb, unsigned int nr_entries,
404 			     unsigned int *pos)
405 {
406 	if (!__pblk_rb_may_write(rb, nr_entries, pos))
407 		return 0;
408 
409 	/* Protect from read count */
410 	smp_store_release(&rb->mem, (*pos + nr_entries) & (rb->nr_entries - 1));
411 	return 1;
412 }
413 
pblk_rb_flush(struct pblk_rb * rb)414 void pblk_rb_flush(struct pblk_rb *rb)
415 {
416 	struct pblk *pblk = container_of(rb, struct pblk, rwb);
417 	unsigned int mem = READ_ONCE(rb->mem);
418 
419 	if (pblk_rb_flush_point_set(rb, NULL, mem))
420 		return;
421 
422 	pblk_write_kick(pblk);
423 }
424 
pblk_rb_may_write_flush(struct pblk_rb * rb,unsigned int nr_entries,unsigned int * pos,struct bio * bio,int * io_ret)425 static int pblk_rb_may_write_flush(struct pblk_rb *rb, unsigned int nr_entries,
426 				   unsigned int *pos, struct bio *bio,
427 				   int *io_ret)
428 {
429 	unsigned int mem;
430 
431 	if (!__pblk_rb_may_write(rb, nr_entries, pos))
432 		return 0;
433 
434 	mem = (*pos + nr_entries) & (rb->nr_entries - 1);
435 	*io_ret = NVM_IO_DONE;
436 
437 	if (bio->bi_opf & REQ_PREFLUSH) {
438 		struct pblk *pblk = container_of(rb, struct pblk, rwb);
439 
440 		atomic64_inc(&pblk->nr_flush);
441 		if (pblk_rb_flush_point_set(&pblk->rwb, bio, mem))
442 			*io_ret = NVM_IO_OK;
443 	}
444 
445 	/* Protect from read count */
446 	smp_store_release(&rb->mem, mem);
447 
448 	return 1;
449 }
450 
451 /*
452  * Atomically check that (i) there is space on the write buffer for the
453  * incoming I/O, and (ii) the current I/O type has enough budget in the write
454  * buffer (rate-limiter).
455  */
pblk_rb_may_write_user(struct pblk_rb * rb,struct bio * bio,unsigned int nr_entries,unsigned int * pos)456 int pblk_rb_may_write_user(struct pblk_rb *rb, struct bio *bio,
457 			   unsigned int nr_entries, unsigned int *pos)
458 {
459 	struct pblk *pblk = container_of(rb, struct pblk, rwb);
460 	int io_ret;
461 
462 	spin_lock(&rb->w_lock);
463 	io_ret = pblk_rl_user_may_insert(&pblk->rl, nr_entries);
464 	if (io_ret) {
465 		spin_unlock(&rb->w_lock);
466 		return io_ret;
467 	}
468 
469 	if (!pblk_rb_may_write_flush(rb, nr_entries, pos, bio, &io_ret)) {
470 		spin_unlock(&rb->w_lock);
471 		return NVM_IO_REQUEUE;
472 	}
473 
474 	pblk_rl_user_in(&pblk->rl, nr_entries);
475 	spin_unlock(&rb->w_lock);
476 
477 	return io_ret;
478 }
479 
480 /*
481  * Look at pblk_rb_may_write_user comment
482  */
pblk_rb_may_write_gc(struct pblk_rb * rb,unsigned int nr_entries,unsigned int * pos)483 int pblk_rb_may_write_gc(struct pblk_rb *rb, unsigned int nr_entries,
484 			 unsigned int *pos)
485 {
486 	struct pblk *pblk = container_of(rb, struct pblk, rwb);
487 
488 	spin_lock(&rb->w_lock);
489 	if (!pblk_rl_gc_may_insert(&pblk->rl, nr_entries)) {
490 		spin_unlock(&rb->w_lock);
491 		return 0;
492 	}
493 
494 	if (!pblk_rb_may_write(rb, nr_entries, pos)) {
495 		spin_unlock(&rb->w_lock);
496 		return 0;
497 	}
498 
499 	pblk_rl_gc_in(&pblk->rl, nr_entries);
500 	spin_unlock(&rb->w_lock);
501 
502 	return 1;
503 }
504 
505 /*
506  * Read available entries on rb and add them to the given bio. To avoid a memory
507  * copy, a page reference to the write buffer is used to be added to the bio.
508  *
509  * This function is used by the write thread to form the write bio that will
510  * persist data on the write buffer to the media.
511  */
pblk_rb_read_to_bio(struct pblk_rb * rb,struct nvm_rq * rqd,unsigned int pos,unsigned int nr_entries,unsigned int count)512 unsigned int pblk_rb_read_to_bio(struct pblk_rb *rb, struct nvm_rq *rqd,
513 				 unsigned int pos, unsigned int nr_entries,
514 				 unsigned int count)
515 {
516 	struct pblk *pblk = container_of(rb, struct pblk, rwb);
517 	struct request_queue *q = pblk->dev->q;
518 	struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd);
519 	struct bio *bio = rqd->bio;
520 	struct pblk_rb_entry *entry;
521 	struct page *page;
522 	unsigned int pad = 0, to_read = nr_entries;
523 	unsigned int i;
524 	int flags;
525 
526 	if (count < nr_entries) {
527 		pad = nr_entries - count;
528 		to_read = count;
529 	}
530 
531 	c_ctx->sentry = pos;
532 	c_ctx->nr_valid = to_read;
533 	c_ctx->nr_padded = pad;
534 
535 	for (i = 0; i < to_read; i++) {
536 		entry = &rb->entries[pos];
537 
538 		/* A write has been allowed into the buffer, but data is still
539 		 * being copied to it. It is ok to busy wait.
540 		 */
541 try:
542 		flags = READ_ONCE(entry->w_ctx.flags);
543 		if (!(flags & PBLK_WRITTEN_DATA)) {
544 			io_schedule();
545 			goto try;
546 		}
547 
548 		page = virt_to_page(entry->data);
549 		if (!page) {
550 			pblk_err(pblk, "could not allocate write bio page\n");
551 			flags &= ~PBLK_WRITTEN_DATA;
552 			flags |= PBLK_SUBMITTED_ENTRY;
553 			/* Release flags on context. Protect from writes */
554 			smp_store_release(&entry->w_ctx.flags, flags);
555 			return NVM_IO_ERR;
556 		}
557 
558 		if (bio_add_pc_page(q, bio, page, rb->seg_size, 0) !=
559 								rb->seg_size) {
560 			pblk_err(pblk, "could not add page to write bio\n");
561 			flags &= ~PBLK_WRITTEN_DATA;
562 			flags |= PBLK_SUBMITTED_ENTRY;
563 			/* Release flags on context. Protect from writes */
564 			smp_store_release(&entry->w_ctx.flags, flags);
565 			return NVM_IO_ERR;
566 		}
567 
568 		flags &= ~PBLK_WRITTEN_DATA;
569 		flags |= PBLK_SUBMITTED_ENTRY;
570 
571 		/* Release flags on context. Protect from writes */
572 		smp_store_release(&entry->w_ctx.flags, flags);
573 
574 		pos = (pos + 1) & (rb->nr_entries - 1);
575 	}
576 
577 	if (pad) {
578 		if (pblk_bio_add_pages(pblk, bio, GFP_KERNEL, pad)) {
579 			pblk_err(pblk, "could not pad page in write bio\n");
580 			return NVM_IO_ERR;
581 		}
582 
583 		if (pad < pblk->min_write_pgs)
584 			atomic64_inc(&pblk->pad_dist[pad - 1]);
585 		else
586 			pblk_warn(pblk, "padding more than min. sectors\n");
587 
588 		atomic64_add(pad, &pblk->pad_wa);
589 	}
590 
591 #ifdef CONFIG_NVM_PBLK_DEBUG
592 	atomic_long_add(pad, &pblk->padded_writes);
593 #endif
594 
595 	return NVM_IO_OK;
596 }
597 
598 /*
599  * Copy to bio only if the lba matches the one on the given cache entry.
600  * Otherwise, it means that the entry has been overwritten, and the bio should
601  * be directed to disk.
602  */
pblk_rb_copy_to_bio(struct pblk_rb * rb,struct bio * bio,sector_t lba,struct ppa_addr ppa,int bio_iter,bool advanced_bio)603 int pblk_rb_copy_to_bio(struct pblk_rb *rb, struct bio *bio, sector_t lba,
604 			struct ppa_addr ppa, int bio_iter, bool advanced_bio)
605 {
606 	struct pblk *pblk = container_of(rb, struct pblk, rwb);
607 	struct pblk_rb_entry *entry;
608 	struct pblk_w_ctx *w_ctx;
609 	struct ppa_addr l2p_ppa;
610 	u64 pos = pblk_addr_to_cacheline(ppa);
611 	void *data;
612 	int flags;
613 	int ret = 1;
614 
615 
616 #ifdef CONFIG_NVM_PBLK_DEBUG
617 	/* Caller must ensure that the access will not cause an overflow */
618 	BUG_ON(pos >= rb->nr_entries);
619 #endif
620 	entry = &rb->entries[pos];
621 	w_ctx = &entry->w_ctx;
622 	flags = READ_ONCE(w_ctx->flags);
623 
624 	spin_lock(&rb->w_lock);
625 	spin_lock(&pblk->trans_lock);
626 	l2p_ppa = pblk_trans_map_get(pblk, lba);
627 	spin_unlock(&pblk->trans_lock);
628 
629 	/* Check if the entry has been overwritten or is scheduled to be */
630 	if (!pblk_ppa_comp(l2p_ppa, ppa) || w_ctx->lba != lba ||
631 						flags & PBLK_WRITABLE_ENTRY) {
632 		ret = 0;
633 		goto out;
634 	}
635 
636 	/* Only advance the bio if it hasn't been advanced already. If advanced,
637 	 * this bio is at least a partial bio (i.e., it has partially been
638 	 * filled with data from the cache). If part of the data resides on the
639 	 * media, we will read later on
640 	 */
641 	if (unlikely(!advanced_bio))
642 		bio_advance(bio, bio_iter * PBLK_EXPOSED_PAGE_SIZE);
643 
644 	data = bio_data(bio);
645 	memcpy(data, entry->data, rb->seg_size);
646 
647 out:
648 	spin_unlock(&rb->w_lock);
649 	return ret;
650 }
651 
pblk_rb_w_ctx(struct pblk_rb * rb,unsigned int pos)652 struct pblk_w_ctx *pblk_rb_w_ctx(struct pblk_rb *rb, unsigned int pos)
653 {
654 	unsigned int entry = pos & (rb->nr_entries - 1);
655 
656 	return &rb->entries[entry].w_ctx;
657 }
658 
pblk_rb_sync_init(struct pblk_rb * rb,unsigned long * flags)659 unsigned int pblk_rb_sync_init(struct pblk_rb *rb, unsigned long *flags)
660 	__acquires(&rb->s_lock)
661 {
662 	if (flags)
663 		spin_lock_irqsave(&rb->s_lock, *flags);
664 	else
665 		spin_lock_irq(&rb->s_lock);
666 
667 	return rb->sync;
668 }
669 
pblk_rb_sync_end(struct pblk_rb * rb,unsigned long * flags)670 void pblk_rb_sync_end(struct pblk_rb *rb, unsigned long *flags)
671 	__releases(&rb->s_lock)
672 {
673 	lockdep_assert_held(&rb->s_lock);
674 
675 	if (flags)
676 		spin_unlock_irqrestore(&rb->s_lock, *flags);
677 	else
678 		spin_unlock_irq(&rb->s_lock);
679 }
680 
pblk_rb_sync_advance(struct pblk_rb * rb,unsigned int nr_entries)681 unsigned int pblk_rb_sync_advance(struct pblk_rb *rb, unsigned int nr_entries)
682 {
683 	unsigned int sync, flush_point;
684 	lockdep_assert_held(&rb->s_lock);
685 
686 	sync = READ_ONCE(rb->sync);
687 	flush_point = READ_ONCE(rb->flush_point);
688 
689 	if (flush_point != EMPTY_ENTRY) {
690 		unsigned int secs_to_flush;
691 
692 		secs_to_flush = pblk_rb_ring_count(flush_point, sync,
693 					rb->nr_entries);
694 		if (secs_to_flush < nr_entries) {
695 			/* Protect flush points */
696 			smp_store_release(&rb->flush_point, EMPTY_ENTRY);
697 		}
698 	}
699 
700 	sync = (sync + nr_entries) & (rb->nr_entries - 1);
701 
702 	/* Protect from counts */
703 	smp_store_release(&rb->sync, sync);
704 
705 	return sync;
706 }
707 
708 /* Calculate how many sectors to submit up to the current flush point. */
pblk_rb_flush_point_count(struct pblk_rb * rb)709 unsigned int pblk_rb_flush_point_count(struct pblk_rb *rb)
710 {
711 	unsigned int subm, sync, flush_point;
712 	unsigned int submitted, to_flush;
713 
714 	/* Protect flush points */
715 	flush_point = smp_load_acquire(&rb->flush_point);
716 	if (flush_point == EMPTY_ENTRY)
717 		return 0;
718 
719 	/* Protect syncs */
720 	sync = smp_load_acquire(&rb->sync);
721 
722 	subm = READ_ONCE(rb->subm);
723 	submitted = pblk_rb_ring_count(subm, sync, rb->nr_entries);
724 
725 	/* The sync point itself counts as a sector to sync */
726 	to_flush = pblk_rb_ring_count(flush_point, sync, rb->nr_entries) + 1;
727 
728 	return (submitted < to_flush) ? (to_flush - submitted) : 0;
729 }
730 
731 /*
732  * Scan from the current position of the sync pointer to find the entry that
733  * corresponds to the given ppa. This is necessary since write requests can be
734  * completed out of order. The assumption is that the ppa is close to the sync
735  * pointer thus the search will not take long.
736  *
737  * The caller of this function must guarantee that the sync pointer will no
738  * reach the entry while it is using the metadata associated with it. With this
739  * assumption in mind, there is no need to take the sync lock.
740  */
pblk_rb_sync_scan_entry(struct pblk_rb * rb,struct ppa_addr * ppa)741 struct pblk_rb_entry *pblk_rb_sync_scan_entry(struct pblk_rb *rb,
742 					      struct ppa_addr *ppa)
743 {
744 	unsigned int sync, subm, count;
745 	unsigned int i;
746 
747 	sync = READ_ONCE(rb->sync);
748 	subm = READ_ONCE(rb->subm);
749 	count = pblk_rb_ring_count(subm, sync, rb->nr_entries);
750 
751 	for (i = 0; i < count; i++)
752 		sync = (sync + 1) & (rb->nr_entries - 1);
753 
754 	return NULL;
755 }
756 
pblk_rb_tear_down_check(struct pblk_rb * rb)757 int pblk_rb_tear_down_check(struct pblk_rb *rb)
758 {
759 	struct pblk_rb_entry *entry;
760 	int i;
761 	int ret = 0;
762 
763 	spin_lock(&rb->w_lock);
764 	spin_lock_irq(&rb->s_lock);
765 
766 	if ((rb->mem == rb->subm) && (rb->subm == rb->sync) &&
767 				(rb->sync == rb->l2p_update) &&
768 				(rb->flush_point == EMPTY_ENTRY)) {
769 		goto out;
770 	}
771 
772 	if (!rb->entries) {
773 		ret = 1;
774 		goto out;
775 	}
776 
777 	for (i = 0; i < rb->nr_entries; i++) {
778 		entry = &rb->entries[i];
779 
780 		if (!entry->data) {
781 			ret = 1;
782 			goto out;
783 		}
784 	}
785 
786 out:
787 	spin_unlock_irq(&rb->s_lock);
788 	spin_unlock(&rb->w_lock);
789 
790 	return ret;
791 }
792 
pblk_rb_wrap_pos(struct pblk_rb * rb,unsigned int pos)793 unsigned int pblk_rb_wrap_pos(struct pblk_rb *rb, unsigned int pos)
794 {
795 	return (pos & (rb->nr_entries - 1));
796 }
797 
pblk_rb_pos_oob(struct pblk_rb * rb,u64 pos)798 int pblk_rb_pos_oob(struct pblk_rb *rb, u64 pos)
799 {
800 	return (pos >= rb->nr_entries);
801 }
802 
pblk_rb_sysfs(struct pblk_rb * rb,char * buf)803 ssize_t pblk_rb_sysfs(struct pblk_rb *rb, char *buf)
804 {
805 	struct pblk *pblk = container_of(rb, struct pblk, rwb);
806 	struct pblk_c_ctx *c;
807 	ssize_t offset;
808 	int queued_entries = 0;
809 
810 	spin_lock_irq(&rb->s_lock);
811 	list_for_each_entry(c, &pblk->compl_list, list)
812 		queued_entries++;
813 	spin_unlock_irq(&rb->s_lock);
814 
815 	if (rb->flush_point != EMPTY_ENTRY)
816 		offset = scnprintf(buf, PAGE_SIZE,
817 			"%u\t%u\t%u\t%u\t%u\t%u\t%u - %u/%u/%u - %d\n",
818 			rb->nr_entries,
819 			rb->mem,
820 			rb->subm,
821 			rb->sync,
822 			rb->l2p_update,
823 #ifdef CONFIG_NVM_PBLK_DEBUG
824 			atomic_read(&rb->inflight_flush_point),
825 #else
826 			0,
827 #endif
828 			rb->flush_point,
829 			pblk_rb_read_count(rb),
830 			pblk_rb_space(rb),
831 			pblk_rb_flush_point_count(rb),
832 			queued_entries);
833 	else
834 		offset = scnprintf(buf, PAGE_SIZE,
835 			"%u\t%u\t%u\t%u\t%u\t%u\tNULL - %u/%u/%u - %d\n",
836 			rb->nr_entries,
837 			rb->mem,
838 			rb->subm,
839 			rb->sync,
840 			rb->l2p_update,
841 #ifdef CONFIG_NVM_PBLK_DEBUG
842 			atomic_read(&rb->inflight_flush_point),
843 #else
844 			0,
845 #endif
846 			pblk_rb_read_count(rb),
847 			pblk_rb_space(rb),
848 			pblk_rb_flush_point_count(rb),
849 			queued_entries);
850 
851 	return offset;
852 }
853